Discovering immune homologs across the eukaryotic tree of life

The first step to understanding the evolution of CD-NTases, STINGs, and viperins was to acquire sequences for these proteins from across the eukaryotic tree. To search for diverse immune homologs, we employed a hidden Markov model (HMM) strategy, which has high sensitivity, a low number of false positives, and the ability to separately analyze multiple (potentially independently evolving) domains in the same protein [22–24]. We used this HMM strategy to search the EukProt database, which has been developed to reflect the true scope of eukaryotic diversity through the genomes and transcriptomes of nearly 1,000 species, specifically selected to span the eukaryotic tree [25]. EukProt contains sequences from NCBI and Ensembl, plus many diverged eukaryotic species not found in any other database, making it a unique resource for eukaryotic diversity [25]. While it can be challenging to acquire diverse eukaryotic sequences from traditional databases due to an overrepresentation of metazoan data [26], EukProt ameliorates this bias by downsampling traditionally overrepresented taxa.

To broaden our searches from initial animal homologs to eukaryotic sequences more generally, we used iterative HMM searches of the EukProt database, incorporating the hits from each search into the subsequent HMM. After using this approach to create pan-eukaryotic HMMs for each protein family, we then added in bacterial homologs to generate universal HMMs (Figs 1A and S1), continuing our iterative searches until we either failed to find any new protein sequences or began finding proteins outside of the family of interest (S1 Fig). To define the boundaries that separated our proteins of interest from neighboring gene families, we focused on including homologs that shared protein domains that defined that family (see Materials and methods for domain designations) and were closer to in-group sequences than the outgroup sequences on a phylogenetic tree (outgroup sequences are noted in the Materials and methods).

Our searches for CD-NTases, STINGs, and viperins recovered hundreds of eukaryotic proteins from each family, including a particularly large number of metazoan sequences (red bars, Fig 1B). It is not surprising that we found so many metazoan homologs, as each of these proteins was discovered and characterized in metazoans and these animal genomes tend to be of higher quality than other taxa (S2 Fig). We also recovered homologs from other species spread across the eukaryotic tree, demonstrating that our approach could successfully identify deeply diverged homologs (Fig 1B). However, outside of Metazoa, these homologs were sparsely distributed, such that for most species in our dataset (711/993), we did not recover proteins from any of the 3 immune families examined (white space, lack of colored bars, Fig 1B). While some of these absences may be due to technical errors or dataset incompleteness (S2 Fig), we interpret this pattern as a reflection of ongoing, repeated gene losses across eukaryotes, as has been found for other innate immune proteins [27–29] and other types of gene families surveyed across eukaryotes [28,30–32]. Indeed, many of the species that lacked any of the immune homologs were represented by high-quality datasets (Ex: Metazoa, Chlorplastida, and Fungi). Thus, although it is always possible that our approach has missed some homologs, we believe the resulting data represents a fair assessment of the diversity across eukaryotes, at least for those species currently included within EukProt.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. HMM searches to find homologs across the eukaryotic Tree of Life.

(A) A schematic of the HMM search process. Starting from initial, animal-dominated HMM profiles for each protein family, we used iterative HMM searches of the EukProt database to generate pan-eukaryotic HMMs. These were combined with bacterial sequences to enable discovery of bacteria-like homologs in eukaryotes. Each set of searches was repeated until few or no additional eukaryotic sequences were recovered which was between 3 and 5 times in all cases. (B) Phylogenetic tree of eukaryotes, with major supergroups color coded. The height of the colored rectangles for each group is proportional to its species representation in EukProt. Horizontal, colored bars mark each eukaryotic species in which we found homologs of STINGs, CD-NTases, or viperins. White space indicates species where we searched but did not recover any homologs. The CD-NTase hits are divided into the 3 eukaryotic superfamilies, defined in Fig 2. Individual data are available in S1 File. CD-NTase, cGAS-DncV-like nucleotidyltransferase; HMM, hidden Markov model; STING, Stimulator of Interferon Genes.

https://doi.org/10.1371/journal.pbio.3002436.g001

Eukaryotes acquired CD-NTases from bacteria through multiple, independent HGT events

We next studied the evolution of the innate immune proteins, beginning with cGAS and its broader family of CD-NTase enzymes. Following infections or cellular damage, cGAS binds cytosolic DNA and generates cyclic GMP-AMP (cGAMP) [33–36], which then activates downstream immune responses via STING [35,37–39]. Another eukaryotic CD-NTase, 2′5′-Oligoadenylate Synthetase 1 (OAS1), synthesizes 2′,5′-oligoadenylates that bind and activate Ribonuclease L (RNase L) [40]. Activated RNase L is a potent endoribonuclease that degrades both host and viral RNA species, reducing viral replication (reviewed here [41,42]). Some bacterial CD-NTases such as DncV behave similar to animal cGAS; they are activated by phage infection and produce cGAMP [8,21,43]. These CD-NTases are commonly found within cyclic oligonucleotide-based antiphage signaling systems (CBASS) across many bacterial phyla and archaea [8,21,44].

In addition to the well-studied cGAS, a number of other eukaryotic CD-NTases have been previously described: the OAS1 paralogs (OAS2/3), Male abnormal 21-Like 1/2/3/4 (MAB21L1/2/3/4), Mab-21 domain containing protein 2 (MB21D2), Mitochondrial dynamics protein 49/51(MID49/51), and Inositol 1,4,5 triphosphate receptor-interacting protein 1/2 (ITPRILP/1/2) [44]. Of these, cGAS and OAS1 are the best characterized and both play roles in immune signaling. Recent work has shown that cGAS and related animal proteins, the cGAS-like Receptors (cGLRs), are present in nearly all metazoan taxa and generate diverse cyclic dinucleotide signals [45]. However, the immune functions of Mab21L1 and MB21D2 remain unclear, although Mab21L1 has been shown to be important for development [46–48].

To analyze the evolutionary history of the eukaryotic CD-NTases, we searched EukProt v3 for homologs and then generated phylogenetic trees. We aligned the homologs with MAFFT and MUSCLE and then generated phylogenetic trees with IQtree and RaxML (see Materials and methods). We considered our results to be robust if they were concordant across the majority of 4 trees generated per gene.

To begin our sequence searches for eukaryotic CD-NTases, we used the Pfam domain PF03281, representing the main catalytic domain of cGAS, as a starting point. As representative bacterial CD-NTases, we used 6,132 bacterial sequences, representing a wide swath of CD-NTase diversity [21]. Following our iterative HMM searches, we recovered 313 sequences from 109 eukaryotes, of which 34 were metazoans (S30, S31 and S32 Files and Fig 1B). Within the phylogenetic trees, most eukaryotic sequences clustered into one of 2 distinct superfamilies: the cGLR superfamily (defined by clade and containing a Mab21 PFAM domain: PF03281) or the OAS superfamily (OAS1-C: PF10421) (Fig 2A). Bacterial CD-NTases typically had sequences matching the HMM for the Second Messenger Oligonucleotide or Dinucleotide Synthetase domain (SMODS: PF18144).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Independent HGT events gave rise to multiple CD-NTase superfamilies.

(A) Maximum likelihood phylogenetic tree generated by IQtree of CD-NTases spanning eukaryotic and bacterial diversity. The cGLR superfamily (red, top left) is largely an animal-specific innovation, with many paralogs including cGAS. In contrast, most other eukaryotic lineages encode CD-NTases from the OAS superfamily (multicolor, top right). The relatively small eSMODS superfamily (pink, bottom left) likely arose from a recent HGT between clade D bacteria and eukaryotes. Bacterial CD-NTase sequences shown in gray. Eukaryotic sequences are colored according to eukaryotic supergroup as in Fig 1B. Tree is arbitrarily rooted on a branch separating bacterial clades A, B, G, and H from the rest of the bacterial CD-NTases. (B) Venn diagrams showing the number of species where we detected at least 1 STING, cGLR, and/or OAS homolog, either within Metazoa (left) or in non-metazoan eukaryotes (right). (C) Magnification of the CD-NTase phylogenetic tree in (A), showing the region where the OAS superfamily branches within clade C bacterial CD-NTases (gray branches). (D) Magnification showing clade D CD-NTases (gray branches), which have been horizontally transferred into eukaryotes multiple times, giving rise to both the cGLR and the eSMODS superfamilies. Ultrafast bootstraps determined by IQtree shown at key nodes. See S4 Fig for full CD-NTase phylogenetic tree. Underlying Newick file is included in S2 File. Additional information on which species encode CD-NTases of a given homolog (Fig 2B) can be found in S1 File. CD-NTase, cGAS-DncV-like nucleotidyltransferase; cGAS, cyclic GMP-AMP synthase; HGT, horizontal gene transfer; STING, Stimulator of Interferon Genes.

https://doi.org/10.1371/journal.pbio.3002436.g002

The cGLR superfamily is composed almost entirely of metazoan sequences, with only a few homologs from Amoebozoa, choanoflagellates, and other eukaryotes (Fig 2A). Indeed, the majority of animal CD-NTases (cGAS, Mid51, Mab21, Mab21L1/2/3/4, Mb21d2, ITPRI) are paralogs within the cGLR superfamily, which arose from repeated animal-specific duplications [49] (S4 Fig). In contrast, unlike the animal-dominated cGLR superfamily, the OAS superfamily spans a broad group of eukaryotic taxa, with OAS-like homologs present in 8/12 eukaryotic supergroups. This distribution makes OAS proteins the most common CD-NTases found across eukaryotes and implies that they arose very early in eukaryotic history, possibly before the LECA.

Given the connections between cGAS and STING in both animals and some bacteria [3,21,50], we asked whether species that encode STING also have cGLR and/or OAS proteins. Because the cGLR superfamily is largely animal specific, we performed this analysis separately in either Metazoa or with all non-metazoan eukaryotes (Fig 2B). In animal species where we found a STING homolog, we also typically found a cGLR superfamily sequence (32/34), and specifically a cGAS homolog (26/34 species) (Fig 2B), consistent with the consensus that these proteins are functionally linked. We also observed 19 metazoan species that had a cGLR-like sequence with no detectable STING homolog. Almost half of these species (10/19) were arthropods, aligning with prior findings of STING sparseness among arthropods [50]. We did find STING homologs in 8/19 arthropod species in EukProt v3, including the previously identified STINGs of Drosophila melanogaster, Apis mellifera, and Tribolium castaneum [50,51]. Outside of animals, we found that species with a STING homolog typically did not have a detectable CD-NTase protein from either superfamily (22/34). While it remains possible that these STING proteins function together with a to-be-discovered CD-NTase that was absent from our dataset, we therefore hypothesize that many eukaryotes outside of metazoans and their close relatives [52] use STING and CD-NTase homologs independently of each other.

What was the evolutionary origin of eukaryotic CD-NTases? Interestingly, the cGLR and OAS superfamilies are only distantly related to one another. Each lies nested within a different, previously defined, bacterial CD-NTase clade (Fig 2C and 2D). The OAS superfamily falls within bacterial Clade C (with the closest related bacterial CD-NTases being those of subclade C02-C03, Fig 2C), while the metazoan cGLR superfamily lies within bacterial Clade D (subclade D12) (Fig 2D). We note that in this tree (Fig 2D), Clade D does not form a single coherent clade, as was also true in the phylogeny that originally defined the bacterial CD-NTase clades [11].

We also observed a number of eukaryotic sequences scattered across different bacterial CD-NTase clades (Fig 2A, colored branches within gray clades). While some of these may reflect additional HGT events, others likely come from technical artifacts such as bacterial contamination of eukaryotic sequences. To minimize such false positive HGT calls, we took a conservative approach in our analyses, considering potential bacteria–eukaryote HGT events to be trustworthy only if: (1) eukaryotic and bacterial sequences branched adjacent to one another with strong support (bootstrap values >70); (2) the eukaryotic sequences formed a distinct subclade, represented by at least 2 species from the same eukaryotic supergroup; (3) the eukaryotic sequences were produced by at least 2 different studies; and (4) the position of the horizontally transferred sequences was robust across all alignment and phylogenetic reconstruction methods used (S3A Fig). For species represented only by transcriptomes, these criteria may still have difficulty distinguishing eukaryote–bacteria HGT from certain specific scenarios such as the long-term presence of dedicated, eukaryote-associated, bacterial symbionts. However, because these criteria allow us to focus on relatively old HGT events, they give us higher confidence these events are likely to be real.

The cGLR superfamily passed all 4 of the HGT thresholds, as did another eukaryotic clade of CD-NTases that were all previously undescribed. We name this clade the eukaryotic SMODS (eSMODS) superfamily, because the top scoring domain from hmmscan for each sequence in this superfamily was the SMODS domain (PF18144), which is typically found only in bacterial CD-NTases (S25 File). This sequence similarity suggests that eSMODS arose following a recent HGT from bacteria and/or that these CD-NTases have diverged from their bacterial predecessors less than the eukaryotic OAS and cGLR families have. Additionally, all of the eSMODS sequences were predicted to have a Nucleotidyltransferase domain (PF01909), and (8/12) had a Polymerase Beta domain (PF18765), which are features shared with many bacterial CD-NTases in Clades D, E, and F (S25 File). The eSMODS superfamily is made up of sequences from Amoebozoa, choanoflagellates, Ancryomonadida, and 1 animal (the sponge Oscarella pearsei), which clustered together robustly and with high support (ultrafast bootstrap value of 99) within bacterial Clade D (e.g., subclade D04, CD-NTase 22 from Myxococcus xanthus) (S4 Fig). The eSMODS placement on the tree was robust to all alignment and phylogenetic algorithms used (S3A Fig), suggesting that eSMODS represent an additional, independent acquisition of CD-NTases from bacteria.

CD-NTases from bacterial Clade C and Clade D are the only CD-NTases to produce cyclic trinucleotides, producing cyclic tri-Adenylate and cAAG, respectively [11,14,53,54]. Interestingly, OAS produces linear adenylates, which is one step away from the cAAA product made by previously characterized Class C CD-NTases, and similarly cGAMP (made by cGAS) is one adenylate away from the Clade D product cAAG. As of this writing, the Clade D CD-NTases closest to the eSMODS and cGLR superfamilies (D04 and D12, respectively) have not been well characterized. Therefore, we argue that these CD-NTases should be a focus of future studies, as they may hint at the evolutionary stepping stones that allow eukaryotes to acquire bacterial immune proteins.

Diverged eukaryotic STINGs bridge the gap between bacteria and animals

We next turned to analyze STING proteins. In animals, STING is a critical cyclic dinucleotide sensor, important during viral, bacterial, and parasitic infections (reviewed here [55]). Structurally, most metazoan STINGs consist of an N-terminal transmembrane domain (TM), made of 4 alpha helices fused to a C-terminal STING domain [56]. Canonical animal STINGs show distant homology with STING effectors from the bacterial CBASS, with major differences in protein structure and pathway function between these animal and bacterial defenses. For example, in bacteria, the majority of STING proteins are fusions of a STING domain to a TIR (Toll/interleukin-1 receptor) domain (Fig 3A). Bacterial STING proteins recognize cyclic di-GMP and oligomerize upon activation, which promotes TIR enzymatic activity [3,57,58]. Some bacteria, such as Flavobacteriaceae, encode proteins that fuse a STING domain to a transmembrane domain, although it is unclear how these bacterial TM-STINGs function [3]. Other bacteria have STING domain fusions with deoxyribohydrolase, α/β- hydrolase, or trypsin peptidase domains [19]. In addition to eukaryotic TM-STINGs, a few eukaryotes such as the oyster Crassostrea gigas have TIR-STING fusion proteins, although the exact role of their TIR domain remains unclear [3,51,59].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Diverse eukaryotic STING proteins bridge the gap between metazoans and bacteria.

(A) Graphical depiction of common domain architectures of STING proteins. (B) Maximum likelihood unrooted phylogenetic tree of STING domains from Metazoa and bacteria, which are separated by 1 long branch. Black dot (•) indicates proteins that have been previously experimentally characterized. Bacterial sequences are in gray and animal sequences are in red. (C) Maximum likelihood unrooted phylogenetic tree of hits from iterative HMM searches for diverse eukaryotic STING domains. The STING domains from blSTINGs from diverse eukaryotes break up the long branch between bacterial and animal STINGs. Structures of the indicated STING proteins are shown above, with those predicted by AlphaFold indicated by an asterisk. Homologs with X-ray crystal structures are from [3,87]. Two domain architectures exist in bacteria and eukaryotes (STING linked to a TIR domain and STING linked to a transmembrane domain), each of which have evolved convergently through domain shuffling. Ultrafast bootstraps determined by IQtree shown at key nodes. Eukaryotic sequences are colored according to the eukaryotic group as in Fig 1B. See S5 Fig for full STING phylogenetic tree. Underlying Newick files are included in Supporting information (S3 File and S4 File). AlphaFold predicted structures are also included in the Supporting information (S6 File, S7 File, and S8 File). blSTING, bacteria-like STING; HMM, hidden Markov model; STING, Stimulator of Interferon Genes.

https://doi.org/10.1371/journal.pbio.3002436.g003

Given these major differences in domain architectures, ligands, and downstream immune responses, how have animals and bacteria evolved their STING-based defenses, and what are the relationships between them? Prior to this work, the phylogenetic relationship between animal and bacterial STINGs has been difficult to characterize with high support [19]. Indeed, when we made a tree of previously known animal and bacterial STING domains, we found that the metazoan sequences were separated from the bacterial sequences by one very long branch, along which many changes had occurred (Fig 3B).

To improve the phylogeny through the inclusion of a greater diversity of eukaryotic STING sequences, we began by carefully identifying the region of STING that was homologous between bacterial and animal STINGs, as we expected this region to be best conserved across diverse eukaryotes. Although Pfam domain PF15009 (TMEM173) is commonly used to define animal STING domains, this HMM includes a portion of STING’s transmembrane domain which is not shared by bacterial STINGs. Therefore, we compared the crystal structures of HsSTING (6NT5), Flavobacteriaceae sp. STING (6WT4), and Crassostrea gigas STING (6WT7) to define a core “STING” domain. We used the region corresponding to residues 145–353 of 6NT5 as an initial HMM seed alignment of 15 STING sequences from PF1500915 (“Reviewed” sequences on InterPro). Our searches yielded 146 eukaryotic sequences from 64 species, which included STING homologs from 34 metazoans (S31 File and Fig 1). Using maximum likelihood phylogenetic reconstruction on the STING domain alone, we identified STING-like sequences from 26 diverse microeukaryotes whose STING domains clustered in between bacterial and metazoan sequences, breaking up the long branch. We name these sequences the bacteria-like STINGs (blSTINGs) because they were the only eukaryotic group of STINGs with a bacteria-like Prok_STING domain (PF20300) and because of the short branch length (0.86 versus 1.8) separating them from bacterial STINGs on the tree (Fig 3C). While a previous study reported STING domains in 2 eukaryotic species (1 in Stramenopiles and 1 in Haptista) [19], we were able to expand this set to additional species and also recover blSTINGs from Amoebozoa, Rhizaria, and choanoflagellates. This diversity allowed us to place the sequences on the tree with high confidence (bootstrap value >70), recovering a substantially different tree than previous work [19]. As for CD-NTases, the tree topology we recovered was robust across multiple different alignment and phylogenetic tree construction algorithms (S3A Fig).

Given the similarities between the STING domains of the blSTINGs and bacterial STINGs, we next asked whether the domain architectures of these proteins were similar using Hmmscan and AlphaFold. The majority of the new eukaryotic blSTINGs were predicted to have 4 N-terminal alpha helices (Fig 3C and S5 File and S6 File), similar to human STING. While bacterial TM-STINGs had superficially similar N-terminal transmembrane domains, these proteins were predicted to have only 2 alpha helices and in all phylogenetic trees the STING domains from bacterial TM-STINGs were more similar to other bacterial STINGs than to eukaryotic homologs (S3A Fig). These results suggest that eukaryotes and bacteria independently converged on a common TM-STING domain architecture through domain shuffling.

Interestingly, a similar pattern of convergent domain shuffling appears to have occurred a second time with the TIR-STING proteins. Some eukaryotes, such as the oyster C. gigas, have a TIR-STING fusion protein [3,51,59]. The STING domain of these TIR-STINGs clustered closely to other metazoan STINGs, suggesting an animal origin (Fig 3B). We also investigated the possibility that C. gigas acquired the TIR-domain of its TIR-STING protein via HGT from bacteria; however, this analysis also suggested an animal origin for the TIR domain (S7 Fig), as the C. gigas TIR domain clustered with other metazoan TIR domains such as Homo sapiens TICAM1 and 2 (ultrafast bootstrap value of 75). Eukaryotic TIR-STINGs are also rare, further supporting the hypothesis that this protein resulted from recent convergence, where animals independently fused STING and TIR domains to make a protein resembling bacterial TIR-STINGs, consistent with previous reports [19]. Overall, the phylogenetic tree we constructed (Fig 3C) suggests that there is domain-level homology between bacterial and eukaryotic STINGs, but due to sparseness and lack of a suitable outgroup, this tree does not definitively explain the eukaryotic origin of the STING domain. However, the data does clearly support a model in which convergent domain shuffling in eukaryotes and bacteria generated similar TM-STING and TIR-STING proteins independently. Interestingly, the non-metazoan, blSTINGs (Fig 3C) that are found in the Stramenopiles, Haptista, Rhizaria, Choanoflagellates, and Amoebozoa have a TM-STING domain architecture similar to animal STINGs but a STING domain more similar to bacterial STINGs.

Viperin is an ancient and widespread immune family

Viperins are innate immune proteins that restrict the replication of a diverse array of viruses by conversion of nucleotides into 3′-deoxy-3′,4′didehdro- (ddh) nucleotides [4,15–17]. Incorporation of these ddh nucleotides into a nascent RNA molecule leads to chain termination, blocking RNA synthesis and inhibiting viral replication [15,18]. While metazoan viperin specifically catalyzes CTP to ddhCTP [15], homologs from archaea and bacteria can generate ddhCTP, ddhGTP, and ddhUTP [4,60]. Previous structural and phylogenetic analysis showed that eukaryotic viperins are highly conserved at both the sequence and structural level and that, phylogenetically, animal and fungal viperins form a distinct monophyletic clade compared to bacterial viperins [4,16,60].

As viperin proteins consist of a single Radical SAM protein domain, we iteratively searched EukProt beginning with domain PF04055 (Radical_SAM). The 194 viperin-like proteins we recovered came from 158 species spanning the full range of eukaryotic diversity, including organisms from all of the major eukaryotic supergroups, as well as some orphan taxa whose taxonomy remains open to debate (Fig 1, Ancyromonadida, Hemimastigophora, Malawimonadida). When we constructed phylogenetic trees from these sequences, we found that the large majority of the eukaryotic viperins cluster together in a single, monophyletic clade, separate from bacterial or archaeal viperins (Fig 4). Within the eukaryotic viperin clade, sequences from more closely related eukaryotes often clustered together (Fig 4, colored blocks), as would be expected if viperins were present and vertically inherited within eukaryotes for an extended period of time. The vast species diversity and tree topology both strongly support the inference that viperins are a truly ancient immune module and have been present within the eukaryotic lineage likely dating back to the LECA.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Viperin is a deeply conserved innate immune module.

Maximum likelihood phylogenetic tree generated by IQtree of viperins from eukaryotes, bacteria, and archaea. All major eukaryotic supergroups have at least 2 species that encode a viperin homolog (colored supergroups). Eukaryotic sequences are colored according to eukaryotic group as in Fig 1B. Bacterial viperin sequences shown in gray and archaeal sequences in dark gray. There are 2 clades of Chloroplastida (a group within Archaeplastida) sequences that branch robustly (>80 ultrafast bootstrap value) within the bacteria clade. Ultrafast bootstraps determined by IQtree shown at key nodes. Tree is arbitrarily rooted between the major eukaryotic and bacterial clades. See S6 Fig for fully annotated viperin phylogenetic tree. Underlying Newick file is included in S8 File under Supporting information.

https://doi.org/10.1371/journal.pbio.3002436.g004

In addition to this deep eukaryotic ancestry, we also uncovered 2 examples of bacteria–eukaryote HGT that have occurred much more recently, both in Chloroplastida, a group within Archaeplastida. The first of these consists of a small clade of Archaeplastida (Clade A) consisting of marine algae such as Chloroclados australicus and Nemeris dumetosa. These algal viperins cluster closely with the marine cyanobacteria Anabaena cylindrica and Plankthriodies (Figs 4 and S6). The second clade (Clade B) includes 4 other Archaeplastida green algal species, mostly Chlamydomonas spp. In some of our trees, the Clade B viperins branched near to eukaryotic sequences from other eukaryotic supergroups; however, the placement of the neighboring eukaryotic sequences varied depending on the algorithms we used; only the Archaeplastida placement was consistent (Figs 4 and S3A and S6). Taken together, we conclude that viperins represent a class of ancient immune proteins that have likely been present in eukaryotes since the LECA. Yet, we also find ongoing evolutionary innovation in viperins via HGT, both among eukaryotes and between eukaryotes and bacteria.

Iterative HMM search

The goal of this work was to search the breadth of EukProt v3 for immune proteins from the CD-NTase, STING, and viperin families that span the gap between metazoan and bacterial immunity. Our overall strategy was to first search with eukaryotes alone (starting from mainly Metazoa). Then, we added in bacterial sequences and searched with a mixed bacterial-eukaryotic HMM search until we either found no new hits, or until we began getting hits from an outgroup gene family. As outgroup sequences, we used Poly(A) RNA polymerase (PAP) sequences for the CD-NTases and molybdenum cofactor biosynthetic enzyme (MoaA) for viperin. We did not have a suitable outgroup for STING domains nor did any diverged outgroups come up in our searches. In parallel to our searches spanning bacterial and eukaryotic protein diversity, we also performed bacteria-only and eukaryote-only searches to ensure that we found as many homologs as possible (schematized in Fig 1A and further in S1A Fig).

Final alignment and tree building

To generate final phylogenetic trees, all eukaryotic search hits and bacterial sequences were aligned using MAFFT (default parameters). We downsampled the CD-NTase bacterial sequences from approximately 6,000 down to 500 using PDA software (options -k 500) on a FastTree (default settings) tree built upon an MAFFT (default parameters) tree, to facilitate more manageable computation times on alignments and tree construction. For the STING and viperin trees, we included all bacterial sequences. These initial alignments were first trimmed manually in Geneious (v2023.1.2) to remove unaligned N- and C-terminal regions, and then realigned with MAFFT (default parameters) or MUSCLE (v5.1) [79] and trimmed with TrimAI (v1.2) [76]. MUSCLE was used with the “-super5” option and otherwise default parameters. MUSCLE was deployed in parallel with MAFFT to generate these final alignments to ensure that the final tree topology would be as robust as possible. MUSCLE is a slightly more accurate but more computationally intensive alignment software [79]. The length of these final alignments were 232, 175, and 346 amino acids long for CD-NTase, STING, and viperin domains, respectively. These alignments represent ≥75% of the length of alignment their respective PFAM domain (PF3281 (Mab-21 protein nucleotidyltransferase domain) for CD-NTases, PF20300 (Prokaryotic STING domain) for STING, and PF404055 (Radical SAM family) for viperin). These alignments were used to generate phylogenetic trees using 3 tree inference softwares: FastTree (v2.1.10) [77], IQtree (2.2.2.7) [80], and RaxML-ng (v0.9.0) [81]. FastTree was utilized with default settings. IQtree was used to determine the appropriate evolutionary model and was run with 1,000 ultrafast bootstraps (IQtree settings: -s, -bb 1000, -m TEST, -nt AUTO). RaxML-ng trees were produced with 100 bootstraps using the molecular model specified from the IQtree analysis (Raxml-ng settings:—all,—model [specified by IQtree],—tree pars{10}—bs-trees 100). Phylogenetic trees were visualized with iTOL [82]. Weighted Robinson–Foulds distances for S3B Fig were calculated with Visual TreeCmp (settings: -RFWeighted -Prune trees -include summary -zero weights allowed) [83].

TIR domain alignment and tree

We used hmmscan to identify the coordinates of TIR domains in a list of 203 TIR domain containing-sequences from InterPro (all 203 proteins from curated “Reviewed” selection of IPR000157 (Toll/interleukin-1 receptor homology (TIR) domain as of 2023-04-04)) and 104 bacterial TIR-STING proteins (the same TIR-STING proteins used in Fig 3) [3]. Next, we trimmed the sequences down to the hmmscan identified TIR coordinates and aligned the TIR domains with MUSCLE (-super5). We trimmed the alignments with TrimAL and built a phylogenetic tree with IQtree (-s, -bb 1000, -m TEST, -nt AUTO).

Venn diagrams

Venn diagrams were generated via DeepVenn [84] using presence/absence information for cGLR, OAS, and STING from each eukaryotic species that encoded at least one of these proteins.

Protein structure modeling

To model 3D protein structures for STING homologs without a published crystal structure, we ran AlphaFold (v2.1.1) [85,86]. We generated 5 ranked models for STINGs from Flavobacteriaceae (IMG ID: 2624319773), Nitzschia sp. (EukProt ID: P007051), and Caveostelium apophsatum (EukProt ID: P019191). Fig 2C shows highest ranked models only. These highest ranked models are provided as S5, S6 and S7 Files under Supporting information for Nitzschia, Caveostelium, and Flavobacteriaceae, respectively.