Evolutionary conservation and lineage-specific expansions of HEPN domains helps predict novel RNA processing and defense systems in eukaryotes

In eukaryotes the distribution of HEPN domains shows two distinct patterns. One group of HEPN families is strongly conserved across all major eukaryotic lineages implying that they were present in the last eukaryotic common ancestor (LECA). This group includes the KEN domains found at the C-termini of serine/threonine kinase domains in Ire1-like proteins, Las1, and the family prototyped by the human protein C6orf70 (DUF4209). The KEN domain is a RNase that is involved in the degradation of rRNAs, mRNAs associated with the endoplasmic reticulum (ER) membrane, and spliceosome-independent splicing as part of the cellular response to the accumulation of unfolded proteins in the ER [53, 75, 76]. Thus, the emergence of the KEN domain appears to have been linked to the origin of the eukaryotic endomembrane system. The C6orf70 family, which we predict to be a catalytically active HEPN domain protein (Figure 1), similar to the Ire1-like proteins, contains a single transmembrane (TM) region and is predicted to localize to the ER membrane (Figure 4). Thus, we predict that, similar to Ire1, these proteins also function in the degradation of RNA at the ER membrane, perhaps as part of the misfolded protein response or similar stress-related regulatory processes. The identification of a HEPN domain in Las1 helps clarify key steps in the remarkably complex, eukaryote-specific processing of the ITS2 linker between the 5.8S and 25S/28S rRNAs in their common precursor [55, 77]. Las1 copurifies with several exoRNases, and cooperates with the exosome and other exoRNases in processing the ITS2 linker to release the mature rRNAs [55]. However, the identity of the endonuclease required for initiating this processing event remains unknown. Based on the presence of intact catalytic residues in the HEPN domain of Las1, we predict that this protein functions as the endoRNase that makes the two initial breaks in this processing event.

The Swt1 endonuclease family, although not confidently traceable to the LECA, is inferred as being present in the common ancestor of animals, plants and fungi (Table 1). This version of the HEPN domain is fused to an N-terminal PIN endoRNase domain (Figure 4) and might be catalytically inactive due to loss of the conserved motif. Hence, we infer that this HEPN domain might have a role in binding and sensing unspliced pre-mRNAs that are specifically targeted by the Swt1 nuclease at the nuclear envelope [56].

The KEN and Las1 families, although traceable to the LECA, do not show specific relationships with any prokaryotic HEPN domains. Conceivably, these quintessentially eukaryotic variants of the HEPN domain originated by rapid sequence evolution from either a bacterial or an archaeal HEPN ancestor at the stem phase of eukaryote evolution antedating the LECA. In contrast, both the C6orf70 and Swt1 families include catalytically active bacterial versions (Table 1; see below), pointing to their origin via lateral transfer at different points in eukaryote evolution.

The other prominent trend in eukaryotes is the independent lineage-specific expansion (LSE [79]) of several families of HEPN domains. The KEN domain, separately from the N-terminal kinase domain found in Ire1-like proteins, shows independent LSEs in the tunicate Oikopleura dioica and several monocot plants, such as rice. Certain eukaryotes, such as the dictyostelid slime molds, the heterolobosean amoeboflagellate Naegleria and the crustacean Daphnia, possess a distinct, catalytically active HEPN domain of the Swt1 family, which is more closely related to the bacterial versions than the conserved version in Swt1 (Additional file 1). In Daphnia this version underwent a massive LSE with 46 distinct copies in the genome. A representative of the paREP1/8 family that appears to be of ultimate crenarchaeal origin was acquired by the nematodes of the genus Caenorhadbditis, where it underwent independent LSE in C. elegans and C. remanei. This pattern of multiple independent LSEs is common among eukaryotic genes with immunity- or defense-related functions: the multiplicity of diverged paralogs provides the means for maximizing the diversity of recognized targets [79–81]. Thus, it appears likely that these LSEs of HEPN domains play specific roles in defense responses. The expanded KEN and Swt1 family HEPN domains are inferred to be catalytically active and can be predicted to function as defensive endoRNases that might be directed against different viral RNAs. Several of the versions from the LSEs in the Caenorhadbditis species might not be active. Hence, we propose that these proteins function as receptors for targeted RNAs rather than as active RNases. Some of the nematode HEPN domains are fused to an inactive ATPase domain closely related to the N-terminal ATPase domain of the LAF-1/Vasa-like RNA helicases (Figure 4). Given that these helicases are components of the P granules of germline precursor cells, which contain RNA-interacting components that silence pseudogenes and transposons [82, 83], it is possible that these HEPN proteins form a line of defense for the germline genome against selfish elements.

Gene-neighborhoods and protein domain architectures suggest that HEPN domains function in multi-pronged defense jointly with prokaryotic restriction-modification systems

The identification of the nuclease domains of PrrC and RloC as HEPN domains is of considerable interest because these nucleases are deployed as part of a multi-pronged defense strategy against the enterobacteriophage T4 (and most likely against other, related phages). Although PrrC and RloC are both anticodon nucleases (ACNases), which target tRNA^Lys of the host cell to inhibit translation during the T4 infection, each of these endoRNases has distinct biochemistry. While PrrC merely cleaves the anticodon loop, RloC excises the wobble nucleotide of tRNA^Lys[50, 59], thereby preempting the RNA-ligase-dependent phage counter-strategy. These endoRNases are part of fine-tuned defense systems that are regulated via interactions with domains in the same polypeptide and/or other proteins encoded in the same operons and whose potentially self-harming activities are deployed only at opportune moments during phage infection [52]. In PrrC and RloC the C-terminal HEPN domain is combined with N-terminal SbcC/Rad50-like ABC NTPase domains [84] (Figure 4) which regulate the activity of the nuclease domain in a manner dependent on NTP hydrolysis (in both) or sensing nucleotides (dTTP in the case of PrrC). Furthermore, PrrC is embedded in a gene-neighborhood that also encodes the three subunits (hsdMSR) of a type Ic R-M system, PrrI. This R-M system, which interacts with PrrC to keep it in a catalytically inactive state, functions as the first line of defense against the phage [51]. However, when T4 inactivates the PrrI R-M system by deploying the Stp anti-restriction peptide that is conserved in T4-like phages, or when the levels of dTTP or unmodified DNA increase, PrrC is relieved of its negative regulation [52] and steps in as a second-line of defense against the virus by inactivating tRNA^Lys. In contrast, RloC is not linked to any R-M system but is normally kept in an inactive state by its own N-terminal ABC-ATPase domain. The HEPN nuclease domain of RloC appears to be activated when the conformation of the ABC-ATPase domain is modified in response to DNA damage from genotoxic stress induced by the virus.

The results of these studies imply that analysis of the gene neighborhoods and domain architectures of the prokaryotic HEPN domains might help uncover multi-pronged defense strategies that evolved through the arms race between viruses and their hosts [4]. Our current analysis showed that at least 16 distinct clades of HEPN domain proteins are encoded by genes that are linked to a diverse array of R-M systems through conserved gene neighborhoods (Table 1 and Figure 5). These associations are primarily represented in bacteria where they comprise one of the most common genomic contexts of HEPN genes (Figure 3). By analogy to the PrrC-PrrI linkage, we propose that these associations between R-M systems and HEPN domains represent different multi-pronged defense strategies. A subset of RloC-like ABC-HEPN proteins are encoded within mobile gene-neighborhoods that in addition to genes for R-M components, also encode a toxin of the DOC superfamily [35] (Figure 5). The DOC domains function by NMPylating serines and threonines in target proteins and are contained in a broad variety of toxins including TA systems, polymorphic toxins and secreted effectors of pathogens [2, 85]. These genomic associations suggest that the respective defense systems exercise a three-level defense strategy which targets invading DNA via the R-M system, RNA via the HEPN protein, probably by inhibition of translation, and proteins via the DOC toxin. In a similar vein, we found that some PrrC-like proteins are encoded by genomic loci that combine genes for R-M system components and those for RhuM-like proteins, which were previously observed in pathogenicity islands of Salmonella[86]. In these gene neighborhoods the RhuM-like protein occupies a position similar to that of the DOC toxin in the neighborhoods discussed above, and indeed the RhuM-like domain is often fused to the DOC domain. Based on this association, we propose that RhuM is also a toxin domain that might function via protein modification as part of a multilevel defense program, jointly with the PrrC-like and RM proteins. We also found that several HEPN domains of the Ymh (Pfam: PF09509) family are fused to the C-termini of ATPases of the GHKL superfamily, known as paraMORCs [87], in proteins encoded by genes embedded in R-M system gene neighborhoods. The paraMORC domains, although unrelated to SbcC/Rad50 ABC-ATPases, appear to function analogous to the latter in both R-M and other contexts [88, 89]. Hence, we propose that these Ymh proteins represent an independent emergence of a domain architecture that is functionally analogous to PrrC and RloC.

Several families of HEPN domains display independent fusions to one or more of four distinct families of endoDNase domains found in R-M systems, namely domains of the REase fold (on multiple independent occasions), HNH/EndoVII fold, ParB-like fold and HKD/phospholipase D fold (Table 1 and Figures 4 and 5). In addition, we identified multiple, independent fusions of the HEPN domain with SWI2/SNF2 helicases, EcoEI-like superfamily(SF)-II helicases and SF-I helicases, which are the helicase subunits found in several distinct R-M systems (Figure 4). In one such group of giant proteins (e.g. gi: 380301041 from Brachybacterium squillarum), in addition to a fusion to the HEPN domain, the SF-I helicase is also fused to a transglutaminase-like peptidase, a REase fold DNase of the very short patch repair (Vsr) family and winged helix-turn-helix (wHTH) domains (Figure 4) [90–92]. In another class of R-M systems, a HEPN domain of the Abi2/SWT1 family is fused to a distinct version of the AAA + ATPase domain (currently labeled DUF499 in the Pfam database, Figure 4). Given that these RM systems typically possess a distinct helicase subunit (Figure 5), we propose that the AAA + domain fused to HEPN functions as an accessory subunit required for DNA-looping, analogous to the AAA + protein GTPase McrB in the McrBC system [93]. The above observations indicate that HEPN domains, associated with R-M systems, fuse only to restriction endonucleases, helicases and other ATPase subunits but not to the methylases. These multiple, convergent fusions imply strong selection for functional linking of the HEPN domains with DNases that cleave the target DNA and other enzymes that facilitate cleavage but not the DNA-modifying enzymes. Thus the functional analogy with PrrC is likely to extend to the HEPN domains that are associated with R-M systems. Specifically, the RNase activity of these HEPN domains is reversibly inhibited by the associated R-M system subunits but is released from this block when the R-M system is neutralized by a virus counter-strategy or in response to a genotoxic stress signal indicating that the defensive capacity of the R-M system is overwhelmed [4]. The above mentioned systems comprised of giant proteins containing HEPN, transglutaminase, SF-I helicase, Vsr DNase and wHTH domains entirely lack associated genes for DNA-modification subunits. Hence these proteins are likely to function independently of any modification, probably by directly recognizing invading DNA via their C-terminal wHTH domains. As in the case of the regular R-M systems, here too the RNase activity of the N-terminal HEPN domain is probably deployed for suicidal action if the associated DNase activity fails against the invading DNA. The fusion to the transglutaminase domain suggests that a further line of defense might involve protein cleavage catalyzed by this domain.

HEPN domains in bacterial RNA-based defense systems

In contrast to PrrC and RloC, the HEPN domain proteins RNase LS and LsoA, which also constitute distinct anti-phage T4 defense systems, are indiscriminate mRNases that cleave both free and ribosome-associated transcripts [49, 94]. Although these endoRNases can degrade host mRNA, they appear to be primarily directed against viral mRNAs. Both RNase LS (RnlA) and LsoA are typically kept in an inactive state via physical interaction with the unstable products of the respective upstream genes, the RnlB proteins [48, 49]. However, when phage T4 inhibits the production of host proteins, the RnlB proteins are removed through degradation, unleashing the RNase activity of the HEPN domain [48]. In this regard, the RnlAB system resembles Type-II TA systems some of which are deployed as defense mechanisms against phages such as P1 [48, 95]. Thus, RnlAB appears to be a defense system that primarily functions at the RNA-level rather than in conjunction with any DNA-level restriction system. In most cases, the RNase LS family HEPN domain is fused to an N-terminal caulimovirus-like RNase H fold domain (e.g. gi: 335428883 from Haloplasma contractile; Figure 4), which in the E. coli RNase LS and LsoA is interrupted by a stop codon, leaving HEPN as the only active nuclease domain. The presence of this RNase H module (in particular its N-terminal caulimovirus viroplasmin domain; Figure 4) suggests that these RNase LS family proteins specifically target RNA in DNA-RNA duplexes, perhaps priming intermediates of viral replication or transcription initiation sites. Other RNase LS family HEPN domains are fused to an N-terminal TBP (TATA Binding Protein)-like domain (e.g. gi: 336234563 from Geobacillus thermoglucosidasius; Figure 4), similar to that fused to an RNase III-like domain in RNase HIII [96]. Given that in RNase HIII this TBP-like domain is involved in binding DNA-RNA hybrids [97], this fusion is additional evidence that a subset of the RNase LS family HEPN domains indeed target RNA in DNA-RNA duplexes.

In addition to the RNase LS family, we identified several other fusions between (predicted) catalytically active HEPN domains and other active RNase domains resulting in “two-headed RNases”. A case in point is the fusion of HEPN with a C-terminal RNase III and a dsRBD domain (e.g. gi: 119489560 from the cyanobacterium Lyngbya) (Figure 4). Given the specificity of RNase III and dsRBD toward RNA-RNA duplexes (e.g. in Dicer, a key component of the eukaryotic RNAi system) [98], it appears likely that these bacterial proteins cleave dsRNA targets, with multiple cleavages catalyzed by the HEPN and RNase III domains. Similarly, a distinct family of HEPN domains (e.g. gi: 389884779 Microcystis aeruginosa; Figure 4), which is distantly related to AbiF and AbiD (see below), shows fusions to the endoRNase L-PSP domain that is known to cleave mRNAs [99]. Hence, these HEPN proteins might also target mRNAs analogously to the members of the RNase LS family.

In addition to the fusions within a single multidomain protein, we identified three groups of HEPN proteins encoded in gene-neighborhoods that also contain a gene coding for an uncharacterized conserved protein (Table 1 and Figure 5). Sequence profile searches showed that this uncharacterized protein contained a conserved domain that it is also present in the Photorhabdus luminescens nematicidal toxin NamA [100]; accordingly, we named it the NamA domain. Profile-profile comparisons using the HHpred program indicated that the NamA domain contains a novel version of RNase H fold with two large inserts within the conserved core of the fold (HHpred probability 82%; Additional file 1). Nevertheless, the NamA domains retain all the key active site residues that are required for the ribonuclease activity of RNase H. Thus, these proteins are likely to be RNA-cleaving toxins. The NamA genes also co-localize, either with or without HEPN genes, with a gene coding for a KorC-like DNA-binding HTH domains, which might again point to an activity towards DNA-RNA hybrids. The NamA-HEPN gene-neighborhoods could represent yet another example of HEPN domains functioning in conjunction with other RNases. This preponderance of functional associations between multiple RNases might be indicative of a strategy of multiple cuts (as observed in the case of RloC) employed by prokaryotes to circumvent phage RNA ligase-dependent repair systems that can easily restore RNAs with single endonucleolytic breaks [51]. Furthermore, such combinations of RNases could be involved in cleavage of RNAs with complex secondary structures.

Major role for HEPN domains in abortive infection systems

The abortive infection (Abi) systems, of which over 22 distinct versions were first characterized in Lactococcus lactis (labeled AbiA to AbiV), represent anti-bacteriophage strategies that limit the spread of infection [101]. Homologous Abi-like systems are found across a wide range of bacterial lineages [102]. They appear to act both by directly targeting phage components and by causing suicide of the infected host before the release of progeny virions [101, 103]. Thus, the Abi systems seem to implement a multi-layer defense strategy that is generally analogous to that of the HEPN-RM system combinations. Here we show that the primary components of 6 Abi systems define distinct groups of HEPN domains, namely the AbiA-C-terminal domain (AbiA-CTD), AbiD (including AbiD1), AbiF, AbiJ, AbiU2 and AbiV families (Table 1). Furthermore, though the originally identified Lactococcus lactis AbiTii protein lacks a HEPN domain, its homologs from several bacteria are found fused to two distinct C-terminal HEPN domains namely of Ymh (gi: 372210551) and the c2405 (gi: 358446093) families. The mode of action of these Abi proteins has remained largely enigmatic to date. The detection of HEPN domains suggests a unified mechanism for their action, based on the predicted RNase activity. For example, AbiD1 has been shown to be toxic to the host cell but also to interfere with the activity of the RuvC-like Holiday junction resolvase of phage bIL66 [92, 101]. Furthermore, L. lactis AbiD1 induces cell death at suboptimal temperatures and is also toxic in heterologous systems such as E. coli[101]. Based on the identification of a HEPN domain in AbiD1, we propose that the wide-range toxicity of this protein is a consequence of its RNase activity. The AbiA and AbiK proteins abrogate the maturation of phage P335, primarily by inhibiting the phage-encoded Erf/Rad52-like single-strand annealing proteins via untemplated synthesis of a DNA molecule that is covalently linked to the reverse transcriptase domain [104, 105]. Although the mechanisms and the targets are completely different, the activity of these proteins is comparable to that of AbiD1, in that both inhibit phage recombination. The detection of a C-terminal HEPN domain in the AbiA proteins suggests that it might also promote cell suicide mediated by the RNase activity of HEPN. AbiF causes delayed DNA replication of phage 936, possibly by interfering with replication initiation [106]. Taken together, all these observations suggest a general two-pronged mode of action for the Abi systems: i) diverse interactions with bacteriophage components resulting in inhibition of phage reproduction and ii) host cell suicide through RNA degradation mediated by the HEPN domains.

The recognition of this general mode of action for the HEPN-containing Abi systems also leads to a hypothesis on the functions of 5 novel protein domains that we detected at the N-termini of different groups of proteins of the AbiJ family and the conserved domain in AbiTii (Table 2, Figures 3 and 4). We named the former domains, which could not be unified with any previously known domains, AbiJ-NTD1-5 (after N-Terminal Domain; Additional file 1). All these domains are predicted to adopt α + β folds, with AbiJ-NTD1, AbiJ-NTD3 and AbiJ-NTD4 displaying similar predicted secondary structures (Additional file 1). Likewise, the AbiTii domain is a novel α + β globular domain with a highly conserved glutamate (Table 2, Additional file 1). We found that, in addition to being fused to the AbiJ family of HEPN domains, the AbiJ-NTD domains also occur in other proteins at the N-termini of several enzymatic domains, thereby presenting an architectural analogy to the AbiJ family. These fusions include other HEPN domains (Ymh family with AbiJ-NTD1 and AbiJ-NTD2; DUF4145 family with AbiJ-NTD1), MRR-type REase fold domains (with AbiJ-NTD1 and AbiJ-NTD2), another REase fold DNase (DUF2726 in Pfam with AbiJ-NTD3), HKD/phospholipase D fold nucleases (with AbiJ-NTD1 and AbiJ-NTD2), TIR domains, some of which might possess nuclease activity [87] (with AbiJ-NTD2, AbiJ-NTD3 and AbiJ-NTD5), protein kinases (with AbiJ-NTD1 and AbiJ-NTD3), nucleoside 2-deoxyribosyltransferases (with AbiJ-NTD1) and a distinct group of ABC-ATPase domains (with AbiJ-NTD3). Although members of the MAE_18760 family of HEPN domains (Table 1) have not been recovered among currently known Abi systems, they display architectures similar to those of Abi proteins, with fusions to three distinct NTD domains. One of these is the previously described Schlafen domain which is also found fused to other domains implicated in intra-genomic conflicts and has an important anti-viral role in metazoans [107, 108] (AMB and LA unpublished observations). The second is the novel HEPN/Toprim-NTD1 domain (Table 2 and Additional file 1) that, in addition to the aforementioned HEPN domain fusion, is also fused to another family of HEPN domains (ERFG_01251; Table 1), a TOPRIM domain nuclease that is found in several defense related-contexts [109, 110] (AMB, VA and LA unpublished observations), and a REase fold DNase domain of Mrr family. A third NTD (Table 2, Additional file 1), in addition the HEPN domain, also occurs fused to RES, which a toxin domain found in several type-II TA and polymorphic toxin systems, and is predicted to function as an RNase [2].

Thus, there is a strong tendency toward multiple, independent fusions of these NTDs with both RNases and DNases, as well as other catalytic domains such as protein kinases that might also function as toxins (Figure 3). Although the nucleoside 2-deoxyribosyltransferase domain is not a nuclease, it cleaves the glycosidic bond between base and deoxyribose [111]; hence this enzyme is likely to act on DNA in a manner similar to the effect of Ricin-like toxins on RNA [112]. Generally, we propose that the Abi-NTDs interact with specific targets, namely viral proteins or nucleic acids, and interfere with their functions. The C-terminal enzymatic domains of these proteins are likely to be deployed as toxins that could cause cell suicide. The NTDs also might function as antitoxins that inhibit the enzymatic activity of the C-terminal toxin domain under normal conditions, and this inhibition is relieved when the NTDs interact with viral components. Notably, AbiJ proteins lacking the NTDs more frequently occur in RM operons suggesting that they are strongly functionally coupled with RM systems as discussed above.

HEPN domains in CRISPR-Cas and other antivirus defense systems

Comparable to the Abi systems, HEPN domain proteins are also major players in Type I and Type III CRISPR-Cas adaptive immunity systems in archaea and bacteria [57]. There are 4 distinct families of HEPN proteins associated with CRISPR-Cas that all show a conserved domain architectural core comprised of a distinct N-terminal Rossmann fold domain, (CRISPR-Cas associated Rossmann fold or CARF), and a C-terminal HEPN domain (Figure 4). In several cases the CARF domains are fused to a different RNase (e.g. RelE, HD or PIN superfamily) or DNase (RecB-like REase fold nuclease) domains instead of the HEPN domain at their C-termini [4] (KSM, VA, A. Burroughs, EVK, LA, unpublished observations)). These nuclease-containing CARF domain proteins do not appear to be involved in spacer acquisition or spacer-sequence-dependent restriction of foreign nucleic acids in the CRISPR-Cas systems. Furthermore, CARF-nuclease proteins are also encoded by standalone genes and in certain cases by other potential anti-phage systems (e.g. a TerY-dependent system), independent of the CRISPR-Cas systems [113]. These parallel domain architectures clearly resemble those of the three AbiJ-NTD domains discussed in the previous section. Hence we propose that the CARF-HEPN proteins function analogously so that the CARF domain is a specific sensor for an invasive component (DNA or RNA) or an infection-induced metabolite, most likely a nucleotide derivative, whereas the HEPN domain acts as a suicidal RNase. Again, it appears likely that in the absence of the infection signal, CARF keeps the toxin activity of the HEPN domain in check. Thus, the CARF-HEPN proteins most likely function as an accessory to the CRISPR-Cas systems, being the final line of defense when the CRISPR-Cas immunity is overwhelmed.

Beyond the PrrC-like and RloC-like families of HEPN proteins, we detected several additional fusions of ABC-ATPases with C-terminal HEPN domains, e.g. those prototyped by the APECO1_4465 (gi: 117623091) protein from avian pathogenic E. coli (Table 1, Additional file 1). This group of HEPN domains is frequently found in mobile genomic islands composed of integrated prophages in several distinct bacteria (Additional file 1). A similar localization was observed for a small subset of the HEPN domains of the RloC family (e.g. gi: 209885297). Previously, prophage-encoded enzymes have been found to be an important source of anti-phage defensive mechanisms (as an extension of the mechanism to preempt super-infection) [114]. Conceivably, the prophage-encoded ABC-HEPN proteins play a comparable role in preventing infection by other phages, probably independent of R-M systems. Another group of ABC-HEPN proteins is typified by ERFG_01251 (gi: 422781011) which couples an N-terminal ABC domain with a classical HEPN domain that is more closely related to the HEPN domains associated with MNTs rather than the versions in PrrC and RloC. Sporadically, these proteins are encoded by genes embedded within CRISPR-Cas gene neighborhoods (e.g. in Neisseria and Kingella). These ABC-HEPN proteins might perform roles similar to those proposed for the CARF-HEPN proteins.

This general principle appears to be compatible with the detection of sporadic linkages between genes encoding HEPN domain proteins and some other dedicated phage resistance systems. For example, in a few instances, members of the pEK499_p136 and RloC families of HEPN domains are embedded within a large predicted operon along with genes encoding the core components of the phage growth limitation (Pgl) system that was first characterized as a defense system against lysogenic phages (e.g. phiC31) in Streptomyces coelicolor (Figure 5). The Pgl system appears to function by “reverse restriction-modification”: here the DNA of progeny virions produced by an infected cell is methylated by the Pgl system methylases and restricted upon reinfection by its DNase components [115]. In the Pgl operons the gene for the HEPN protein is combined with genes for the core Pgl system components, namely the phosphatase PglZ, the AAA + ATPase PglY and DNA methylase PglX and several other genes which might encode a thermonuclease-like RNase, an OLD family ABC ATPase and a Lon-type AAA + ATPase (Figure 5) [102, 116] (KSM, AMB, LA, EVK, unpublished). Given the delayed action of the Pgl system, it provides immunity only after the death of the initially infected cell. Accordingly, the Pgl system is likely to spring into action in advanced stages of infection after those defense mechanisms that could potentially save the cell have failed. Hence the sporadic couplings with HEPN domain and the thermonuclease could induce cell suicide but additionally or alternatively might cleave phage RNAs to limit the phage burst size. Comparable roles can be proposed for the HEPN genes (Pfam DUF4145) that are, on a few occasions, coupled with the TerD-dependent anti-phage system that also includes a McrBC-like RM system [113].

HEPN proteins in MNT-associated toxin-antitoxin systems, other mobile elements and regulatory systems

The organization of genetic elements which encode the originally identified HEPN domains and MNTs clearly resembles Type II TAs suggesting that these elements are novel TA systems [10] although their mode of action remained obscure. It has been shown that the DrrA effector of Legionella pneumophila, that is secreted through a type-IV secretion system and contains a nucleotidyltransferase-HEPN fusion, functions as a toxin that targets eukaryotic host cells [117]. The nucleotidyltransferase activity of this protein adenylates a specific tyrosine in the host Rab1b GTPase and is essential for the toxicity of this protein [118], suggesting that the MNT is the toxic moiety. However, a recent genome-scale assay for bacterial toxins implicated the HEPN domain [43]. In several Type-II TA gene dyads the antitoxin gene occupies the 5’ position upstream of the toxin gene [2, 7, 10, 35]. This operon organization ensures that the antitoxin is produced first and is available to inactivate the toxin as soon as the latter is synthesized. In the MNT-HEPN gene dyads, the MNT almost always occupies the 5’ position. Taken together with the predicted RNase activity of the HEPN domains, these observations strongly suggest that HEPN is the toxin and the MNT is the antitoxin in these distinct TA systems. The antitoxin activity of MNT might involve nucleotidylation of the HEPN toxin, possibly at a conserved tyrosine that is present in the C-terminal region of most HEPN domains associated with MNTs (Figure 1). Given that toxins and antitoxins of Type II TA systems typically strongly interact with each other, it is not surprising that the MNT and HEPN domains tightly interact to form a complex [119]. This interaction appears to have been exapted to utilize the HEPN domain as a substrate-binding or regulatory domain for the MNTs. Indeed, the HEPN domains that are fused to MNT domains within a multidomain protein typically lack the predicted RNase catalytic residues and accordingly are most likely inactive. Thus, “domestication” of former TA systems appears to have given rise to protein-modifying regulatory enzymes such as the nucleotidyltransferases, which regulate glutamine synthetase [120], a potential adenylyl cyclase and several enzymes such as kanamycin nucleotidyltransferase which are used as defense against antibiotics [42]. Under this scenario, the protein-modifying activity of the MNT domain was secondarily recruited as a toxin directed against eukaryotic proteins in the case of DrrA [117].

We also uncovered a similar but less common gene dyad that combines a HEPN gene of the MAE_18760 family (Table 1, Figure 5) with a gene coding for a ParA/Soj-like ATPase [121]. Given that the ATPase gene occupies a position equivalent to that of the MNT in the MNT-HEPN modules, we postulate that its product is likely to be the antitoxin whereas the HEPN protein is the RNase toxin of these novel TA systems. The antitoxin activity of the ParA/Soj-like ATPase could either involve a nucleotide-dependent conformational change in the HEPN protein or direct phosphorylation, which is consistent with the kinase activity observed in some members of this family [121, 122].

A further wrinkle regarding the MNT-HEPN systems relates to the mode of action of the HEPN toxins. A subset of the HEPN domains found in these systems preserve the Rx4-6H motif (in large part represented by the DUF86 family in Pfam) or have alternative histidines and and are likely to function as endoRNases, similar to toxins in various TA systems. However, no conserved, potential active site residues are found in several HEPN domains from the MNT-HEPN systems (Additional file 1 and Figure 1). Nevertheless, the genome-scale scan for toxin proteins revealed that even HEPN proteins lacking this motif are effective as toxins [43]. Unless these proteins have evolved an alternative nuclease active site (a phenomenon observed in the BECR fold of RNases that includes the RelE superfamily [2]), it is possible that these HEPN domains exert their toxic activity via a non-catalytic mode, conceivably by binding RNA and blocking translation. Such a non-catalytic, regulatory action also might be a feature of another family of HEPN domains, which we identified in this study, the MtlR family. Although some members of this family are predicted to function as active RNases (Figure 1), a large fraction is likely to be inactive on account of the loss of the conserved motif (Table 1). The gene coding for MtlR is frequently found in an operon with mannitol utilization genes, and has been shown to function as the repressor of this operon [123]. However, it has been shown that MtlR is unlikely to act as a conventional DNA-binding transcription factor and shows no detectable interaction with the promoter/operator region of the mannitol operon [124]. Hence, inactive HEPN domains of the MtlR family might function as RNA-binding proteins that repress the mannitol operon by blocking either transcription elongation or translation.

Expansion of the MNT-HEPN systems in Archaea, along with the frequent transfer of these operons to thermophilic bacteria [10], suggests these TA systems might play some role in the thermal stress adaptation. Several chromosomally encoded TA systems are important players in stress adaptation such as dormancy and stationary phase survival in various bacteria [125–127]. Therefore, the MNT-HEPN systems that are widespread in archaeal and bacterial thermophiles might perform comparable functions. One interesting possibility is that the recovery from the accumulation of unfolded proteins resulting from high temperature or low pH shock requires translational arrest that could “buy time” for the clearance of protein aggregates by chaperones and proteolytic systems. Such translational arrest could be mediated by the MNT-HEPN module when the activity of the HEPN domain is unmasked by degradation or misfolding of the MNT component. In this regard, it is of interest to note that in extremophilic crenarchaea these systems occasionally cluster with multiple MNT and HEPN genes (Figure 5). Each HEPN protein encoded in these loci might interact with a specific set of target RNAs thereby allowing a more precise regulation of the response.

This hypothesis seems to be consistent with the presence of a HEPN domain in Sacsin from animal and slime molds (Dictyostelium and Polysphondylium; Table 1). The Sacsin gene is mutated in human patients with spastic ataxia of Charlevoix-Saguenay, a degenerative disorder of the cerebellum and spinal cord [128]. Sacsin is a gigantic multidomain protein (Figure 4) that contains a N-terminal ubiquitin-like domain, three Hsp90-like ATPase modules followed by a DnaJ domain, which recruits Hsp70 [129], and a C-terminal HEPN domain. Sacsin has been shown to function as a chaperone aiding protein folding [130] but the role of its HEPN domain has been enigmatic. We hypothesize that, in addition to acting at the protein level to relieve aggregation via chaperone action, sacsin also acts at the RNA-level via the HEPN domain. The HEPN domain in Sacsin orthologs from several animals preserves the conserved motif (Figure 1); however, in organisms like humans it is lacks the conserved motif. Thus, depending on the lineage, the Sacsin HEPN domains might either act as RNases or as non-catalytic RNA-binding domains. In either case they could inhibit translation by cleaving or binding tRNA or mRNA thereby limiting the amount of unfolded protein in the cell under stress conditions. In certain animals, there are Sacsin paralogs with N-terminal DEATH domains that are major apoptosis-mediating adaptor domains (Figure 4). It is conceivable that these proteins are part of a suicidal response that is perhaps triggered by overwhelming unfolded protein stress.

We also identified HEPN domains (e.g. the integron family; PDB: 3jrt) that are associated with certain mobile elements, such as integrons, which are major vehicles in the spread of drug resistance determinants among proteobacterial pathogens [131]. The integron cassettes are known to be activated by stress conditions, thereby allowing swapping of genetic material that might be of adaptive value [132, 133]. We hypothesize that the HEPN domains present in some integron cassettes contribute to the stress response by functioning as RNases that induce dormancy by probably inhibiting translation and thus enabling survival of harsh conditions. Notably, integron cassettes often encompass also other toxin RNases such as RelE and Cas2-like proteins (VA and LA, unpublished observations) that are likely to play similar roles.

Bacterial membrane-associated HEPN domains and stimulus-dependent RNA degradation

In the present work, we identified at least three distinct groups of HEPN domains that are combined with TM segments. The first of these belongs to the family that overlaps with the Pfam DUF4145 family (e.g. gi: 153825856 from Vibrio cholerae) and contains a distinctive N-terminal domain with a single TM helix with a strictly conserved WP signature (Figures 3, 4 and Additional file 1). This TM domain is also found in several bacterial proteins where it is fused to C-terminal receiver domains (e.g. gi: 158314014) of two-component signaling systems [134] in place of the HEPN domain. A distinct group of catalytically active HEPN domains of the Abi2/SWT1 family are fused to the C-terminus of a single, well-conserved TM helix, which in turn is preceded by another conserved globular all α-helical domain (e.g. gi: 16762698 from Salmonella enterica). In the third group the N-terminal HEPN domain is separated from the C-terminal Zn-ribbon domain by a pair of TM helices (e.g. gi: 325108440 from Planctomyces brasiliensis). All these HEPN domains are predicted to be the cytoplasmic globular domains of inner membrane proteins. This localization suggests that, similar to the Ire1 and C6orf70 proteins in eukaryotes, these HEPN proteins process RNAs on the inner side of the membrane. The specialized TM segments with the WP signature and the potential external domains could act as sensors for stimuli on the cell surface, and the resulting signal could affect the HEPN domain conformation and hence RNA stability. We also identified HEPN domains that are fused to CBS domains, in some cases together with additional HD phosphohydrolase domains [135] (e.g. gi: 118580987 from Pelobacter propionicus; Figure 4). Given that the CBS domains sense nucleotides and their derivatives [136], these proteins might respond to such ligands to regulate RNA stability. Thus, sensing of cell-surface and soluble signals resulting in RNA degradation could be a poorly appreciated signaling pathway in diverse bacteria.

HEPN domains in eukaryotic host-pathogen conflicts: evidence from domain architectures

Analysis of phyletic patterns suggests that, beyond RNase L, several other distinct HEPN domains might be key players in host-pathogen conflicts in eukaryotes. This possibility was also suggested by several eukaryote-specific domain architectures that we recovered as part of this study (Figure 4). For example, in the sponge Amphimedon queenslandica, there are multiple Sacsin-like proteins (e.g. gi: 340377463) fused to DEATH domains, the key adaptors in metazoan apoptosis and immunity [137, 138]. Proteins of the CXorf38 family, one of the novel families of HEPN domains identified in this work, are fused to double-stranded RNA-binding (dsRBD) domains in vertebrates, cephalochordates and hemichordates, and in cephalochordates and cnidarians they are fused to NACHT NTPases [139] and DEATH domains (Figures 3 and 4). Furthermore, the human CXorf38 is strongly overexpressed in B lymphoblasts and CD56+ NK cells which are key player in the vertebrate immune response [140]. The DEATH domains and NACHT NTPase modules could link the action of the HEPN domain to an apoptotic and/or defensive response in which either cellular RNAs are degraded by analogy with RNase L, or else viral RNAs are targeted. The presence of the dsRBD containing versions of the CXorf38 family is suggestive of activity on dsRNA substrates which could include RNA viral replication intermediates. Some of these eukaryotic domain architectures are also reminiscent of bacterial proteins (e.g. gi: 229028907 from Bacillus cereus) that often combine an N-terminal HEPN domain with NTPase modules of the STAND superfamily (which includes the NACHT NTPase [141]), and in some cases C-terminal Cold-shock protein like OB-fold RNA-binding domains [142].

We also found evidence of deployment of HEPN domains as effectors directed against their eukaryotic hosts by apicomplexan parasites. A group of HEPN domains of the Swt1 family prototyped by the MAL13P1.321 protein from Plasmodium falciparum was found to be conserved throughout apicomplexa. These proteins combine a pair of N-terminal aegerolysin domains with C-terminal HEPN domains (Figure 4). Aegerolysin domains perforate membranes and could facilitate protein translocation across the lipid bilayer [143]. Given that in P. falciparum MAL13P1.321 is expressed during intraerythrocytic development [144], these proteins might be employed by apicomplexan in the host cells. The aegerolysin domains could ensure trafficking across the bounding vacuolar membrane and thus enable the interaction between the host RNAs and the HEPN domains. Given that these HEPN domains lack the conserved motif, they probably function non-catalytically by binding specific host RNAs.