Reductive evolution and unique predatory mode in the CPR bacterium Vampirococcus lugosii

Identification of ultra-small cells associated with blooms of anoxygenic phototrophic gammaproteobacteria

The Salada de Chiprana (NE Spain) is the only permanent athalassic hypersaline lake in Western Europe. It harbors thick, conspicuous microbial mats covering its bottom (Fig. 1a, b) and exhibits periodic stratification during which the deepest part of the water column becomes anoxic and sulfide-rich, favoring the massive development of sulfide-dependent anoxygenic photosynthetic bacteria¹⁶. We collected microbial mat fragments that were maintained in culture in the laboratory. After several weeks, we observed a bloom of anoxygenic photosynthetic bacteria containing numerous intracellular sulfur granules (Fig. 1c). Many of these bacterial cells showed one or several much smaller and darker non-flagellated cells attached to their surface (Fig. 1d, e). The infected photosynthetic cells were highly mobile, swimming at high speed with frequent changes of direction (Supplementary Movie 1), in contrast with the non-infected cells that displayed slower (approximately half speed) and more straight swimming. Sometimes, two or more photosynthetic cells were connected through relatively short filaments formed by stacked epibiont cells (Fig. 1f). Although the photosynthetic cells carrying these epibionts were often actively swimming, in some cases the epibionts were associated to empty ghost cells where only the photosynthesis-derived sulfur granules persisted. Closer scrutiny of the epibionts revealed that they actually consisted of piles of up to 10 very flattened cells of 550 ± 50 nm diameter and 220 ± 20 nm height (n = 100). These characteristics (size, morphology, and specific attachment to sulfide-dependent anoxygenic photosynthetic bacteria) perfectly matched the morphological description of the genus Vampirococcus observed over forty years ago in sulfidic freshwater lakes¹⁵.

Since the first Vampirococcus description included transmission electron microscopy (TEM) images, to further ascertain this identification we examined our Chiprana Lake samples under TEM and scanning electron microscopy (SEM). SEM images confirmed the peculiar structure of the epibionts, with multiple contiguous cells separated by deep grooves (Fig. 1g). Thin sections observed under TEM confirmed that the cells were actually separated by a space of ~20–50 nm filled by a fibrous material (Fig. 1h, i). The space between epibiont and host cells was larger (~100 nm) and also filled by dense fibrous material (Fig. 1h). The sections also showed that, in contrast with the typical Gram-negative double membrane structure of the host, the epibiont cells had a single membrane surrounded by a thick layer of fibrous material that conferred a rugose aspect to the cells (Fig. 1i). In sharp contrast with the often highly vacuolated cytoplasm of the host, the epibiont cells showed a dense, homogeneous content. These observations were also in agreement with those published for Vampirococcus, reinforcing our conclusion that the epibionts we observed belonged to this genus, although most likely to a different species, as the first described Vampirococcus occurred in a non-hypersaline lake¹⁵.

Using a micromanipulator coupled to an inverted microscope, we collected cells of the anoxygenic photosynthetic bacterium carrying Vampirococcus attached to their surface (Supplementary Fig. 1) and proceeded to amplify, clone, and sequence their 16 S rRNA genes. We were able to obtain sequences for both the epibiont and the host for ten infected cells and, in all cases, we retrieved the same two sequences. The host was found to be a Halochromatium-like gammaproteobacterium (Supplementary Fig. 2). Phylogenetic analysis of the epibiont sequence showed that it branched within the CPR radiation close to the Absconditabacteria (Supplementary Fig. 3), previously known as candidate phylum SR1². Since all host and epibiont cells we analyzed had identical 16 S rRNA gene sequences, suggesting that they were the result of a clonal bloom, we collected three sets of ca. 10 infected cells and carried out whole genome amplification (WGA) before sequencing (Illumina HiSeq; see Methods). This strategy allowed us to assemble the nearly complete genome sequence of the Vampirococcus epibiont (see below). In contrast with the completeness of this genome, we only obtained a very partial assembly (~15%) of the host genome, probably because of the consumption of the host DNA by the epibiont. To make more robust phylogenetic analyses of Vampirococcus, we retrieved the protein sequence set used by Hug et al. to reconstruct a multi-marker large-scale phylogeny of bacteria². The new multi-gene maximum likelihood (ML) phylogenetic tree confirmed the affiliation of our Vampirococcus species to the Absconditabacteria with maximum support, and further placed this clade within a larger well-supported group also containing the candidate phyla Gracilibacteria and Peregrinibacteria (Fig. 2a and Supplementary Fig. 4). Therefore, our epibiotic bacterium represents the first characterized member of this large CPR clade and provides a phylogenetic identity for the predatory bacterial genus Vampirococcus described several decades ago. We propose to call this new species Candidatus Vampirococcus lugosii (see Taxonomic appendix).

Genomic evidence of adaptation to predatory lifestyle

We sequenced DNA from three WGA experiments corresponding each to ~10 Halochromatium–Vampirococcus consortia. Many of the resulting (57.2 Mb) raw sequences exhibited similarity to those of available Absconditabacteria/SR1 metagenome-assembled genomes (MAGs) and, as expected, some also to Gammaproteobacteria (host-derived sequences) as well as a small proportion of potential contaminants probably present in the original sample (Bacillus– and fungi-like sequences). To bin the Vampirococcus sequences out of this mini-metagenome, we applied tetranucleotide frequency analysis on the whole sequence dataset using emergent self-organizing maps (ESOM)⁶. One of the ESOM sequence bins was enriched in Absconditabacteria/SR1-like sequences and corresponded to the Vampirococcus sequences, which we extracted and assembled independently. This approach yielded an assembly of 1,310,663 bp. We evaluated its completeness by searching i) a dataset of 40 universally distributed single-copy genes¹⁷ and ii) a dataset of 43 single-copy genes widespread in CPR bacteria⁸. We found all them as single-copy genes in the Vampirococcus genome, with the exception of two signal recognition particle subunits from the first dataset which are absent in many other CPR bacteria¹⁸. These results supported that the Vampirococcus genome assembly was complete and did not contain multiple strains or other sources of contamination. Manually curated annotation predicted 1151 protein-coding genes, a single rRNA gene operon, and 38 tRNA coding genes. As already found in other Absconditabacteria/SR1 genomes¹⁹, the genetic code of Vampirococcus is modified, with the stop codon UGA reassigned as an additional glycine codon.

A very large proportion of the predicted proteins (48.9%) had no similarity to any COG class and lacked any conserved domain allowing their functional annotation (Fig. 2b). Thus, as for other CPR bacteria, a significant part of their cellular functions remains inaccessible. A comparison with three other Absconditabacteria genomes revealed a very small set of only 390 genes conserved in all them (Fig. 2c), suggesting a highly dynamic evolution of gene content in these species. Comparison with more distantly related CPR groups (Gracilibacteria and Peregrinibacteria) showed that gene loss has been a dominant trend in all these organisms, which have lost 30–50% of the 1124 genes inferred to have existed in their last common ancestor (Fig. 2a). Nevertheless, this loss of ancestral genes was accompanied by the acquisition of new ones by different mechanisms, including horizontal gene transfer (HGT). In the case of Vampirococcus, we detected, by phylogenetic analysis of all individual genes that had homologs in other organisms, the acquisition of 126 genes by HGT from various donors (Supplementary Data 1).

The set of genes that could be annotated provided interesting clues about the biology and lifestyle of Vampirococcus. The most striking feature was its oversimplified energy and carbon metabolism map (Fig. 3). ATP generation in this CPR species appeared to depend entirely on substrate-level phosphorylation carried out by the phosphoenolpyruvate kinase (EC 2.7.1.40). In fact, Vampirococcus only possesses incomplete glycolysis, which starts with 3-phosphoglycerate as first substrate. This molecule is the major product of the enzyme RuBisCO and, therefore, most likely highly available to Vampirococcus from its photosynthetic host. Comparison with nearly complete MAG sequences available for other Absconditabacteria/SR1 showed that Vampirococcus has the most specialized carbon metabolism, with 3-phosphoglycerate as the only exploitable substrate, whereas the other species have a few additional enzymes that allow them to use other substrates (such as ribulose-1,5 P and acetyl-CoA) as energy and reducing power (NADH) sources (Supplementary Fig. 5). This metabolic diversification probably reflects their adaptation to other types of hosts where these substrates are abundant. Vampirococcus also lacks all the enzymes involved in some Absconditabacteria/SR1 in the 3-phosphoglycerate-synthesizing AMP salvage pathway²⁰, including the characteristic archaeal-like type II/III RuBisCO²¹.

The genomes of Absconditabacteria/SR1 and Vampirococcus encode several electron carrier proteins (e.g., ferredoxin, cytochrome b₅, several Fe-S cluster proteins) and a membrane F₁F_O-type ATP synthase. However, they apparently lack any standard electron transport chain and, therefore, they seem to be non-respiring^19,20,22. The electron carrier proteins may be related with the oxidative stress response and/or the reoxidation of reduced ferredoxin or NADH²⁰. In the absence of any obvious mechanism to generate proton motive force (PMF), the presence of the membrane ATP synthase is also intriguing. It has been speculated either that CPR bacteria might tightly adhere to their hosts and scavenge protons from them or that the membrane ATP synthase might work in the opposite direction as an ATPase, consuming ATP generated by substrate level phosphorylation to extrude protons and drive antiporters⁹. However, in the case of Vampirococcus the direct transport of protons from the host is unlikely since, as observed in the TEM sections (Fig. 1h), it seems that there is no direct contact with the host cell membrane. In fact, parasite and host cell membranes are separated by a relatively large space of ~100 nm, which would be largely conducive to proton diffusion and inefficient transfer between cells. Alternatively, since the Chiprana lake has high Na⁺ concentration (1.6 g l⁻¹), it might be possible that the ATP synthase uses Na⁺ instead of protons. However, the Na⁺-binding domain of the subunit c of typical Na⁺-dependent ATP synthases exhibited several differences with that of Vampirococcus (Supplementary Fig. 6). Similar differences have been considered indicative of the use of protons instead of Na⁺ in other organisms²³. Although the proton/cation antiporters (e.g., for Na⁺, K⁺, or Ca²⁺) encoded by Vampirococcus and the other Absconditabacteria/SR1 may serve to produce some PMF, it is improbable that this mechanism represents a major energy transducing system as cells would accumulate cations and disrupt their ionic balance; these antiporters are most likely involved in cation homeostasis.

These observations prompted us to investigate other ways that these cells might use to generate PMF usable by their ATP synthase. We found a protein (Vamp_33_45) with an atypical tripartite domain structure. The N-terminal region, containing 8 transmembrane helices, showed similarity with several flavocytochromes capable of moving electrons and/or protons across the plasma membrane (e.g., ²⁴). The central part of the protein was a rubredoxin-like nonheme iron-binding domain likely able to transport electrons. Finally, the C-terminal region, containing an NAD-binding motif, was similar to ferredoxin reductases involved in electron transfer²⁵. This unusual Vampirococcus 3-domain protein is well conserved in the other Absconditabacteria/SR1 genomes sequenced so far, suggesting it plays an important function in this CPR phylum. Its architecture suggests that it can transport electrons and/or protons across the membrane using ferredoxin as electron donor and makes it a strong candidate to participate in a putative new PMF-generating system. Alternatively, this protein could play a similar role to that of some oxidoreductases in the strict anaerobic archaeon Thermococcus onnurineus, including a thioredoxin reductase, which couple reactive oxygen species detoxification with NAD(P) + regeneration from NAD(P)H to maintain the intracellular redox balance and enhance O₂-mediated growth despite the absence of heme-based or cytochrome-type proteins²⁶.

Although our Vampirococcus genome sequence appears to be complete, genes encoding enzymes involved in the biosynthesis of essential cell building blocks such as amino acids, nucleotides and nucleosides, cofactors, vitamins, and lipids are almost completely absent (Fig. 2b). Therefore, the classical bacterial metabolic pathways for their synthesis²⁷ do not operate in Vampirococcus. Such simplified metabolic potential, comparable to that of intracellular parasitic bacteria such as Mycoplasma²⁸, implies that Vampirococcus must acquire these molecules from an external source and supports the predatory nature of the interaction with its photosynthetic host. An intriguing aspect of this interaction concerns the transfer of substrates from the host to Vampirococcus, especially considering that, despite examination of several serial ultrathin sections, the cell membranes of these two partners do not appear to be in direct contact (Fig. 1h). Vampirococcus encodes several virulence factors, including divergent forms of hemolysin and hemolysin translocator (Vamp_11_169 and Vamp_9_166, respectively), a phage holin (Vamp_5_129), and a membrane-bound lytic murein transglycosylase (Vamp_144_2). These proteins are likely involved in the host cell wall and membrane disruption leading to cell content release. Hemolysin has also been found in Saccharibacteria (formerly candidate phylum TM7), the only CPR phylum for which an epibiotic parasitic lifestyle has been demonstrated so far^13,14. Recent coupled lipidomic-metagenomic analyses have shown that CPR bacteria that lack complete lipid biosynthesis are able to recycle membrane lipids from other bacteria²⁹. In Vampirococcus, also devoid of phospholipid synthesis, a phospholipase gene (Vamp_34_196) predicted to be secreted and that has homologs involved in host phospholipid degradation in several parasitic bacteria³⁰, may not only help disrupting the host membrane but also to generate a local source of host phospholipids that it can use to build its own cell membrane. Two Vampirococcus peptidoglycan hydrolases (Vamp_68-56_103 and Vamp_145_30), also predicted to be secreted, most probably contribute to degrade the host cell wall. The Vampirococcus genome also encodes two murein DD-endopeptidases (Vamp_311_38 and Vamp_41_33). As in other predatory bacteria, such as Bdellovibrio, one probably acts to degrade the prey cell wall whereas the other is involved in self-wall remodeling³¹. Despite their high sequence divergence, we could align both Vampirococcus sequences with those of Bdellovibrio and other bacteria (Supplementary Fig. 7). Both sequences conserved the characteristic active site serine residue of DD endopeptidases and, in contrast with the Bdellovibrio “predatory” enzymes, also the regulatory domain III. The deletion of this regulatory domain has been associated with the capacity of the Bdellovibrio “predatory” DD endopeptidases to act promiscuously on a wide variety of peptidoglycan substrates³¹. This difference most likely reflects that, whereas Bdellovibrio is able to prey on very diverse bacteria, Vampirococcus is a specialized predator of Chromatiaceae that has evolved specialized enzymes to degrade the wall of its particular prey. The Vampirococcus enzymes could also be aligned with the region where the ankyrin-repeat-containing self-protective regulatory inhibitor Bd3460 binds the Bdellovibrio “predatory” DD endopeptidases³², although only partially for the C-terminal part of Vamp_41_33, like in the self-wall Bdellovibrio enzyme Bd3244 (Supplementary Fig. 7). In that sense, the Vamp_311_38 enzyme seems more similar to the “predatory” Bdellovibrio ones. Interestingly, the “predatory” endopeptidase Bd3459 and the regulatory inhibitor Bd3460 are contiguous in the genomes of Bdellovibrio and other periplasmic predators but not in epibiotic predators³². Vampirococcus confirms this pattern since, although it possesses several ankyrin-repeat-containing proteins, none of them is encoded adjacent to the DD endopeptidase genes.

Vampirococcus also possesses a number of genes encoding transporters, most of them involved in the transport of inorganic molecules (Fig. 3). One notable exception is the competence-related integral membrane protein ComEC (Vamp_67_106)³³ which, together with ComEA (Vamp_21_186) and type IV pili (see below), probably plays a role in the uptake of host DNA that, once transported into the epibiont, can be degraded by various restriction endonucleases (five genes encoding them are present) and recycled to provide the nucleotides necessary for growth (Supplementary Fig. 8). These proteins are widespread in other CPR bacteria where they may have a similar function³⁴. Vampirococcus also encodes an ABC-type oligopeptide transporter (Vamp_40_40) and a DctA-like C4-dicarboxylate transporter (Vamp_41_97), known to catalyze proton-coupled symport of several Krebs cycle dicarboxylates (succinate, fumarate, malate, and oxaloacetate)³⁵. The first, coupled with the numerous peptidases present in Vampirococcus, most likely is a source of amino acids. By contrast, the role of DctA is unclear since Vampirococcus does not have a Krebs cycle.

In sharp contrast with its simplified central metabolism, Vampirococcus possesses genes related to the construction of an elaborate cell surface, which seems to be a common theme in many CPR bacteria^9,12. They include genes involved in peptidoglycan synthesis, several glycosyltransferases, a Sec secretion system, and a rich repertoire of type IV pilus proteins. The retractable type IV pili are presumably involved in the tight attachment of Vampirococcus to its host and in DNA uptake in cooperation with the ComEC protein. Other proteins probably play a role in the specific recognition and fixation to the host, including several very large proteins. In fact, the Vampirococcus membrane proteome is enriched in giant proteins. The ten longest predicted proteins (between 1392 and 4163 aa, see Supplementary Table 1) are inferred to have a membrane localization and are probably responsible of the conspicuous fibrous aspect of its cell surface (Fig. 1i). Most of these proteins possess domains known to be involved in the interaction with other molecules, including protein-protein (WD40, TRP, and PKD domains) and protein–lipid (saposin domain) interactions and cell adhesion (DUF11, integrin, and fibronectin domains). Two other large membrane proteins (Vamp_6_203, 2368 aa, and Vamp_19_245, 1895 aa) may play a defensive role as they contain alpha-2-macroglobulin protease-inhibiting domains that can protect against proteases released by the host. Several other smaller proteins complete the membrane proteome of Vampirococcus, some of them also likely involved in recognition and attachment to the host thanks to a variety of protein domains, such as VWA (Vamp_41_85) and flotillin (Vamp_11_100). We did not detect genes coding for flagellar components, confirming the absence of flagella observed under the microscope (Fig. 1).

New CRISPR-Cas systems and other defense mechanisms in Vampirococcus

Although most CPR phyla are devoid of CRISPR-Cas³⁶, some have been found to contain new systems with original effector enzymes such as CasY³⁷. In contrast with most available Absconditabacteria genomes, Vampirococcus possesses two CRISPR-Cas loci (Fig. 4a and Supplementary Fig. 9). The first is a class II type V system that contains genes coding for Cas1, Cas2, Cas4, and Cpf1 proteins associated to 34 spacer sequences of 26–32 bp. Proteins similar to those of this system are encoded not only in genomes of close relatives of the Absconditabacteria (Gracilibacteria and Peregrinibacteria) but in many other CPR phyla. These sequences form monophyletic groups in phylogenetic analyses (e.g., Cas1, see Fig. 4b), which suggests that this type V system is probably ancestral in these CPR. The second system found in Vampirococcus belongs to the class I type III and contains genes coding for Cas1, Cas2, Csm3, and Cas10/Csm1 proteins associated to a cluster of 20 longer (35–46 bp) spacers. In contrast with the previous CPR-like system, the proteins of this second system did not show strong similarity with any CPR homolog but with sequences from other bacterial phyla, suggesting that they have been acquired by HGT. Phylogenetic analysis confirmed this and supported that Vampirococcus gained this CRISPR-Cas system from different distant bacterial donors (Supplementary Fig. 10). Interestingly, these two CRISPR-Cas systems encode a number of proteins that may represent new effectors. A clear candidate is the large protein Vamp_48_93 (1158 aa), located between Cpf1 and Cas1 in the type V system (Fig. 4a), which contains a DNA polymerase III PolC motif. Very similar sequences can be found in a few other CPR (some Roizmanbacteria, Gracilibacteria, and Portnoybacteria) and in some unrelated bacteria (Supplementary Fig. 11). As in Vampirococcus, the gene coding for this protein is contiguous to genes encoding different Cas proteins in several of these bacteria, including Roizmanbacteria, Omnitrophica, and the deltaproteobacterium Smithella sp. (Supplementary Fig. 11). This gene association, as well as the very distant similarity between this protein and Cpf1 CRISPR-associated proteins of bacterial type V systems, supports that it is a new effector in type V CRISPR-Cas systems. Additional putative new CRISPR-associated proteins likely exist also in the Vampirococcus type III system (Fig. 4a). Three proteins encoded by contiguous genes (Vamp_21_116, Vamp_21_127, and Vamp_21_128) exhibit very distant similarity with type III-A CRISPR-associated Repeat Associated Mysterious Proteins (RAMP) Csm4, Csm5, and Csm6 sequences, respectively, and most probably represent new RAMP subfamilies. To date, Absconditabacteria³⁸ and Saccharibacteria³⁹ are the only CPR phyla for which phages have been identified. Because of its proximity to Absconditabacteria, Vampirococcus is probably infected by similar phages, so that the function of its CRISPR-Cas systems may be related to the protection against these genetic parasites. Nevertheless, we did not find any similarity between the Vampirococcus spacers and known phage sequences, suggesting that it is infected by unknown phages. Alternatively, considering that Vampirococcus -as most likely many other CPR bacteria- seems to obtain nucleotides required for growth by uptaking host DNA, an appealing possibility is that the CRISPR-Cas systems participate in the degradation of the imported host DNA.

Although CPR bacteria have been hypothesized to be largely depleted of classical defense mechanisms⁴⁰, we found that Vampirococcus, in addition to the two CRISP-Cas loci, is endowed with various other protection mechanisms. These include an AbiEii-AbiEi Type IV toxin-antitoxin system, also present in other CPR bacteria, which may offer additional protection against phage infection⁴¹ and several restriction-modification systems, with three type I, one type II and one type III restriction enzymes and eight DNA methylases. In addition to a defensive role, these enzymes may also participate in the degradation of the host DNA. As in its sister-groups Absconditabacteria and Gracilibacteria^{5,19,20,42,43}, Vampirococcus has repurposed the UGA stop codon to code for glycine. The primary function of this recoding remains unknown but it has been speculated that it creates a genetic incompatibility, whereby these bacteria would be “evolutionarily isolated” from their environmental neighbors, preventing their potential competitors from acquiring their genomic innovations by HGT¹⁹. However, the opposite might be argued as well, since the UGA codon reassignment can protect Vampirococcus from foreign DNA expression upon uptake by leading to aberrant protein synthesis via read-through of the UGA stop with Gly insertion. This can be important for these CPR bacteria because they are not only impacted by phages³⁸ but they most likely depend on host DNA import and degradation to fulfill their nucleotide requirements. In that sense, it is interesting to note that the Vampirococcus ComEA protein likely involved in DNA transport⁴⁴ is encoded within the class I type III CRISPR-Cas system (Fig. 4a).

Source: Ecology - nature.com