Harnessing "A Billion Years of Experimentation": The Ongoing Exploration and Exploitation of CRISPR-Cas Immune Systems.

The famed physicist-turned-biologist, Max Delbrück, once remarked that, for physicists, "the field of bacterial viruses is a fine playground for serious children who ask ambitious questions." Early discoveries in that playground helped establish molecular genetics, and half a century later, biologists delving into the same field have ushered in the era of precision genome engineering. The focus has of course shifted-from bacterial viruses and their mechanisms of infection to the bacterial hosts and their mechanisms of immunity-but it is the very same evolutionary arms race that continues to awe and inspire researchers worldwide. In this review, we explore the remarkable diversity of CRISPR-Cas adaptive immune systems, describe the molecular components that mediate nucleic acid targeting, and outline the use of these RNA-guided machines for biotechnology applications. CRISPR-Cas research has yielded far more than just Cas9-based genome-editing tools, and the wide-reaching, innovative impacts of this fascinating biological playground are sure to be felt for years to come.

Bacterial viruses, or bacteriophages (phages for short), have been the focus of scientific exploration and inspiration for just over a century, beginning with Twort and d'Herelle's landmark phage discoveries published during World War I.[1] In the ensuing years, researchers pursued the development of phages as a new way to treat pathogenic bacterial infections because of their remarkable ability to selectively kill bacterial hosts, but phage therapy quickly took a backseat to the rise of cheap and effective antibiotic treatments.[2] Yet phages would eventually transform biology and medicine in a far more momentous way: by serving as incredibly versatile model systems that, together with their bacterial hosts, shed light on some of biology's most fundamental questions. Beyond providing direct evidence that DNA is the genetic material (Hershey and Chase's famous “Waring blender experiment”), phage research revealed the fine structure of genes and the processes of genetic recombination and transduction, helped prove the triplet nature of the genetic code, revealed fundamental mechanisms of gene regulation, and catalyzed the rise of the biotechnology industry through the discovery of phage- and host-encoded DNA-manipulating enzymes, among many other breakthroughs.[3,4]

Beyond their utility for laboratory experiments, phages are the most abundant biological entity in the biosphere.[5] With an estimated 1031 total phage particles, one can calculate that there are ∼1 trillion phages for every grain of sand on Earth.[3] They are also expected to outnumber bacteria on the order of 10-to-1 and can be isolated from every environment in which bacteria exist. In the oceans, phages play major roles in carbon and nitrogen cycling by killing somewhere between 20% and 40% of all bacteria every day,[6] highlighting the immense ecological and environmental pressures they exert. On top of that, phages are known to contribute to bacterial pathogenicity and evolution by functioning as efficient gene transfer vectors, alongside plasmids, transposons, and other mobile genetic elements. From this incessant genetic battle (Fig. 1A), an arms race ensued, in which prokaryotes and phages evolved numerous elaborate and orthogonal strategies to protect themselves from viral infection and to combat antiviral mechanisms, respectively. The impressive extent of this genetic pressure is clearly visible in prokaryotic genomes, of which a substantial fraction (up to 10%) is dedicated to their own defense.[7]

(A) Transmission electron micrograph of bacteriophages infecting Escherichia coli. The evolutionary arms race between viruses and their microbial hosts has resulted in the adaptation of numerous defense mechanisms, including CRISPR–Cas systems, as well as counter-attack strategies. ©Graham Beards (https://upload.wikimedia.org/wikipedia/commons/5/52/Phage.jpg). (B) Schematic overview of the three stages of CRISPR–Cas immunity. During spacer acquisition, protospacers are excised from foreign nucleic acids and integrated at the leader end of the growing CRISPR array. crRNA biogenesis involves transcription of the entire array, followed by enzymatic processing of the pre-crRNA into mature crRNAs, often by dedicated Cas proteins. In the interference stage, the crRNA and Cas protein(s) form an effector complex that targets complementary nucleic acids for degradation. Class 1 systems rely on multiple Cas proteins during interference (depicted), whereas Class 2 systems require only a single protein effector.

The resulting range of bacterial and archaeal defense mechanisms (reviewed in Refs.[7,8]) can be broadly classified into three functionalities: (1) impeding virus adsorption to the host cell, primarily by alteration of host receptors, (2) dormancy induction and/or programmed cell death, which protect the population, and (3) nucleic acid-targeting immune systems. This last category can be further divided into innate immunity—including the well-characterized restriction-modification systems[9] and the recently discovered system involving prokaryotic Argonaute proteins[10]—and the adaptive (acquired) immunity provided by CRISPR[*]–Cas systems.

It is remarkable that, although most archaea and almost half of sequenced bacterial species contain CRISPR–Cas systems,[11,12] detection of these pervasive modules eluded researchers for decades. CRISPR–Cas research is now advancing at an incredible pace, leading to thousands of CRISPR-related articles being published in 2017 alone. Within the past decade, CRISPR–Cas biology has been in the scientific limelight time and time again, and “CRISPR” has all but become a household term due to its development for revolutionary genome engineering applications.[13,14] (As a testament to its growing popularity, CRISPR is already at risk of losing its privileged acronym status in the popular media, much like laser and radar.) The focus has of course largely been on Cas9-containing immune systems, but recent studies have highlighted the incredible diversity of CRISPR–Cas systems, both in terms of their distinguishing mechanistic features and their potential for novel biotechnological uses.

The scope and pace of contemporary CRISPR research are astounding, and it is impossible to adequately review the field in a single article. Nevertheless, we attempt to broadly summarize the current understanding of CRISPR–Cas biology while focusing on mechanisms of RNA-guided nucleic acid targeting and the development of CRISPR-based tools. Wherever possible, we include more focused review articles for readers interested in particular topics, and we offer a sincere apology to the many colleagues whose work we could not mention due to space constraints.

General Mechanism of CRISPR–Cas Immunity

CRISPR refers to specialized genomic regions found in bacteria and archaea that consist of alternating repeat-spacer units, often neighboring conserved sets of CRISPR-associated (cas) genes.[15] Observations in 2005 that CRISPRs harbor some spacers matching extrachromosomal elements, namely viruses and plasmids, led to the first hypotheses of an immune system function.[16-18] Barrangou et al. were the first to provide experimental evidence for CRISPR–Cas-mediated antiviral immunity,[19] and subsequent work highlighted the essential role of noncoding CRISPR RNAs (crRNAs).[20]

Research in the ensuing years has greatly expanded our understanding of the molecular mechanisms of CRISPR–Cas adaptive immunity (reviewed in Refs.[21-23]), which can be broadly grouped into three distinct stages: adaptation, crRNA biogenesis, and interference (Fig. 1B). During adaptation,[24] also known as spacer acquisition, foreign DNA segments called protospacers are processed and integrated into one end of the growing CRISPR array, accompanied by duplication of the terminal repeat. CRISPR arrays are flanked by a leader sequence, which both helps define the site of spacer integration and also serves as a promoter for the transcription of precursor crRNAs that span the length of the entire array. During crRNA biogenesis,[25] these pre-crRNAs are enzymatically processed into libraries of short mature crRNAs, each of which contains a single spacer (or guide) sequence flanked by fragments of the repeat sequence. Mature crRNAs are then bound by one or more Cas proteins to form large ribonucleoprotein (RNP) effector complexes, which target complementary nucleic acids for degradation during the interference stage.

Although most well-studied CRISPR–Cas systems obey this general paradigm, as with much else in biology, exceptions are the norm. For example, some systems encode reverse transcriptase fusions that promote acquisition directly from RNA precursors,[26] diverse effector complexes can cleave DNA, RNA, or both DNA and RNA targets during interference,[27] and some effectors not only cleave foreign nucleic acids identified through complementary base-pairing, but also unleash potent, nonspecific “collateral” cleavage upon target binding.[28-31] And beyond mediating prokaryotic antiviral immunity, CRISPR–Cas systems have been co-opted by bacterial hosts to promote pathogenesis through endogenous gene regulation,[32,33] by selfish genetic elements to possibly facilitate transposition,[34] and by phages themselves to evade host innate immunity.[35]

CRISPR–Cas Diversity and Classification

Most steps during adaptation and interference are carried out by dedicated Cas proteins expressed from cas genes that neighbor the CRISPR array and typically number on the order of 4–8. Efforts to define a general mechanism for adaptive immunity were initially hampered by the fact that, although certain cas gene cassettes exhibit highly conserved arrangements, cassettes can differ almost entirely when comparing CRISPR–Cas systems between organisms or even within single organisms. To better understand this diversity, Makarova et al. introduced a polythetic classification scheme in 2011 that combined phylogenetic, structural, and comparative genomic analyses to arrive at three major system types (I, II, and III), each of which exhibits a signature gene, cas3, cas9, and cas10, respectively.[36] The contemporary classification defines two broad classes of CRISPR–Cas systems and six major types (I–VI),[11,37] which are presently delineated into roughly 30 subtypes in total.[38]

Class 1 and Class 2 systems are differentiated based on the nature of the protein–RNA effector complex involved in nucleic acid targeting. Class 1 effector complexes[39] contain crRNA and multiple protein subunits, typically encoded by three to six cas genes, whereas Class 2 effector complexes[40] comprise a single Cas protein together with crRNA, sometimes accompanied by an additional trans-activating crRNA molecule known as tracrRNA. Both classes are further divided into three different types: types I, III, and IV (Class 1) and types II, V, and VI (Class 2).

Because Class 2 systems exhibit simpler cas gene cassettes and a more streamlined effector complex architecture than Class 1 systems, they have been more rapidly adopted for biotechnology applications involving heterologous expression or delivery. Nevertheless, Class 2 systems are far less abundant in nature, comprising roughly 10% of CRISPR–Cas loci in sequenced bacterial genomes and being almost completely absent in archaea.[11,41] In contrast, Class 1 systems are widespread, with type I alone making up ∼50% of the loci in both bacteria and archaea.

Based on both the relative prevalence of Class 1 and Class 2 systems as well as phylogenetic analyses of the cas1 gene, which is present in all autonomous CRISPR–Cas systems, the current scenario for the evolution of prokaryotic adaptive immune systems posits that the ancestral CRISPR–Cas system was of the Class 1 type.[37] After insertion of a Cas1-encoding transposon (casposon)[42] next to a primitive innate immune system, further duplication of ancient RNA recognition motif domains led to ancestral type I and III systems containing multisubunit targeting complexes. Type II, V, and VI systems then evolved independently, largely through the co-option of genes and enzymatic domains from transposons.[37] The observation that Class 2 systems appear to derive almost completely from different mobile genetic elements beautifully highlights the extent to which enzymes have been co-opted as a means of both offense and defense within the evolutionary “guns for hire” paradigm.[43]

It is worth noting that the precise annotation of CRISPR–Cas subtypes continues to evolve as new cas genes are discovered and validated. New discoveries will undoubtedly follow from the availability of novel genomic or metagenomic sequences,[41] as well as from new bioinformatics strategies to search for CRISPR-associated or CRISPR-linked genes.[37,44,45]

Mechanisms of RNA-Guided Nucleic Acid Targeting

In the following sections, we review our current understanding of nucleic acid-targeting mechanisms across different CRISPR–Cas immune systems. Given most readers' familiarity with Cas9, we begin by discussing type II systems, followed by type I systems wherein many hallmark features of RNA-guided targeting were first described. After reviewing the remaining four types, we highlight the various ways in which Cas enzymes have been harnessed for tool development.

Type II – Cas9: Facile double-strand breaks

Type II systems feature the well-known Cas9 endonuclease,[46] a protein that formerly went by the monikers Cas5 and Csn1.[19,36,47,48] Guided by a dual-RNA substrate comprising crRNA and tracrRNA,[49] Cas9 introduces double-strand breaks (DSBs) in DNA targets that contain a flanking protospacer-adjacent motif (PAM)[50] and harbor complementarity to the ∼20-nucleotide crRNA guide sequence[49,51] (Fig. 2). The PAM requirement ensures that Cas9 avoids cleaving complementary targets within the CRISPR array itself, because these sequences lack flanking PAMs. Interestingly, Cas9-mediated PAM recognition is also critical for the acquisition of new spacer sequences.[52,53] After R-loop formation, cleavage of the target strand (base-paired to the crRNA) and nontarget strand (displaced within the R-loop complex) is mediated by HNH and RuvC nuclease domains of Cas9, respectively.[49,51]

A schematic of RNA-guided nucleic acid-targeting mechanism and effector complex structure for each CRISPR–Cas type. In the schematic, crRNAs are shown 5′-to-3′ (left-to-right), and large and small red arrows indicate specific and nonspecific cleavage sites, respectively. For type III, the direction of transcription is indicated with a black arrow. In the structures, nuclease active sites are indicated with opaque circles, target nucleic acids are omitted, and crRNAs are highlighted in red for clarity. PDB accession numbers by type: 4QYZ and 4QQW (I), 4OO8 (II), 3X1L/4W8Y and 5FSH (III), 5B43 (V), and 5XWP (VI).

The apo form of Streptococcus pyogenes Cas9 (SpyCas9) is conformationally flexible, and guide RNA (gRNA) binding drives large-scale conformational rearrangements that lock Cas9 into a bilobed architecture.[54-56] Once complexed with gRNA—either a natural crRNA-tracrRNA dual-RNA guide or an engineered single-RNA guide[49]—the Cas9–gRNA RNP complex interrogates double-stranded DNA for potential target sites matching the guide sequence. Numerous experimental studies point to a model for target search, in which PAM recognition acts as an obligate first step during protospacer recognition and leads directly to local DNA unwinding.[57-60] The search is largely three-dimensional,[57,61,62] and sampling times at off-target sites are a function of the degree of gRNA-DNA complementarity.[63-65] Targeting by Cas9–gRNA is most sensitive to mismatches in the PAM-proximal region of the DNA target site, a phenomenon that results at least, in part, from preordering of the gRNA “seed” sequence.[55,66]

Numerous in vitro and in vivo experiments have revealed that DNA binding is significantly more promiscuous than cleavage, and recent biophysical studies have highlighted the conformational rearrangements that are required for driving DNA-bound SpyCas9 into a catalytically active state.[67-70] Target-strand cleavage by the HNH domain precedes nontarget-strand cleavage by the RuvC domain,[71,72] and mutations that modulate either DNA interactions or Cas9 conformational dynamics can significantly enhance the fidelity of RNA-guided DNA cleavage.[69,73-76]

Two other observations regarding DNA cleavage deserve emphasis. First, although Cas9 is often described as an enzyme that generates blunt DSBs, biochemical experiments with SpyCas9 clearly demonstrate that the RuvC nuclease domain catalyzes additional “trimming” of the nontarget strand in the 3′→5′ direction, upstream of the cleavage site.[49,77,78] The production of asymmetric, recessed ends may explain, in part, the observation that DNA repair outcomes of cellular Cas9-mediated cleavage often involved templated insertions.[79,80] Second, Cas9 lacks multiple-turnover catalytic activity in biochemical cleavage assays, due to its high affinity for DNA cleavage products[57,71,72,81] and may persist at cleaved sites in cells as well; whether and how this impacts both adaptive immunity in bacteria and genome-editing outcomes in eukaryotic cells has not been systematically explored, although kinetic analyses indicate that in vivo repair of DSBs occurs more rapidly than the long Cas9–DSB half-life measured in vitro.[82]

In addition to DNA, Cas9 can also be naturally or artificially programmed to bind and cleave RNA targets. Whereas SpyCas9 has been specifically manipulated to enable RNA targeting in vitro and in vivo,[83-85] other Cas9 orthologs possess the intrinsic capability to target single-stranded RNA molecules,[86-88] but the biological import of this activity is unclear. In at least one case, though, Cas9 from Francisella novicida has been shown to directly downregulate an endogenous transcript using tracrRNA and an alternative CRISPR-associated RNA, as a mechanism for promoting pathogenesis during host infection.[32]

Although type II CRISPR–Cas systems provide robust protection against viral infection, phages can escape interference using multiple evasion strategies. Spontaneous mutations in the PAM or protospacer preclude efficient targeting,[89] and DNA modifications such as glucosylation can impair DNA recognition and/or lead to enhanced mutation frequencies that enable escape.[90,91] Multiple phage-encoded inhibitors of Cas9 (“anti-CRISPRs,” reviewed in Ref.[92]) have also been discovered, which arrest nucleic acid targeting at various stages either by preventing DNA binding or by blocking the conformational rearrangements required for DNA cleavage.

Type I – Cascade/Cas3: DNA cut and run

In contrast to single-effector type II systems, type I CRISPR–Cas systems—the most abundant in sequenced genomes—encode multisubunit effector complexes[93] (Fig. 2). Many mechanistic discoveries on interference, such as the existence of a seed sequence,[94,95] the nature of R-loop formation,[96] and inhibition by anti-CRISPRs,[97] were first described for type I. An incredible variety of type I systems is known, spanning at least six different subtypes (A–F), all of which utilize a DNA binding complex known as Cascade and harbor the signature protein, Cas3.[38]

The effector complex in type I systems requires a mature crRNA that is formed upon pre-crRNA processing. A designated ribonuclease (usually Cas6) cleaves a specific stem-loop structure formed by portions of the repeat sequences,[98,99] resulting in a spacer sequence flanked by two repeat fragments. The 5′ end of the crRNA lies at the “base” of Cascade, where it binds Cas5 and neighbors the large subunit (Cas8) that is responsible for PAM recognition.[100,101] The crRNA spacer sequence is spanned by multiple copies of Cas7 that define the helical “backbone” of the overall architecture, whereas small subunit proteins (Cas11, or a Cas8-fusion thereof) constitute the “belly” and help stabilize the crRNA–DNA heteroduplex upon target binding. In some systems, Cas6 remains bound to the 3′ end of the crRNA after processing and caps the “top” of the RNP complex.[20] Beyond the general overview presented here, multiple groups have published high-resolution X-ray crystal and cryoelectron microscopy structures of Cascade in different functional states,[101-107] providing a wealth of information about complex assembly, gRNA presentation, target DNA binding, Cas3 nuclease recruitment, and subtype-specific variations.

DNA targeting depends on PAM recognition to avoid self-targeting of the host CRISPR array, although PAM specificity is far more relaxed in type I systems than type II.[108] As with Cas9, PAM binding leads to local dsDNA melting and directional unwinding of the DNA upon formation of the growing crRNA–DNA heteroduplex,[58,107,109,110] with complementarity being most critical within the seed sequence. Interestingly, because type I crRNAs contain repeat fragments flanking both ends of the spacer, Cascade is sterically prevented from forming a continuous RNA–DNA double helix upon target binding. This results in a characteristic protein-induced kink at every sixth base-pair.[102] The impact of this organization on target specificity throughout the target and PAM sequence has been revealed in a number of high-throughput experiments.[108,111-113]

Unlike most other CRISPR–Cas types, in which the effector complex possesses intrinsic nuclease activity, Cascade generally functions as an RNA-guided DNA binding complex only; PAM and protospacer recognition lead to downstream, Cas8-mediated recruitment of Cas3.[104,114] Cas3 is essential for type I immunity and harbors both helicase activity and a nucleolytic HD domain.[20,115,116] Recruitment to the target leads to initial nicking of the nontarget strand, followed by extensive DNA degradation outside of the target region through ATP-dependent DNA unwinding and concomitant ssDNA nuclease activity[117-119] (Fig. 2). Recent single-molecule experiments have demonstrated that, in contrast to earlier models, Cascade remains tightly associated with Cas3 during DNA unwinding and translocation, leading to a looped intermediate.[109,120,121] In addition, type I interference creates a positive feedback loop that facilitates the acquisition of new spacers against the same invader, a process known as priming, by recruiting the adaptation machinery and providing degradation products that serve as spacer precursors.[24,120,122]

Type I systems may be the most diverse CRISPR–Cas grouping, and one generalized description does not adequately encompass their multiplicity. Apart from having a characteristic large subunit protein, each subtype possesses additional particularities, including the presence of a split Cas3 organization (I-A); the functional substitution of Cas6 with Cas5 or the presence of their fusion (I-C, I-U); the functional replacement of Cas8 with Cas5 (I-F variant); and fusions between Cas2 and Cas3 (I-F), and between Cas8 and Cas3 (some I-E systems). The level of variation among these complexes highlights the plasticity of CRISPR–Cas immune systems and demonstrates the many possibilities that nature has explored over evolutionary time.

Type III – Csm/Cmr: Targeting by triple threat

Nucleic acid targeting within type III systems (reviewed in Ref.[27]) is arguably the most complicated among CRISPR–Cas systems, involving three distinct nuclease modalities (Fig. 2). The interference mechanism appeared inconsistent for years, with in vivo experiments on a type III-A system reporting DNA targeting[123] whereas in vitro experiments on a type III-B system demonstrated RNA targeting.[124] Only recently has this conundrum been resolved[125,126] and further extended by remarkable reports on an additional nuclease activity activated by the production of second messengers.[29,30]

Type III systems are divided into four subtypes (A–D), and the multisubunit effectors are known as Csm (III-A, III-D) or Cmr (III-B, III-C) complexes. The type III effector complex architecture shows striking similarities to the type I effectors. Both systems require a mature crRNA that is processed by Cas6[127]; however, type III systems require further processing of the 3′ end by non-Cas nucleases[128] and as such do not maintain Cas6 as a subunit of the mature complex. The “base” of the complex is again composed of a Cas5 family member that accommodates the 5′ end of the crRNA and a characteristic large subunit, which in this case is Cas10. The helical “backbone” filament is likewise composed of multiple Cas7 family proteins that span the spacer sequence but also cap the 3′ end of the crRNA. The “belly” of the complex is composed of multiple small subunit proteins that belong to the Cas11 family.[129-131]

Type III complexes function foremost as RNA-guided RNA-targeting effectors by identifying single-stranded target RNAs through base-pairing with the crRNA. However, the ensuing RNA duplex is systematically disrupted by a conserved β-hairpin present in Cas7, resulting in the outward flipping of a nucleotide at 6-nt intervals.[131,132] These specific nucleotides are thereby positioned for cleavage, which is facilitated by a conserved aspartate residue present in Cas7.[131,132] Rigid positioning of the crRNA handle within the effector complex orchestrates cleavage at fixed distances from the 5′ end, resulting in a characteristic, ruler-like degradation pattern.[132-135]

Elegant experiments revealed that target RNA binding (but not necessarily cleavage) activates orthogonal deoxyribonuclease activity of the HD domain in Cas10, which nonspecifically cleaves ssDNA.[123,136-138] Targeting does not depend on the presence of a PAM, as previously described for types I and II, but rather on the absence of complementarity between the crRNA 5′ handle and the sequence flanking the RNA target.[139] Although Cas10 can cleave ssDNA within mismatched bubble regions in vitro,[136] the primary substrate in vivo is likely to be the nontemplate strand of DNA within transcription bubbles, where the effector complex is spatially restricted during targeting in cis of the nascent transcript. This DNase activity is also temporally controlled, because rapid cleavage of target RNA returns Cas10 to an inactive state.[138]

Two recent studies[29,30] uncovered perhaps the most remarkable twist on type III interference. Beyond triggering DNase activity of Cas10, target RNA binding by the Csm complex also induces Cas10-mediated synthesis of cyclic oligoadenylates, an activity catalyzed by the Palm domain. These oligoadenylates function as signaling molecules that bind the CARF domains of homodimeric Csm6 or Csx1, Cas proteins often encoded within type III CRISPR–Cas loci but that do not associate with the effector complex.[140,141] Binding to the CARF domains allosterically activates the nonspecific RNA degradation activity of the HEPN ribonuclease domains also present in Csm6/Csx1. In addition to revealing an ingenious adaptation that allows type III systems to target foreign nucleic acids on three fronts, these studies also discovered the first examples of oligoadenylate molecules functioning as second messengers in prokaryotes, and reveal a fascinating parallel with eukaryotic innate immune systems that similarly synthesize oligoadenylate second messengers in response to viral RNA detection.[142]

Apart from employing an impressive collection of weapons for stand-alone defense, type III immune systems have been shown to target escape mutants from other CRISPR systems.[143] In addition, they differentiate between lysogenic and lytic infections,[144,145] thereby allowing potentially beneficial traits of lysogenic infection to be enjoyed by the host. Collectively, these observations underscore the marvelous polishing effect of natural selection on the composition and function of CRISPR–Cas systems.

Type V – Cas12: Staggering cuts

Type V systems were first assigned as a putative grouping by Makarova et al. in 2015,[11] based on the signature gene cpf1 that had been detected in several prokaryotic genomes adjacent to adaptation genes and a CRISPR array.[146,147] Experiments published soon thereafter by Zetsche et al. firmly substantiated this new assignment within Class 2 systems, demonstrating that, like Cas9, Cpf1—now known as Cas12a—functions as a single-effector, RNA-guided endonuclease that catalyzes double-stranded DNA cleavage[148] (Fig. 2).

Yet, there are critical mechanistic features that distinguish Cas9 and Cas12a.[148] Rather than utilizing a dual-RNA guide, Cas12a, like Cascade and Csm/Cmr complexes, naturally functions with just crRNA and no tracrRNA. DNA targets are recognized in a reversed orientation as compared with Cas9, with a T-rich PAM located upstream (5′) of the target region and the seed sequence located at the 5′ end of the crRNA guide region. Instead of cleaving target DNA toward the PAM-proximal end, at similar positions on both strands, Cas12a introduces staggered cuts toward the PAM-distal end, leaving 5′ overhangs. In addition to possessing the associated deoxyribonuclease cleavage activity for target DNA cleavage, Cas12a also contains a ribonuclease domain that is responsible for enzymatic pre-crRNA processing.[149]

Cas12a possesses a RuvC-like nuclease domain that is homologous to the respective Cas9 nuclease domain; however, it lacks the HNH nuclease domain. Early structural work identified a novel domain sandwiched between discontinuous RuvC motifs, which was assigned putative nuclease function, and initial biochemical experiments suggested that the novel nuclease domain and RuvC domains were responsible for cleaving the target and nontarget strands, respectively.[150] More recently, additional structures have provided tantalizing evidence that both strands within the DNA-bound R-loop are sequentially threaded through and cleaved by the very same RuvC active site,[151,152] a conclusion that is well supported by careful mutational studies.[153] Perhaps most intriguing, target DNA binding also activates potent, single-stranded DNase activity,[31] a behavior that is reminiscent of the collateral damage effect first observed with Cas13a[28] (see Type VI). Whether target-activated ssDNA shredding is biologically meaningful during an adaptive immune response, such as to degrade ssDNA phages or to target exposed ssDNA regions during replication or transcription, is not known.

Type V systems recently expanded to include subtypes A–E, which encode Cas12a–e, previously referred to in the literature as Cpf1 (A), C2c1 (B), C2c3 (C), CasY (D), and CasX (E).[37,38,41] All Cas12 homologs share a RuvC nuclease domain (but otherwise show low sequence similarity), and, with the exception of Cas12c, have been shown to mediate biochemical DNA cleavage and/or DNA interference activity in Escherichia coli. Lastly, the Type V group also includes uncharacterized loci (subtype V-U), which encode Class 2 candidate proteins that are substantially smaller than other Cas12 family members.[154] These putative RNA-guided effectors possess RuvC-like nuclease domains and are predicted to be active, based on the observation that they flank CRISPR arrays that often contain spacers matching phage genomes. It will be interesting to learn whether they provide the same degree of adaptive immunity as their larger counterparts, and how the molecular mechanism of DNA targeting is similar or different.

Type VI – Cas13: Collateral cleavage

Before 2016, Csm/Cmr complexes (type III) were the only crRNA-guided effectors known to naturally target RNA during an immune response. Then, Abudayyeh et al. demonstrated that C2c2—now known as Cas13a—functions as a crRNA-guided RNA-targeting nuclease,[28] expanding single-effector Class 2 systems to encompass a new type VI grouping. There are presently four distinct subtypes within type VI (A–D) that encode Cas13a–d.[155,156]

Cas13 family members possess two HEPN domains with predicted single-stranded RNase activity, and experiments in which Type VI systems were heterologously expressed in E. coli demonstrated that these systems provide resistance against RNA phages. Instead of cleaving within the targeted sequence itself, Cas13 becomes activated upon target RNA binding, unleashing multiple-turnover nonspecific RNA cleavage activity in trans (Fig. 2). Targeting is stimulated by the presence of a specific protospacer flanking sequence,[28,157] although not in all scenarios,[158,159] and collateral, nonspecific RNA cleavage is both sensitive to secondary structure and has specific nucleotide preferences, depending on the Cas13 homolog.[160]

A set of recent structural studies have provided elegant insights into the mechanisms of RNA targeting and RNA cleavage for Cas13a.[161-163] Similar to Cas9 and Cas12, Cas13a adopts a bilobed architecture in which the nuclease domains form part of a lobe that is distinct from the crRNA recognition lobe. Target RNA binding drives large-scale conformational changes that bring the two HEPN domains into proximity, forming a composite catalytic center that is competent for RNA cleavage. This active site is exposed on the exterior of the protein, explaining how RNAs in solution, other than the target-bound RNA, can be bound and cleaved in trans after Cas13 activation. Interestingly, Cas13, like Cas12, employs an additional ribonuclease domain for pre-crRNA processing,[164] and inactivation of either catalytic activity does not impair the other.

Although Cas13 members can confer specific protection against RNA phages when expressed heterologously in E. coli,[28,157] the collateral, nonspecific cleavage of other cellular RNAs impedes bacterial growth and may have evolved as a strategy to induce cell dormancy and/or programmed cell death. Understanding the physiological consequences of interference in Type VI systems will require additional studies, ideally including experiments in which Cas13 function is tested in native organisms.

Type IV: To be determined

Type IV systems are a recent addition to Class 1 and encode known components of multisubunit protein–crRNA complexes but often lack adaptation genes, a candidate DNA nuclease, and even CRISPR arrays.[11] They have not been experimentally studied, and their function is unknown.

Expanding the CRISPR–Cas Toolbox

The mechanisms employed in CRISPR–Cas immunity have proven to be nothing less than a goldmine for biotechnological tool development, with researchers utilizing machinery from all three different stages—adaptation, crRNA biogenesis, and interference—for a wide range of applications. As early as 1993, microbiologists appreciated the inherent value of using CRISPR arrays, one of the fastest evolving regions of bacterial and archaeal genomes,[165] for high-resolution genotyping and strain differentiation.[166] More recently, processing ribonucleases involved in the crRNA biogenesis stage have been employed for applications ranging from tagged RNA isolation[167,168] to multiplexed gRNA processing,[169,170] owing to their exquisite sequence specificity for repeat-derived RNA substrates. Yet, the interference stage of CRISPR–Cas immune systems has clearly been the richest source of biotechnology tool development, with RNA-guided Cas proteins proving invaluable for next-generation methods to control cellular genotype and phenotype. Many of these advances would not have been possible without the solid foundations laid by heroic efforts to develop earlier genome-editing platforms.[171,172]

In the sections that follow, we summarize some of the major areas of CRISPR tool development, providing reviews where possible for expanded content. We refer interested readers elsewhere for work on gene drives,[173] model organism-specific applications,[174,175] multiplexing,[176] genome-wide screening,[177] in vivo delivery methods,[178] and development of human therapeutics.[179]

CRISPR-based genome editing

S. pyogenes Cas9 has seen the widest use as a genome engineering platform since it was first harnessed for DNA editing in eukaryotic cells.[180-184] Together with a natural dual-RNA guide or engineered single-RNA guide,[49] Cas9 can be programmed to introduce DSBs at DNA target sites, leading to DNA repair outcomes that fall into two major classes: nonhomologous end joining (NHEJ) and homology-directed repair (HDR)[185] (Fig. 3A). Repair outcomes from NHEJ, although not random,[79] typically result in small insertions or deletions at the target site that often create loss-of-function phenotypes when introduced within exons. Precise genetic alterations through HDR can be accessed through the combined use of Cas9–gRNA and a donor template. Yet, NHEJ is generally the dominant repair pathway in mammalian cells, and significant ongoing efforts are aimed at increasing HDR efficiency.[186]

Major tool categories in the CRISPR–Cas toolbox. (A) DSBs are introduced by Cas9–gRNA complexes, leading to permanent genomic edits through repair by either NHEJ or HDR. Cas9 orthologs, Cas9/gRNA variants, and alternative Cas nucleases can increase the fidelity of editing and expand targetable space in the genome. Base editing is achieved using nickase Cas9 variants fused to either cytidine or adenosine deaminases (bottom). (B) CRISPR-mediated transcriptional repression and activation (CRISPRi/CRISPRa) rely on deactivated Cas9 nucleases (dCas9) fused to various activator or repressor domains. Fluorescent protein fusions allow for imaging of specific chromosomal loci, and recruitment of histone- and DNA-modifying enzymes enables epigenome editing (bottom). (C) Cas13 has been harnessed for both targeted RNA knockdown and RNA editing, wherein the latter tool utilizes deactivated nucleases fused to adenosine deaminase domains. (D) Lineage tracing and molecular recording have been achieved using both Cas9-based strategies (top) and CRISPR arrays themselves (bottom). (E) Cas enzymes have proven useful outside of the cell, such as for Cas9-mediated removal of undesirable molecules within high-throughput DNA sequencing libraries (top), and for Cas13-mediated fluorescence detection of specific nucleic acid molecules in complex mixtures (bottom). DSBs, double-strand breaks; gRNA, guide RNA; HDR, homology-directed repair; NHEJ, nonhomologous end joining.

A major challenge of genome-editing applications is mitigating off-target effects. Cas9 readily cleaves genomic sequences that differ from the gRNA at one or more positions,[187-189] and the development of systematic approaches to interrogate off-target effects in an unbiased way, at the genome-wide scale, has been the focus of numerous laboratories.[190] Available methods use high-throughput sequencing and may broadly be grouped into two categories, cell-based methods and in vitro methods. In parallel, considerable efforts are underway to generate higher fidelity Cas9–gRNA variants, either by modifying the gRNA[191-193] or by creating improved Cas9 mutants through rational engineering or directed evolution.[69,74-76,194] Multiple strategies have improved accuracy through general destabilization of Cas9–gRNA DNA binding affinity, and recent theoretical models provide compelling kinetic explanations for the origins of this enhancement.[65,195]

SpyCas9 continues to be improved as a genome-editing reagent, but so too are other crRNA-guided nucleases being harnessed as alternatives (Fig. 3A). Some Cas9 orthologs are smaller and more amenable to viral packaging,[196] others may have higher intrinsic targeting fidelity,[197] and still others derive from thermophilic bacteria and exhibit elevated stability.[198,199] The recent discovery that human serum contains antibodies against two of the most commonly used Cas9 variants[200] is sure to stimulate additional Cas9 ortholog screening. Beyond Cas9, Cas12—the other family of single-effector Cas deoxyribonucleases—has been rapidly adopted for genome-editing experiments.[148,201] In addition to recognizing a distinct PAM and exhibiting high specificity,[202,203] initial studies suggested that the staggered nature of DNA cleavage products generated by Cas12 might naturally lend themselves to enhanced repair by the homologous recombination machinery; thus far, supporting data are lacking. Finally, efforts to engineer or evolve modified Cas9 and Cas12 variants that recognize novel PAM sequences have succeeded in substantially expanding the available “targetable” space of the genome.[194,204]

In recent years, an alternative approach to introduce single base-pair changes has been developed by Komor et al., termed base editing[205] (Fig. 3A). Rather than relying on host machinery for the repair of DSBs, base editors—a fusion of a Cas9 nickases to a nucleoside deaminase—achieve direct chemical conversion of one nucleobase to its deaminated counterpart, ultimately resulting in a permanent mutation. C·G-to-T·A base editors have been constructed using various cytidine deaminases,[206,207] including APOBEC and AID, and more recently, adenine base editors that mediate A·T-to-G·C mutations were created through extensive directed evolution and protein engineering.[208] In addition to enabling high-efficiency edits without the semistochastic outcomes characteristic of NHEJ, base editing also avoids adverse consequences of introducing mutagenic DSBs.

Last but not least, Class 1 systems—thus far not leveraged outside of prokaryotes, but by far the more abundant class in sequenced prokaryotic genomes—also show potential for certain genome-editing applications. In E. coli, types I and III systems have been harnessed for targeted plasmid elimination and programmable removal of specific strains from mixed populations.[209-211] In addition, by programming Cascade (together with Cas3) or Csm/Cmr complexes with self-targeting spacers in native hosts, researchers have achieved genome editing of bacterial, archaeal, and even phage genomes.[212-215] Whether similar strategies can be ported into eukaryotic cells remains to be determined, although one effort currently underway aims to treat antibiotic-resistant bacterial infections in human hosts by harnessing the programmable cell-killing activity of Cascade/Cas3.[216]

Deactivated nucleases for genome manipulation

As revolutionary as the Cas9 nuclease has been for genome editing, the utility of nuclease-deactivated Cas9 (dCas9) as a programmable DNA-binding protein has made nearly as large an impact on the research community. By virtue of its ability to associate with nearly any payload imaginable—either through protein–protein fusions or via RNA aptamers fused to gRNA scaffolds—dCas9–gRNA can achieve a wide variety of cellular outputs by ferrying diverse machinery to specific genomic loci (Fig. 3B; reviewed in Ref.[217]).

At the transcriptional level, dCas9 was first leveraged for repression, also known as CRISPR interference (CRISPRi), by simply occluding transcriptional machinery in bacteria[218] or by recruiting repressor domains like KRAB to promoter regions in mammalian cells.[219] CRISPR-mediated transcriptional activation (CRISPRa) was developed soon thereafter, employing either dCas9–activator fusions, the recruitment of tagged activators to gRNAs through RNA aptamers, or a combination of these approaches.[220-222]

The same basic recruitment strategies have allowed researchers to transform dCas9–gRNA into a robust system for editing the epigenome.[223] dCas9 fusions to histone demethylases like LSD1 or histone acetyltransferases like p300 have enabled up- and downregulation of gene expression through altered epigenetic marks at the histone level.[224,225] Parallel approaches have succeeded in achieving exquisite control over DNA epigenetic marks as well, whereby dCas9 fusions to either Tet1 or Dnmt3a have led to sustainable DNA demethylation or methylation and corresponding activation or silencing of gene expression, respectively.[226,227] The recent demonstration that methylation-edited cells can lead to persistent changes in heterochromatin status that reverses a disease phenotype suggests that epigenome editing will have therapeutic promise.[228]

dCas9 also offers a powerful new approach to image the three-dimensional organization of chromosomes in living cells through RNA-guided recruitment of fluorescent proteins to specific genomic loci.[229] As with CRISPRi/CRISPRa, various molecular strategies for recruitment have been explored, and the use of orthogonal reagents has allowed for multiplexed imaging of multiple loci simultaneously.[230] Although signal improvement remains a challenge for imaging-based approaches, particularly for nonrepetitive loci, a recent study developed a tandem gRNA assembly strategy to tile dCas9–GFP along a dozen or more sites within genomic regions of interest.[231]

Beyond chromosome imaging and gene expression control, the possibilities for recruiting other factors using dCas9–gRNA seem limitless. Large gRNA fusions have enabled locus-specific targeting of long noncoding RNAs;[232] dCas9–Spo11 fusion proteins have been harnessed to stimulate meiotic recombination at novel sites;[233] orthogonal dCas9 heterodimers have been constructed to enforce DNA looping in E. coli;[234] and dCas9–recombinase fusion proteins have been developed toward the eventual goal of programmable, recombinase-based genome editing.[235]

RNA-targeting applications

Two of the six immune system types (III and VI) are presently known to recognize RNA as their primary target, and effectors from two others (I and II) retain the capability to target RNA in vivo.[32,33] Naturally, then, there has been eager interest in exploring applications that harness crRNA-programmed Cas proteins for RNA manipulation.

The realization that SpyCas9 could be engineered to bind and cleave RNAs in vitro[83] led to some of the earliest demonstrations of programmable RNA targeting in mammalian cells, first for live cell RNA imaging[84] and later for the specific elimination of toxic RNA species directly in primary patient cells.[85] And although the multisubunit nature of Csm/Cmr complexes has thus far precluded biotechnological applications in eukaryotic cells, a recent preprint describes the use of recombinant Csm complexes to achieve targeted RNA knockdown in zebrafish embryos.[236]

Cas13, which naturally targets RNA within type VI systems, is proving to be a far more powerful enzyme for RNA-specific applications (Fig. 3C). In addition to harnessing Cas13 for nucleic acid detection (see CRISPR-based biochemical tools), the Zhang laboratory has achieved targeted knockdown of endogenous transcripts in mammalian and plant cells using Cas13, with comparable specificity and potency as RNA interference; fluorescent tracking of RNA transcripts using nuclease-inactive Cas13 (dCas13); and direct adenosine-to-inosine editing of RNA targets using ADAR2 fusions to an improved dCas13 variant from subtype VI-B.[158,159] It will be fascinating to see future applications of Cas13 for genome-wide RNA knockdown screens, direct manipulation of RNA splicing and protein synthesis, isolation and characterization of cellular RNA–protein complexes, and interrogation of lncRNA function.

Lineage tracing and molecular recording

Cas enzymes have also enabled new methods of molecular recording and cell lineage tracing (Fig. 3D). Rather than perturbing RNA or DNA sequences as a means of understanding the resulting impact on a cell or organism, these approaches instead seek to reveal the history of a cell through some form of genetic barcoding. Subsequent analysis reveals the comprehensive lineage relationships between cells within a population, and/or the past stimuli that a given cell and its ancestors have experienced.

Genome editing of synthetic target arrays for lineage tracing, developed by McKenna et al., leverages Cas9 to stochastically introduce indels within arrays of tandem target sequences.[237] Specific barcodes are inherited during each cell division, allowing for reconstruction of cell lineages along a developmental pathway by either high-throughput sequencing or, in an alternative approach, by in situ single-molecule FISH.[238] An alternative method for cellular recording leverages self-targeting gRNAs whose encoding loci can themselves be edited in a manner that reveals lineage history or specific biological events of interest, such as lipopolysaccharide-induced inflammation.[239,240] More recently, an analog recording strategy known as CAMERA, which harnesses either Cas9 or Cas9 base editors for DNA removal or DNA editing, respectively, has allowed multiple stimuli to be recorded, including exposure to viruses, light, and nutrients.[241]

Remarkably, molecular recording has also been achieved with actual CRISPR arrays, aided not by interference enzymes but by Cas1 and Cas2, the adaptation machinery responsible for spacer integration. After all, CRISPR arrays are nothing but a molecular memory of past infections that grow directionally, thus lending themselves naturally to function like a “biological tape recorder.” Shipman et al. were the first to apply a Cas1–Cas2 expression system[242] for molecular recording and CRISPR-archived data storage, which they used to encode a digital movie in a living bacterial population.[243,244] Whereas these studies required user-defined protospacers to be electroporated into cells, Sheth et al. have since engineered a system whereby the availability and selection of intracellular protospacers for CRISPR integration directly reflect past biological stimuli, such as the availability of metabolites in the growth medium.[245] Future improvements in CRISPR-based molecular recording methods will not only advance our understanding of development and cellular differentiation but may also shed light on the emergence and spread of cancer and other diseases.

CRISPR-based biochemical tools

The last category in the CRISPR toolbox we discuss revolves around the biochemical use of Cas enzymes (Fig. 3E). In an analogous manner to restriction enzymes, Cas9 has been combined with Gibson assembly and other DNA manipulations to facilitate molecular cloning of recombinant plasmids with nucleotide precision.[246-248] Potentially more broadly useful are molecular biology applications that harness Cas9 for the targeted depletion of abundant, unwanted sequences from high-throughput sequencing library preparations. For example, Wu et al. removed mitochondrial DNA sequences from ATAC-seq libraries in their study of chromatin accessibility,[249] and Gu et al. adopted a similar approach while further demonstrating that Cas9 could deplete wild-type alleles of KRAS from patient cancer samples, facilitating the detection of rare mutant KRAS alleles.[250] Cas9 may also be useful to directly enrich for desired sequences, either through dCas9-based affinity purification or through the excision and subsequent isolation of genomic regions of interest.[251]

Cas9 has even proven useful for physical mapping of genomic DNA. In one study, a Cas9 mutant was used to nick specific genomic sequences, followed by DNA polymerase-mediated local incorporation of fluorescent nucleotides; the resulting sites could then be directly visualized within nanochannel arrays using fluorescence microscopy to identify structural variants.[252] Unlabeled, DNA-bound Cas9–gRNA particles have also been directly imaged by high-speed atomic force microscopy, enabling a new form of “nanomapping” that can fill gaps not addressable by traditional sequencing.[253]

An exciting development within just the past year has revolved around the use of Cas12 and Cas13 enzymes for the in vitro detection of target nucleic acid sequences with attomolar (10−18 M) sensitivity and single-mismatch specificity. The two related detection platforms, SHERLOCK[254,255] and DETECTR,[31] take advantage of the target-activated, nonspecific nucleic acid degradation activity that was first observed for Cas13 with ssRNA and later for Cas12 with ssDNA. After the demonstration that quenched fluorescent reporter substrates could be employed for optical detection of specific RNA transcripts,[164] Gootenberg et al. adopted a similar approach but, importantly, incorporated isothermal amplification, reverse transcription, and in vitro transcription steps to both increase the sensitivity and generalize the approach to allow for both RNA and DNA detection.[254] The latest developments in SHERLOCK technology include the use of orthogonal enzymes for multiplexing and lateral flow read-out, such that Zika and Dengue viral ssRNAs, as well as specific cancer mutations, could be detected without the requirement of any additional instrumentation beyond the SHERLOCK cocktail itself.[255]

Collectively, these methods highlight the remarkable functional diversity and utility of CRISPR–Cas proteins, as well as their robustness and stability in biochemical assays. In light of the growing availability of recombinant Cas enzymes and gRNAs from commercial vendors, in vitro tool development is sure to continue growing alongside cell-based applications in the coming years.

Concluding Thoughts

The surge of innovative technologies harnessing CRISPR–Cas—which can no longer even be confined within the “genome engineering” umbrella term—is proceeding unabated, and it is clear that biologists, regardless of model system or research problem, will increasingly turn to CRISPR-based tools for help. Yet, while the toolbox expands and grows in complexity, so too are the adaptive immune systems that inspired it in the first place becoming ever richer and more interesting than previously appreciated. Indeed, far from the study of CRISPR–Cas biology decelerating or approaching a point of saturation, new discoveries continue to lurk just around the corner. From the discovery of novel effectors[155] and virally encoded CRISPR–Cas inhibitors,[256] to the identification of entirely new families of putative, CRISPR-linked accessory genes,[44,45] there will undoubtedly be ample subject matter for future experimental exploration and exploitation.

In the grander scheme of the microbial defense arsenal, though, CRISPR–Cas adaptive immune systems and other well-studied innate systems may be just the tip of the iceberg. Inspiring work from Sorek and colleagues have uncovered entirely new families of defense systems,[257-259] many of which have unknown or poorly understood mechanisms of action. In addition, the future mining of metagenomic data sets is sure to turn up more novelties buried within the genomes of uncultivated species.[41] Given that the present sampling of microbes and viruses is many orders of magnitude below the true diversity in our biosphere,[260,261] plenty remains to be discovered.

Although our understanding of phage biology and host defenses has exploded in recent years, it is remarkable that one of the fundamental assays used time and time again—infect a bacterial culture with virus and count plaques or resistant colonies—is hardly different than it was nearly a century ago. Luria and Delbrück used this approach with great success in their famous Fluctuation Test experiment in the 1940s, which provided support for the “hypothesis of mutation to immunity” and confirmed Darwin's theory of natural selection.[262] Now imagine if they had been using a bug other than the immunocompromised E. coli B strain (which lacks cas genes): would a fully functioning CRISPR–Cas system have provided evidence for the “hypothesis of acquired immunity” instead, and thus supported a Lamarckian model of evolution?[263,264] How might this fictitious outcome have affected the trajectory of molecular genetics research?

Delbrück was never one to shy away from new experimental problems, and from his own writings, one can envision how he might have tackled the mystery of CRISPRs: “I could not do it in a few months. Perhaps it will take a few decades, and perhaps it will take the help of a few dozen other people. But listen to what I have found, perhaps you will be interested to join me.”[265]