Search for Origins of Anti-CRISPR Proteins by Structure Comparison

Abstract

Many bacterial and archaeal viruses encode anti-CRISPR proteins (Acrs) that specifically inhibit CRISPR-Cas systems via various mechanisms. The majority of the Acrs are small, non-enzymatic proteins that abrogate CRISPR activity by binding to Cas effector proteins. The Acrs evolve fast, due to the arms race with the respective CRISPR-Cas systems, which hampers the elucidation of their evolutionary origins by sequence comparison. We performed comprehensive structural modeling using AlphaFold2 for 3693 experimentally characterized and predicted Acrs, followed by a comparison to the protein structures in the Protein Data Bank database. After clustering the Acrs by sequence similarity, 363 high-quality structural models were obtained that accounted for 102 Acr families. Structure comparisons allowed the identification of homologs for 13 of these families that could be ancestors of the Acrs. Despite the limited extent of structural conservation, the inferred origins of Acrs show distinct trends, in particular, recruitment of toxins and antitoxins and SOS repair system components for the Acr function.

Introduction

CRISPR-Cas systems are adaptive immune mechanisms of bacteria and archaea that detect and destroy foreign nucleic acids.^1–4 Anti-CRISPRs (Acrs) are proteins encoded by many bacterial and archaeal viruses and proviruses that counter-act the CRISPR-Cas immunity via diverse mechanisms.^5–8 Most of the Acrs are highly specific toward a particular CRISPR-Cas subtype or variant. Notably, some viruses encode entire arrays of Acrs with distinct specificities.^9,10

The majority of the Acrs are small proteins with a broad structural variety, including all α, all β, mixed α/β, and separated α+β folds.^11,12 Mechanisms of CRISPR-Cas inhibition by Acrs show remarkable functional diversity as well, including, but not limited to, inhibition of DNA binding by the CRISPR effectors, inhibition of CRISPR (cr) RNA-Cas complex formation, and DNA cleavage.^11–13

The ability of Acrs to interfere with CRISPR-Cas systems can be harnessed to develop various biotechnological applications, potentially, improving the accuracy of CRISPR-Cas based manipulations. Reduction of cytotoxicity as well as off-target, and off-tissue effects of CRISPR-based tools using various Acrs has been reported.^14–16 Moreover, Acrs can substantially improve the efficacy of phage therapy efficiency against antibiotic-resistant bacteria that have natural CRISPR-Cas immunity.^12,17

The genes encoding Acrs evolve rapidly as part of the arms race with the adapting CRISPR-Cas systems.⁸ Given this high rate of change combined with the small size, the capacity of sequence similarity analysis for the prediction of new Acrs, identification homologs, and inference of their origins is limited. With respect to the identification of Acrs, this challenge has been partially met through the development of various algorithms and machine-learning approaches based on the signature features of the Acrs and Acr-associated genes.^18–21

However, the origins and evolution of the Acrs remain poorly understood because in many cases, neither sequence nor structural comparisons yielded any significant similarities to the Acrs.¹¹

Here, we took advantage of the recently developed powerful approach for protein structure prediction, AlphaFold2 (AF2),²² to comprehensively predict the structures of Acrs and used structural similarity search combined with sensitive sequence similarity search to detect proteins homologous to Acrs that could be their evolutionary ancestors. We show that Acrs share folds with diverse, unrelated proteins, which suggests their origins from multiple sources. Nevertheless, consistent trends in the apparent recruitment of proteins with particular functions as Acrs were identified.

Methods

We predicted structures for Acrs from Anti-CRISPRdb v2.2 database²³ (http://guolab.whu.edu.cn/anti-CRISPRdb) using AF2 v2.2.0 with locally installed complete databases.²² The protein sequences from the Anti-CRISPRdb were clustered with 50% identity and 80% bidirectional coverage using mmseqs2.²⁴ Then, from each cluster, the protein model with the most confident structural prediction by AF2 was selected based on the highest average predicted local distance difference test (pLDDT) for further analysis.²⁵

The structures predicted with AF2 were used to search for structurally similar proteins using DALI v5 against the PDB90 subset of the protein structure database.²⁶ For this comparison, regions with pLDDT <80 were removed from the models.

To complement the structure similarity search with a sensitive search for sequence similarity, HHpred search²⁷ was run against the PDB70, Pfam A, and Conserved Domains Database (CDD) with default parameters for a representative of each Acr family. For proteins structure visualization and structures comparison, ChimeraX²⁸ and ICM-browser²⁹ were used.

Data availability

Structures of all Acrs predicted with AF2 are available at https://doi.org/10.5281/zenodo.7747008. The rest of the data are available in the article and the Supplementary Material.

Results

Using AF2, we predicted the structures for all experimentally validated and predicted Acrs (n = 3693). As a quality control, we compared the structures predicted with AF2 with all available experimental structures of Acrs. BLASTP search in the Protein Data Bank (PDB) database detected 41 unique Acr structures in a complex with CRISPR proteins or in a monomeric/dimeric form (Supplementary Table S1). We superimposed structures of these proteins with AF2 models using least-squares fitting and calculated root-mean-square deviation (RMSD) between C_α atoms (Supplementary Fig. S1).

Generally, for average pLDDT >95, structures aligned near perfectly, with RMSD <1 Å, and for pLDDT >80, there were minor differences, with RMSD <2 Å, but in each case, the fold was predicted correctly, with deviations in the loop regions only. Structures with pLDDT <80 often are misfolded and cannot be considered reliable. Thus, for 41 Acrs with the experimental structure, AF2 reproduced 37, where the mean pLDDT was greater than 80.

In addition, we predicted structures for 63 Acrs without experimentally solved structures, with mean pLDDT >80, and for 11 Acrs with mean pLDDT <80. After clustering the Acrs by sequences similarity and discarding unreliable structural models (Methods section), the final set contained models for 363 representative Acrs. We then compared these models to PDB90 using DALI and collected for further examination all hits with Z-score >5 (n = 2311).

In addition, we ran HHpred against the PDB, Pfam A, and CDD databases and collected the best hits with a Probability score greater than 80. Overall, we analyzed >300 hits from these databases (Supplementary Table S2). As a result of this analysis, we were able to identify homologs for 13 Acr families (Table 1 and Supplementary Table S3). For 12 of these, the result was reproduced at the sequence level using HHpred. The key findings from these searches are presented and discussed next.

Table 1.

Acr homologs identified by structural similarity search

No.	Acr/Accession	PDB	Z	RMSD	PDB description
1	AcrIB/ACV38859.1	5fvj-A	17.3	2.3	Acetyltransferase TacT
2	AcrVA5/WP_046699157.1	5c82-A	11.2	1.7	Nourseothricin acetyltransferase
3	AcrIF11/OHU91773.1	6gw6-A	6.3	3.2	RES toxin
4	AcrIF11/MBO08902.1	6d0h-A	5.8	2.8	ParT, COG5654 (RES domain)
5	AcrIF11/BAG80428.1	2q5t-A	6.4	3.1	Cholix toxin
6	AcrIF11/KHN50706.1	4ow6-B	5.7	3.4	Diphtheria toxin
7	AcrVIA6/ETD74580.1	4pu4-D	5.3	2.3	Antitoxin HipB
8	AcrIIA1/MCN72780.1	4yg1-A	5.3	2.3	Antitoxin HipB
9	AcrIIA1/KTA35071.1	6cf1-B	5.7	3.8	Antitoxin HigA
10	AcrIIA1/OPX53912.1	3vwb-A	6.3	3.3	VirB
11	AcrIIA1/OEH82315.1	2grm-A	7.3	3.9	PrgX
12	AcrIF9/PMT73868.1	5i8j-A	5.7	2.9	Dmd
13	AcrIIC3/CWS40682.1	5zyf-A	6.9	2.7	Cas2, CRISPR-Cas protein involved in adaptation step
14	AcrIA/AAR27922.1	5ean-A	13.1	2.4	Dna2
15	AcrIA/AAR27922.1	7lw7-A	12.7	3.0	Exonuclease V
16	AcrIA/AAR27922.1	3h4r-A	11.3	3.2	Exodeoxyribonuclease 8
17	AcrIIA21/WP_000384271.1	3k2z-B	8.5	1.7	LexA
18	AcrIIA21/WP_000384271.1	4pt7-B	15	1.5	RepA
19	AcrIIA11/WP_118225168.1	3nct-D	6.4	2.8	PsiB, negative regulator of RecA
20	AcrIIA11/WP_118225168.1	5oqm-D	6.7	4.4	CTD of Mediator of RNA polymerase II transcription subunit
21	AcrIC7/EWC40192.1	1ghh-A	7.5	2.3	DNA-damage-inducible protein I (DinI)
22	AcrIIA8/WP_038352084.1	2kz4-A	10.2	1.3	Putative head-tail adaptor
23	AcrIIA8/WP_035290260.1	6te9-E	10.1	2.5	Adaptor protein RCC01688
24	AcrIIA8/WP_084113462.1	3f3b-A	8.7	1.8	Phage-like element PBSX protein xkdH
25	AcrIIA8/WP_066523678.1	2pp6-A	8.3	1.7	ATP-binding sugar transporter-like protein
26	AcrIIA8/WP_002597422.1	6zlv-A	7.5	2.0	Rod shape-determining protein MreC
27	AcrVA3/STZ14554.1	1u3e-M	6.6	2.5	HNH endonuclease DNA binding domain

CTD, C-terminal domain; PDB, Protein Data Bank; RMSD, root mean squared deviation between C_α atoms coordinates of aligned query and target structures; Z, DALI Z-score.

Structural similarity between toxins-antitoxins and Acrs

A notable trend that we identified among the Acrs is structural similarity to bacterial toxins-antitoxins (TA). AcrIB from Leptotrichia buccalis C-1013-b prophage³⁰ shares nearly identical folds with a group of GNAT toxins with N-Acetyltransferase activity, such as TacT, ItaT, KacT, and AtaT, with sequence identity of 17–20% (Fig. 1A and Table 1, No. 1).

FIG. 1.

Superimposed structures of AcrIB and AcrVA5 with acetyltransferases. (A) AcrIB and TacT, (B) AcrVA5 and Nourseothricin acetyltransferase. Acrs are colored purple, and acetyltransferases are yellow.

The fold of these acetyltransferases, as well as the fold of AcrIB and AcrVA5, belong to the superfamily of N-acyltransferases (SCOP ID: 4001800), which adopts a mixed α/β structure with six β-strands in the center and α-helices surrounding the β-sheet. GNAT N-Acetyltransferases are the toxins of type II TA system that are widely spread among bacteria and inhibit translation by acetylating the α-amino group of amino acids in aminoacyl-tRNAs.³¹ An example is TacT, which controls Salmonella typhimurium growth through this mechanism.³²

Many residues that comprise the Acetyl-CoA binding site are conserved in AcrIB, including L117, F168, and F167, the mutation of which in TacT2 (Y143F) completely abolished the toxicity in Salmonella³² (Supplementary Fig. S2). Thus, it appears most likely that AcrIB is an active acetyltransferase although the acetylated substrate remains unknown.

The structural core of AcrIB, which consists of five β-strands and two neighboring α-helices, is also structurally similar to AcrVA5 (Z-score 8.2, RMSD 2.4 A) although the sequences are only 16% identical. AcrVA5, found originally in Moraxella bovoculi prophages,³³ is an acetyltransferase that inhibits the target DNA binding by acetylating Cas12a, the subtype V-A CRISPR-Cas effector, at the Lys635 residue, which is crucial for protospacer adjacent motif (PAM) recognition.³⁴ AcrVA5 shares high structural similarity with acetyltransferases (Fig. 1B; Table 1, No. 2) and contains a conserved active site.

However, the acetyltransferases that are most similar to AcrVA5 are not toxins. For instance, nourseothricin (NTC), produced by Streptomyces noursei, is a streptothricin-class aminoglycoside antibiotic that targets protein synthesis. NTC acetyltransferase acetylates the beta-amino group of the beta-lysine residue inactivating NTC.³⁵ Therefore, it appears likely that, although AcrIB and AcrVA5 are homologs, these proteins evolved from distinct acetyltransferases that were independently recruited by viruses for anti-CRISPR activity, and only ancestors of AcrIB were toxins.

AcrIF11 from a Pseudoalteromonas amylolytica prophage³⁶ is structurally similar to the RES toxin, a component of the RES-Xre type II TA system (Fig. 2A; Table 1, No. 3). Activation of this toxin leads to depletion of intracellular NAD+ level, inhibiting cell growth.³⁷ The same NAD-cleaving domain is also present in the toxin of the ParST TA (Table 1, No. 4).³⁸

FIG. 2.

Superimposed structures AcrIF11 with (A) RES toxin, (B) Cholix toxin and (C) Diphtheria toxin. AcrIF11 is colored purple, and toxins are colored yellow.

Besides the NAD+-cleaving RES toxin of type II TAs, AcrIF11 structures also showed similarity to the Cholix toxin from Vibrio cholerae (Fig. 2B; Table 1, No. 5) and Diphtheria Toxin from Corynebacterium diphtheriae (Fig. 2C; Table 1, No. 6). These bacterial toxins are members of a diphthamide-specific family of ADP-ribosyl transferases that possess specific activity against eukaryotic translation elongation factor 2 (eEF2).^39–41

Both toxins are large proteins that, in addition to the catalytic domain that is similar to AcrF11, contain a receptor-binding domain, and an anti-parallel β-jellyroll domain. Similar to the AcrVA5, AcrIF11 inhibits dsDNA binding by the effector complex of subtype I-F. In this case, AcrIF11 ADP-ribosylates the Asn250 residue of Cas8, the large subunit of effector complex, which is crucial for PAM recognition.³⁶ The structure of the binding site is conserved although not all the key amino acids are retained.

For instance, in AcrIF11 compared with Diphtheria toxin (PDBID: 1TOX), histidine H21 is conserved; two tyrosines, Y54 and Y65, are substituted with phenylalanine (F32) and histidine (H43), respectively, and glutamate E148 is substituted with a serine (S146).

AcrVIA6 from Rhodobacter capsulatus prophage³⁰ and the homologous N-terminal domain of AcrIIA1 showed structural similarity to Antitoxin HipB. The mechanism of HipA-HipB TA system is unusual among the type II TAs. HipA autophosphorylates and, in the phosphorylated state, forms a HipA-HipB-DNA complex in which the HipB dimer binds dsDNA (PDB: 5K98).⁴² HipB consists of a small helix-turn-helix (HTH) domain where four alpha helices interact with DNA and share the domain architecture with AcrVIA6 (Fig. 3A; Table 1, No. 7).

FIG. 3.

Structures of HipB antitoxin (yellow) superimposed with (A) AcrVIA6 and (B) N-terminal domain of AcrIIA1 (purple).

AcrIIA1 from a Listeria monocytogenes prophage⁴³ consists of two domains, with the N-terminal HTH domain also similar to HipB and HigA (Fig. 3B, Table 1, No. 8), and the C-terminal HTH domain similar to VirB and PrgX (Table 1, Nos. 9–11). HigA is a small HTH protein that binds to its own operator to repress transcription of the higBA operon on the Rts1 plasmid of Proteus vulgaris.^44,45

The apparent homology of AcrIIA1 to HigA has been previously reported.⁴⁶ AcrIIA1 is also similar to PrgX and VirB, which activates the transcription of virulence and conjugation genes.⁴⁷ Although the similarities between proteins with the simple HTH fold potentially could be generic and not indicative of homology, in these cases, the low RMSD (Table 1) and the near perfect superposition appear to reveal an ancestral relationship.

AcrIF9 from Vibrio parahaemolyticus prophage⁴⁸ showed significant structural similarity to Dmd, the discriminator of mRNA degradation of Escherichia Phage RB69 (Fig. 4; Table 1, No. 12). Dmd interacts with RnlA, the toxin of the type II TA system RnlAB, the activation of which results in the phage mRNA degradation.⁴⁹ Both proteins consist of a single domain with a TBP-like fold (SCOP ID: 2000457), which, in the case of Dmd, mimics the antitoxin RnlB and forms a stable complex with RnlA preventing its activation.^50–52 AcrIF9 also inhibits dsDNA binding to the effector complex, in this case, in contrast to AcrIF11 and AcrVA5, simply, by sterically blocking crRNA–dsDNA duplex formation.⁵³

FIG. 4.

Structure of AcrIF9 (purple) superimposed with the Dmd discriminator of mRNA degradation (yellow).

AcrIIC3 from Neisseria meningitidis prophage⁵⁴ is structurally similar to VapD toxin of the VapXD TA of Haemophilus influenzae and other mRNA interferase toxins with the RNA recognition motif fold as well as Cas2 (Fig. 5; Table 1, No. 13), as previously noticed.⁵⁵ Cas2 is a structural subunit of the CRISPR adaptation complex, but it also exhibits endoribonuclease activity in most of the CRISPR systems.^56,57 Based on the homology with VapD, it has been suggested that Cas2 was a toxin to which Cas1 would serve as the antitoxin,⁵⁸ but so far there is no experimental evidence of Cas2 toxicity.

FIG. 5.

Acrs homologous to Cas proteins. (A) Structure of AcrIIC3 superimposed with Cas2 and (B) Structure of AcrIA superimposed with Dna2. Acrs are colored purple, and Cas2 (A) and Dna2, a homolog of Cas4 with the highest structural similarity to AcrIA (B), are colored yellow.

Notably, AcrIA from Sulfolobus virus Ragged Hills⁵⁹ is a homolog of another Cas protein, Cas4, a PD-DExK family exonuclease that participates in adaptation in several CRISPR-Cas subtypes.^60,61 However, it has been shown that AcrIA inhibits spacer acquisition, thus acting as an antagonist of Cas4.⁵⁹ AcrIA shares high structural similarity with the C-terminal domain of DNA replication factor Dna2, which has ATP-dependent nuclease activity,⁶² and to several other exonucleases, including human Exonuclease V and Exodeoxyribonuclease 8 (Fig. 5B; Table 1, Nos. 14–16).

Proteins involved in bacterial SOS response

Several Acrs are structurally similar to proteins that are directly or indirectly involved in the DNA damage-inducible SOS response.⁶³ In particular, AcrIIA21 from Streptococcus agalactiae prophage¹⁸ shares a closely similar fold with the N-terminal DNA binding domain of LexA (Fig. 6A; Table 1, No. 17), which belongs to the superfamily of the winged-HTH DNA-binding domain (SCOP ID: 3000034).⁶⁴ AcrIIA21 shares an even closer similarity with the C-terminal domain of RepA, which is homologous to the N-terminal domain of LexA (Fig. 6B; Table 1, No. 18) and mediates oligomerization of this protein.⁶⁵

FIG. 6.

Acrs with structural similarity to proteins involved in SOS response. (A) Structure of AcrIIA21 with LexA, (B) Structure of AcrIIA21 superimposed with RepA, (C) Structure of AcrIIA11 superimposed with PsiB, and (D) Structure of AcrIC7 superimposed with DinI. Acrs and their accessions are colored purple; acetyltransferase and their PDB IDs are colored yellow.

AcrIIA11 from a Clostridium sp. prophage from human gut metagenome⁶⁶ is similar to the bacterial conjugation factor PsiB, which interacts with RecA, another critical component of SOS response (Fig. 6C; Table 1, No. 19). PsiB is a negative RecA regulator that binds free RecA and prevents the interaction of RecA with DNA during F plasmid conjugation.⁶⁷

In addition, AcrIIA11 is similar to the C-terminal domain (CTD) of the mediator of RNA polymerase II transcription subunit 17 (Table 1, No. 20). AcrIIA11 interacts with dsDNA, but its mechanism of CRISPR inhibition remains unclear.⁶⁶

AcrIC7 from Pseudomonas stutzeri prophage⁶⁸ is similar to another RecA inhibitor, DinI (DNA-Damage-Inducible Protein I) (Fig. 6D; Table 1, No. 21). DinI, which is widely conserved across bacteria, inhibits RecA and regulates the cellular response to DNA damage during SOS response.^69,70 AcrIC7 and DinI are small, single-domain proteins and share a distinct, two-layer fold (SCOP ID: 2000700) consisting of three β-strains and two α helixes. AcrIC7 apparently inhibits type I-C systems by blocking target DNA binding.⁶⁸

Other proteins structurally similar to Acrs

Phage head-to-tail adaptor

As previously noticed, AcrIIA8 from human gut metagenomic libraries is highly structurally similar to head-to-tail adaptor/stopper proteins from phages and gene transfer agents⁷¹ (Table 1, Nos. 22–26). These proteins form a hexamer ring in the neck of the virus particle and presumably function as stoppers that lock viral DNA in capsids. These proteins adopt a split barrel-like fold (SCOP ID: 2000543) containing six main beta-strands and additional secondary structure elements on the sides of the barrel.⁷²

DNA binding domain of HNH homing endonuclease

The C-terminal domain of AcrVA3 identified in M. bovoculi prophages⁷³ is structurally similar to the C-terminal domain of HNH homing endonuclease I-HmuI (Table 1, No. 27). This domain interacts with the major groove of DNA and is similar to the functionally analogous domain of the GIY-YIG homing endonuclease of I-TevI.⁷⁴

The N-terminal domain of AcrIIA7 from human gut metagenomic libraries⁷¹ is even more closely related to the N-terminal domain of another HNH endonuclease (UniRef accession: UPI000DCACB76) from Latilactobacillus sakei, which is an uncharacterized domain with PGDYG motif (PF14083) as demonstrated by an HHpred. However, different members of the AcrIIA7 family contain distinct N-terminal domains or even lack such domains.

Discussion

The Acrs are highly diverse, typically, small proteins that evolve rapidly, apparently, due to the arms race with the CRISPR systems. Because of this rapid evolution, protein sequence comparisons are not particularly informative toward elucidating the origins of Acrs. Moreover, it is hardly surprising that even structural comparisons using DALI, the most powerful current approach for detecting homologous relationships among proteins on the basis of significant structural similarity, reveal similar structures only for a small fraction of the currently known Acrs.

Nevertheless, even these limited observations present evidence that Acrs have highly diverse origins as demonstrated by the detection of structural similarity to a variety of unrelated, functionally diverse proteins. Formally, searches for structural similarity allow inference of homology when the observed similarity is highly significant, but not direct inference of the directionality of evolution that has to be explored separately.

However, with regard to the Acrs, two lines of reasoning suggest the origin of these virus encoded proteins from the host homologs. First, the recruitment of host proteins for functions in virus reproduction is the mainstream route of evolution whereas recruitment of viral proteins for host functions is far rarer, however important.^75,76 Second, in each case when we detected homologs of Acrs by structure comparison, the homologous proteins, in particular, toxins and antitoxins, were widespread in bacteria and/or archaea, whereas the Acrs were limited to a narrow group of viruses.

Therefore, it appears that in all these cases, the origin of Acrs from the corresponding cellular homologs is by far the most likely evolutionary scenario. These observations are compatible with exaptation of various host proteins for the anti-CRISPR function by bacterial and archaeal viruses on multiple, independent occasions.⁷⁵ It should be noted, however, that a different route of Acr evolution, by serial duplication of an Acr gene, has been reported as well, in particular, in archaeal viruses, some of which encompass families of paralogous Acrs.⁷⁷

Even if limited in scale, analysis of the detected structural similarities for Acrs reveals notable trends. The most prominent of such observations is the apparent origin of six unrelated Acrs (AcrIB, AcrIF9, AcrIF11, AcrIIA1, AcrVA5, and AcrVIA6 [it should be noted that the activity of AcrVIA6 and other Acrs purported to target type VI CRISPR systems has not been reproduced in an independent study⁷⁸]) from different bacterial toxins and antitoxins, mostly, from components of type II TA systems.

This trend falls within the framework of the “guns for hire” concept whereby homologous proteins, in particular, nucleases, are employed for defense by cellular life forms and for counter-defense by mobile genetic elements (MGEs).⁷⁹ A plausible evolutionary scenario seems to be that viruses first capture TA systems that can be readily repurposed to inhibit the corresponding host toxins and are subsequently more radically exapted for the Acr functions.⁷⁵ Moreover, toxins and antitoxins that typically are small, compact proteins appear “pre-adapted” for the recruitment as Acrs.

Another category of proteins that appear to have spawned several Acrs are components of the DNA damaged-induced SOS repair pathway.⁶³ Again, after the initial capture by viruses, these proteins might have acted as SOS pathway regulators, before being repurposed for CRISPR inhibition.

Evolution of many viruses, particularly, those infecting eukaryotes, such as poxviruses, involved capture of components of host defense systems that act as specific decoys (dominant negative inhibitors) of the respective antivirus mechanisms.⁷⁵ It could be tempting to hypothesize that co-option of CRISPR systems components or their homologs in the decoy capacity was a major source of Acrs.

However, our analysis identified only two cases, those of AcrIIC3 (Cas2 homolog) and AcrIA (Cas4 homolog) that seem to fit this scenario. Previously, it has been reported that AcrIF3 structurally mimicked the “nuclease recruitment helix” of the Cas8f effector protein, interacting with the Cas2/3 nuclease of the I–F system and preventing the formation of the recruitment of Cas2/3 for interference.⁸⁰

The similarity between AcrIF3 and Cas8f failed to come up in our structural searches, suggesting that, in this case, molecular mimicry might reflect convergence rather than homology. Generally, the recruitment of Cas protein for the Acr function might be hampered by the (relatively) large size of most of the former. Small proteins, such as toxins and antitoxins, could represent a more facile pool for Acr evolution.

Finally, some Acrs appear to have evolved via “intramural exaptation,”⁷⁵ that is, repurposing of viral proteins with other, in particular, structural roles, such as head-to-tail adaptors. In this case, the likely evolutionary scenario is duplication of the gene encoding a virus structural protein followed by neofunctionalization into an Acr that was accompanied by substantial sequence and structural divergence.

Apart from these observations, a substantial majority of the Acrs remain “dark matter” with obscure origins. Given that most Acrs are small and fast evolving proteins, it seems likely that the origins of some of these unknowns will become traceable once more viral genomes are sequenced so that these Acrs form large families providing for higher quality of structure modeling.

Much of the dark matter, however, might never be illuminated, either because host proteins undergo major structural rearrangements while being neofunctionalized as Acrs or because Acrs evolve de novo from non-coding sequences; the de novo origin is particularly likely for the smallest Acrs. It seems likely that both these routes of evolution contributed to the origins of Acrs and other viral proteins involved in counter-defense.

Authors' Contributions

H.S. collected the data and performed the analysis; H.S., K.S.M., and E.V.K. analyzed the results; and H.S. and E.V.K. wrote the manuscript that was edited and approved by all authors.

Author Disclosure Statement

No competing financial interests exist.

Funding Information

The authors' research is supported through the Intramural research Program of the National Institutes of Health of the USA (National Library of Medicine).

Supplementary Material

Supplementary Figure S1

Supplementary Figure S2

Supplementary Table S1

Supplementary Table S2

Supplementary Table S3