Cornerstones of CRISPR–Cas in Drug Discovery PDF

REVIEWS Cornerstones of CRISPR–Cas in drug discovery and therapy Christof Fellmann1*, Benjamin G. Gowen1,2*, Pei-Chun Lin1–3, Jennifer A. Doudna1,4–7 an...

REVIEWS Cornerstones of CRISPR–Cas in drug discovery and therapy Christof Fellmann1*, Benjamin G. Gowen1,2*, Pei-Chun Lin1–3, Jennifer A. Doudna1,4–7 and Jacob E. Corn1,2 Abstract | The recent development of CRISPR–Cas systems as easily accessible and programmable tools for genome editing and regulation is spurring a revolution in biology. Paired with the rapid expansion of reference and personalized genomic sequence information, technologies based on CRISPR–Cas are enabling nearly unlimited genetic manipulation, even in previously difficult contexts, including human cells. Although much attention has focused on the potential of CRISPR–Cas to cure Mendelian diseases, the technology also holds promise to transform the development of therapies to treat complex heritable and somatic disorders. In this Review, we discuss how CRISPR–Cas can affect the next generation of drugs by accelerating the identification and validation of high-value targets, uncovering high-confidence biomarkers and developing differentiated breakthrough therapies. We focus on the promises, pitfalls and hurdles of this revolutionary gene-editing technology, discuss key aspects of different CRISPR–Cas screening platforms and offer our perspectives on the best practices in genome engineering. 1 Department of Molecular The central dogma of molecular biology posits a flow molecular toolbox for mammalian genome engineer- and Cell Biology, University of California, Berkeley, Berkeley, of information from gene to mRNA to protein1. The ing 2–6. Gene-editing technologies in the form of clus- California 94720, USA. genome serves as the blueprint of life, setting the stage tered regularly interspaced short palindromic repeat 2 Innovative Genomics for all downstream activity. Although approaches to (CRISPR)–CRISPR-associated protein (Cas) systems Institute, University of treat human disease predominantly target the end of the stand poised to transform many stages of drug discovery California, Berkeley, Berkeley, California 94720, USA. information cascade (for example, by inhibiting signal- and development by enabling fast and accurate alterations 3 Present address: Helen Diller ling pathways, supplementing metabolites or interfering of genomic information in mammalian model systems Family Comprehensive Cancer with viral polymerases), the discovery and validation of and human tissues. In addition, direct somatic editing 7 Center, University of California, therapeutic targets often takes place at the level of genes in patients will, eventually, radically change the drugga- San Francisco, San Francisco, California 94158, USA. and transcripts. The discovery of human mutations that ble space8 by enabling the targeting of nearly any entity, 4 Department of Chemistry, are directly linked to disease, such as somatic breakpoint including the introduction of corrective mutations and the University of California, cluster region–Abelson tyrosine kinase 1 (BCR–ABL1) modification of regulatory elements or splicing patterns. Berkeley, Berkeley, California 94720, USA. fusions in chronic myeloid leukaemia or inherited Following the description of a two-component single 5 Howard Hughes Medical BRCA1 mutations in breast cancer, or of mutations guide RNA (sgRNA)–Cas9 complex to introduce DNA Institute, University of associated with a survival benefit, including proprotein double-strand breaks (DSBs) in an RNA-guided manner2, California, Berkeley, Berkeley, convertase subtilisin/kexin type 9 (PCSK9) mutations many studies have demonstrated ingenious applications California 94720, USA. 6 Li Ka Shing Biomedical and in minimizing cardiovascular disease, is considered by and uncovered orthogonal immune systems, together Health Sciences Center, many to be the ‘gold standard’ for drug target identifica- enabling nearly unlimited genome engineering opportu- University of California, tion. However, the paucity of scalable genetic engineer- nities (FIG. 1). Berkeley, Berkeley, ing tools in mammalian cell culture and model systems The technological domestication of CRISPR–Cas sys- California 94720, USA. 7 MBIB Division, Lawrence has necessitated that many discovery efforts that link tems and molecular mechanisms of Cas-based genome Berkeley National genotype with phenotype are either observational, such editing have been thoroughly covered elsewhere9–11. Laboratory, Berkeley, as genome-wide association studies (GWAS), or take Briefly, a sgRNA directs the Cas9 endonuclease to induce California 94720, USA. *These authors contributed place in genetically malleable invertebrate models such DSBs at homologous sites2. During genome editing, the equally to this work. as the fruitfly Drosophila melanogaster and the nematode DSBs are fixed by cellular DNA repair mechanisms, Correspondence to J.E.C. Caenorhabditis elegans. including the predominant error-prone non-homologous [email protected] The recent development of easily programmable end joining (NHEJ)12–14 and the less-frequent templated doi:10.1038/nrd.2016.238 RNA-guided nucleases, which are derived from micro- homology-directed repair (HDR)15–19 pathways. NHEJ Published online 23 Dec 2016 bial adaptive immune systems, has revolutionized the is most often leveraged to disrupt genetic sequences, NATURE REVIEWS | DRUG DISCOVERY VOLUME 16 | FEBRUARY 2017 | 89 © 2 0 1 7 M a c m i l l a n P u b l i s h e r s L i m i t e d , p a r t o f S p r i n g e r N a t u r e. A l l r i g h t s r e s e r v e d. REVIEWS CRISPR–Cas gene editing Target identification Cell-based and validation Safety models therapies SNP CAR T cells Screening CCR5 KO cells Disease models Figure 1 | Pipeline of CRISPR–Cas-assisted drug discovery. Unmet medical needs for numerous diseases and the rapid progress of CRISPR–Cas gene editing can feed into a drug discovery and development pipeline, which leads to improved Nature therapies. The CRISPR–Cas system allows for improved target identification and validation Reviews as well as faster| Drug Discovery generation of safety models. CRISPR–Cas can also be used to develop cell-based therapies, such as chimeric antigen receptor (CAR) T cells for immunotherapy and C-C motif chemokine receptor 5 (CCR5)-knockout (KO) cells for HIV treatment. CRISPR–Cas-assisted drug discovery will yield innovative therapies and treatment paradigms for patients. SNP, single-nucleotide polymorphism. whereas HDR can be used to introduce or alter infor- ‘personalized’ or ‘precision’ medicine, which combines mation at a specific locus with properly designed repair classical patient information with personal genetic templates. In addition, a catalytically inactive mutant of data to directly inform individual treatment strategies. Cas9 can be fused to various effector domains to acti- However, hypotheses that are generated by large-scale vate or inhibit the transcription of target genes, strategies observational ‘omics’ efforts often demand testing with known as CRISPRa and CRISPRi, respectively 20–22. Most precise genetic models, particularly to evaluate variants Non-homologous end studies to date have used Cas9 from Streptococcus pyo- of unknown significance, optimize patient stratification, joining genes (SpyCas9), which is the default Cas9 referenced in reassign approved drugs to new indications and develop (NHEJ). The repair of double-strand DNA breaks by this Review. Cas9 molecules from other species, Cas9‑like alternative treatment paradigms. direct ligation of the broken CRISPR nucleases and engineered versions of Cas9 with Even when a single factor (such as the mutational sta- ends. No homology is required novel functions have also been established and can convey tus of TP53, MYC or KRAS) is compared between cells, to promote the end-joining particular advantages in various settings (Supplementary there are often many confounding features that obscure reaction, and it can result in the introduction of small information S1 (table)). Although we focus on SpyCas9, in a direct relationship between genotype and disease phe- non-templated insertions or particular its use in therapeutic discovery and the building notype. Researchers might use matched patient samples deletions (indels). of the next generation of transformational drugs, the gen- from diseased and normal tissues to tease apart such rela- eral outline described here applies to the larger ensemble tionships. However, large collections of matched samples Homology-directed repair of CRISPR–Cas tools. can be difficult to obtain and are not available in many (HDR). The repair of double-strand DNA breaks cases. Although overexpression of appropriate (often using an endogenous or CRISPR–Cas as a tool for drug discovery mutant) cDNA can partially address this issue, such exogenous DNA template with Precision cellular models. Advances in DNA sequenc- constructs are often expressed at non-native levels and homology to regions flanking ing and their large-scale application have provided in the presence of the wild-type protein. The generation the break. insight into genetic variation across groups of patients of mutant or knockout clones via classical homologous CRISPRa and populations, which has expanded our understand- recombination led to a limited set of isogenic cell lines, The activation of transcription ing of the links between genetic variation and disease in which a derived line differs from the parent by a min- through RNA-guided predisposition, disease development and the treatment imal, defined mutation32–35. These resources have proved recruitment of a catalytically response. For example, integrated information from to be incredibly useful, but initial techniques for their inactive Cas9 fused to transcriptional activators. The Cancer Genome Atlas (TCGA)23–28, the Cancer Cell generation were very labour intensive and time consum- Line Encyclopedia (CCLE)28 and the Encyclopedia of ing, which hindered their widespread adoption for drug CRISPRi DNA Elements (ENCODE)29,30 led to improvements in development. The inhibition of transcription the standard of care for patients with glioblastoma, ena- The advent of CRISPR–Cas genome editing2 has dras- through RNA-guided recruitment of a catalytically bling stratification based on the methylation status of the tically altered this landscape (FIG. 2). The generation of inactive Cas9 fused to O6-methylguanine DNA methyltransferase (MGMT) isogenic knockout human (and other) cell lines for com- transcriptional repressors. promoter 31. Such advances have stimulated interest in parative genomics is now so straightforward that, in just 90 | FEBRUARY 2017 | VOLUME 16 www.nature.com/nrd © 2 0 1 7 M a c m i l l a n P u b l i s h e r s L i m i t e d , p a r t o f S p r i n g e r N a t u r e. A l l r i g h t s r e s e r v e d.  REVIEWS Cell of origin Gene editing (Cas–sgRNA) Physiologically normal cells Mutate gene of interest Isogenic cell lines Disease-specific cells Chromosomal translocation Target validation Patient-derived cells or inversion Mechanistic analysis Transgene expression Stratification studies Organoid culture Differentiation Cell self-organization Spatially restricted lineage commitment Matched organoids Disease modelling Drug efficacy testing Organ replacement Large scale screens Synthetic lethal screens or Drug target discovery Combinatorial therapies Cas9 and sgRNA library Pooled screens Arrayed screens Viral vectors Positive or negative selection High content Plasmid collections Genetic synthetic lethal Synthetic lethal RNPs Figure 2 | CRISPR–Cas in the generation of cellular models and large-scale screens. CRISPR–Cas gene editing can be used to generate isogenic cell lines for drug target validation, mechanistic analysis and patient Naturestratification studies. Reviews | Drug Discovery Isogenic cell lines can also be used to generate organoids, which are particularly useful for modelling differentiation and self-organization processes. Large-scale single guide RNA (sgRNA) libraries can be used for high-throughput pooled or high-content arrayed screens, either in unmodified or in CRISPR–Cas-edited cell lines. RNPs, ribonucleoproteins. 4 years, the practice has become commonplace36 and is phosphatidylinositol‑4,5‑bisphosphate 3‑kinase catalytic being carried out by researchers around the globe. Gene subunit alpha (PIK3CA) and isocitrate dehydrogenase 1 knockout via CRISPR–Cas has proved to be efficacious (IDH1), or tumour suppressors including TP53, RB1 and in virtually all cell types, including induced pluripotent von Hippel–Lindau (VHL)46,47. More generally, isogenic stem cells (iPSCs), cancer-specific organoids and primary series can be used to analyse the effect of mutants on immune cells37–40. Knockout-based target discovery efforts disease development or to query the specificity of are thus no longer limited to specialized cell lines, such as mutant-targeting therapeutic candidates. From a tech- the haploid lines that were previously used for gene trap nical perspective, HDR requires delivery of the Cas9– experiments41,42, and can instead be performed in the cell sgRNA complex — in the form of a viral vector or type that is most appropriate for the disease of interest. For plasmid (that encodes Cas9 and the sgRNA), or Cas9– example, if a panel of tumour-derived lines are thought sgRNA ribonucleoprotein (RNP) complexes (compris- to be sensitized to a drug candidate via a genetic lesion, ing Cas9 protein and the sgRNA) — along with a DNA CRISPR–Cas-mediated gene knockout can directly test the repair template. The HDR template can also take on hypothesis of synthetic lethality 43–45. Such isogenic knock- different forms, and its exact design can substantially outs allow researchers to rapidly establish causative roles affect repair efficiency 48,49. In mammalian cells, short for oncogenes, tumour suppressors and other factors in a single-stranded DNA (ssDNA) oligonucleotides can be defined context, thereby removing secondary differences. designed to take advantage of the molecular nature of Similarly, ‘knocking in’ mutant alleles by HDR the Cas9–target architecture and these oligonucleotides allows researchers to test the effects of disease-associ- have been shown to effectively introduce small muta- ated mutations in an isogenic background. For exam- tions50. Additional control over the efficiency of mutant ple, HDR can serve to generate mutant allelic series introduction and zygosity can be achieved by varying the to compare the effects of each variant found across distance between the DSB and the site of the mutation patients, as is the case for oncogenes such as KRAS, on the repair template 51. NATURE REVIEWS | DRUG DISCOVERY VOLUME 16 | FEBRUARY 2017 | 91 © 2 0 1 7 M a c m i l l a n P u b l i s h e r s L i m i t e d , p a r t o f S p r i n g e r N a t u r e. A l l r i g h t s r e s e r v e d. REVIEWS Despite this promise, although CRISPR–Cas knock- simplicity of designing potent sgRNAs and the ability to outs are effective in nearly any cell, rates of HDR can apply the system to nearly any cell type or tissue (FIG. 2). vary across cell types. As one example, it has been dif- Large-scale screens typically rely on pooled lentiviral ficult to achieve even moderate levels of HDR in non- libraries of sgRNAs, often achieving robust hit identi- mitotic human cells, including neurons. These barriers fication by including 3–10 sgRNAs per gene20,63–67. The are particularly frustrating, because sequence insertion procedure of CRISPR–Cas-based screens is very similar or replacement in these contexts could be used to model to that of short hairpin RNA (shRNA) screens. A pool or to treat many genetic diseases. New approaches that of cells that co-express Cas9 and the sgRNA library use non-homologous or microhomology-mediated is subjected to the desired phenotypic selection, and integration of cassettes48,52–55 offer routes to bypass high-throughput DNA sequencing of the sgRNA cas- HDR pathways that are inactive in non-mitotic human sette is used to identify sgRNAs that were enriched or cells and in organisms in which HDR has proved to depleted during the treatment. be difficult. Another exciting new development is the Genome-scale CRISPR–Cas knockout, inhibition engineering of Cas enzymes with additional function- and activation screens have identified essential genes in alities to enable precise, template-less introduction of various cancer cell lines63,64,68–70, uncovered genes that are specific mutations by direct alteration of target bases. involved in the response to small-molecule inhibitors60,65 A first step towards this goal was the fusion of various and cellular toxins20,66, and dissected the relative impor- cytidine deaminases to Cas9, which resulted in hybrid tance of viral host factors71. They have also been used in enzymes that are capable of RNA-guided ‘base editing’ a xenograft mouse model of tumour growth and metas- (REFS 56,57), and one can anticipate a dramatic increase tasis to assay gene phenotypes in cancer evolution72. in the number of new Cas derivatives developed using Although CRISPR–Cas screens for cell growth or survival similar strategies. have been quite successful (except when targeting genet- ically amplified regions64,68,69), screens for more complex Functional screening with CRISPR–Cas. Large-scale phenotypes are still in the process of being optimized. functional screening with CRISPR–Cas is simultane- Recent comparisons with microRNA-based shRNA ously expanding and evolving, as researchers uncover screens have found comparable performance60,70, and the the advantages and disadvantages of different screening complementary strengths of both approaches should be systems. Until recently, systematic loss‑of‑function stud- carefully weighed when choosing a screening platform ies focused on genome-wide RNA interference (RNAi) (TABLE 1). screens58–60 or insertional mutagenesis screens in hap- In CRISPR nuclease (CRISPRn) screens, stably loid human cell lines41,42,61,62. CRISPR–Cas screens have expressed Cas9–sgRNA complexes continue to oper- rapidly been adopted in various contexts owing to the ate on a target site until it is ablated and can therefore Table 1 | Comparison of screening platforms Characteristic CRISPRn CRISPRi RNAi* CRISPRa Effect Knockout Knockdown Knockdown Activation Mechanism Mutation-causing indel Transcriptional Transcript Transcriptional interference degradation activation and/or translational interference Guide target choice Anywhere in the genome TSS with a PAM Exons TSS with a PAM with a PAM Target selectivity Can distinguish any target Depends on TSS, Can distinguish Depends on TSS, cannot distinguish splice variants cannot distinguish products derived products derived from the same from the same transcript transcript Highly amplified Off-target effects: DSBs Can be targeted Can be targeted Can be targeted regions (genes) evoke DNA damage repair, if all use the same if all use the same resulting in cell cycle arrest TSS TSS independently of target Distinguish Possible Yes Possible Yes alternative TSSs Distinguish transcript Possible No Possible No splice variants CRISPR nuclease Performance of Most work Many work Requires good Many work (CRISPRn). Targeting a DNA individual sgRNAs or prediction tools or sequence with catalytically shRNAs testing active Cas9 to generate a DSB, double-strand break; indel, insertion and/or deletion; PAM, protospacer adjacent motif; RNAi, RNA interference; sgRNA, single double-strand break or a nick. guide RNA; shRNA, short hairpin RNA; TSS, transcription start site. *MicroRNA-based shRNAs. 92 | FEBRUARY 2017 | VOLUME 16 www.nature.com/nrd © 2 0 1 7 M a c m i l l a n P u b l i s h e r s L i m i t e d , p a r t o f S p r i n g e r N a t u r e. A l l r i g h t s r e s e r v e d.  REVIEWS generate homozygous knockout phenotypes at high fre- We expect that CRISPR–Cas-based screens will quency in most cell types. Conversely, high copy num- continue to improve, especially as they are used for an ber genomic amplifications can be a barrier to CRISPRn increasingly broad array of phenotypes. Most of the screens, mainly because of the large numbers of DNA pioneering CRISPR screens simply looked for growth breaks that are generated in high copy number regions. advantages and disadvantages, leading to the identifi- The large number of DNA breaks can lead to reduced cation of genes that are essential for proliferation, or cell growth triggered by the DNA damage response and resistance or sensitivity to certain toxins. Going for- cell cycle arrest, which are activated independently of ward, there will be more CRISPR screens to examine the targeted gene or genomic region (thus representing the sensitivity of cancer cells to candidate therapeutics, a systematic, sequence-independent off-target resistance to pathogen infections, or the regulation and effect)64,68,69. As CRISPRn generally depends on sequence cellular localization of a gene or protein of interest 42,60,71. frame-shifting to generate knockouts, phenotype pen- CRISPR screens in human pathogens can also be used etrance can be affected if in‑frame deletions are pref- to identify candidate drug targets82. The relatively low erentially created. This can be overcome by targeting cost of sgRNA library design facilitates creative screen- functional domains43, although this approach requires ing approaches, such as efforts to identify non-coding pre-existing knowledge of target proteins. Furthermore, sequences that control expression of B-cell CLL/lym- sgRNAs that target the 5ʹ end of the coding region may phoma 11A (BCL11A), TP53 and oestrogen receptor 1 be ineffective if alternative downstream start codons are (ESR1) using target-tiling CRISPRn screens83,84, in present 68. which a genomic region is targeted with multiple guide CRISPRi screens do not rely on frame shifting and RNAs, and we expect future screens for non-coding can offer certain advantages over CRISPRn screens regulatory elements to examine even larger regions of from a drug discovery perspective, because knocking DNA sequence. However, more systematic analyses are down gene expression (using CRISPRi or RNAi) can needed to compare CRISPRn, CRISPRi and various mimic the effects of a small-molecule inhibitor more types of RNAi screens (including microRNA-based closely than does complete gene ablation73. CRISPRi shRNAs)70. Such comparisons will define the relative screens can also identify the contributions of tran- strengths and weaknesses of each platform and allow scripts arising from different transcription start sites researchers to choose the best type of screen to address (TSSs), whereas RNAi screens can uniquely distinguish their question (TABLE 1). different splice variants59,74. CRISPRa screens, which assess gene targets whose Rapid generation of animal models. Beyond cell culture overexpression leads to a given phenotype20,21, are an applications, genome editing has dramatically altered emerging and particularly exciting area of recent devel- our ability to generate animal models of disease (FIG. 3). opment. They have an array of benefits and trade-offs It will soon be common for early ‘go’ or ‘no‑go’ decisions compared with cDNA screens, which have previously in a drug development campaign to be based on results been used in this area. Construction and use of cDNA from rapidly created mutant animals of the most rele- screening resources are labour intensive owing to the vant model species for a disease. Indeed, shortly after complex nature of cDNAs. By contrast, the resources their initial development, CRISPR–Cas tools were used necessary to perform CRISPRa screens are similar to generate mice with multiple genetic lesions in a single to those required by CRISPRn or CRISPRi screens20. editing step85, as well as for one-step knock‑in of reporter Moreover, cDNA expression screens can only interro- and conditional alleles into mouse zygotes86. gate the transcripts present in the library, which may In general, efficient CRISPR–Cas editing techniques, lack certain genes or transcript variants. Conversely, including NHEJ and short HDR, can be achieved by by stimulating expression from the endogenous locus, microinjection or simple electroporation of zygotes CRISPRa screening can activate expression of alternative instead of proceeding through traditional embryonic transcripts from secondary TSSs as easily as it activates stem (ES) cell manipulation87–89. This is a crucial devel- expression of the primary transcript, and sgRNAs can be opment in two ways. First, as multiple genes can be tar- designed to target each TSS within each gene. However, geted in a single step, double- and triple-mutant mice CRISPRa screens are subject to their own set of false can be rapidly generated without the need for crossing negatives. For example, CRISPRa will have no effect if single-mutant strains, although it must be noted that the target gene contains loss‑of‑function mutations or is such alleles follow Mendelian segregation upon breed- missing entirely in the cell line of interest. ing. Second, genome editing in zygotes eliminates the A substantial technical barrier for CRISPRa screen- requirement to derive, culture and edit ES cells, which ing is the activation of highly repressed genes. To over- has slowed the generation of mutants and has been come this challenge, a range of CRISPRa systems have a major barrier to widespread genetic tractability in sev- Backcrossing The process of breeding a been developed that recruit multiple and/or diverse tran- eral model organisms relevant to the process of therapeu- hybrid organism with an scriptional activation domains to increase the potency tic discovery, such as rats. Zygote editing also accelerates individual genetically similar to of gene activation20,21,75–80. Ultimately, an ideal CRISPRa the generation of additional mutations in pre-existing one of its parents, with the screening platform would use the fewest necessary exo­ animal models of disease by eliminating the need for objective of diluting the genetic contribution of the other genous parts to potently activate any gene target; addi- ES cell derivation or lengthy backcrossing. Nevertheless, parent to subsequent tional developments and systematic comparisons are the introduction of large transgenes or complex multi- generations. needed in order to achieve this goal81. component systems via zygote editing remains inefficient, NATURE REVIEWS | DRUG DISCOVERY VOLUME 16 | FEBRUARY 2017 | 93 © 2 0 1 7 M a c m i l l a n P u b l i s h e r s L i m i t e d , p a r t o f S p r i n g e r N a t u r e. A l l r i g h t s r e s e r v e d. REVIEWS a Ex vivo editing Somatic editing Lentivirus or retrovirus AAV or lentivirus In vivo screening Transfection Transposon vectors Tissue microenvironment Electroporation Functional immune system Development or regeneration b Cas9–sgRNA Plasmid mRNA and sgRNA RNP ES cell editing Or Blastocyst injection or tetraploid Transgenic animal models complementation Rodent models Large mammals Non-human primates Zygote preparation and editing Animal model Harvest zygotes Re-derive ES cells Experimental cohort Disease modelling Tissue toxicity Development Figure 3 | Applications of CRISPR–Cas in in vivo screens and the generation of animal models. a | Ex vivo editing can be used to generate a library of modified cells for transplantation into recipient animals. Alternatively, Nature Reviews editing reagents | Drug Discovery can be delivered to host animal tissues directly for somatic in situ editing. b | CRISPR–Cas has also revolutionized the generation of transgenic animal models through facile editing of embryonic stem (ES) cells for traditional gene targeting and by enabling direct zygote editing in most species. Zygote editing can be done ex vivo by electroporating or microinjecting zygotes with CRISPR–Cas constructs in the form of plasmids, RNA preparations or ribonucleoproteins (RNPs). AAV, adeno-associated virus; sgRNA, single guide RNA. and for now, gene targeting in ES cells is likely to remain months or weeks. A large range of models can now be the method of choice for generating animals that harbour generated in a timescale relevant to early go or no‑go such mutations90,91. decisions in a modern drug discovery campaign. Drug Founder animals from zygote editing or conven- discovery implications of gene editing in additional tional blastocyst injection of modified ES cells can species are discussed at the end of this section. exhibit mosaicism (BOX 1). Mosaicism in ES cell injec- Pairing CRISPR–Cas with viral or transposon-based tion studies can be reduced by the tetraploid com- vectors has allowed researchers to directly introduce plementation method92,93 in which modified ES cells somatic mutations in certain tissues, such as lung and are introduced into developmentally compromised liver tissues, in adult animals. This approach has been blasto­c ysts, although this is a technically complex used to create numerous cancer and other disease procedure that requires amenable ES cells. Conversely, models94–96 and to correct disease mutations and phe- in zygote electroporation studies, mosaicism is due to notypes97–100. One illustrative example of the power of the fact that the single-cell zygote occasionally divides CRISPR–Cas tools is the in vivo engineering of onco- before editing occurs. Hence, replacing Cas9 mRNA genic chromosomal rearrangements that mimic fusion and sgRNA or Cas9–sgRNA-encoding plasmids with proteins found in patients (for example, EML4–ALK, Cas9–sgRNA RNP complexes, which can immedi- KIF5B–RET and CD74–ROS1), which lead to in situ ately act on their targets, increases the fraction of tumour initiation from edited somatic cells101,102. The non-mosaic founders but does not completely solve ability to introduce disease-associated alleles in live the problem87,88. Overall, CRISPR–Cas promises to rev- animals is particularly transformative compared with olutionize mouse genetics by reducing the time that is xenograft models that require immunosuppressed recip- necessary to generate targeted models from years to ients and mostly rely on implantation at non-native sites. 94 | FEBRUARY 2017 | VOLUME 16 www.nature.com/nrd © 2 0 1 7 M a c m i l l a n P u b l i s h e r s L i m i t e d , p a r t o f S p r i n g e r N a t u r e. A l l r i g h t s r e s e r v e d.  REVIEWS Box 1 | Mosaicism of the biggest challenges in the genetic manipulation of new mammalian model organisms. CRISPR–Cas zygote Mosaicism is the presence of cells that have multiple different genotypes within a editing should soon eliminate this hurdle. single animal or cell population. In cell culture Specificity of CRISPR systems In most cases, a population of edited cells will contain numerous mutations, even if all Although CRISPR-based tools are easily programmed of the alleles within the cell population are edited. This is because DNA double-strand to target basically any genomic location, they can also break (DSB) repair by the predominant non-homologous end joining (NHEJ) pathway lead to low rates of off-target editing or sequence-inde- can lead to different insertions and/or deletions (indels) in different cells. Depending on pendent cell cycle arrest if highly amplified loci are tar- the experiment, mosaicism in cultured cells may or may not be problematic. If the geted64,68,69. At first glance, one might assume that a less editing efficiency is sufficiently high and all mutations cause the same phenotype (for example, loss of function due to mutations in the active site of an enzyme), the than perfect gene-editing reagent would prevent substan- mosaicism is functionally irrelevant. In other cases, some indels might result in an tial adoption of the tool. However, for non-therapeutic in‑frame deletion that has no phenotype, leading to a variegated population. use, such stringency might not always be needed and Mosaicism can be eliminated by deriving single-cell clones. can be compensated for with proper controls. Hence, the In animal models most important aspect is a thorough understanding of When edited embryonic stem (ES) cells are injected into a blastocyst for model off-target events, their biological consequences and how generation, the resulting animal can be a mosaic of the donor ES cells and the cells of these effects can be mitigated. the recipient blastocyst. Tetraploid embryo complementation, a method that renders Sequence-dependent off-target propensities are a recipient blastocyst developmentally compromised, can reduce this risk. Mosaicism best understood for the SpyCas9 enzyme, for which can also be a result of zygote editing if editing takes place after the one-cell stage. combinations of systematic and unbiased experiments Hence, editing methods that act on their DNA targets directly upon transduction (such have begun to shed light on potential liabilities111–113. as Cas9–single guide RNA ribonucleoprotein complexes) may reduce mosaicism in Several excellent reviews have extensively discussed founder animals. Mosaicism at a given locus can be eliminated by backcrossing founder CRISPR–Cas off-target effects114–116. Nevertheless, our animals for a single generation, but can nonetheless be problematic if multiple genes understanding of the molecular mechanisms through are targeted simultaneously. For example, founder animals with mutations in three targeted genes will not necessarily carry all three mutations in every individual cell. which Cas9 can sometimes inappropriately bind to If this is the case, multiple generations of breeding are needed to generate non-mosaic and cut off-target sequences is still in its infancy, and animals with mutations in all three genes. indeed such tolerance may be built into naturally evolved CRISPR–Cas systems as part of the immunological ‘arms race’ between the phage and its bacterial host. The in situ introduction of mutations with CRISPR–Cas For SpyCas9, phenomenological data revealed that the allows researchers to accurately recapitulate disease ini- 8–10 nucleotides neighbouring the protospacer adjacent tiation, development and maintenance in an autoch- motif (PAM) are most stringently recognized, whereas thonous and immunocompetent setting, including the one or two mismatches can be tolerated in the remaining native microenvironment and tissue structure. This nucleotides67,117,118. ability will be transformative for many diseases, particu- Off-target sites are determined by the nuclease and larly for cancer, in which the interaction of tumour cells the sgRNA sequence. Thus, several algorithms have with immune cells can have a drastic effect on disease been developed to predict sgRNA efficiency and off- outcome96,103. target sites67,119–123. Although comparison with unbiased The large and rapidly growing number of organisms genome-wide assessment of off-target sites has revealed targeted by CRISPR–Cas holds great promise, as tradi- the limited predictive power of many algorithms for tional gene targeting has remained difficult in preclinical distantly related off-target sites111, likely off-target sites models other than mice. CRISPR–Cas editing has been and clearly risky sgRNAs can still be identified. From performed in rats104, dogs105 and cynomolgus monkeys106, the perspective of research use for target identification which are all commonly used during preclinical drug and validation, any candidate sequence that is identified discovery and development. As with mouse zygote tar- through a CRISPR–Cas knockout experiment should be geting, many of the edited animals exhibit mosaicism. validated with orthogonal strategies to rule out off-target The generation of disease models in primates, such as effects. These strategies might include the use of multiple a model of Duchenne muscular dystrophy in rhesus sgRNAs, isolation of multiple clonal lines, validation by monkeys107, further emphasizes how gene editing can alternative transcript knockdown methods (for exam- not only accelerate therapeutic development but also ple, CRISPRi or RNAi), and cDNA or CRISPRa comple- test the efficacy and safety of therapeutic compounds. mentation studies. This methodology mirrors follow‑up Protospacer adjacent motif CRISPR–Cas may even drive the development of por- experiments that are required for comparable RNAi (PAM). Short genomic cine xenotransplant platforms through the inactivation approaches. In a research setting, the ability to perform sequence adjacent to the of endogenous retroviruses108. such validation experiments makes the extensive iden- sequence targeted by the guide RNA that is required for CRISPR–Cas editing will also be a boon to infec- tification of rare off-target sites relatively superfluous. recognition by Cas effectors. tious disease research. Many human pathogens are best Off-target analyses and de‑risking strategies are far more This sequence varies based on modelled in hosts other than mice, such as influenza critical for therapeutic CRISPR–Cas gene editing than the identity of the effector (for (ferrets)109, leptospirosis (hamsters)109 and tuberculo- for preclinical investigations. example, Cas9 versus Cpf1) and species (for example, sis (guinea pigs)110, and we expect zygote editing to be Much effort has been put into the development of Streptococcus pyogenes versus proven feasible in these organisms in the near future. strategies to systematically minimize CRISPR–Cas Francisella novicida). The optimization of conditions for ES cell work was one off-target effects. One tactic requires two Cas9 nickases NATURE REVIEWS | DRUG DISCOVERY VOLUME 16 | FEBRUARY 2017 | 95 © 2 0 1 7 M a c m i l l a n P u b l i s h e r s L i m i t e d , p a r t o f S p r i n g e r N a t u r e. A l l r i g h t s r e s e r v e d. REVIEWS (Cas9‑D10A or Cas9‑H840A2, which cleave or ‘nick’ CRISPR–Cas can contribute to the field. We focus on only a single DNA strand) to bind at neighbouring therapeutic applications other than in vivo gene editing, sites, thereby increasing the effective stringency that as this topic has been covered by several recent reviews. is due to the low probability of adjacent off-target sites within a genome124,125. Similarly, a dimerizing FokI nucle- Creating CAR T cell-based therapies with gene editing. ase domain (used by other DNA-editing tools, such as The application of gene editing for somatic diseases has zinc-finger nucleases (ZFNs) and transcription activa- begun to overlap with the rapidly expanding field of can- tor-like effector nucleases (TALENs)) has been fused to cer immunotherapy, with immediate interest centring a catalytically inactive Cas9 to ensure that only paired on the production of next-generation chimeric antigen binding can induce DSBs126,127. Although they reduce receptor (CAR) T cells. These modified T cells, which off-target events, paired-nickase strategies also reduce the express tumour-targeting receptors, have shown promise targetable space because they require two sgRNAs to bind in the treatment of various leukaemias and lymphomas, to their targets within a relatively short stretch of DNA. In and may eventually be used to treat solid cancers135. pooled screening scenarios, paired nickases also require a CARs comprise an extracellular binding domain (cur- combinatorial library or tandem sgRNA vectors. rently a single-chain variable fragment), which recog- A second strategy to reduce off-target events relies on nizes an antigen that is strongly expressed on — and sgRNA or protein engineering to enforce higher specific- specific to — tumour cells, and an intracellular chimeric ity. Truncated guide RNAs can remove a few of the rela- signalling domain that activates the T cell upon recep- tively permissive bases from the 5ʹ end of the guide RNA, tor engagement and promotes T cell-mediated killing of which results in both decreased on-target and off-target tumour cells. The first battery of CAR T cell-mediated activity 111,128. SpyCas9 has also been mutagenized to therapies targeted CD19, an antigen expressed by B cells more specifically recognize only a single PAM129 or to and related cancer cells; several of these therapies have abrogate nonspecific binding and thereby reduce the entered clinical trials (Juno Therapeutics: NCT02535364 cleavage of non-target sequences130,131; although, again, and NCT02631044; Kite Pharma: NCT02601313 and the mechanism of action is still under investigation. NCT02348216; and Novartis: NCT02030834 and A third strategy to reduce off-target events adds strict NCT02445248). control over the amount of active Cas9 in cells. So far, Currently, most CAR T cells are generated by using such approaches have used tightly regulated induction each patient’s own T cells, an expensive and time- of Cas9 activity 132 and even reversible small-molecule- consuming process that involves isolating, modifying induced or photo-induced Cas9 activity 133,134. These and expanding T cells for every new patient. This pro- methods reduce off-target effects and also enable tempo- cess is limited by current manufacturing capabilities. ral control of genome editing. In appropriate scenarios, Hence, the economics of CAR T cells are less favour- use of carefully titrated amounts of Cas9–sgRNA RNP able than those of antibody-based checkpoint cancer complexes, which are rapidly degraded, can have similar immunotherapies such as ipilimumab, pembrolizumab benefits. Ultimately, many of the strategies outlined here and nivolumab. CAR T cell therapy could become much might even be modularly combined for further gains in faster and less expensive if universal donor CAR T cells specificity, although this has yet to be experimentally could be generated, as ‘off-the-shelf ’ cells would sub- tested. SpyCas9 has naturally high fidelity, and a vari- stantially increase the number of patients that could be ety of approaches have been able to improve its speci- treated by a single CAR T cell product. However, graft- ficity. It is easy to foresee how additional engineering versus-host disease (GVHD) and host rejection, caused approaches, combined with a more detailed mechanistic by recognition of recipients’ cells by the CAR T cells and understanding of the conformational changes that occur recognition of the CAR T cells by the host, respectively, during target binding and cleavage, will advance editing remain major barriers to an off-the-shelf approach. In precision. this context, ZFNs and TALENs have been used to knock out endogenous T cell receptor genes in T cells, which Using CRISPR–Cas to make therapeutics could prevent unwanted graft-versus-host reactivity 136,137 In addition to generating powerful research tools, (Servier: NCT02808442). Genome-editing strategies genome editing with CRISPR–Cas technology holds could also be used to prevent or delay the rejection of great promise to make therapeutic agents or as a thera- CAR T cells by the recipient’s immune system through peutic itself. In principle, any DNA-editing technology the elimination of or a decrease in the expression of could be used for the therapeutic strategies described histo­compatibility antigens on the donor T cells138. in this section. Although ZFNs have advanced the fur- In addition to enabling the generation of off-the-shelf thest in clinical trials to date, there is currently insuffi- CAR T cells, genome editing could be used to boost cient evidence to declare whether the clinical utility of CAR T cell efficacy by knocking out the genes encoding CRISPR–Cas, ZFNs or TALENs will be superior. The T cell inhibitory receptors or signalling molecules, such fast and inexpensive reprogramming of Cas9 gives it as cytotoxic T lymphocyte-associated protein 4 (CTLA4) a clear advantage in contexts in which rapid exper- or programmed cell death protein 1 (PD1)139,140. Indeed, imental iteration is beneficial or in contexts in which the US National Institutes of Health (NIH) Recombinant many different loci need to be targeted. Here, we briefly DNA Advisory Committee (RAC) recently approved discuss the current state of therapeutic gene editing a clinical trial that will be carried out at the University (mostly in the context of ZFNs and TALENs) and how of Pennsylvania in which Cas9 will be used to knock 96 | FEBRUARY 2017 | VOLUME 16 www.nature.com/nrd © 2 0 1 7 M a c m i l l a n P u b l i s h e r s L i m i t e d , p a r t o f S p r i n g e r N a t u r e. A l l r i g h t s r e s e r v e d.  REVIEWS out the genes encoding PD1 and the endogenous T cell the T cells themselves may not be. Researchers have receptor in melanoma-targeting CAR T cells (see Further recently focused on disrupting CCR5 in HSCs in order to information). China recently began the first clinical trial produce long-term self-renewing HIV-resistant cells147. of CRISPR–Cas. The trial uses Cas9 to knock out PD1 CRISPR–Cas could also be applied to the same workflow in T cells of individuals with lung cancer, although no of extracting, editing and re‑implanting cells, and several CAR will be introduced in that trial141. Similar trials with groups have edited CCR5 with Cas9 (REFS 6,148,149). PD1‑knockout T cells for prostate and bladder cancer, as The haemoglobinopathies sickle cell disease (SCD) well as renal cell carcinoma, are also being initiated (see and β-thalassaemia have been targeted for ex vivo gene Further information). In the future, gene editing might correction instead of disruption. All patients with SCD even be used to introduce the CAR itself via HDR. Site- carry the same causal mutation in the haemoglobin specific knock‑in would eliminate the need for randomly subunit beta (HBB) gene, which causes a glutamate-to- integrating viral delivery vectors and allow for control over valine substitution, ultimately leading to the aggregation where the CAR integrates142,143. Future CAR T cell thera- of haemoglobin and misshapen red blood cells. ZFNs pies could benefit from combined modification of endog- have been used to correct the sickle allele in HSCs via enous T cell receptor genes, histocompatibility genes and HDR by using an integrase-defective lentiviral vector or components of signalling pathways. Still, it will be impor- a single-stranded oligonucleotide donor 150. CRISPR–Cas tant to establish that the removal of inhibitory signals does has rapidly caught up to ZFNs, demonstrating correction not enable uncontrolled proliferation of the CAR T cells. of the sickle allele using either an adeno-associated virus Compared with other gene-editing reagents, such as 6 (AAV6) or oligonucleotide donor 151,152. By contrast, ZFNs and TALENs, CRISPR–Cas allows for extremely β-thalassaemia is caused by a variety of null or hypomor- rapid testing of any newly proposed genetic modifica- phic mutations in HBB153, thereby requiring a plethora tions. Several industry partnerships have been announced of case-specific targeting complexes and repair donors. between developers of CAR T cell therapies and com- CRISPR–Cas could be superior to other nucleases in such panies specializing in gene editing, including Novartis’ situations, as designing new sgRNAs is much faster and collaboration with Intellia Therapeutics and Caribou cheaper than engineering new TALENs or ZFNs. The Biosciences, and Juno Therapeutics’ collaboration with regulatory landscape that surrounds such personalized Editas Medicine. CAR T cell producer Cellectis acquired a approaches is in flux. Currently, even though multiple licence to use TALENs from the University of Minnesota. editing reagents might revert mutations to the same sequence, they would be classified as separate investiga- Therapeutic ex vivo gene editing. Drug delivery to the tional new drugs (INDs). appropriate cells or tissue in situ is challenging in many Regardless, individually correcting all disease-caus- fields and is certainly a major limitation for therapeutic ing HBB mutations could be unnecessary, as β-thalassae- applications of CRISPR–Cas. Ex vivo manipulation of tar- mia and SCD may be correctable by reactivation of fetal get cells circumvents this issue. The haematopoietic sys- γ-globin (also known as haemoglobin γ-subunit) tem is an excellent target for ex vivo gene editing, because expression. The transcription factor BCL11A represses cells are readily obtained from peripheral blood samples fetal γ-globin in adults, and Sangamo and Biogen ini- and can be re‑injected after manipulation and expansion. tially sought to systemically disrupt BCL11A to increase Therapeutic ex vivo gene editing of haematopoietic stem fetal globin expression in patients with β-thalassaemia. cells (HSCs) has previously been explored using ZFNs However, several groups have now used TALENs, and TALENs, and some of these therapies are showing ZFNs and CRISPR–Cas to identify an erythroid- promise in clinical trials. The most advanced strategy specific enhancer that controls BCL11A expression83,154,155. uses ZFNs to target the C-C motif chemokine receptor Notably, tiling sgRNA libraries, in which a genomic 5 (CCR5) gene in cells from patients with HIV144. CCR5 region is targeted with many sgRNAs, took advantage of is a co‑receptor for HIV entry, and individuals with the ease with which CRISPR–Cas can be reprogrammed loss‑of‑function mutations in CCR5 are highly resistant to probe more than 500 sites in the enhancer region, and to HIV infection but are otherwise healthy. Importantly, identified a minimal target sequence for disruption83. transplantation of bone marrow from a CCR5‑deficient As disruption of the enhancer leads to an erythroid- donor to an HIV-infected individual with leukaemia, specific decrease in BCL11A and an increase in fetal known as the ‘Berlin patient’, reduced the patient’s viral globin production, Biogen and Sangamo have combined load to undetectable 145. Although CCR5‑deficient, their B