Engineering Synthetic TALE and CRISPR/Cas9 Transcription Factors for Regulating Gene Expression

Abstract

Engineered DNA-binding proteins that can be targeted to specific sites in the genome to manipulate gene expression have enabled many advances in biomedical research. This includes generating tools to study fundamental aspects of gene regulation and the development of a new class of gene therapies that alter the expression of endogenous genes. Designed transcription factors have entered clinical trials for the treatment of human diseases and others are in preclinical development. High-throughput and user-friendly platforms for designing synthetic DNA-binding proteins present innovative methods for deciphering cell biology and designing custom synthetic gene circuits. We review two platforms for designing synthetic transcription factors for manipulating gene expression: Transcription Activator-Like Effectors (TALEs) and the RNA-guided Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR)/Cas9 system. We present an overview of each technology and a guide for designing and assembling custom TALE- and CRISPR/Cas9-based transcription factors. We also discuss characteristics of each platform that are best suited for different applications.

1. Introduction

The inspiration for many of the technological advances that have transformed biomedical research, such as polymerase chain reaction or RNA interference, has come from natural biological processes. Similarly, recent discoveries of the principles of protein-DNA interactions in various species and systems has guided the development of methods for engineering designer proteins that can be targeted to any DNA target site. These proteins can serve as a scaffold for building enzymes that can modify DNA sequence, transcriptional regulation, or the epigenetic status at any site in the genome. Three main classes of natural biomolecules have been engineered to target new DNA sequences and manipulate gene expression: zinc finger proteins (ZFPs), Transcription Activator-Like Effectors (TALEs), and the Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR)/CRISPR-associated protein (Cas) system. Each of these programmable DNA-binding proteins can be genetically fused to an effector domain to create custom enzymes that localize the effector function to the DNA target site. Various effectors domains have been widely used to create targeted changes to genome sequence, including nucleases, integrases, and recombinases (1–5). Alternatively, the fusion of these DNA-binding proteins to transcriptional activation and repression domains enables the control of gene regulation at targeted promoter or enhancer elements (6–8). The applications of these gene regulation technologies are diverse, including stimulating expression of endogenous therapeutic factors (9–16), controlling cell differentiation (17–20), and regulating synthetic gene circuits (21–23). This article discusses the origins of these technologies for targeted gene regulation, the methods for engineering programmable transcriptional activators and repressors, and design considerations for implementing the various approaches.

The first engineered proteins designed to target new DNA sequences were based on the Cys₂His₂ zinc finger domain (5, 24), the most common DNA-binding motif in the human proteome (25). Despite many notable successes with this technology (1, 3, 5, 26), engineering new ZFPs with high activity and specificity remains technically challenging for most researchers. Since predictions of ZFP DNA-binding specificity and affinity are complex (27, 28), it is typically necessary to build and screen many rationally designed proteins or use high-throughput selections to find functional proteins within large libraries (24, 29, 30). Consequently, academic laboratories are adopting the newer TALE and CRISPR/Cas9 platforms that have straightforward DNA-recognition properties (31, 32). Nevertheless, two decades of studies on targeted gene regulation with engineered ZFPs have provided much of the foundation on which these new platforms are being built (5).

The DNA recognition code for TALE proteins was first reported in 2009 (33, 34) and presented a new modular DNA-binding domain that is more easily reprogrammed to target new sequences compared to ZFPs (35–38). Because of their simple DNA recognition code, the frequency of engineering active TALE-based enzymes is very high (4, 39). Consequently, there have been many recent successes modulating mammalian gene expression with synthetic TALE enzymes (14, 18,19, 21, 22, 35, 36, 40–42). In 2012, an engineered version of the CRISPR/Cas9 system was developed as the first platform for targeting proteins to DNA target sites through RNA:DNA interactions rather than direct protein-DNA interactions (43). Since novel proteins do not need to be engineered for each DNA sequence, the CRISPR/Cas9 system greatly expedites the process of molecular targeting to new sites by simply modifying the expressed RNA molecule. Therefore the CRISPR/Cas9 system is typically easier, faster, and more economical to implement in comparison to the ZFP and TALE technologies. After the CRISPR/Cas9 system was shown to be effective as a nuclease platform for genome editing (44–48), it was quickly reengineered for transcriptional regulation (20, 49–56). The early studies on both of these systems for gene regulation have demonstrated that there are advantages and limitations of each approach, and further work is necessary to elucidate properties of each of these systems. Here we present an overview and user’s guide for applying the TALE and CRISPR/Cas9 systems to control gene expression.

1.1 Transcription Activator Like Effectors

Transcription Activator Like Effectors (TALEs) are modular DNA-binding proteins derived from the plant pathogenic bacteria Xanthomonas (33, 34) and Ralstonia (57). As a defense mechanism, these organisms produce TALEs to modulate host gene expression. The TALE DNA-binding domain consists of multiple repeats of 34 amino acids where variability in positions 12 and 13, referred to as the repeatvariable di-residues (RVDs), confer binding specificity for one specific DNA base (33, 34, 58–60) (Figure 1A). Multiple TALE monomers can be linked in tandem to recognize the desired DNA sequence (35–38) (Figure 1B). The array of TALE domains is then fused to an effector domain to induce a specific action at a user-determined genomic locus (61).

Figure 1. The TAL Effector DNA-binding Domain.

(A) Through a DNA-protein interaction, each TALE repeat binds one bp of DNA. The TALE repeat is shown in blue, and the repeat variable diresidue (RVD) at the 12^th and 13^th position are shown in green and red, respectively. (B) TALEs can be linked in tandem to recognize virtually any DNA sequence. The desired string of TALEs is then fused to an effector domain to induce a specific action at a predetermined DNA sequence. Crystal structure adapted from (60).

1.2 The CRISPR/Cas System

Bacteria and archaea have evolved the CRISPR/Cas system as an RNA-guided defense mechanism against viral parasites that detects and silences foreign nucleic acids (62). In the naturally-occurring system, bacteria and archaea integrate short fragments of foreign nucleic acids (termed protospacers) into the CRISPR genomic loci. Functioning as molecular memory of previous invaders, the CRISPR locus is transcribed and processed into short CRISPR-derived RNAs (crRNAs). Thus each crRNA contains sequence complementarity to a prior nucleic acid invader. In the type II system, crRNAs associate with transactivating crRNAs (tracrRNAs) and the Cas9 endonuclease. Through complementary base pairing, the crRNA localizes the Cas9 complex to the foreign DNA sequence to induce a double strand break. Characterization of CRISPR cleavage sites identified a short sequence directly downstream from the protospacer required for Cas9-mediated cleavage. This sequence is called the protospacer adjacent motif (PAM) and the identity of the sequence is highly variable between CRISPR systems from different species. For ease of use, a single transcript chimeric guide RNA (gRNA) has been engineered to recapitulate the function of both the crRNA and tracrRNA (43–45). The chimeric gRNA consists of three regions: a 20 bp protospacer which confers targeting specificity through complementary base pairing with the desired DNA target, a nucleotide hairpin which mimics the crRNA:tracrRNA structure required for Cas9 protein binding, and a transcriptional termination sequence (63). It was recently demonstrated that the crRNA, tracrRNA, and Cas9 nuclease were all necessary and sufficient to program the RNA-guided Cas9 nuclease activity to new sequences outside of the native host (43). The gRNA could also substitute for the crRNA and tracrRNA in these experiments. Subsequently, this engineered type II CRISPR system was shown to function effectively for RNA-guided programmable nuclease activity in numerous other hosts, including human cells (44–47), mouse cells (64–66), C. elegans (67, 68), zebrafish (48, 69), and bacteria (70).

Cas9 catalyzes DNA double-stranded breaks via RuvC and HNH endonuclease domains, each of which cleaves one strand of the target DNA. Both of these enzymatic domains can be inactivated by a single amino acid substitution (D10A and H840A), generating a Cas9 protein that has no endonuclease activity but maintains its RNA-guided DNA-binding capacity (43). This deactivated Cas9 (dCas9), in conjunction with the gRNA, functions as a modular DNA-binding scaffold similar to ZFPs and TALEs. Therefore the discovery of CRISPR-based immunity has led to the development of a new class of modular synthetic enzymes where DNA-binding is directed by an RNA:DNA interaction rather than a protein:DNA interaction (Figure 2). This dCas9 scaffold has been used to create both RNA-guided repressors (20, 49, 50) and transcriptional activators (51–56, 71).

Figure 2. The CRISPR/Cas9 DNA-binding Domain.

The Cas9 protein forms a complex with the gRNA, which recognizes a specific 20 bp DNA target sequence, known as the protospacer. A short sequence directly downstream from the protospacer, the protospacer adjacent motif (PAM), is required for Cas9-mediated cleavage. The PAM sequence is highly variable between different organisms (Table 2). With only two amino acid substitutions (D10A and H840A), Cas9 endonuclease activity can been eliminated while simultaneously maintaining its RNA-guided DNA-binding activity. This deactivated Cas9 (dCas9) functions as a modular DNA-binding domain, similar to TALEs. RNA-guided transcriptional activators and repressors have been created by fusing dCas9 with different effector domains.

2. Synthetic Repressors

Transcriptional repression typically occurs through one of three main mechanisms: inhibiting formation of the pre-initiation complex, decreasing activator function, or chromatin remodeling (72). Synthetic repressors are typically engineered to block formation of the pre-initiation complex or induce the formation of heterochromatin. The simplest mechanism of transcriptional repression is physically blocking the binding of activators to the target promoter or inhibiting the progression of RNA polymerase, thus impeding either formation of the pre-initiation complex or elongation of the nascent RNA chain. Depending on the specific target, localizing a DNA-binding domain to the transcriptional start site of a gene of interest may be sufficient to physically block transcription. In general, fusing a modular DNA binding domain to a repressor domain that recruits heterochromatin-forming complexes has proven more effective for knocking down gene expression in eukaryotes. The most potent and commonly used is the Krüppel-associated box (KRAB) repressor domain that occurs naturally in many eukaryotic transcription factors (73, 74) and was shown to be effective at silencing target genes when fused to the early ZFP-based transcription factors (8, 75). KRAB facilitates heterochromatin formation by recruiting KAP1 (76), which serves as a scaffold for various heterochromatin-forming complexes including histone deacetylases and histone methyltransferases (77–79). Although KRAB robustly silences target genes when directed to the proximal promoter (6), it is not clear to what extent the heterochromatin formation spreads along the chromosome or is heritable, which may be specific to the target, cell type, and experimental conditions (80, 81).

2.1 TALE Repressors

TALE repressors have been used to successfully suppress multiple endogenous genes in human cells (82), model organisms (83), and plants (84). In addition to targeting endogenous genes, TALE repressors are valuable tools for the development of synthetic gene circuits (21, 22). Algorithms have been developed to identify potential TALE targets that have minimal homology to promoters in the genome of interest such that they can be used in orthogonal gene circuits with minimal effects on the host genome (21). Despite these notable advances, TALE-repressors have not yet been widely used, which may be in part due to their recent development (82). This may also be due to competition with other technologies, including creating knockouts via genome editing with engineered nucleases (1, 3) and gene silencing by RNA interference (RNAi). Genome editing has the advantage of permanently disabling expression of the target gene, however for some applications it may be desirable to use a reversible mechanism that targets transcription. Genome editing to create gene knockouts also typically requires the clonal derivation of the cells with the targeted modification, whereas transcriptional repression technologies should act efficiently in each cell that expresses the repressor. Although RNAi is currently the most widely used gene silencing technology, it can be challenging to find specific target sequences and RNAi can interfere with endogenous microRNA processing pathways (85). Additionally, transcription repression with KRAB domains may ultimately prove more efficient as it is only necessary to target 1–2 copies of a gene, whereas RNAi needs to prevent translation of typically hundreds or thousands of mRNA molecules.

2.2 CRISPR Interference

dCas9 and fusion proteins of dCas9 and repressor domains can also be used to direct gene silencing. S. pyogenes Cas9 is a large protein (~190 kDa), which once localized to the desired target DNA sequence, can sterically block the target promoter from transcriptional activation machinery and prevent transcript elongation in both bacteria an mammalian cells (49, 50, 86). The fusion of dCas9 with repression domains increases the level of gene knockdown in eukaryotic systems, presumably via the recruitment of heterochromatin in addition to steric hindrance of endogenous factors (49). As demonstrated by RNA-seq, CRISPR/Cas9-based repression can be highly specific with only the target gene exhibiting a significant level of knockdown (49). An active field of research is determining the optimal CRISPR targets for transcriptional repression. In bacteria, gRNAs targeted to the −10 and −35 elements in promoters achieve the most significant repression, regardless of whether the non-template or template strand is targeted (50, 86). However, when the gRNA was targeted to the coding region, greater knockdown was achieved by targeting the non-template strand (50, 86). It has been hypothesized that gRNAs targeted to the coding region of the non-template strand may bind mRNA and thus physically block translation (50). Guidelines for designing gRNAs for efficient repression in eukaryotes are not yet clear, although targeting key enhancer motifs may be helpful (49). One study of bacterial gene regulation showed that expression of two gRNAs targeted to the same coding region can lead to synergistic knockdown as long as the two target sequences are non-overlapping (50). To our knowledge, there have not been any reports of synergistic effects of multiple repressors in eukaryotic systems.

CRISPR-mediated repression is introducing new and exciting possibilities for manipulating gene expression. Although RNAi produces a similar effect, RNAi is restricted to eukaryotic systems that contain the necessary endogenous processing pathways. Therefore CRISPR has created opportunities for reversible gene silencing in prokaryotic organisms. CRISPR-based systems will also be useful for generating synthetic signaling pathways with rapid kinetics. In contrast to the slow degradation kinetics of proteins that typically determine the rate of pathway dynamics, turnover of cellular RNA is relatively fast. Therefore more complex synthetic pathways may be studied on timescales shorter than the cell cycle, which is a confounding factor in many other experiments.

3. Synthetic Transcriptional Activators

Transcription activators are responsible for inducing tissue-specific patterns of gene expression. Through binding to gene promoter and enhancer regions, transcriptional activators facilitate assembly of the pre-initiation complex by recruiting co-activators and chromatin remodeling proteins to specific genomic loci. Synthetic transcriptional activators can be designed as components of gene circuits or gene regulation systems (21, 22, 87), or to modulate endogenous gene expression. There are several potential advantages to controlling the expression of genes from their natural chromosomal context, in contrast to conventional overexpression of transgenes. For example, activating endogenous expression allows for natural mRNA processing ensuring that different splice variants are expressed at the correct ratios and trafficked properly throughout the cell and tissue (9). This approach could also be used for genetic screens (88, 89), activating silent tumor suppressors (15) or genes to compensate for genetic defects (10–14), or controlling endogenous gene networks that direct cell fate (17–20).

Synthetic transcriptional activators are constructed by fusing a programmable DNA-binding domain to a modular activation domain. There are two main classes of transcriptional activation domains, proximal activation domains and general activation domains (90). In response to remote enhancer regions, proximal activation domains stimulate transcription from positions close to the TATA box. General activation domains are more versatile and can activate transcription from enhancer and proximal promoter regions. The most commonly used activation domains VP16, VP64, and p65 are general activation domains. VP16 is a portion of a natural transcription factor originally isolated from the Herpes Simplex Virus (91). VP64 is a more potent version of VP16 constructed by fusing of four copies of VP16 in tandem (8). Due to their potency, VP16 and VP64 can induce cytotoxicity in certain cell types. Additionally, these proteins are potentially immunogenic and may not be suitable for therapeutic applications. For these reasons, researchers sometimes opt for the weaker p65 activation domain, derived from the human NF-κB RelA transcription factor (92).

3.1 TALE Transcriptional Activators

Synthetic TALE activators have been used to successfully induce expression of many endogenous genes (14, 18,19, 35, 36, 40–42, 93). TALEs are typically designed to target sites in the proximal promoter such that the activation domain will recruit the pre-initiation complex in close proximity to the transcriptional start site. In most instances targeting the proximal promoter region induces gene expression from the endogenous locus. However, for certain tightly regulated genes, targeting the proximal promoter alone induces very low expression levels that may not be sufficient for functional effects. One example is activation of the Oct4 locus. Targeting the human Oct4 proximal promoter with a single TALE activator led to modest changes in gene expression (36). A subsequent study showed greater levels of gene activation when cells were co-treated with a TALE activator and inhibitors of either DNA methyltransferases or histone deacetylases, which presumably facilitated non-specific dechromatinization of this locus and others (18). Interestingly, a more recent study showed that targeting TALE activators to the distal enhancer of Oct4 leads to demethylation of the proximal promoter and elevated levels of Oct4 expression (19). These TALE activators could substitute for Oct4 overexpression during genetic reprogramming and the epigenetic profile of these reprogrammed cells resembled the epigenetic profile of ES cells, similar to cells reprogrammed with Oct4, cMyc, Klf4, and Sox2 (19).

Although endogenous gene activation by TALEs has been reproducible across many studies, the induced expression levels of target genes are often too low to generate functional effects. Whereas these studies have typically focused on delivery of a single TALE activator, natural gene regulation often occurs by the cooperative action of many transcription factors at a single promoter. In an effort to recapitulate this complexity, it was found that targeting multiple TALE activators to a single endogenous promoter has substantial synergistic effects on gene activation (40, 41). Furthermore, the levels of gene activation could be finely tuned based on the number and identity of TALE proteins (40).

3.2 CRISPR/Cas9-Based Transcription Factors

The most common strategies for engineering transcription factors targeted to custom sequences have been based on programmable DNA-binding proteins. Although these methods have been widely successful for a variety of applications, the associated protocols for engineering new proteins for each target site are labor intensive and require screening multiple constructs. Similar to ZFP- and TALE-based activators, the CRISPR/Cas9 system has been successfully engineered to activate endogenous gene expression in both E. coli and human cells. In this approach, transcriptional activators are created by fusing dCas9 with modular activation domains, including the omega subunit of RNAP (86), p65 (49), and VP64 (49, –54, 56).

The early studies with CRISPR transcription factors suggest that many genes can be regulated by a variety of gRNAs targeted to either strand (51, 52). Other studies indicate that the most potent gRNA targets tend to be within the proximal promoter regions (56, 94). The effects of CRISPR activators targeted to enhancer regions are yet to be explored. Importantly, the CRISPR activators are able to enhance the expression of genes from promoters that are already moderately expressed, as well as induce the expression of closed promoters in heterochromatic regions (51). Interestingly, it has been reported that CRISPR transcription factor activity does not correlate with the non-homologous end joining indel mutation rates produces by the same gRNAs co-expressed with an active Cas9 nuclease (52). This suggests that there are additional factors besides binding that influence the activity of CRISPR transcription factors at specific loci. Currently, it is difficult to predict CRISPR transcription factor activity prior to testing. Larger sets of gRNAs need to be examined to determine the most effective DNA targets.

In general, the magnitude of transcriptional activation induced by CRISPR activators has been lower than gene activation induced by TALE activators directed against the same promoters (40, 41,51–53). The magnitude of CRISPR transcriptional activation can be increased by targeting multiple gRNAs to the same promoter to induce synergistic gene activation (51, 52, 56). Although off-target DNA cleavage by CRISPR nucleases has been readily-detectable (56, 95–97), RNA-seq (51) and microarray (53) analyses have demonstrated that CRISPR activators have limited off target effects in mammalian cells. This is probably because individual gRNAs induce minimal gene activation and multiple gRNAs at a single promoter are required to induce detectable levels of expression (51).

4. Considerations for Building and Using TALEs

4.1 Selection of TALE Targets

Predicting the optimal TALEs for regulating target genes is a very active area of research. Thus far, two rigid rules for the selection of potential TALE DNA targets have emerged. First, engineered TALE DNA-binding proteins require a 5’ T, referred to as the 0^th repeat, although this requirement can be circumvented by using TALE domains that have undergone directed evolution (98) or are obtained from other species (57). Second, a 3’ flanking half-repeat is required for proper DNA binding (33, 34). As these appear to be the only restrictions on binding site selection, TALEs offer a large sequence space of potential genomic targets. As described above, variation in the 12^th and 13^th amino acid position (the RVD) of each TALE repeat confers its DNA-binding specificity. The first amino acid in each RVD does not contact a DNA base but rather forms hydrogen bonds that stabilize the TALE structure (59, 60). The second residue contacts the top strand of DNA through sequence-specific hydrogen bonds and van der Waals interactions. The original RVDs with specificity for each nucleotide, based on their occurrence in naturally-occurring TALEs, were reported as NI, HD, NG, and NN for the base pairs A, C, T and G/A, respectively (33, 34). Concerns about the wobble recognition of both G and A by NN led to subsequent studies demonstrating that the NK RVD is more specific for G (93), but unfortunately NK also has a lower DNA-binding affinity (99). Later NH was found to be more specific for G than NK, while maintaining a similar DNA-binding affinity in comparison to NN (82). Regardless, most publications thus far have focused on the original set of RVDs (NI, HD, NG, and NN).

Engineered TALE DNA-binding domains are sensitive to 5-methylcytosine (5mC) (18). About 80% of CpG loci are methylated in the human genome (100). To address this limitation of TALE DNA-binding, it was discovered that either the TALE repeat N* or thymidine-targeting NG can efficiently accommodate binding to methylated cytosines (101, 102). Therefore it is advisable to avoid methylated cytosines if working with the standard set of four RVDs, but it is possible to target these regions if necessary for certain applications.

It is common to build TALE proteins ranging from 14.5 to 17.5 RVDs in length to target sequences that are statistically unique in most genomes (35, 36, 40, 41). However the optimal TALE length that achieves both a high level of binding affinity and specificity still remains poorly understood. Using a synthetic reporter system in plants, the minimal number of RVDs required to activate gene expression was 6.5 RVDs with the level of activation reaching a plateau after 10.5–12.5 RVDs (34). This suggests that maximum binding affinity is reached around 10.5 RVDs. In fact, potent transcriptional regulators have been used in mammalian cells with as few as 13.5 RVDs (22).

Based on in vitro binding assays (35, 103) and interrogation of predicted likely off-target sites, it is assumed TALEs act primarily at their intended target sequence. However, off-target effects have been reported. TALEs recognizing 18–20 bp targets can often tolerate point mutations in their binding site (21, 56, 103). However, three or more mutations in target sequences 18–20 bp in length decreases TALE activity by greater than nine-fold (21). On the other hand, shorter TALEs recognizing 10–14 bp targets are much more sensitive to single base mismatches and therefore are more specific (56, 103). Mismatches at the 5’ end of the target site have a larger disruptive effect on TALE DNA-binding affinity compared to mismatches in the 3’ end of the target site (104). Collectively, these data suggest that specificity and affinity should be carefully balanced, and users of the TALE technology should carefully scan the genome sequence for potential off-target recognition sites and select targets that have the fewest sites with significant sequence homology, particularly at the 5’ end of the target. Additionally, mutations to the constant terminal region of the TALE protein have been recently reported to decrease excess binding energy and increase specificity (103).

Additional studies are necessary to directly examine the binding specificity and affinity of designed TALE proteins and develop predictive frameworks for selecting optimal TALE target sites. In fact, ongoing bioinformatics efforts are leading to improved algorithms for both designing TALENs with high activity (105) and specificity (106). Nonetheless, we and others have already found TALEs to be a very effective platform for designing custom enzymes. In our experience, the majority of custom TALEs are active at the desired target locus. Through careful target design and screening, it is possible to identify highly specific and active TALE DNA-binding domains. Advances in high-throughput sequencing have made it possible to global gene expression profiles and analyze genome-wide protein:DNA binding patterns, which have shown remarkable specificity in early studies (107). These methods will be pivotal for determining off-target effects of synthetic DNA-binding proteins.

4.2 Construction of TALES

There are multiple platforms available for constructing TALEs. The optimal platform for a particular user will depend on the number of unique TALE proteins desired and available resources. For generating a small number of TALE proteins in a typical academic laboratory, the Golden Gate Assembly platforms are particularly straightforward and user-friendly (Figure 3) (36, 37, 108). Golden Gate cloning is a method of assembling multiple DNA fragments in a single reaction. The protocol utilizes type IIS restriction enzymes that cleave sequences adjacent to their recognition sequence. In this manner, one restriction enzyme creates multiple unique overhangs that can only ligate into the destination vector in the correct orientation. Golden Gate Assembly is a very efficient process where most colonies resulting from the ligation contain the desired insert. Golden Gate platforms are user-friendly, allow for rapid assembly of custom TALEs in under a week, and make use of commercially available reagents distributed by the Addgene non-profit plasmid repository. The kit developed by the Voytas group is specifically designed for cloning TALE nucleases anywhere between 12–31 repeats in length (37). The advantage of this platform is that it does not require a PCR step, minimizing the likelihood of obtaining point mutations. The Voytas Golden Gate TALEN and TAL effector kit 2.0 is available on Addgene (kit # 1000000024) and the associated software is available at https://tale-nt.cac.cornell.edu/. Other plasmids that are compatible with this system but contain alternative effector domains, such as activators and repressors, are also available through Addgene (https://www.addgene.org/Talen/). The kit developed by the Zhang group utilizes a PCR step and the maximum number of repeats that can be cloned is 24. The kit comes with destination vectors for cloning both TALE nucleases and transcriptional activators by fusion to the FokI catalytic domain and VP64, respectively. The TALE Toolbox from the Zhang lab is Addgene kit # 1000000019 and the associated software can be found at http://taleffectors.genome-engineering.org/tools/.

Figure 3. Golden Gate Assembly of TALEs.

Golden Gate assembly makes use of type IIS restriction enzymes, including BsaI, BsmBI, and Esp3I, that cleave outside their recognition sequence to create unique overhangs. Therefore it is possible to digest and ligate multiple inserts into a destination plasmid with a single restriction enzyme in a single reaction. In step 1, single RVDs are excised from module plasmids and ligated into the desired array plasmid (sample overhangs are shown). This platform allows for construction of up to 10 RVDs into each array plasmid. Importantly, the array plasmids confer spectinomycin resistance (Spec^R) rather than tetracycline resistance (Tet^R). This ensures that only successfully assembled array plasmids are propagated. In step 2, the array plasmids and the last repeat (LR) plasmid are assembled in a second Golden Gate reaction to obtain the final desired TALE construct. Similar to step 1, in step 2 the final backbone vector confers ampicillin resistance (Amp^R), rather than spectinomycin or tetracycline resistance, to ensure that only successfully assembled vectors are propagated. Replacement of the β-galactosidase expression cassette (LacZ) in the final step allows for blue-white screening of successful ligations. Figure adapted from (37).

For labs needing to build many TALE enzymes, there are several automatable high-throughput solid phase assembly platforms. Fast ligation-based automatable solid-phase high throughput (FLASH) assembly developed by the Joung Lab (39) and Iterative Capped Assembly (ICA) developed by the Church group (109) offer rapid and cost effective platforms for large-scale assembly of TALEs. The FLASH system contains a library of 376 plasmids that encode every permutation of one, two, three, or four TAL effector repeats. These monomers are assembled via an iterative ligation process on magnetic beads. Since each fragment can encode one to four repeats, this system allows for the creation of TALEs of any desired length. With automated liquid handlers, this procedure can produce up to 96 TALEs in less than one day, whereas a manual process using multichannel pipettes takes 1–2 days (39). Once cleaved from the magnetic bead the TALE domain is subsequently cloned into the desired expression vector. All reagents necessary for conducting FLASH can be requested directly from the Joung Lab (http://www.talengineering.org/.) Similar to FLASH, ICA allows for custom assembly of any number of RVD repeats on magnetic beads. An advantage of ICA is that is does not rely on a library of preassembled RVDs. Instead, each individual RVD is added sequentially. This design makes it simple to incorporate additional optimized RVDs as the technology advances. The ICA ligation takes place in the presence of “capping” oligonucleotides that terminate the elongation of incomplete chains thus increasing the ligation efficiency and occurrence of full-length products. In under three hours, users can construct TALEs up to 21 RVDs in length. Following synthesis, the TALE domains are PCR amplified and ligated into the desired destination vector. Finally, for even greater levels of throughput and fidelity, ligation-independent cloning (LIC) methods can be used (110). LIC relies on a library of DNA constructs encoding different combinations of TAL effectors. These long constructs are processed to produce long (10–30 bp) single stranded overhangs that anneal to the overhangs of other fragments. Without a ligation step, the annealed fragments are directly transformed into bacterial cells and the endogenous bacterial ligase repairs the nicks.

4.3 Delivery of TALEs

Delivery of TALE proteins to cells has some unique challenges. ZFPs have been delivered as proteins by fusion to cell-penetrating peptides (111), tethering to ligands of cell-surface receptors (112), and using the innate cationic properties of the protein to transverse the cell membrane (113). However, the neutral charge of TALEs suggests it will be more challenging to transport these molecules as proteins into cells (113), but it can be delivered as a fusion to cationic peptides (114). Most efforts thus far have focused on delivery of the gene or mRNA encoding the engineered TALE proteins into cells. However the larger size of these genes restricts the number of unique TALE genes that could be packaged in a single virus. The repetitive nature of TALEs has also created challenges in expressing TALEs from lentivirus (115). Although refactoring of the TALE repeat sequences have made it possible to package TALEs in lentiviral particles (116), these RVDs are not publicly available and may need to be repeated for each new TALE. Consequently most work in this area uses transient delivery of plasmid DNA or mRNA by electroporation or lipid-mediated transfection.

5. Considerations for Building and Using the CRISPR/Cas9 System

5.1 Selection of CRISPR/Cas9 Targets

Although several studies have made advances in developing methods for optimal DNA target selection, there is still much to be learned about CRISPR target site recognition and the formation of the Cas9, gRNA, and DNA target site complex. A strict rule for CRISPR/Cas9 target selection is that the 20 bp protospacer must be adjacent to a PAM sequence corresponding the particular Cas9 being used (43, 56, 71, 117). PAM recognition is an obligate first step for target recognition and off-target binding sites positively correlate with PAM density (117). Cas9-gRNA complexes collide with DNA and only upon binding to a PAM does the Cas9 interrogate the surrounding DNA for sgRNA complementarity (117). Cas9 proteins isolated from different organisms have varying 3–5 nucleotide PAM sequences (71). The most commonly used Cas9 protein was isolated from Streptococcus pyogenes and its corresponding PAM is 5’-NGG-3’.

Although most studies have focused on 20 bp protospacers, corresponding to the protospacer size in the natural CRISPR system, the effects of protospacer length and sequence composition on CRISPR activity and specificity have not been comprehensively assessed. Extending the 5’ end of the gRNA may lead to more active CRISPR/Cas9 complexes, perhaps due to an increased half-life of the gRNA (56). The specificity of gRNA targeting may be increased by extending the 5’ end of the protospacer with one or two guanine nucleotides (118) or by truncating the protospacer to 17 or 18 nucleotides to eliminate excess binding energy to the target site (119). The large sequence space of possible protospacers makes it challenging to determine the positions that are most important for DNA binding (56, 120). Studies both in bacterial (70, 86) and mammalian (45, 56, 120) systems suggest that the 10–12 bases directly adjacent to the PAM sequence (termed the seed region) are the most important for DNA-binding.

A concern for the CRISPR/Cas9 system in applications requiring very precise genomic modification is the potential for off-target effects. Studies of Cas9 nuclease activity have shown that the gRNA:Cas9 complex generally tolerates 1–3 mutations in the protospacer target sequence while still maintaining activity (56, –97, 121). However, it is relatively easy to find target sites with limited sequence homology to any other potential target site in the genome such that this potential problem can be avoided (44, 45, 118). Although the Cas9/gRNA complex may interact with many potential off-target sites in the genome, this weak interactions do not appear to be sufficient for nuclease activity at these sites (117, 122, 123). Furthermore, because it is typically necessary for multiple gRNAs to target adjacent sites to achieve robust levels of gene activation (51, 52), CRISPR/Cas9-based activators are highly specific in modulating target genes (51, 53). Our experience has been that, like TALEs, the majority of CRISPR enzymes are active in mammalian cells. CRISPR nucleases cut with efficiencies that are comparable to ZFP and TALE nucleases. Although we have found functional CRISPR activators at almost every locus tested, we and others have observed that the levels of gene activation induced by CRISPR are typically lower than levels induced by TALE activators targeting the same promoters (40, 41, 51, 52). It is yet to be determined why this is the case; the differences in how CRISPR/Cas9 and TALEs interact with their DNA target sites may play a significant role.

The CRISPR design software developed by the Zhang lab is particularly useful for selecting gRNA targets (www.CRISPR.MIT.edu). This software allows the user to input the DNA sequence of the target region and returns a list of potential protospacers scored based on the number of potential off target binding sites found within the genome of interest. Other web servers for gRNA design with various properties have also been reported (124–126). In general, there is still much to be learned about the interactions between CRISPR/Cas9 and its genomic DNA target site. While some trends have emerged, the field is new and the characterization is less substantial than has been completed for the ZFP and TALE technologies. Until more specific guidelines for choosing optimal target sites emerge, it is advisable to screen multiple gRNAs. For designing CRISPR-based repressors, we recommend designing gRNAs targeted to the template and non-template strands along both the promoter and the gene’s coding region. Both dCas9-KRAB and dCas9 can be tested depending on whether the goal is to epigenetically silence the gene or simply block activators or RNA polymerase.

5.2 Construction and Delivery of CRISPR/Cas9-Based Transcription Factors

Most CRISPR/Cas9 expression plasmids are publicly available through Addgene (http://www.addgene.org/CRISPR/) (Table 1). There are many variations available and it is important to obtain the vectors that are well suited for a specific application. As mentioned before, the Cas9 protein from S. pyogenes is currently the most widely used and there are a variety of S. pyogenes dCas9 plasmids available through Addgene. However other variants such as S. thermophilus dCas9 and N. meningitidis dCas9 are also available (71, 127) (Table 2). Importantly, each dCas9 variant must be paired with its unique corresponding tracrRNA/crRNA or gRNA expression plasmid. The tracrRNAs, crRNAs, and chimeric gRNAs are typically expressed from DNA expression cassettes via RNA polymerase III promoters, most commonly the human U6 promoter, although they can also be transcribed in vitro and transfected to cells as RNA (47). Several studies have found delivery of chimeric gRNAs along with dCas9 to be more effective than co-delivery of the tracrRNA, crRNA, and dCas9 (51). Thus many efforts are now exclusively focused on using gRNAs to direct dCas9 targeting. Furthermore a recent study showed improved transcriptional repression using a modified gRNA architecture (128). The gRNA expression cassettes can be generated by custom synthesis of each complete gRNA to be ligated into the desired expression vector (44). An alternative simple and cost effective method of cloning gRNAs is to use gRNA expression plasmids that facilitate direct cloning of new protospacers (44, 45). In this system, two short oligonucleotides corresponding to the protospacer sequence are annealed and directly ligated into BbsI restriction sites (Figure 4). The final plasmid expresses the complete gRNA from the hU6 promoter. Plasmids using this strategy are currently available through Addgene (Table 1). This gRNA cloning method can easily be adapted for other CRISPR systems, such as those from S. thermophilus and N. meningitidis gRNAs (Table 2) (71). To date, most studies deliver dCas9 and gRNAs transiently using conventional transfection methods. Both dCas9 and the gRNAs have also been expressed from lentiviral vectors for gene regulation (50).

Table 1.

Examples of CRISPR/Cas9-Related Plasmids Publicly Available Through Addgene

Transgene	Vector Description	Depositing Group	Addgene Plasmid Number
S. pyogenes Cas9	Active Cas9, human codon optimized	Church	48668
S. pyogenes Cas9	dCas9	Gersbach	47107
S. pyogenes Cas9	dCas9-VP64	Gersbach	47106
S. pyogenes Cas9	dCas9-KRAB	Qi-Weissman	46911
S. pyogenes chimeric gRNA	hU6-gRNA expression vector Allows for cloning of protospacers between two BbsI sites	Gersbach	47108
S. thermophilus Cas9	Active Cas9, human codon optimized	Church	48669
S. thermophilus Cas9 and tracrRNA	dCas9	Church	48659
S. thermophilus Cas9	dCas9-VP64, human codon optimized	Church	48675
N. meningitidis Cas9 and tracrRNA	Active Cas9	Church	48646
N. meningitidis Cas9	dCas9-VP64, human codon optimized	Church	48676

Table 2.

gRNA and PAM Sequences Corresponding to Various Cas9 Proteins

Cas9	Chimeric gRNA	PAM
S. pyogenes	(N)20TGTTTTAGAGCTAGAAATAGCAAGTTAAAATAAGGC TAGTCCGTTATCAACTTGAAAAAGTGGCACCGAGTC GGTGCTTTTTT	NGG
S. thermophilus	(N)20GTTTTTGTACTCTCAGAAATGCAGAAGCTACAAAGAT AAGGCTTCATGCCGAAATCAACACCCTGTCATTTTAT GGCAGGGTGTTTTTTT	NNAGAA NNAGGA NNGGAA
N. meningitidis	(N)20GTTGTAGCTCCCTTTCTCGAAAGAGAACCGTTGCTACA ATAAGGCCGTCTGAAAAGATGTGCCGCAACGCTCTGCC CCTTAAAGCTTCTGCTTTAACGGGCTTTTTTT	NNNNGANN NNNNGTTN NNNNGNNT

Figure 4. Custom gRNA Cloning.

The most common gRNA cloning methods make use of the BbsI type IIS restriction enzyme that cleaves outside its recognition sequence to create unique overhangs. Single stranded oligonucleotides containing each protospacer are annealed to create overhangs that are compatible with the BbsI sites in the destination vector. Upon ligation, the protospacer is inserted directly following the human U6 promoter and in front of the remainder of the chimeric gRNA sequence.

6. Future Directions

TALE- and CRISPR/Cas9-based transcription factors have broad potential in both medicine and research. Although technologies based on engineered ZFPs are further along in clinical development, including clinical trials led by Sangamo BioSciences currently underway for HIV therapies (129) and previously for diabetic neuropathy (130, 131), TALE and CRISPR/Cas9 technologies are already on a path to preclinical development and will capitalize on the knowledge gained from almost 20 years of ZFP research. However both the TALE and CRISPR/Cas9 technologies are still new and additional work is necessary to fully characterize their potential and limitations. For example, studies are needed to develop effective means of delivering large TALE and Cas9 proteins, determine the immunogenicity of these proteins, and further characterize DNA-binding specificity in the context of large complex genomes. Both TALEs and the S. pyogenes Cas9 are large proteins that will be difficult to efficiently package with the necessary regulatory elements into size-restricted vectors, such as the conventional AAV that has had very promising results in other gene therapy clinical trials. The tendency for genetic recombination of the highly repetitive TALE sequences is also a concern (115). Currently, a primary concern of the both systems is specificity. However, in the context of gene regulation, specificity may not be a major problem given that multiple activators must work together to generate significant changes in expression, and genome-wide expression analysis has demonstrated exceptionally specific effects using this approach (49, 51). The CRISPR/Cas9 technology is in its infancy, and modified gRNA and Cas9 architectures are likely to be developed in the near future to address some of these concerns. Next-generation Cas9/gRNA pairs that are smaller and more specific will likely facilitate the development of human therapies (71, 128, 132).

Currently, the CRISPR/Cas9 technology is a powerful tool for genomics and genetics. The simplicity of gRNA design enables previously intractable methods for high-throughput genomic research. Using conventional DNA synthesis, libraries of gRNAs targeting every gene in any sequence genome can be inexpensively and efficiently produced (133, 134). In addition to the activation and repression domains described here, effectors that control epigenetic modifications will also be useful in interrogating genome architecture and pathways of endogenous gene regulation (55, 107, 135). As endogenous gene regulation is a delicate balance between multiple enzymes, multiplexing TALE and Cas9 systems with different functionalities will allow for examining the complex interplay among different regulatory signals. Aptamer-modified gRNAs (56) and orthogonal Cas9s (71, 136) will enable multiplex gene regulation using single set of gRNAs. Finally, methods for achieving inducible control of these proteins, either by chemical (137, 138) or optogenetic (55, 139) regulation, will facilitate investing the role of dynamics of gene regulation in both time and space.

Although synthetic enzymes based on ZFPs and TALEs have stimulated the fields of gene therapy, synthetic biology, and fundamental genomics, the future of genomic exploration has been simplified and spread to the general biomedical research community by the CRISPR/Cas9 technology. Applying these tools to reverse engineering of the genome is proving to be highly valuable for understanding the complexity of gene regulatory pathways. Continuing to develop these tools for modulating gene expression is necessary to meet the needs of the next era of high throughput biomedical research.