Although the structure of Wnt was described over a decade prior (Bazan et al., 2012; Janda et al., 2012; Chu et al., 2013), its origins have been a mystery as it is phylogenetically restricted to Metazoa. Much attention has been focused on the three extended β-hairpins and a poorly structured loop extruding out of the core, their stabilizing cysteine residues, and the absolutely conserved serine residue, the site of palmitoleoylation (Janda et al., 2012; Zhong et al., 2021; Figure 1A). Our discovery of the first prokaryotic Wnt domains helped define its ancestral α-helical core, revealing the cysteine-rich extensions as Metazoa-specific insertions. Comparison of the core of the metazoan Wnt with AlphaFold structural models of the prokaryotic versions (Burroughs and Aravind, 2020) revealed a shared globular domain composed of five α-helices (Figure 1A). As the prokaryotic homologs retained just the conserved core of the Wnt proteins, we named these the minimal Wnt (Min-Wnt) family. The core helices of the Min-Wnt family contained absolutely conserved sequence motifs (Figure 2), consistent with the enzymatic function we had earlier proposed for them (Burroughs and Aravind, 2020) (see below).

Having defined this shared core, we initiated sequence-based homology searches in an effort to identify remote homologs. Iterative position-specific sequence matrix (PSSM)-based searches (see Methods) initially recovered animal and bacterial versions of the Serum Amyloid A (SAA) proteins, and further rounds of searching initiated from this set of sequences further recovered a vast collection of additional homologous families. As an example, a search initiated with a bacterial SAA-like sequence from Bdellovibrio bacteriovorus (Genbank acc: AHZ84906.1) retrieved a sequence overlapping with the Pfam models for ‘Domain of unknown function,’ DUF2279 (acc: WP_146898260.1, iteration: 5, e-value: 0.004), DUF4056 domain (acc: MBW8016507.1, iteration: 5, e-value: 0.005), and sequences automatically annotated as ‘YfiM’ in the GenBank database (acc: WP_019077413.1, iteration: 4, e-value: 0.004). Sequence profile-profile searches with HHpred confirmed these relationships and captured more distant ones. For instance, a HHpred search initiated with the Bacteriovorax stolpii Min-Wnt domain (acc: WP_102242990.1, residues 1–109) recovered the Pfam Wnt profile (PF: PF00110.23, p-value: 1.5e-6) and the Pfam SAA profile (PF: PF00277.22, p-value: 3.7e-5). Similarly, a HHpred search initiated with a Gemmatimonadetes sequence (acc: PYP94660.1, residues 75.170) recovered the DUF2279 Pfam profile (PF: PF10043.12, p-value: 5.4E-21), the DUF2238 profile (PF: PF09997.12, p-value: 1.3E-07), and the DUF4056 Pfam profile (PF: PF13265.9, p-value: 2.5E-05), among others.

Exhaustion of these searches, followed by clustering and manual inspection of the multiple sequence alignments of the retrieved sequences (see Methods), revealed a shared four-helix core across all of them, hereinafter referred to as H1 through H4 (Figure 1A). This 4-helix core of the domain was further confirmed by inspection of AlphaFold structural models constructed for representatives of the individual families, along with the rare instances of experimentally determined structures. These comparisons established that the above-mentioned fifth C-terminal helix in the Wnt core is a synapomorphy (shared derived character) restricted to the Wnts and closely related families like SAA (Figure 1A). In all, the results of our clustering analysis tallied 30 distinct families constituting a large superfamily. Remarkably, of these, 17 families had no pre-existing annotations. Phyletic analysis of individual families revealed a range of distributions, ranging from broad conservation in multiple superkingdoms of Life to those restricted to a small number of lineages (see below, Figure 1—figure supplement 1). A relationship network for the superfamily was constructed based on p-value and e-value scores using alignments of each family as a query in HHalign profile-profile searches against the rest (see Methods, Figure 1B). The Leiden community detection algorithm (Traag et al., 2019) was then applied to this network to identify higher-order assemblages (see Methods). These groupings were also supported by structural synapomorphies, such as a circular permutation and versions with a two-stranded ‘handle’ (see below).

The four helices conserved across the superfamily constitute a cone-like structure (Figure 1A), with the helices tending to coalesce on one end and opening out into a pocket on the other, lined by the conserved sequence positions (Figure 1A, Figure 1—figure supplement 2). The core is also marked by a linker between H1 and H2, which adopts characteristic extended conformations in certain families and higher-order groups. While the linkers joining H2 and H3 and H3 and H4 tend to be more constrained, there are some exceptions; for example, the extended loop insert housing the palmitoleoylated serine residue between H2 and H3 in the metazoan Wnt family (Figure 1A).

We observed that these Wnt-related families dramatically varied in their hydrophobicity. Using an index for transmembrane propensity (Zhao and London, 2006) (see Methods) and comparing that to known transmembrane (TM) segments, we predict that the α-helices in 18 of the 30 families are hydrophobic enough to qualify as TM domains, and show a statistically significant tendency to group to the exclusion of the other families (Figure 1C, Figure 1—figure supplement 3). Thus, these are predicted to be integral membrane domains. Further, these ‘hydrophobic families’ often evince a broader and deeper phyletic distribution pattern than the less-hydrophobic families (Figure 1—figure supplement 1, methods), implying that the ancestral version of the superfamily was likely an integral membrane domain. Thus, their association with the lipid membrane, combined with the cone-like shape of the conserved core (Figure 1A), leads us to refer to the whole superfamily hereinafter as the Lipocone superfamily.

AlphaFold 3-assisted transmembrane topology prediction (Jumper et al., 2021) revealed that 14 of the 17 integral membrane families are consistently oriented with the aperture of the cone-like structure opening toward the outer face of the membrane. This predicted TM topology is also generally consistent with the domain fusions when present: e.g., domains that are typically cytoplasmic and those that have extracellular or periplasmic functions are, respectively, predicted as projecting either inside or outside the membrane (see below). However, three families in the cpCone clade (see below) did not yield consistent orientation predictions, potentially owing to the diversity of structural variations observed in the clade, including a circular permutation event.

Of the 30 identified families, 26 display a striking conservation pattern of polar residues associated with the pocket of the Lipocone domain (Figure 2, Figure 1—figure supplement 2). Of these, a set of three positions, one mapping to each of H2, H3, and H4, can be inferred as being ancestrally present and were likely occupied by a histidine (H2), glutamate (H3), and aspartate (H4), though in some families their identities have secondarily changed (Figure 2, Figure 1—figure supplement 2). A fourth well-conserved polar position is observed at or near the end of H3; while its ancestral identity is difficult to establish, it is frequently an aspartate or glutamate (Figure 2). Two further well-conserved positions are often seen in H4: a polar position downstream of the broadly conserved aspartate residue and a glycine residue near the C-terminus of H4 (Figure 2) that likely caps the said helix. Although the ancestral pattern is noticeably degraded in the metazoan Wnt (Met-Wnt) family, it is strongly preserved in the prokaryotic Min-Wnt family (Figure 1A). In experimentally determined and modeled structures, the above set of 4 conserved positions forms a predicted active site in the aperture of the Lipocone domain. This, in turn, implies a shared biochemistry across the superfamily, with secondary inactivation in some families like Met-Wnt (see below, Figure 2). At the same time, the differences in the specific residues in the conserved positions between different families point to a range of distinct but related activities across the superfamily (Bastard et al., 2014; Glasner et al., 2006; Zhang et al., 2014).

Consistent with these observations, two of the families with intact active sites, the PTDSS1/2 (Stone and Vance, 1999; Tomohiro et al., 2009) and TelC (Whitney et al., 2017), which we identified in this work as members of the Lipocone superfamily, have been characterized as active enzymes operating on different lipid substrates (Figure 3A). The eukaryotic PTDSS1/2 localizes to the endoplasmic reticulum (ER) membrane and catalyzes a reaction on the polar head group of phosphatidylcholine or phosphatidylethanolamine (Saito et al., 1996; Stone and Vance, 2000; Miyata and Kuge, 2021). PTDSS1 and PTDSS2, respectively, exchange the phosphate-linked choline or ethanolamine head groups with L-serine (Stone and Vance, 1999; Figure 3A). The toxin domain of TelC acts on lipid II (Whitney et al., 2017), the final intermediate in peptidoglycan biosynthesis, which couples an undecaprenyl diphosphate tail to a head group comprised of a N-acetylmuramic acid-N-acetylglucosamine disaccharide, with a pentapeptide further linked to the former sugar (Anderson et al., 1967; Higashi et al., 1967). TelC cleaves the bond between the undecaprenol and the diphosphate coupled to the head group (Whitney et al., 2017; Figure 3B). The reaction is comparable to that catalyzed by PTDSS1/2, as both attack phosphate linkages in lipid head groups. However, TelC apparently directs a water molecule for the attack in lieu of the hydroxyl group of serine directed by PTDSS1/2 (Figure 3B).

Combining the above observations, we infer the unified biochemistry for the catalytically active families thus: (1) They act on the head groups of lipids either by removing or swapping phosphate-linked head groups (Figure 3A–B). These would be comparable to the phospholipase D (PLD), transphosphatidylation, or polyisoprenol phosphoesterase reactions (English, 1996). (2) Given the cone-like cavity and the hydrophobicity of the helices, the lipid tail is predicted to be housed within the lipocone with the head group positioned in the active site. (3) In the case of the integral membrane versions, their orientation would predict the targeting of the head groups of the outer leaf of the bilayer.

The extreme sequence divergence of the superfamily, coupled with the small size of the domain, prevents the use of simple phylogenetic tree analyses to resolve its deep evolutionary history. Hence, we combined community finding algorithms applied on profile-profile similarity networks, comparison of structural features and motifs, and phyletic patterns (Figures 1B and 2, Figure 1—figure supplement 1) to reconstruct the most parsimonious evolutionary scenario for the diversification of the Lipocone superfamily (Figure 4, see Methods). In the below sections, we survey the higher-order clades, highlighting their specific features.

This clade consists of four families, with the two prokaryotic families (Min-Wnt and prok-SAA) (Burroughs and Aravind, 2020; Zámocký and Ferianc, 2023), respectively, giving rise to their counterpart eukaryotic families (Met-Wnt and Met-SAA; Figures 1A and 4). This clade is structurally unified by the presence of a fifth helix that stacks in the space between the H2 and H4 helices (Figure 1A, Figure 1—figure supplement 2). In the Wnt families, this helix is comparable in length to the core helices, while in the SAA families, it is usually shorter (Figure 1A). The clade is further unified by the pronounced conservation of a sNxxGR motif (where ‘s’ is a small residue) encompassing the conserved active site position in H4 (Figure 2). SAW clade Lipocones show low overall hydrophobicity and are known or predicted to be soluble domains. Outside of the clearly inactive eukaryotic Wnt family, the remaining three families largely conserve the core active site residues (Figure 2).

This clade unites seven families: the two VanZ families, VanZ-1 and VanZ-2, prototyped by the bacterial VanZ protein originally identified in the context of vancomycin resistance, and the five Skillet families, which form a distinct subclade. These are unified by a ‘handle’-like structure (hence, ‘Skillet’), adopting a helical conformation in the H1-H2 linker (Figure 1A, Figure 1—figure supplement 2). Strikingly, a symmetric helical handle is present in the H3-H4 loop of the Skillet-DUF2809 and Skillet-3 families (Figure 1—figure supplement 2) of this clade. VanZ-1 features a conserved asparagine residue in the H2 position and a DxDDxxxN motif in H4, while VanZ-2 features RKxxH and DxxxD motifs in these respective positions (Figure 2). The Skillet families are largely unifiable in their conservation of an ExxQ motif in H3, an aspartate three positions upstream of the canonical H4 aspartate, and another aspartate in the H2 contributing to the active site. These first two features specifically ally them with the VanZ-1 family (Figure 2).

While the VanZ domain was previously reported as including a fifth TM helix, which is C-terminal to the 4-helix Lipocone core defined here (Woods et al., 2018; Sur et al., 2021), our survey instead reveals a striking diversity of configurations around the core 4-helix Lipocone domain (Figure 5—source data 2). These range from standalone Lipocone configurations to one or more TM-helices adorning the domain at its N- and/or C-terminus. This variation is consistent with a further tendency for the VanZ families to feature an extensive diversity of domain fusions to both soluble globular domains and discrete TM modules (see below).

The VanZ families are deep-branching, as suggested by their wide phyletic spread (Figure 1—figure supplement 1). VanZ-2 is the most widespread individual Lipocone family in bacteria, with several genomes encoding multiple paralogs (Figure 5—source data 2; Woods et al., 2018; Stogios and Savchenko, 2020). It is also found in certain eukaryotes, including a pan-fungal presence and in some representatives of the SAR clade. Both VanZ-1 and VanZ-2 are particularly well-represented in Gram-positive bacterial lineages like Actinomycetota and Firmicutes, while VanZ-2 is nearly universally conserved in the Bacteriodetes/Chlorobi lineage (Figure 1—figure supplement 1). In contrast, only one of the Skillet families, Skillet-DUF2809, is widely but sporadically distributed, with the four others being more restricted (Figure 4, Figure 1—figure supplement 1).

This clade includes three families that are consistently centrally located in the profile-profile similarity network (Figure 1B). This is likely due to their being close in sequence conservation to the ancestral state of the superfamily (Figure 2). Consistent with this, the YfiM-1 family also presents a structurally minimal Lipocone domain, comprised of just the 4-helix configuration with no further elaborations. Notably, this also extends to a lack of domain fusions in this family. In contrast, YfiM-DUF2279 and YfiM-Griddle (DUF3943) are structurally distinguished by an unusual H1-H2 linker (Figure 1A), which wraps around the outside and stacks against the H3-H4 linker (Figure 1—figure supplement 2). The YfiM-Griddle family further features a unique ‘flattened’ surface around the aperture of the Lipocone formed by protruding loops (hence, ‘Griddle;’ Figure 1—figure supplement 2). This leaves the active site pocket more accessible relative to families with more elaborately structured inter-helix linkers. The Griddle family also features a C-terminal extension with a two-helix hairpin (with a hhsP motif in the turn between the two helices, where ‘h’ is a hydrophobic residue and ‘s’ is a small residue) (Figure 1—figure supplement 2). The three YfiM families straddle the membrane-propensity boundary in the plot (Figure 1C). Furthermore, the YfiM-DUF2279 and Griddle families are strikingly absent in Gram-positive bacterial lineages (Figure 1—figure supplement 1). Concurrent with these features, they are often predicted by the deep-learning-based localization predictor deepTMHMM as outer-membrane proteins, suggesting a role in this subcellular location (see below).

Members of this clade are unified by an elaborated H1-H2 linker that often contains one or more helical segments that are typically predicted to guard the aperture of the Lipocone domain (Figure 1—figure supplement 2). This linker ends in a ‘clasp’-like element, which forms a range of structures in different families of the clade before leading into H2 (Figure 1—figure supplement 2). The clade is also unified by a striking reduction of overall hydrophobicity, predicting that the members of this clade are soluble domains (Figure 1C). Outside of the divergent TelC subclade, most of the families in this clade conserve a serine residue three positions upstream of the active site aspartate in H4, often preceded by an aromatic residue, which is typically phenylalanine. H4 also usually features a conserved asparagine four positions downstream of the conserved aspartate active site position, immediately preceded by a small residue (Figure 2). The second H3 active site position is generally poorly conserved, though when present, it is usually an aspartate residue. Finally, H2 contains either a DK or xD motif four positions upstream of the canonical H2 active site histidine residue (Figure 2).

The most rudimentary clasps are found in the ClaspCone-1,–2, and –3 families, where it is little more than a rounded loop, though, in ClaspCone-1, a small β-hairpin emerges within it. The three ClaspCone families are further unified by the presence of a two-helix insert leading into H2 that stacks against the Lipocone core (Figure 1—figure supplement 2). The three CapCone families, CapCone-DUF4056, CapCone-1, and CapCone-2, are named so for an encasing structure over the active site resembling a cap (Figure 1—figure supplement 2). They share a conserved glycine residue six positions upstream of the active site H2 histidine and a S/GxxSxx motif upstream of the conserved H4 aspartate (Figure 2). They are further unified by a pronounced β-hairpin clasp augmented by an additional strand (Figure 1—figure supplement 2). They also display varying degrees of degeneration of H1, along with family-specific structural elaborations.

The TelC group of this clade, prototyped by the streptococcal TelC toxin (Whitney et al., 2017), is divided into two families featuring prokaryotic (prok-TelC) (Whitney et al., 2017) and metazoan versions (Met-TelC) (Dziarski and Gupta, 2006a). Both TelC families feature a ‘cap’ with contributions from inserts in the H1-H2 and H3-H4 loops (Figure 1A, Figure 1—figure supplement 1). Unique to these families is the conservation of an aspartate residue located six positions downstream of the canonical active site aspartate of H4 (Figure 2). This aspartate points away from the center of the Lipocone and interacts with a conserved arginine from a synapomorphic C-terminal helical extension.

A widespread yet sporadically distributed clade of seven families emerging as a stable community in the profile-profile similarity network (Figure 1B) is defined by a unique structural synapomorphy: a circular permutation (Bliven and Prlić, 2012) (hence, cpCone) placing the normally N-terminal H1 at the C-terminus of H4 (Figure 1—figure supplements 2 and 4). This clade is also united by unique sequence features, viz., a polar residue (typically aspartate) six positions upstream of the conserved H2 histidine and a second glutamate three positions downstream of the conserved H3 glutamate (Figure 2). While the circular permutation is shared across the clade, several structural variations are seen, often within the same family (Figure 1—figure supplement 4). These include: (1) versions containing a duplication of the Lipocone domain. While the second copy in these versions is catalytically inactive, the H1’ from the second duplicate displaces the H1 from the first copy, suggestive of an intermediate to the circular permutation. (2) Versions retaining a candidate H1 that has been displaced by H1’ in a five-helix arrangement. (3) Those containing just the circularly permuted core. (4) Versions showing a degradation of the H1 helix, preserving just a 3-helix core (Figure 1—figure supplement 4). Despite this propensity for structural variation, the active site residues are strongly conserved, with the exception of the cpCONE-i family, which we infer to be catalytically inactive (Figure 2). The core helices of the cpCone clade are strongly hydrophobic, and they are all predicted to be integral membrane domains (Figure 1C). Consistent with this, the eukaryotic PTSSD1/2 domains reside in the ER membrane (Saito et al., 1996; Stone and Vance, 2000).

The Wok family (partly covered by the Pfam DUF2238 model) shows a higher order grouping with the above circularly permuted clade (Figures 1B and 4) but has a phyletic distribution only rivaled by the VanZ-2 family (Figure 1—figure supplement 1), suggesting a deep-branching origin. The shape of this family is reminiscent of a wok formed by two distinguishing structural synapomorphies: a 2-TM helix N-terminal extension and a unique ‘handle’ formed by the linker between the H3 and H4 (Figure 1—figure supplement 2). It additionally features a C-terminal, rapidly diversifying cytoplasmic tail. Despite these elaborations, it retains the inferred ancestral active site configuration (Figure 2). The strongly hydrophobic core helices of the Wok family predict it to be an integral membrane enzyme (Figure 1C).

Given our inference of shared general biochemistry across the Lipocone superfamily in targeting phosphate-containing linkages in head groups of both classic phospholipids and polyisoprenoid lipids, we next used contextual information from conserved gene-neighborhoods, domain architectures and phyletic pattern vectors, a powerful means of deciphering gene function (Aravind, 2000), to narrow down the predictions for specific families (Figure 5, Figure 5—source data 1 and Figure 5—source data 2). To this end, we constructed a graph (network) wherein the nodes are individual domains and edges indicate adjacency in domain architectures or conserved gene-neighborhoods (Figure 6, see Methods). We then identified cliques in these networks and merged the individual cliques containing a particular Lipocone domain to define its dense subgraph (Figure 6—figure supplements 1–3). We then analyzed these subgraphs to identify statistically significant functional categories represented in them (Figure 6—source data 1; see Methods). This data was combined with existing experimental results and the sequence and structure analyses outlined above to arrive at the functional themes surveyed in the below sections.

Across different Lipocone families, we found statistically significant connections to roles in modifying lipid head groups in various membranes and in lipids involved in the synthesis of extracellular matrix polymers such as peptidoglycan and lipopolysaccharides (Figure 6, Figure 6—source data 1, Figure 5—source data 2).

The Skillet-1 Lipocone is strongly coupled in an operon with a downstream gene coding for a protein with an unusual lipobox-like sequence followed by one of several extracellular domains (e.g., concanavalin, β-jelly roll, OB-fold, Ig-like, β-propeller) predicted to bind carbohydrates or other ligands (Kadirvelraj et al., 2008; Flint et al., 2004; Murzin, 1993; Williams and Barclay, 1988; Chen et al., 2011; Figures 5J and 6, Figure 6—figure supplement 1, Figure 5—source data 2). The lipobox-like sequence features a conserved GS motif at its C-terminus instead of the usual GC of the classic lipobox of bacterial lipoproteins (Babu et al., 2006; Figure 5—figure supplement 1). In the canonical lipoprotein processing pathway, a thioether linkage is formed between the sulfhydryl of the cysteine and a diacylglycerol lipid embedded in the inner membrane by the lipoprotein diacylglyceryl transferase (lgt) enzyme, followed by the cleavage of the signal peptide at the GC motif junction by the signal peptidase (Sankaran and Wu, 1994; Tjalsma et al., 1999). Given the serine in place of the cysteine in these lipobox-like sequences, we propose that it undergoes non-canonical lipidation by the associated Skillet-1 Lipocone protein in lieu of lgt. We propose that, comparable to PTDSS1/2, which act on free serine, the Skillet-1 family links the conserved serine from the lipobox-like sequence to a phospholipid (Figure 3A and D).

VanZ-1 proteins (Figure 1A) were initially identified as encoded by a gene linked to that coding for the VanY D-alanyl-D-alanine carboxypeptidase involved in resistance to glycopeptide antibiotics like vancomycin and teicoplanin (Arthur et al., 1995; Wright et al., 1992; Arthur et al., 1998; Arthur et al., 1994; Figure 6). These antibiotics bind the terminal D-Ala-D-Ala in the peptide moiety of peptidoglycan, preventing the transpeptidase cross-linking reaction necessary for its maturation. Upon detection of these antibiotics, enzymes encoded by the core vancomycin resistance operon re-engineer the exported peptidoglycan by inserting a D-Ala-D-Lac in place of the D-Ala-D-Ala linkage, precluding antibiotic binding (Stogios and Savchenko, 2020). The VanY peptidase, while not strictly required for antibiotic resistance, acts as an accessory to this system by cleaving any remaining D-Ala-D-Ala linkages generated via the canonical pathway (Stogios and Savchenko, 2020; Arthur et al., 1998). However, the role of VanZ in this system has so far remained unknown. While only a small fraction of the VanZ-1 genes are found in these antibiotic resistance contexts (Figure 5—source data 2), interestingly, other Lipocone genes, namely those of the VanZ-2 and the Skillet families, might also be linked to VanY in lieu of VanZ-1. Furthermore, VanY might be replaced by a structurally unrelated secreted D-Ala-D-Ala carboxypeptidase of the metallo-beta-lactamase fold (Palomeque-Messia et al., 1991) in operonic contexts with VanZ-1 (Figure 5V). Hence, given our above prediction regarding VanZ acting in peptidoglycan and/or exopolysaccharide metabolism, VanZ-1 and the Lipocones displacing it might indeed play an accessory role with VanY at the membrane (Wright et al., 1992; Arthur et al., 1998) in antibiotic resistance. We posit that, in these contexts, it likely acts on the head group of Lipid II to recycle canonical peptidoglycan intermediates for their accelerated or more thorough replacement with the resistant versions (Figure 3G).

We also identified a conserved five-gene operon featuring a YfiM-1 family Lipocone that might play a role in resistance to antibacterial agents (Figure 5W, Figure 6—figure supplement 1). Other than YfiM-1, this operon contains genes for: (i) a thioredoxin domain protein Qi and Grishin, 2005; (ii) A DTW clade RNA modifying enzyme of the SPOUT superfamily Meyer et al., 2016; Burroughs and Aravind, 2014; (iii) a protein with acyl-CoA ligase, GNAT superfamily N-acetyltransferase and ATP-grasp domains Dong et al., 2007; Fraser et al., 2002; Iyer et al., 2009; (iv) a PssA-like phosphatidylserine synthetase of the HKD superfamily (Raetz and Kennedy, 1974; Figure 5W, Figure 5—source data 2). Of these enzymes, the phosphatidylserine synthetase is predicted to act in its usual capacity to generate a lipid with a serine head group (Raetz and Kennedy, 1974). We propose that this would then function as a substrate for the YfiM-1 Lipocone domain, which might exchange the serine for another moiety via a reaction paralleling PTDSS1/2 (Figure 3A). This moiety could then be modified by aminoacylation, further acylation, and a redox modification by the third protein listed above, together with the thioredoxin. Indeed, such peptide modifications of lipid head groups by lysine, alanine, or arginine aminoacylation catalyzed by derived tRNA synthetases fused to GNATs have been shown to be a key resistance mechanism against breaching of the membrane by antibacterial peptides (Fields and Roy, 2018; Hancock, 1997). Hence, we predict the modifications catalyzed by this system might play a comparable role. The presence of a tRNA-modifying DTW domain suggests that in parallel to the tRNA synthetases, the GNAT in this system might use a tRNA-linked acyl group as a substrate, as seen in peptidoglycan biosynthesis (Benson et al., 2002; Schneider et al., 2004).

Lipocone domains have been transferred on several occasions from bacteria to eukaryotes (Figure 4, Figure 1—figure supplement 1). While there is predicted functional overlap with the above-described, predominantly bacterial versions, we discuss these separately as the inferred biological contexts of their deployment are often distinct from the above.

No single well-defined Lipocone clade is universally conserved across the three superkingdoms of Life (Figure 4, Figure 1—figure supplement 1, Figure 1—figure supplement 5). However, the VanZ and Wok clades are both found across all major bacterial phyla (notwithstanding sporadic losses in certain lineages) and in some archaeal lineages (Figure 1—figure supplement 1). At the same time, the cpCone clade is found across most major archaeal lineages and is nearly universally conserved in the eukaryotes (absent in Ascomycota and some choanoflagellates) (Figure 1—figure supplement 1). Notably, the cpCone and Wok clades tend to group together in the profile-profile similarity network (Figure 1B). These observations suggest that at least a single version of the Lipocone superfamily was likely present in the Last Universal Common Ancestor (LUCA). The phyletic patterns suggest that the LUCA Lipocone gave rise to the VanZ/Wok precursor in the bacterial lineage on the one hand and the cpCONE clade via a circular permutation event in the archaeo-eukaryotic lineage on the other (Figure 4). Based on the features of these deep-branching clades, the LUCA version is inferred to feature a hydrophobic domain with a 4TM helix core, with the active site facing the outer leaf of the lipid bilayer (Figure 1C). Given that extant versions operate both on classic phospholipids and isoprenoid lipids, it is difficult to infer which of these might have been substrates for the LUCA version. It is not impossible that this early version had a generic specificity that became specialized in the descendant clades.

The early diversification of the Lipocone domain appears to have had different drivers in the two prokaryotic superkingdoms. The presence of an extensive repertoire of exopolysaccharides in the cell wall (peptidoglycan, teichoic acids), cell surface (e.g., ECA), and outer membrane (e.g., lipopolysaccharide), synthesized via isoprenoid lipid-linked intermediates, like lipid-II, was the primary driver in the bacterial superkingdom (Egan et al., 2020). Here, this diversification yielded four monophyletic groups: the VanZs, Wok, YfiM, and Skillet (Figure 4). The deeper VanZ and Wok branches, which were likely recruited first for lipid-II-related functions, were probably the predecessors of the more restricted bacterial families with specialized functions. For instance, the emergence of the outer membrane in certain bacteria was potentially coupled with the origin of the YfiM-like clade (Figure 4). Similarly, our predictions suggest that within these clades, further diversification accompanied the acquisition of specialized functional roles in antibiotic resistance, secondary sensor roles in single and multicomponent signaling, and lipoprotein processing. The interoperability of Lipocone domains on lipid carriers shared across different biosynthetic pathways (see above, Figure 3, Figure 6—figure supplement 1, Figure 6—figure supplement 2) appears to have been a key factor leading to this versatility.

In the ancestral archaeo-eukaryotic lineage, the absence of peptidoglycan and an apparently lower diversity of structures with exopolysaccharides was reflected in the lesser diversification of the Lipocone clades (Figure 4). There are open questions regarding the biochemical functions of the primary archaeo-eukaryotic Lipocone clade, the cpCONE. Although the eukaryotic cpCone PTDSS1/2 family has been shown to swap serine for ethanolamine or choline in lipid head groups (Stone and Vance, 1999; Vance, 2018), their archaeal counterparts remain uncharacterized. Archaea have their own lipid with a serine in the head group (archaeophosphatidyserine), but to date, its synthesis has been shown to depend on a patchwork of different CDP-alcohol phosphatidyltransferase enzymes (CaPs) in different archaeal species (Koga and Morii, 2005; Koga and Morii, 2007; Daiyasu et al., 2005). While the CaPs are also integral membrane enzymes with a 6TM helix core, catalyzing comparable reactions as the Lipocones on lipid head groups in archaea and eukaryotes (Daiyasu et al., 2005), they are evolutionarily unrelated. Nevertheless, we suggest that the archaeal cpCones, like their eukaryotic counterparts, could contribute to distinct, as yet uncharacterized, pathways for the generation of cell membrane phospholipids like archaeophosphatidylserine or those with other head groups.

One of the remarkable aspects of the Lipocone superfamily is the loss of ancestral hydrophobicity in several families (Figure 1C), transforming them from integral membrane proteins to diffusible domains. While unexpected, such a transition in integral membrane enzymes acting on lipid substrates is not unprecedented. The PAP2 superfamily of integral membrane enzymes (e.g., diacylglycerol diphosphate phosphatase) (Stukey and Carman, 1997; Carman and Han, 2006) also contains several soluble versions (Neuwald, 1997) that appear to have emerged from an integral membrane ancestor (AMB and LA, unpublished observations). Most of the soluble Lipocone domains retain their active site conservation (Figure 2) and, at least in one experimentally characterized case, catalyze a comparable reaction as the TM version (Whitney et al., 2017; Figure 3B). The weight of the evidence presented here, including the profile-profile similarity network (Figure 1B), phyletic patterns (Figure 1—figure supplement 1), functional contexts (Figures 5–6), and the broadly shared structural features (Figure 1A, Figure 1—figure supplement 2, Figure 1—figure supplement 4), suggests that the loss of hydrophobicity occurred on a single occasion in the Lipocone superfamily, followed by diversification of these diffusible versions.

Our analysis of the diffusible Lipocone families reveals repeated recruitment as toxins/effectors in anti-viral and polymorphic toxin and allied systems (Zhang et al., 2012), suggesting that their diversification was driven by the arms races arising from the biological conflicts where they are deployed. Recruitment of a representative of the VanZ-1 family as a polymorphic toxin on rare occasions (Figures 5N and 6) suggests a possible evolutionary pathway for their recruitment as toxins: the effector version of Lipocones attacking lipids in competing bacteria likely emerged from an ancestral version that catalyzed endogenous lipid-head-group modifications on the same lipids in metabolic pathways. Once versions with reduced hydrophobicity emerged, they could be deployed as diffusible effectors that were shared across extracellular and intracellular conflict systems, a trend previously recognized in many other effector domains (Aravind et al., 2022).

Unlike bacteria, eukaryotes as a whole do not possess a rich repertoire of Lipocone domains. The PTDSS1/2 family, vertically inherited from the archaeal progenitor, is the only version that can be inferred as being present in the Last Eukaryotic Common Ancestor (Figure 4). However, distinct Lipocone families of ultimately bacterial provenance were acquired early and fixed in certain eukaryotic lineages: (i) YfiM-1 in the plant lineage; (ii) the fungal VanZ-2 domains typified by the Saccharomyces cerevisiae YJR112W-A; (iii) Met-Wnt (discussed further below) (Figure 1—figure supplement 1). The early fixation of these versions in the eukaryotic lineages possessing them suggests that they were recruited for definitive ‘housekeeping’ or developmental roles in the respective lineages. Beyond these, the fungal and metazoan lineages show more sporadically distributed versions, which have all been acquired from bacterial secreted-toxin or antiviral systems: (i) Min-Wnt independently in fungi and certain Metazoa; (ii) SAA; (iii) TelC; the latter two are absent in the basal-most metazoans, the sponges, but are present in Cnidaria, suggesting a relatively early acquisition (Figure 4). The weight of the evidence suggests that they have retained certain aspects of the ancestral bacterial effector function for anti-pathogen immunity in eukaryotes. This is consistent with both their episodic loss and lineage-specific expansion, the tendency to show rapid sequence divergence and, in the Met-TelC family, loss of catalytic activity (Figure 2, Figure 1—figure supplement 1, Figure 5—source data 2).

This independent acquisition of at least three distinct Lipocone families in metazoan immunity from polymorphic and allied effector systems of prokaryotes points to a persistent evolutionary trend. Notably, the Lipocone domains participating in animal immunity have been drawn from secreted effectors rather than the intracellular versions (bacterial intracellular Min-Wnts) predicted to participate in bacterial anti-selfish element immunity. More generally, this adds to a growing list of components drawn from secreted effector systems of prokaryotes in eukaryotic immune systems (Zhang et al., 2012; Aravind et al., 2012; Aravind et al., 2024). For example, this closely parallels another structurally unrelated effector domain, the Zn-dependent deaminase (e.g., metazoan AID/APOBEC deaminases) (Krishnan et al., 2018). Hence, these observations add further support to our hypothesis that the extensive expansion of effectors in diverse prokaryotic inter-organismal conflict systems served as a reservoir from which eukaryotic immune systems repeatedly acquired components (Aravind et al., 2012; Aravind et al., 2024). We propose that symbiotic associations between the early animals and bacteria resulted in potential interactions via secreted effectors of the latter that aided the former against antagonistic bacteria. This probably led to their eventual acquisition by animals and incorporation into their immune processes.

Earlier considerations on the evolution of Wnt signaling indicated that it emerged at the base of the metazoan lineage and incorporated a wide range of components of different origins (e.g., the HMG domain transcription factor TCF/LEF, the HEAT repeat protein β-catenin and the 7TM receptor Frizzled) (Richards and Degnan, 2009). However, the provenance of Met-Wnt itself had been mysterious and was seen as a possible example of a metazoan innovation (Holzem et al., 2024). While the Met-Wnt domains possess peculiar structural elaborations (Janda et al., 2012), its conserved core is a Lipocone domain (Figure 1A). We establish that the progenitor of Met-Wnt emerged as part of the radiation of Lipocone domains in bacteria as effectors deployed in both intracellular and inter-organismal conflict – the Min-Wnt proteins.

Whereas the Min-Wnt proteins are predicted to be secreted toxins, the Met-Wnts underwent an ancestral inactivation through loss of the catalytic residues (Figure 2). However, they retained their ancient involvement in cell-cell interactions as secreted agents. The Met-Wnt residues recognized as essential for the receptor (Frizzled) binding, including the absolutely conserved palmitoleoylated serine residue, are found in the aforementioned Metazoa-specific hairpins and loops (Janda et al., 2012; Zhong et al., 2021). However, despite their inactivation, the Met-Wnts retain the ancestral substrate-binding pocket (Figure 1A, Figure 1—figure supplement 2). This raises the possibility that they might be involved in as-yet unexplored interactions with ligands such as lipids.

Our tracing of the provenance of Wnt back to an effector in secreted bacterial toxin systems adds it to a growing list of components in metazoan signaling networks that have been acquired from such systems. For instance, this is also the case with components of the other key metazoan signaling pathway, Hedgehog (Zhang et al., 2011). Here, the Hedgehog protein itself contains an autoproteolytic HINT peptidase domain that was likely drawn from a structurally and functionally cognate domain observed in polymorphic toxin systems (Zhang et al., 2012; Zhang et al., 2011). Further, an intracellular component of the same signaling pathway, Suppressor of Fused (SuFu), was derived from a common immunity protein found in polymorphic toxin systems (Zhang et al., 2011). Similarly, the Teneurin/Odd Oz proteins mediating signaling in cell migration, neuronal pathfinding, and fasciculation in Metazoa descended from a polymorphic toxin protein with a C-terminal HNH endonuclease toxin tip (Zhang et al., 2012). In a similar vein, the immunity protein of certain CapCone toxins identified in this study might have given rise to the β-sandwich domain in the eukaryotic centriolar assembly factor SAS-6. These observations suggest that, in addition to immune system components, interactions with symbiotic bacteria also potentially furnished the progenitors of components of eukaryotic signaling and cytoskeletal networks that were central to the emergence of Metazoa as a clade of multicellular eukaryotes (Kaur et al., 2020; Kaur et al., 2021).

Using sensitive sequence and structure analysis, we unify a large, hitherto unrecognized superfamily of enzymatic domains, the Lipocone. By combining analysis of the active site and the structure of the Lipocone domain with contextual information from conserved gene-neighborhoods and domain architectures, we present evidence that members of this superfamily target phosphate linkages in head groups of both classical phospholipids and polyisoprenoid lipids. Specifically, they catalyze reactions such as head group exchange or severing of the head group-diphosphate linkage from the polyisoprenol. We present evidence that these activities have been recruited in a wide range of biochemical contexts, including cell membrane lipid modification, metabolism of peptidoglycan and exopolysaccharide lipid-carrier linked intermediates, lipoprotein modifications, bacterial outer membrane modification, sensing of membrane-associated signals, effector activity in antiviral and inter-organismal conflicts, and resistance to antimicrobials. Furthermore, catalytically inactive versions like Met-Wnt have been recruited for signaling roles in Metazoa. We predict the catalytic activity and potential biochemical pathways of numerous representatives for the first time, including some proteins that have remained enigmatic for over two decades, like VanZ.

We identify three notable trends in Lipocone evolution. First, although we reconstruct the ancestral member of the superfamily as being a 4TM integral membrane domain, a large monophyletic subset underwent a dramatic loss of hydrophobicity, transforming them into diffusible versions, including the Wnts and the SAAs (Figure 1C). Second, the superfamily expanded in two major functional niches in bacteria, namely peptidoglycan/exopolysaccharide metabolism and effector domains of both secreted toxins and immune systems (Figure 4). Finally, members of the Lipocone superfamily were acquired on multiple occasions from bacteria by Metazoan and were reused in new functional contexts as signaling messengers and immune factors (Figure 4).

Importantly, our predictions in this regard underscore that much remains unexplored in terms of lineage-specific cell wall and membrane metabolism in prokaryotes. We present several testable biochemical, functional hypotheses for the many poorly understood branches of the superfamily, several of which are being recognized as enzymatic for the first time here. We hope this will also open new avenues of research to fill key gaps in our understanding of lipid metabolism.

Sequence similarity searches were performed using PSI-BLAST (Altschul et al., 1997) and JackHMMER (Johnson et al., 2010) against the NCBI non-redundant protein database (nr) (Sayers et al., 2022) or a version clustered down to 50% sequence identity (nr50). The searches were initiated using the previously identified prokaryotic Wnt (Burroughs and Aravind, 2020), with multiple rounds of searches conducted, each using seeds collected from the preceding searches (Figure 1—figure supplement 5 file 5). Clusters based on sequence similarity (percentage identity or bit-score) were generated using MMseqs (Hauser et al., 2016). The clustering parameters were adjusted according to specific goals, enabling redundancy removal, the definition of homologous groups, and the creation of new profiles. Multiple sequence alignments (MSA) were generated using the MAFFT program (Katoh and Standley, 2013) with the local-pair algorithm, combined with the parameters –maxiterate 3000, –op 1.5, and –ep 0.2, and were manually refined based on structural superpositions and profile-profile comparisons.

The HHalign program (Steinegger et al., 2019) was used to perform profile-profile comparisons, with the resulting p-value and e-value scores serving as edges for constructing a superfamily relationship network. This was then analyzed using the Leiden community finding algorithm (Traag et al., 2019) to detect sub-networks. Network analysis and visualization were performed using the R igraph (Csardi and Nepusz, 2006) or Python networkX libraries (Hagberg et al., 2008).

Genomic neighborhoods were obtained from genomes available in the NCBI Genome database (Sayers et al., 2022) using in-house scripts written in Perl and Python. Conservation analysis of these genomic neighborhoods was performed by clustering the protein products of neighboring genes. Domain identification was conducted using a collection of HMMs and PSSMs maintained by the Aravind lab, along with HMMs from the Pfam database (Finn et al., 2016), utilizing the RPSBLAST (Schäffer et al., 1999) and HMMSCAN (Eddy, 2011) programs (Figure 1—figure supplement 1—l file 5). To further refine detection, domain identification was extended through remote homology analysis using the HHpred (Söding et al., 2005) program, against profiles built from the Pfam (Finn et al., 2016) and PDB70 (Berman et al., 2007) databases. Phylogenetic analyses were performed using FastTree (Price et al., 2010) and iqTREE2 (Minh et al., 2020). Experimental functional data for characterized members of the superfamily were collected with the assistance of the ChatGPT language model (https://chat.openai.com). Structural comparisons, along with shared genomic associations, were used to further refine the interrelationships within and between the groups of the superfamily.

Families with broader presence across multiple major lineages (‘phyla’) and deeper conservation within each of those lineages were inferred to be more ancient. In contrast, those with a more limited phyletic spread and/or limited depth of occurrence within each major lineage were likely later derivations (Figure 4, Figure 1—figure supplement 1). We formalized this inference by calculating a phyletic metric for the Lipocone clades (Figure 1—figure supplement 1) comprised of both the phyletic spread and depth. The phyletic spread Si\begin{document}$S_{i}$\end{document} of the ith Lipocone clade was computed thus:Si=miM,\begin{document}$$\displaystyle S_{i}=\frac{m_{i}}{M},$$\end{document}

Where mi\begin{document}$m_{i}$\end{document} is the number of lineages with at least one representative of the Lipocone clade Si\begin{document}$S_{i}$\end{document}, and M is the total number of lineages examined. The phyletic depth Di\begin{document}$D_{i}$\end{document} of the ith Lipocone clade was computed as a weighted average of its occurrence within each lineage in the form of the mediant:Di=∑j=1Mnj∑j=1MNj,\begin{document}$$\displaystyle D_{i}=\frac{\sum _{j=1}^{M}n_{j}}{\sum _{j=1}^{M}N_{j}},$$\end{document}

where n_j is the number of species in lineage j with a Lipocone domain of the ith Lipocone clade and N_j is the total number of species sampled in lineage j. Si\begin{document}$S_{i}$\end{document} and Di\begin{document}$D_{i}$\end{document} are plotted as a bar graph with Si\begin{document}$S_{i}$\end{document} as its width and Di\begin{document}$D_{i}$\end{document} its height.

Each domain architecture and conserved gene neighborhood was decomposed into its constituent domains. These domains were then labeled for their biochemical function and stored as a YAML file (Figure 6—source data 2). The contextual connections were then rendered as a graph with the domains as its nodes and the adjacency relationships as its edges. Cliques containing a given Lipocone domain were detected in this graph and merged to constitute their respective dense subgraphs. These subgraphs were then examined for the statistically significant prevalence of particular labeled functions using the Fisher exact test. Network analysis was performed using the functions of the R igraph or Python networkX libraries.

Protein structures were modeled using Alphafold3 (Abramson et al., 2024), with visualization and manipulation performed using either MOL* (Sehnal et al., 2021) or PyMOL. Structural similarity searches were conducted using the DALIlite (Holm, 2019) and FOLDSEEK (van Kempen et al., 2024) programs. DALIlite was also used to generate structural alignments.

To create the membrane propensity plots, for each protein P_i in a given family, we compute the average TM-propensity of its amino acids using the TM tendency scale (Zhao and London, 2006). This score H_i for P_i is calculated as:Hi=1n∑j=1nhj\begin{document}$$\displaystyle H_{i}=\frac{1}{n}\sum _{j=1}^{n}h_{j}$$\end{document}

where h_j is the TM tendency of the j-th amino acid in the protein P_i, and n is its total length in amino acids (Figure 1—source data 1). The Kruskal–Wallis nonparametric test was applied to assess whether TM propensity scores differed across the 30 groups. As the Kruskal–Wallis test indicated a significant difference (p<0.05), we performed post-hoc pairwise comparisons using Dunn’s test with Bonferroni correction to control for multiple testing. Group-wise visualizations were presented using critical difference diagrams, where groups not connected by horizontal bars are significantly different (adjusted p<0.05) (Figure 1—figure supplement 3).