Bacteriophages (phages) and their preys are engaged in an evolutionary arms race driving the co-adaptation of their attack and defense mechanisms. In this context, phages have evolved diverse anti-CRISPR proteins to evade the bacterial CRISPR–Cas immune system, and propagate. Anti-CRISPR proteins do not share much resemblance with each other and with proteins of known function, which raises intriguing questions particularly relating to their modes of action. In recent years, there have been many structure–function studies shedding light on different CRISPR–Cas inhibition strategies. As the anti-CRISPR field of research is rapidly growing, it is opportune to review the current knowledge on these proteins, with particular emphasis on the molecular strategies deployed to inactivate distinct steps of CRISPR–Cas immunity. Anti-CRISPR proteins can be orthosteric or allosteric inhibitors of CRISPR–Cas machineries, as well as enzymes that irreversibly modify CRISPR–Cas components. This repertoire of CRISPR–Cas inhibition mechanisms will likely expand in the future, providing fundamental knowledge on phage–bacteria interactions and offering great perspectives for the development of biotechnological tools to fine-tune CRISPR–Cas-based gene edition.
In the battle for survival, bacteria and archaea have evolved mechanistically diverse defense mechanisms to resist their predators . The prokaryotic CRISPR–Cas immune system uniquely provides a memory of past infections in the form of small pieces of DNA, taken from foreign genetic material and integrated into a CRISPR locus as spacers flanked by short repeats. These spacers are transcribed into CRISPR RNAs (crRNAs) that associate with Cas proteins to form effector complexes that patrol the cells for potential invasion. Upon recognition and binding to nucleic acid sequences complementary to crRNAs (referred to as protospacers), the Cas nuclease function is activated and the invading genetic material is destroyed  (Figure 1). CRISPR–Cas systems are highly diverse and currently classified into two broad classes, based on the composition of Cas proteins, six types and 25 subtypes. Class 1 systems (types I, III and IV) use multi-protein effector complexes to bind and cleave target DNA, whereas Class 2 systems (types II, V and VI) use a single, multi-domain Cas protein for nucleic acid binding and cleavage .
Different routes to inactivate CRISPR–Cas immunity.
Without much surprise, bacteriophages (phages) — the most abundant biological entities on the planet  — have evolved different mechanisms to protect against CRISPR–Cas immunity. For instance, they can hide from effector complexes through modifications, deletions or mutations in regions of their genome that require perfect complementarity with crRNAs [2,5–7]. Additionally, phages produce anti-CRISPR proteins (Acr) at the early infection stage that directly interact with, or modify, CRISPR–Cas components, and block their activity. Anti-CRISPR proteins were first identified in Pseudomonas aeruginosa prophages in 2013 . However, the first 3D structure of an anti-CRISPR protein was actually determined in 2009 with the crystal structure of the ORF 99 from the archaeal virus Acidianus filamentous virus 1 (AFV1) . Indeed, He et al.  characterized the first anti-CRISPR protein encoded by the archaeal viruses Sulfolobus islandicus rod-shaped viruses 2 and 3 (SIRV2 and SIRV3), AcrID1, which was none other than a homolog of AFV1 ORF 99. To date, nearly 50 anti-CRISPR proteins inhibiting Class 1 and Class 2 CRISPR–Cas systems have been identified in prophages, virulent viruses and other mobile genetic elements [11–14]. Strikingly, these proteins are different from one another and do not share much sequence similarities with proteins of known function. The diversity and broad the distribution of anti-CRISPR proteins raise many questions relating to their origins and evolution, their roles in the diversification of CRISPR–Cas systems and emergence of anti-anti-CRISPR mechanisms, and their modes of action.
In recent years, the highly dynamic anti-CRISPR field of research has emerged and accumulated important knowledge on the molecular mechanisms deployed by these natural CRISPR–Cas inhibitors. Many reviews that thoroughly present the discovery of anti-CRISPR proteins and their molecular mechanisms have recently been published [11,15–17]. Here, we intend to provide a complementary, global perspective on their mode of action. We summarize the latest findings focusing on the different routes, and their interplay, to interfere with CRISPR–Cas immunity, and on the variety of molecular tactics to inactivate CRISPR–Cas machineries, which can be differentiated into orthosteric, allosteric and enzymatic inhibition mechanisms. In the current context where the characterization of anti-CRISPR proteins is rapidly growing, these few functional and mechanistic parameters can be used to simply sort out diverse CRISPR–Cas inhibitors.
Different routes to inactivate CRISPR–Cas immunity
CRISPR–Cas immunity divides into (1) the adaptation step, to build up the immunological memory through the acquisition of foreign DNA sequences, (2) the expression and assembly step, to produce functional effector complexes and (3) the interference step, to detect and cleave foreign nucleic acids, provided they are flanked by a protospacer-adjacent motif (PAM), with the exception of type III and type VI CRISPR–Cas systems, that allows self and non-self discrimination by the host [5,18] (Figure 1). In type I and type II CRISPR–Cas systems, the PAM is also important for appropriate spacers selection during adaptation [19,20]. Anti-CRISPR proteins have, therefore, the possibility to inactivate CRISPR–Cas immunity by interfering with one, or several, of these steps. However, most of the anti-CRISPR proteins characterized so far are anti-interference molecules that directly interact with effector complexes. The anti-CRISPR field of research having emerged recently, this trend may change in the future.
Inhibition of CRISPR–Cas adaptation
Cas1 and Cas2 are essential proteins for spacer acquisition in all studied CRISPR–Cas systems , but they do not seem to have any role in the expression or interference steps. Therefore, any anti-CRISPR protein inhibiting the production or the activity of these Cas effectors would specifically block CRISPR–Cas adaptation. Cas1 and Cas2 being highly conserved Cas proteins present in most known CRISPR–Cas systems , it is almost certain that phages have evolved anti-CRISPR proteins that directly inhibit their activity and whose discovery was limited by the approaches used so far to uncover CRISPR–Cas inhibitors. In addition, components of the interference machinery are also involved in the adaptation step. The type I helicase/nuclease Cas3 and effector Cascade–crRNA complex (glossary), and the type II effector Cas9–sgRNA complex (glossary), associate with the Cas1–Cas2 complex for naïve, primed and interference-driven spacer acquisition (glossary) [20,22–25]. Noteworthy, Cas2 and Cas3 are fused into a single protein (Cas2/3) in the type I-F CRISPR–Cas system , pointing on the functional link between the adaptation and interference steps. A consequence of this molecular cross-talk between adaptation and interference is that any anti-CRISPR protein targeting the interference machinery could also interfere with the adaptation step .
AcrIF3 is an example of such anti-CRISPR protein with a dual anti-interference and anti-adaptation activity . This CRISPR–Cas inhibitor binds to the helicase/nuclease Cas3, and prevents its recruitment to the effector Cascade–crRNA complex [26–28]. AcrIE1 inactivates the type I-E Cas3 nuclease  and thereby could also inhibit the adaptation step, although further experiments are required to confirm its anti-adaptation activity. The protospacer, bound to the Cascade–crRNA complex, cannot be cleaved, which aborts the interference step and, in turn, stops the generation of spacer precursors for primed and interference-driven adaptation .
In type II-A CRISPR–Cas systems, the Cas9–tracrRNA complex (glossary) associates with the other components of the spacer acquisition machinery, Cas1, Cas2 and Csn2, to ensure that new spacers are flanked by the correct PAM . Moreover, Cas9-mediated interference activity and protospacer cleavage have recently been shown to prime the acquisition of new spacers . Therefore, type II-A anti-CRISPR proteins that inhibit Cas9 interference activity could also perturb the adaptation step.
Inhibition of CRISPR–Cas expression and assembly
Cas proteins together with crRNA molecules form effector complexes competent for the recognition and binding to specific targets. Disrupting the production or assembly of these ribonucleoprotein machineries necessarily abrogates the interference and interference-driven adaptation steps. Anti-CRISPR proteins acting as transcriptional or translational repressors of CRISPR–Cas components, which would, therefore, prevent their expression, remain to be uncovered. Although such anti-CRISPR proteins could not rapidly inactivate CRISPR immunity as required upon virulent phage infection, they could be useful to protect prophages from CRISPR–Cas degradation. To date, AcrIIA1 and AcrIIC2 are the only examples of anti-CRISPR proteins that prevent the formation of effector complexes. While AcrIIA1 binds to a broad-spectrum of type II-A and type II-C Cas9s and triggers their degradation within infected cells  AcrIIC2 binds to the apo form of Neisseria meningitidis Cas9 (unliganded NmeCas9) and blocks sgRNA loading [32,33].
Inhibition of CRISPR–Cas interference
The anti-interference activity can be achieved by the inhibition of target binding, or nuclease activity. To date, inhibition of target binding is the prevalent strategy. Thirteen anti-CRISPR proteins interfere with target recognition and binding (type I-F AcrIF1, AcrIF2 and AcrIF10 [34–37]; type II-A AcrIIA2, AcrIIA4, AcrIIA5 and AcrIIA6 [38–47]; type II-C AcrIIC3, AcrIIC4 and AcrIIC5 [48–51]; type V-A AcrVA1, AcrV4A and AcrVA5 [52–58]), while only five-block target cleavage (type I-E AcrIE1 [29,59]; type III-B AcrIIIB1 ; type I-F AcrIF3 [26, 27]; type II-C AcrIIC1 and AcrIIC3 [48–50]) (Figure 1). Given that DNA binding is the rate-limiting step of Cascade and Cas9-mediated interference activities [60,61], altering this step is, therefore, an efficient way to inactivate CRISPR–Cas interference. Although the type I-D AcrID1, type I-F AcrIF4 and type II-A AcrIIA7-11 were shown to block the interference step, it is not yet known whether they inhibit target binding or cleavage [8,10,13,14].
Molecular mechanisms used by anti-CRISPR proteins
Structure-function studies of anti-CRISPR proteins have revealed a variety of molecular mechanisms that can be grouped into three categories. First, anti-CRISPR proteins can act as inhibitors that directly bind and sterically occlude the functional sites of Cas effectors. Second, anti-CRISPR proteins can act as allosteric inhibitors that associate with regions distinct from the functional sites, and modify the conformational dynamics and structural transitions of CRISPR–Cas machineries. Third, anti-CRISPR proteins can be enzymes that degrade or permanently modify CRISPR–Cas components. Noteworthy, most of the 3D structures of anti-CRISPR proteins available to date display unique folds.
Steric occlusion of Cas effector functional sites
To date, the most common molecular tactic used by anti-CRISPR proteins is to sterically block the access to the functional sites of Cas machineries, including the target-binding site, sgRNA-binding site and catalytic site. Such anti-CRISPR proteins that bind to highly conserved CRISPR–Cas functional sites limit the chance for the host to escape inhibition. This could be one of the reasons making these steric inhibitors the prevalent anti-CRISPR proteins so far. Nevertheless, further investigation of the molecular mechanisms used by anti-CRISPR proteins is required to determine whether this trend is biologically relevant or whether it reflects sampling bias.
The AcrIF10, AcrIIA2 and AcrIIA4 anti-CRISPR proteins act as DNA mimics to prevent target DNA from binding to effector Cascade–crRNA and Cas9–sgRNA complexes, respectively. However, they block the access to effector complexes by different means. AcrIF10 binds to the DNA binding site at the junction between the Cas8f and Cas5f subunits, and induces a DNA-bound conformation of the effector Cascade–crRNA complex . As for AcrIIA2 and AcrIIA4, they are both small acidic proteins that interact with Cas9 PAM binding elements, thereby preventing the necessary primary recognition of the PAM next to the target DNA sequence [39–43] (Figure 2A). Interestingly, AcrIIA2 and AcrIIA4 illustrate a case of functional convergence from structurally distinct inhibitors with overlapping binding sites.
Molecular mechanisms used by anti-CRISPR proteins.
AcrIF1 and AcrIF2 anti-CRISPR proteins use different strategies to block the access to DNA binding sites of the effector Cascade–crRNA complex. Several copies of AcrIF1 bind to the Cas7f backbone of Cascade and block the target DNA to access and hybridize with the crRNA [36,37,63]. As for AcrIF2, its binding site located between the Cas7.6f and Cas8f subunits partially overlaps with the DNA binding site (Figure 2A). Moreover, AcrIF2 binding to the effector Cascade–crRNA complex induces conformational changes incompatible with target DNA binding .
Interestingly, AcrIF3 inactivates the type I-F CRISPR–Cas defense also through molecular mimicry. However, AcrIF3 functions as a mimic of a Cas protein, in contrast with the DNA mimics presented above. Its 3D structure is similar to the C-terminal helical bundle of Cas8, which is exposed upon target DNA binding to Cascade–crRNA complex to recruit Cas2/3 [28,37]. AcrIF3 forms homodimers that bind to Cas2/Cas3 (Figure 2A), thereby preventing its recruitment to the target-bound Cascade–crRNA complex [26,27].
Lastly, AcrIIC1 and AcrIIC2 are so far unique examples of inhibitors that bind to the catalytic site of the Cas9 HNH nuclease domain (glossary) and to the NmeCas9 sgRNA-binding site, respectively. AcrIIC1 directly interacts with the conserved catalytic residues of the HNH domain, thereby trapping the target-bound effector complex in a catalytically inactive state . AcrIIC2 forms homodimers with an acidic groove that strongly interacts with the NmeCas9 arginine-rich bridge helix, thereby impeding sgRNA loading [32,33] (Figure 2A).
Allosteric inhibition and clustering of effector complexes
Recently, the AcrIIA6, AcrIIC3 and AcrVA4 anti-CRISPR proteins have been shown to be allosteric inhibitors of the Streptococcus thermophilus CRISPR1-Cas9 (St1Cas9), NmeCas9 and Lachnospiraceae bacterium Cas12a (LbCas12a) effector complexes, respectively. Interestingly, these anti-CRISPR proteins can also bind to and inactivate two effector complexes at a time (Figure 2B). Noteworthy, AcrID1 directly interacts with Cas10d, the large subunit of the type I-D CRISPR–Cas effector complex, and induces its dimerization. However, the molecular basis of AcrID1 inhibition mechanism remains to be determined . Knowing that anti-CRISPR proteins gradually immunosuppress bacteria through multiple failed phage infections [64,65], lowering the critical concentration of a given anti-CRISPR protein required for the complete inactivation of CRISPR–Cas immunity may be advantageous for a phage population to rapidly propagate.
AcrIIA6 and AcrVA4 both function as dimers that tightly associate with a mixed protein–RNA region distinct from the DNA-binding crevasse and catalytic domains [47,54,57,58]. However, regions of both subunits that compose the AcrIIA6 dimer form each binding interface, while every subunit of the AcrV4A dimer contains one binding interface (Figure 2B). These anti-CRISPR proteins perturb the conformational changes required to bind to target DNA, thus locking St1Cas9 and LbCas12a effector complexes in a nonfunctional state. AcrIIA6 impairs St1Cas9 conformational rearrangements associated with PAM binding, which leads to the inhibition of target DNA recognition and binding . In contrast, AcrVA4 does not affect PAM binding but impairs LbCas12a dynamics and structural changes required for the R-loop formation and stabilization (glossary), which prematurely stops the hybridization between the cRNA and target DNA . Additionally, AcrVA4 can also associate with DNA-bound effector complexes. When it binds to the LbCas12a–crRNA–dsDNA complex with a complete R-loop, it induces the release of the bound DNA before cleavage [54,58]. When it binds to the post-cleavage Cas12a–crRNA–dsDNA complex, it likely blocks the recycling of the enzyme .
As for AcrIIC3, one monomer is able to tether two NmeCas9 effector complexes (Figure 2B). It interacts with the NmeCas9 HNH domain of one effector complex, at the opposite face of the catalytic site, and the NmeCas9 REC lobe (glossary) of another effector complex [33,50,66]. The interaction between AcrIIC3 and the REC lobe, in the vicinity of the DNA–sgRNA hybridization site, likely perturbs the conformational dynamics of NmeCas9 that assists DNA binding, which would explain the reduction in NmeCas9 DNA binding affinity in vitro and the inhibition of DNA binding within cells [48,49]. Besides, AcrIIC3 is able to associate with DNA-bound NmeCas9 effector complexes in vitro . The interaction between AcrIIC3 and the HNH domain blocks the structural changes triggered by target DNA binding, and therefore locks the HNH and RuvC nuclease domains (glossary) in inactive conformations . These data suggest that AcrIIC3 could inactivate NmeCas9 effector complexes in the presence or absence of target DNA.
AcrIIA6, AcrIIC3 and AcrVA4 all bind to two CRISPR–Cas effector complexes, although using different molecular strategies. What is the benefit for CRISPR–Cas allosteric inhibitors to dimerize or harbor multiple binding sites, and cluster effector complexes? One explanation could be that the ‘weakness’ of allosteric inhibitors that leave accessible the CRISPR–Cas functional sites, would be compensated by their ability to sequester multiple effector complexes, thereby reducing the amount of functional effector complexes within infected cells.
Enzymatic modification of CRISPR–Cas components
The two anti-CRISPR proteins with an enzymatic activity that have been characterized so far, AcrVA1 and AcrVA5, inhibit Cas12a effector complexes through different mechanisms (Figure 2C). Interestingly, AcrVA1 and AcrVA5 bind to overlapping regions in the PAM-interacting domain and compete with one another [55,57]. However, their substrate and enzymatic activity are different. AcrVA1 is a multiple-turnover endoribonuclease that cleaves off the Cas12a-bound crRNA spacer sequence to irreversibly inactivate the effector assembly . Noteworthy, AcrVA1 mimics the PAM to position its catalytic residues close to the crRNA substrate . Interestingly, the type II-A AcrIIA5 anti-CRISPR protein was recently shown to lead to sgRNA cleavage at multiples sites out of the crRNA spacer sequence . However, whether or not this anti-CRISPR protein also has a nuclease activity remains to be determined. In contrast, AcrVA5 is an acetyltransferase that permanently modifies one LbCas12a lysine residue required for PAM recognition . This lysine acetylation not only abolishes the interaction with PAM nucleobases, but also generates steric hindrance with the whole PAM. Such enzymatic strategies allowing the permanent inactivation of many interference complexes are likely beneficial to phages to decrease the number of failed phage infections required to immunosuppress their host [64,65], and thereby rapidly evade CRISPR–Cas immunity.
• Cascade: CRISPR-associated complex for antiviral defense. Multi-subunit Cas protein complex that associates with crRNA to form the effector complex.
• HNH domain: endonuclease domain named after its catalytic histidine and asparagine residues.
• Interference-driven adaptation: integration into CRISPR arrays of DNA fragments surrounding targeted protospacers.
• Naïve adaptation: integration into CRISPR arrays of DNA fragments from newly encountered mobile genetic elements.
• Primed adaptation: integration into CRISPR arrays of DNA fragments partially matching to spacers that are not subject to efficient interference.
• REC lobe: Recognition lobe. The REC lobe of Class 2 CRISPR–Cas effectors is a functional region involved in crRNA recognition, target binding and nuclease activation.
• R-loop: nucleic acids structures consisting of a DNA–RNA heteroduplex and displaced ssDNA. R-loops are formed upon CRISPR–Cas effector complex binding to target dsDNA.
• RuvC domain: endonuclease domain named after the RuvC protein, a Holliday junction resolvase from Escherichia coli.
• sgRNA: single-guide RNA. Engineered RNA construct that assembles the crRNA and tracrRNA molecules into a single sequence.
• tracrRNA: trans-activating CRISPR RNA. Small non-coding RNA molecule that forms base pairs with the crRNA and is required for type II and type V-B CRISPR–Cas interference.
Importance of the field: The highly dynamic anti-CRISPR field of research is fueled by the constant discovery of remarkably diverse CRISPR–Cas inhibitors, widespread amongst bacterial and archaeal viruses and with little similarity with proteins of known function. Deciphering their molecular modes of action is having a massive impact on our understanding of phage biology, bacterial evolution and host–pathogen interactions. Moreover, anti-CRISPR proteins are scrutinized for their potential as biotechnological tools to fine-tune CRISPR–Cas-based gene edition.
Summary of current thinking: Based on the structure-function studies of anti-CRISPR proteins that are currently available, four major outcomes stand out. First, because of biases associated with anti-CRISPR protein identification pipelines, our current knowledge mainly covers anti-CRISPR proteins inhibiting the CRISPR–Cas interference step. Second, the inhibition of target binding seems to be the strategy favored by anti-CRISPR proteins to inactivate both Class1 and Class 2 CRISPR–Cas systems. Third, the direct binding to CRISPR–Cas functional sites and their steric occlusion is so far the molecular tactic that prevails over allosteric inhibition modes and enzymatic modifications. Nevertheless, whether this current prevalence of steric inhibition reflects a biological reality or a sampling bias needs to be determined. To finish, nearly all 3D structures of anti-CRISPR proteins reveal new folds. All in all, anti-CRISPR proteins are remarkably diverse at the levels of their sequences, structures and inhibition mechanisms.
Future directions: The characterization of anti-CRISPR proteins that interfere with the adaptation and expression steps of CRISPR–Cas immunity will be key to provide a complete picture of the molecular mechanisms used by these CRISPR–Cas inhibitors. Additionally, phage genomes often encode for several anti-CRISPR proteins, which raises the question of the interplay between different anti-CRISPR proteins to inactivate CRISPR–Cas systems. Besides, whether molecular mechanisms are favored for lytic or lysogenic cycles also remains to be addressed. Our current knowledge of anti-CRISPR inhibition strategies indicates that the ‘best’ anti-CRISPR proteins from a mechanistic and biochemical point of view, such as enzymes that rapidly modify multiple CRISPR–Cas effector complexes, may not be widely used by phages. It will be important to examine the molecular strategy used by a given phage in light of the phage population context and of the phage-host interactions. Lastly, investigations of the mechanisms evolved by bacteria and archaea to resist anti-CRISPR proteins will tackle the next step in the CRISPR-based molecular arms race.
The authors declare that there are no competing interests associated with the manuscript.
Conceptualization, A.G.; Writing — original draft, A.G.; Writing — review and editing, P.H., A.G.; Visualization, P.H. and A.G.
A.G. acknowledges the French National Research Agency for a grant related to this work (ANR-18-CE11-0016-01). UCSF ChimeraX that was used for molecular graphics is developed by the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco, and receive support from NIH R01-GM129325 and P41-GM103311.