Coevolution of Eukaryote-like Vps4 and ESCRT-III Subunits in the Asgard Archaea

The discovery of Asgard archaea has changed the existing ideas on the origins of eukaryotes. Researchers propose that eukaryotic cells evolved from Asgard archaea. This hypothesis partly stems from the presence of multiple eukaryotic signature proteins in Asgard archaea, including homologs of ESCRT proteins that are essential components of the endomembrane system in eukaryotes. However, structural and functional features of Asgard ESCRT remain unknown. Our study provides evidence that Asgard ESCRT is functionally comparable to the eukaryotic counterparts, suggesting that despite the apparent absence of endomembranes in archaea, eukaryotic ESCRT was inherited from an Asgard archaeal ancestor, alongside the emergence of endomembrane system during eukaryogenesis.

tween the ESCRT-III subunits and Vps4 of Asgard archaea and eukaryotes, to the exclusion of other archaea.

RESULTS
Eukaryotic-like ESCRT-III subunits and Vps4 in Asgard archaea. Given that the ESCRT-III subunits are tightly linked to the functional complexity of ESCRT (12), we first performed a detailed sequence comparison and phylogenetic analysis of the Vps2/ 24/46 and Vps20/32/60 as well as the Vps4 ATPase from Asgard archaea based on the available genomic data (18,19). In the unrooted maximum likelihood phylogenetic tree Vps2/24/46 and Vps20/32/60, the Asgard proteins form a cluster with eukaryotic homologs that is separated from the archaeal (TACK) CdvB cluster by a long branch ( Fig. 1A; see also Fig. S1 and Table S1 in the supplemental material), supporting the notion that Asgard archaea possess "eukaryote-like" ESCRT-III subunits. All the Asgard Vps20/32/60 proteins form a strongly supported clade with the eukaryotic Vps20/32/60 which is compatible with a direct ancestral relationship. The Asgard Vps2/24/46 proteins formed three clades, one of which (Odinarchaeota, Lokiarchaeota, and Thorarchaeota) clustered with the eukaryotic homologs, whereas the remaining two (Heimdallarchaeota) placed near the root of the Asgard-eukaryote branch (Fig. 1A). This tree topology probably resulted from the acceleration of evolution in Heimdallarchaeota.
In addition to the phylogenetic results, we found that the Asgard Vps2/24/46 contained leucine-rich motifs located in the C-terminal helix and resembling the C-terminal MIM1 that are conserved in eukaryotes although some leucine residues were  Table S1 in the supplemental material. Bootstrap values are shown for some nodes. (B) Predicted MIM1 and MIM2 in Asgard Vps2/24/46 and Vps20/32/60, respectively. Additional information of proteins used here can be found in Table S1. The ESCRT-III core domain, C-terminal helix, and MIM1 and MIM2 are presented.
Vps4 and ESCRT-III in Asgard Archaea ® substituted by isoleucine in the Asgard homologs ( Fig. 1B) (9). The C-terminal regions of the Asgard Vps20/32/60 contain proline-rich motifs that resemble MIM2, although they do not fully conform to the MIM2 consensus in eukaryotes and TACK archaea (10,11). Taken together, the results of phylogenetic analysis and motif search for ESCRT-III subunits not only demonstrate the Asgard-eukaryote affinity but also show that the ancestors of the Vps2/24/46 and Vps20/32/60 groups had already diverged in Asgard archaea, antedating eukaryogenesis.
It appears likely that Vps4 structurally and functionally coevolved with ESCRT-III subunits in Asgard archaea. To explore the evolution of Vps4, an unrooted maximum likelihood phylogenetic tree was constructed for the group of ATPases, including CdvC from the TACK superphylum, Asgard Vps4, and the so-called eukaryotic "meiotic clade" comprised of Vps4, katanin 60, and spastin (23). As in the ESCRT-III subunit tree, the Asgard Vps4 formed a branch with the eukaryotic homolog that was separated by a long, strongly supported branch from the archaeal CdvC branch ( Fig. 2A and Fig. S2). The Asgard Vps4 did not form a single clade, but rather four clades, all of which were located close to the root of the Asgard-eukaryote branch.
Despite their high divergence demonstrated by the lack of monophyly in the phylogenetic tree ( Fig. 2A and Fig. S2), all Asgard Vps4 contain the eukaryotic-like "arginine collar" that consists of three conserved arginine residues (Fig. S3A). In eukaryotes, this motif is located in the pore loop 2 of Vps4 and is involved in the ESCRT-III filament translocation to the central pore of the Vps4 hexamer for disassembly ( Fig. S3B) (24,25).
Because Vps4 recognizes ESCRT-III subunits via the MIT domain, we specifically analyzed the phylogeny of the MIT domains of the Vps4 proteins from Asgard archaea, eukaryotes, and TACK archaea. The tree demonstrates a clear affiliation of Asgard archaea with eukaryotes that, in this case, form a clade with one of the MIT domains from Heimdallarchaeota ( Fig. 2B and Fig. S4). Affiliation with Heimdallarchaeota has been previously observed for many Asgard archaeal genes (19,26,27).  Table S1. Bootstrap values are shown for some nodes. To further structurally characterize the MIT domain in Asgard Vps4, we constructed stable models of full-length Vps4 from Heimdallarchaeota_LC_3 (Heimdall_LC_3), Odi-narchaeota_LCB_4 (Odin_LCB_4), Thorarchaeota_AB_25 (Thor_AB_25), and Lokiarchae-um_GC14_75 (Loki_GC14_75) using homology modeling and molecular dynamics simulation, and compared these with the S. cerevisiae Vps4 structure. For a control for the Asgard Vps4, we include CdvC from Sulfolobus solfataricus_P2 (cluster I in Fig. 2A) and Bathyarchaeota (cluster II in Fig. 2A). All MIT domains of Asgard Vps4 and TACK CdvC adopted a three-helix bundle structure that is closely similar to the S. cerevisiae MIT domain structure, although the helices in both the Asgard and TACK structures are somewhat shorter than in the S. cerevisiae structure ( Fig. 2Bb and Fig. S5).
Taken together, the above data suggest that the evolution of Asgard Vps4, especially their MIT domain, was accompanied by the functional divergence of the ESCRT-III subunits. Thus, although the Asgard Vps4 proteins are highly diverged, the results of sequence comparison, phylogenetic analysis, and structural modeling are compatible with coevolution of Vps4 with ESCRT-III subunits and an ancestral relationship between the membrane remodeling machineries of Asgard and eukaryotes. Furthermore, it can be predicted that Asgard Vps2/24/46 and Vps20/32/60 form ESCRT-III-like filaments similar to those in eukaryotes.
We further analyzed the structural basis for the MIT domain of Asgard Vps4 binding to the putative MIM1 and MIM2 of Vps2/24/46 and Vps20/32/60, respectively, by using MM-GBSA calculations (28). The key amino acids that contribute to the Vps4 MIT domain binding to the Vps2/24/46 MIM1 in Heimdall_LC_3, Odin_LCB_4, and Thor_ AB_25 are mainly located in helix 2 and helix 3 of the MIT domain similar to the locations of MIM1-interacting residues in S. cerevisiae Vps4 ( Fig. 3A and Table S3). These findings are consistent with the MIM1 peptide binding at the interface between helix 2 and helix 3 of the MIT domain as observed in eukaryotes (10,30). In Loki_GC14_75, the key amino acid residues are located in helix 1 and helix 2, suggesting a distinct interaction mode.
The key residues involved in the MIT-Vps20/32/60 interactions are spread among helix 1, helix 2, and helix 3, in positions closely similar to those involved in the MIT-Vps20 interactions in S. cerevisiae ( Fig. 3B and Table S4). Thus, the MIM2 peptide is predicted to bind the grooves formed by the three-helix bundle rather than helix 1 and helix 3 only as also observed for the eukaryotic ESCRT-III (10, 11). Taken together, these findings indicate that the interactions of the Asgard Vps4 MIT domain with the MIM1 (in Vps2/24/46) and MIM2 (in Vps20/32/60) closely resemble the corresponding interactions in eukaryotes.
Asgard Vps4 phenotypically complements the vps4 null mutation in S. cerevisiae. We further sought to determine whether the Asgard and eukaryotic Vps4 ATPases were functionally interchangeable. To this end, Heimdall_LC_3, Odin_LCB_4, Thor_ AB_25, and Loki_GC14_75 were tested for the ability to complement the S. cerevisiae vps4 null mutation. For a control for the Asgard Vps4, we performed the complementation assays with CdvC from S. solfataricus_P2 and Bathyarchaeota. Briefly, we performed codon optimization again for the coding sequences of Asgard Vps4 and TACK CdvC for expression in S. cerevisiae and assembled the coding sequences into transcription units of the pPOT-RFP vector that contains a native promoter region of S. cerevisiae BY4741 vps4 (a 500-bp DNA sequence region upstream from the ATG start codon of this gene) and an S. cerevisiae cytochrome c isoform 1 (CYC1) terminator using the YeastFab Assembly method (31), respectively. The assembly products were transformed into an S. cerevisiae vps4Δ strain by the lithium acetate (LiAc)/polyethylene glycol (PEG) method (32). As previously described, in S. cerevisiae, vps4 null mutation resulted in temperature-sensitive growth defect, causing growth arrest at 39°C (33,34). We found that the Asgard Vps4 could slightly suppress the growth defect of vps4⌬ strain at 39°C (Fig. 4A). Remarkably, however, after incubation at 39°C for 96 h, the growth of a vps4⌬ strain bearing Asgard Vps4 showed substantial, although variable, restoration at 30°C, in sharp contrast with a vps4⌬ strain for which no restoration was observed (Fig. 4A). Nevertheless, both CdvCs showed only minimal growth restoration of the vps4⌬ strain at 39°C and a limited enhancement of viability at 30°C; complementation with these proteins was substantially less efficient than that observed with their Asgard counterparts. Furthermore, the S. cerevisiae Vps4, Asgard Vps4, and CdvCs were re-codon-optimized, synthesized, and cloned into a pCold-TF vector (TaKaRa Bio Co. Ltd., Japan), respectively. After expression in Escherichia coli BL21, proteins were purified by Mag-Beads His-Tag Protein purification kit (BBI Co., Ltd., China). The biochemical experiments in vitro show that these purified proteins are active ATPases both Vps4 and ESCRT-III in Asgard Archaea ® at 30°C and 39°C (Fig. 4B). This observation eliminates the possibility that the poor complement result of CdvCs was due to the lack of ATPase activity at 39°C and is compatible with the involvement of the ATPase activity of Asgard Vps4 in sustaining the viability of the S. cerevisiae vps4⌬ mutant under nonpermissive conditions.
As previously described, vps4 null mutation could induce formation of an aberrant prevacuolar compartment adjacent to the vacuoles, known as class E compartment, due to the block of intracellular protein trafficking (3,30,33). To further demonstrate that Asgard Vps4 is functionally analogous to its eukaryotic counterpart, we observed the vacuoles in the S. cerevisiae cells bearing Asgard Vps4. As expected, the characterized class E compartment vacuolar morphology was clearly observed in the S. cerevisiae vps4⌬ cells, and this defect was almost completely rescued by S. cerevisiae Vps4 (Fig. 4C). We found that the Vps4 of Heimdall_LC_3, Odin_LCB_4, Thor_AB_25, and Loki_GC14_75 also partially complemented the aberrant vacuoles in the vps4 null mutant, with reduction of the class E compartment to about 80% of that observed with the native S. cerevisiae Vps4 (Fig. 4C). However, the enlarged vacuoles induced in the vps4⌬ strain were not markedly eliminated by the Asgard Vps4. Taken together, these findings show that the Asgard Vps4 are functionally more similar to the eukaryotic homologs than homologs from other archaea.

DISCUSSION
In this work, we combined computational approaches, including sequence comparison, phylogenetic analysis, and structural modeling, with genetic and biochemical experiments to investigate the evolutionary and functional relationships between the ESCRT-III machineries of Asgard archaea and eukaryotes. Phylogenetic analyses of both the ESCRT-III subunits and Vps4 ATPase show a clear affinity between Asgard archaea and eukaryotes, to the exclusion of the other archaea. Moreover, the divergence of the two groups of ESCRT-III subunits already occurred in Asgard archaea.
The results of amino acid sequence analysis and structural modeling are best compatible with the coevolution of Vps4 with the ESCRT-III subunits. In particular, the interaction between the MIT domain of Vps4 and the MIM1-and MIM2-like of the ESCRT-III subunits appears to have evolved already in Asgard archaea.
The findings of the computational analysis are complemented by our experimental results. In particular, we show that Asgard Vps4 is capable of complementing the S. cerevisiae vps4 null mutant much more efficiently than homologs from Crenarchaeota and Bathyarchaeota. This enhanced functionality might be underpinned by the evolution of distinct, "eukaryote-like" structural features, such as the arginine collar that is involved in the disassembly of ESCRT-III polymers.
Taken together, all these findings are compatible with the direct origin of the eukaryotic ESCRT machinery from the Asgard ancestor. In a broader evolutionary context, the ESCRT complex likely evolved in the common ancestor of the TACK and Asgard superphyla, whereas its further elaboration occurred in the Asgard lineage. The key event apparently was the duplication of CdvB that seems to combine features of Vps2/24/46 and Vps20/32/60 (12), with subsequent functional diversification of the subunits and coevolution with Vps4.
An intriguing outstanding question is the function of the ESCRT machinery in the Asgard archaea. There is no indication that these (or any other) archaea possess intracellular membranes (22), so the ESCRT-III proteins and Vps4 are likely to be involved in cell division as demonstrated for the Cdv proteins of Crenarchaeota. However, the specialization of the ESCRT-III subunits might provide for the formation of eukaryote-like filaments that could be involved not only in the inside-out fission to produce membrane vesicles that have been observed in the MK-D1 strain, but also the outside-in fission that allows the Asgard archaea to engulf their bacterial metabolic partners. The latter capability is critical for the "Entangle-Engulf-Enslave model" of eukaryogenesis (22). Further molecular and cell biological study of the Asgard membrane remodeling apparatus, even if challenging due to the recalcitrance of these organisms to growth in culture, should shed light on the origin of the eukaryotic endomembrane system, one of the key aspects of eukaryogenesis.

MATERIALS AND METHODS
Bioinformatics analysis. All the protein sequences were obtained either by NCBI accession number or by BLAST search (35) (39); subunit B was chosen for the modeling template, and the missing residues (1 to 118) were built at the I-TASSER server (http://zhanglab .ccmb.med.umich.edu/I-TASSER). Sequence alignments and homology modelings of Vps4 for Lokiarchaeota, Thorarchaeota, Heimdallarchaeota, and Odinarchaeota with unknown structures were conducted using the MODELLER program (40), downloaded and installed from the salilab server (https://salilab.org/modeller/download_installation .html). The three-dimensional structures of Vps2/24/46 and Vps20/32/60 for S. cerevisiae and four Asgard archaea were also built at the I-TASSER server. Among several three-dimensional models generated using homology modeling and ab initio method, the best model was selected after a series of refining and minimization and molecular dynamics simulation employing ff14SB force fields parameters by AMBER 16.0 package (41). Then the complexes of Vps2/24/46 and Vps20/32/60 against Vps4 were simulated using the ZDOCK server (42). Ten top docking poses were generated.
Molecular dynamics simulation. The parallel version of AMBER 16.0 package was used to prepare the complex files and conduct molecular dynamics (MD) simulations employing ff14SB force field parameters. The ionizable residue default protonation states in AMBER 16.0 were assigned. All MD simulations were conducted by applying cubic periodic boundary conditions (PBC) and in an explicit water box of TIP3P (transferable intermolecular potential with three points) water molecules (43) with a minimum distance of 10.0 Å between the complex surface and water box boundary. The Na ϩ or Cl Ϫ counterions were added in sufficient number to neutralize any net charges of the structures above. All of the chemical bond lengths of hydrogen-heavy atoms were restrained by the SHAKE algorithm (44). A cutoff radius of 10.0 Å was set for both nonbonded electrostatic and van der Waals interactions. Long-range electrostatic forces were taken into account using the particle mesh Ewald (PME) method (45). Langevin dynamics and Langevin piston methods were applied to keep the temperature (300 K) and pressure (1 bar) of the system constant, respectively. The time step was set at 2.0 fs.
The solvated systems were minimized using the PMEMD.CUDA module enabled with NVIDIA graphics processing units (GPUs) (46,47) in three stages: (i) keeping the solute fixed and minimizing the positions of the water and counterions first with 100 kcal/(mol·Å 2 ) restraints and (ii) reducing to 10 kcal/(mol·Å 2 ), and (iii) last, for the entire system without any restraining force. Each stage was conducted with 10,000 steps of steepest descent algorithm, followed by 1,000 steps of conjugate gradient minimization to get rid of any unfavorable steric contacts for both solvent and protein Vps4 and ESCRT-III in Asgard Archaea ® molecules. Then, a NVT (amount of substance [N], volume [V], and temperature [T]) simulation was conducted to slowly heat the systems temperature from 0 K to 300 K over a period of 500 ps, and density was equilibrated for 2,000 ps with a weak restraint applied to the whole protein at 1 atm and 300 K. Finally, all restraints were removed, and production MD simulations were conducted at constant pressure (1 atm) and temperature (300 K) in the NPT (amount of substance [N], pressure [P], and temperature [T]) ensemble. For each system, MD simulation was performed for 500 ns and repeated thrice with different random number, and a total of 1.5-s trajectory was analyzed by using a CPPTRAJ module (48).
Calculations of binding free energy. The binding free energies of Vps2/24/46 and Vps20/32/60 against Vps4 were calculated by molecular mechanics-generalized Born surface area (MM-GBSA) method (28,29). All energy components were calculated using 500 snapshots that were extracted every 200 ps during the last 100 ns of each MD simulation trajectory. The configurational entropy was not considered in the approach, as it is extremely time-consuming. Therefore, the binding free energy in the solvent environment can be expressed as follows: The ΔE ele , ΔE vdw , ΔG np , and ΔG ele represented electrostatic energy in the gas phase, van der Waals energy in the gas phase, nonpolar solvation energy, and polar solvation energy, respectively. All energy terms were calculated using MM-GBSA calculations, and the ΔG ele is estimated by the generalized Born (GB) model (29), and the ΔG np is calculated from the solvent-accessible surface area (SASA) of the molecules by molsur with the 0.00542 and 0.92 values for SURFTEN and SURFOFF, respectively (49). The decomposition of binding free energies were calculated at the residue pair level for a further investigation of the interactions between complexes using the MM-GBSA decomposition program (50,51)  ), and respectively, cloned into a pCold-TF vector (TaKaRa Bio Co. Ltd., Japan) that includes an N-terminal His tag and a soluble trigger factor chaperone tag. The E. coli BL21 (TaKaRa Bio Co. Ltd., Japan) bearing the recombinant vectors were inoculated in LB medium containing 100 g/ml carbenicillin, and incubated at 37°C until the optical density at 600 nm (OD 600 ) reached 0.6 to 0.8, and then isopropyl-D-1thiogalactopyranodside was added at the final concentration of 0.5 mM, followed by incubation at 15°C for 18 to 24 h. The cell pellets were collected and resuspended in 20 ml binding buffer (20 mM phosphate buffer [pH 7.4], 500 mM NaCl, 50 mM imidazole, 1 mM dithiothreitol, 1 mM lysozyme, and 1 mM phenylmethylsulfonyl fluoride), followed by ultrasonic decomposition. Next, the target proteins were purified by Mag-Beads His-Tag Protein purification kit (BBI Co., Ltd., China) with wash buffer (20 mM phosphate buffer [pH 7.4], 500 mM NaCl, 100 mM imidazole, and 0.1% NP-40) and elution buffer (20 mM phosphate buffer [pH 7.4], 500 mM NaCl, and 500 mM imidazole). Finally, the purified proteins were concentrated to 1 to 2 ml in phosphatebuffered saline (PBS) (pH 7.4) by 30K Amicon Ultra-15 (Millipore Co. Ltd., USA). The concentrations of these proteins were determined by Bradford Protein assay kit (Beytotime Bio Co. Ltd., China). The purified Vps4, Vps2/24/46, and Vps20/32/60 belonging to S. cerevisiae, Lokiarchaeum_GC14_75, Thorarchaeota_AB_25, Heimdallarchaeo-ta_LC_3, and Odinarchaeota_LCB_4 were used for isothermal titration calorimetry assay. The purified Vps4 belonging to S. cerevisiae, Lokiarchaeum_GC14_75, Thorar-chaeota_AB_25, Heimdallarchaeota_LC_3, and Odinarchaeota_LCB_4, and Cdv belonging to Sulfolobus solfataricus_P2, and Bathyarchaeota were used for ATPase activity assay.
Isothermal titration calorimetry assay. The isothermal titration calorimetry (ITC) assay was conducted at 25°C using an ITC200 system (MicroCal, USA). The Vps4 MIT domain (3 M in PBS buffer) was placed in the cell and titrated with 19 injections of 10 l of Vps2/24/46 or Vps20/32/60 (33 M in PBS buffer) at 2-min intervals. The heat of ligand dilution into buffer was subtracted from the reaction heat, after removing the data of first injection. Data analysis was conducted using Origin 7.0 (MicroCal, USA).
ATPase activity assay. The ATPase activity was determined by a slightly modified malachite green assay (52). In short, the purified proteins (4 M) were incubated with reaction buffer (1 mM ATP, 20 mM HEPES [pH 7.4], 100 mM NaCl, 10 mM MgCl 2 , 1 mM dithiothreitol [DTT]) in a total volume of 50 l at the indicated temperature for 90 min, and was immediately stopped by liquid nitrogen. Then, the reaction mixture was added with 100 l of malachite green color buffer (14 mM ammonium molybdate, 1.3 M HCl, and 1.5 mM malachite green) and 50 l of 21% (wt/vol) citric acid, followed by incubation at room temperature for 30 min. Finally, the reaction mixture that turned green was attributed to the free phosphate released by Vps4 ATP hydrolysis. Additionally, the control experiments were identical to the treatment group, except that the mixture of Vps4 and reaction buffer was immediately treated with liquid nitrogen before the addition of malachite green color buffer and citric acid, and these experiments were performed to eliminate the interference of irrelevant free phosphate. Also, the empty vector was used to prove that the ATP hydrolysis is due to Vps4.
Complementation assay. The Vps4 coding sequences belonging to Lokiarchaeum _GC14_75, Thorarchaeota_AB_25, Heimdallarchaeota_LC_3, and Odinarchaeota _LCB_4, and CdvC coding sequences belonging to Sulfolobus solfataricus_P2, and Bathyarchaeota (NCBI:protein accession numbers KKK42121.1, OLS30569.1, OLS27542.1, OLS18192.1, AAK41192.1, and WP_119819537.1, respectively) were codon optimized by GeneDesign (http://54.235.254.95/gd/) for expression in S. cerevisiae, before synthesis by BGI Genomics Co., Ltd. (54). To eliminate the interference of transcriptional level factors, a native promoter region of S. cerevisiae BY4741 vps4 (a 500-bp DNA sequence region upstream from the ATG start codon of this gene) was used to drive the coding sequences. Then, we assembled the coding sequences, the S. cerevisiae vps4 native promoter, and an S. cerevisiae CYC1 (cytochrome c isoform 1) terminator into a pPOT-RFP vector according to a developed YeastFab Assembly protocol (31). In addition, the pPOT-RFP vector containing the entire S. cerevisiae BY4741 vps4 with its native promoter and the CYC1 terminator were transformed into the S. cerevisiae vps4 null mutant (32), and this reconstituted strain was designated the "ϩS. cerevisiae" strain shown in the figures. In this study, both the S. cerevisiae and S. cerevisiae vps4⌬ were transformed with the pPOT-RFP vector as the control.
FM-64M staining. S. cerevisiae cells of each strain were cultured in SC-Ura medium at 30°C and normalized to an OD 600 of 0.5 to 0.8. Then, the S. cerevisiae cells were stained with 80 M FM-64M (AAT Bioquest Co. Ltd., China) at 30°C for 20 min, and next cultured for 120 min after washes with medium. Finally, the S. cerevisiae cells were examined under an N-STORM fluorescence microscope (Nikon Co. Ltd., Japan).

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.   We declare that we have no conflicts of interest.