Main

Across the three domains of life, organisms have evolved diverse macromolecular machines for motility and propulsion1. In Archaea, motility is driven by the archaellum, a rotary, ATP-powered nanomachinery unrelated to the bacterial flagellum despite functional resemblance1. The archaellum belongs to the type IV filament (TFF) superfamily, which includes various archaeal surface structures such as the Ups and Aap pili and the bindosome2,3,4,5. TFF systems share a conserved four-protein core with system-specific accessory components, and the archaellum is the only rotary member2. Its filament is composed of archaellins (ArlB or ArlA), processed by a class III signal peptidase6. ArlI is the ATPase that powers the assembly and rotation of the archaellum filament, probably in interaction with the membrane platform protein ArlJ7,8,9 (Fig. 1a). ArlH is a KaiC homologue and can regulate ArlI via autophosphorylation while ArlF and ArlG form the stator complex for torque generation10,11,12,13 (Fig. 1a). In Thermoproteota, ArlX supports the core complex14, while in Euryarchaeota, ArlCDE are important to link the motor complex with the chemotaxis arrays3,15. These genes cluster into 7–11-gene loci essential for archaellation and motility, with gene loci differences defining arl1 and arl2 variants16,17.

Fig. 1: Cell surface machineries in Chloroflexota.
figure 1

a, Schematic representation of searched macromolecular systems and associated proteins. Each system with the corresponding proteins was searched in Chloroflexota genomes using MacSyFinder2. The present subunits are coloured accordingly, and grey subunits were not found. Of the 3,780 genomes searched, 233 potentially flagellated members were found. Around 1,234 encoded a Tad pilus, and 340 have loci for a type IVa pilus. A total of 243 genomes encoded a complete archaellum machinery. b, Reference phylogeny of Chloroflexota at the order level with the relative frequency of each TFF and flagellum systems by order. Order names correspond to GTDB taxonomy. The full species tree of Chloroflexota is in source data. Archaellum machineries are found in most of the orders of Chloroflexota.

Source data

So far, it was thought that the archaellum was exclusive to Archaea1. However, cultivation of members of the SAR202 clade (Chloroflexota) revealed a complete archaellum operon, including multiple arlB copies18; despite this, no surface filaments were observed in these isolates. Although archaellum clusters have been found in some metagenome-assembled genomes, these lacked the processing peptidase PibD/ArlK, suggesting a non-functional system18,19.

In this study, we screened a curated selection of public prokaryotic genomes using MacSyFinder2 and identified complete archaellum gene loci for several Chloroflexota20. An in-depth analysis of 3,780 genomes of Chloroflexota revealed the presence of the archaellum operon in 243 members, including the cultivated species Litorilinea aerophila. We showed the expression and assembly of a functional archaeal-like archaellum structure by L. aerophila, which is used for swimming motility. Cryogenic electron microscopy (CryoEM) single-particle analysis of the purified archaella filaments showed remarkable structural similarity to the archaellum. The preservation of the structural features of the bacterial archaellum, analogous to that of its archaeal equivalent, suggests a conserved mechanism for archaellum-driven swimming in L. aerophila. Phylogenomic analysis of the archaeal type IV pili (T4P) in archaea and bacteria revealed one horizontal gene transfer (HGT) of the archaeal pilus from Archaea to few members of Chloroflexota, and one transfer of the archaellum to members from Dehalococcoidia and Anaerolineae, with Methanotecta as their closest relatives. The monoderm envelope architecture found in Chloroflexota might have eased the successful incorporation of the archaellum machinery into the bacterial envelope.

Results

Mutually exclusive assembly of archaella and flagella in Chloroflexota

Previous studies on Chloroflexota metagenomes and the first cultivated member of the SAR202 clade (‘Candidatus Lucifigimonas marina’) indicated the presence of genes related to archaella in Bacteria18,19. A locally maintained database of archaeal and bacterial genomes with a representative taxon sampling was screened for the presence of TFF and flagellar macromolecular systems using MacSyFinder2 (ref. 20). We found that Bacteria lack the genes encoding the archaellum machinery, with the notable exception of a few members of Chloroflexota. We, therefore, focused on a more detailed analysis of 3,780 available genomes of Chloroflexota. We found that 243 genomes appear to encode the archaellum machinery, while 233 genomes contain flagellar-related genes. Ten genomes that encoded a complete archaellum machinery included a second locus that was incomplete in nine of them. Around 1,234 members encode a Tad pilus, and we detected 340 genetic loci coding for a type IVa pilus (Fig. 1a,b and source data). When the archaellum machinery genes were present, flagellar-related genes were not found; however, some genomes additionally encode a Tad or type IVa pilus (Fig. 1b and source data).

Using TXSScan with MacSyFinder, we identified core components of the bacterial flagellum in 233 Chloroflexota genomes. Among these, FlgE, FlgD, FlgK, FlgL, MotA and MotB were found in 183, 213, 233, 233, 148 and 221 genomes, respectively (source data). The Tad pilus was present in roughly one-third of the genomes, and both Tad and type IVa pili systems lacked secretins, consistent with the absence of an outer membrane in this phylum21 (Fig. 1a,b and source data). Remarkably, archaellum loci were the most complete, often containing all genes required for a functional motility system (Fig. 1a and source data). When core genes (arlJIH, arlB) were present, accessory components such as arlFG and arlCDE were typically also detected. These findings indicate that many Chloroflexota encode bona fide archaellum machineries.

Some Chloroflexota encode a complete archaellum assembly machinery

We identified 244 complete archaellum clusters in Chloroflexota across various orders (source data). Representative clusters from SAR202, Anaerolineales, Dehalococcoidales and Thermofilales were compared with those of known motile archaeal species22. All archaellum-encoding Chloroflexota possess arlB in single or multiple copies (Fig. 2b and source data), with downstream genes forming a single locus for the machinery. Like Euryarchaea, most Chloroflexota encode arlCDE homologues instead of the Thermoproteota-specific arlX. However, unlike species such as Haloferax volcanii or Pyrococcus furiosus, Chloroflexota have arlG and arlF in reverse order. This distinction reflects arl1 (arlFG) versus arl2 (arlGF) locus organization17. Some Chloroflexota show further divergence, such as duplication of arlF and variation in arlHIJ gene order—although all genes remain within one locus. The class III signal peptidase (PibD/PilD), essential for ArlB processing, was found outside the archaellum locus; SAR202 members have archaeal-like PibD, while others have bacterial-type PilD. Additional nearby genes include pilN and pilO with a LysM domain in L. aerophila both typically part of the bacterial type IV pilus alignment complex (Fig. 2)23. Overall, the gene arrangement and co-occurence in Chloroflexota mirror those in Archaea, suggesting an assembly of functional archaella in Bacteria.

Fig. 2: Schematic representation of identified archaellum loci in Chloroflexota compared with archaeal archaellum loci.
figure 2

a, Schematic of the archaellum machinery with its subunits. b,c, Exemplary identified archaellum loci in Chloroflexota (b) and Archaea (c). Chloroflexota of different orders harbour archaellum machinery genes. The machinery is complete and has all the necessary genes for archaellation and motility. The order of the genes and their homology are more similar to those of the arl2 cluster found in S. acidocaldarius, ‘Candidatus Nitrososphaera gargensis’ and the euryarchaeon Archaeoglobus fulgidus. Homologous genes are indicated in the same colour according to the schematic in a. kb, kilobase.

Source data

Structure-guided bioinformatic analysis of the core machinery suggests an archaellum-like rotary mechanism

ATP hydrolysis by ArlI, through interaction with ArlJ, powers archaellum filament assembly and rotation7. To test whether the bacterial archaellum operates similarly, we compared predicted structures of ArlI and ArlJ from L. aerophila. ArlI shares domain architecture and fold with the crenarchaeal ArlI from Sulfolobus acidocaldarius, especially at the conserved C-terminal ATPase domain (Fig. 3b and Supplementary Fig. 1). While the N-terminal domain is more variable, its organization and the presence of the archaellum-specific three-helix bundle resemble the archaeal counterpart and support its dual role in assembly and rotation7. Compared with Methanocaldococcus jannaschii ArlI, the bacterial version contains a longer N-terminus with 36 additional amino acids typical of euryarchaeal ArlIs (Fig. 3a). ArlJ, predicted to have seven to nine transmembrane helices, shows high structural similarity to crenarchaeal ArlJ and retains conserved cytosolic positive charges thought to mediate interaction with ArlI7,9. Together, the structural features of bacterial ArlJI suggest a rotary mechanism similar to that of the archaeal archaellum8,9.

Fig. 3: Structure-guided analysis of the core archaellum machinery.
figure 3

a, Alphafold3-predicted structures of the motor protein ArlI and ArlJ of the bacterial organism L. aerophila, the euryarchaeon M. jannaschii and S. acidocaldarius (PDB: 4ii7 for ArlI). A comparison of all three structures reveals remarkable structural similarity and similar domains in the N-terminal domain (NTD) and C-terminal domain (CTD). b, Conservation of residues in ArlI and ArlJ of bacterial archaella machineries was calculated using ConSurf. The scale bar indicates calculated conservation scores per residue. Predicted local distance difference test (pLDDT) plots of predicted structures are found in Supplementary Fig. 2.

L. aerophila, an archaellated bacterium

Our bioinformatic analyses and structural predictions suggest that Chloroflexota bacteria may assemble a functional archaellum. To test this, we cultivated the thermophilic filamentous bacterium L. aerophila (Caldilineales)24. Although previously described as nonmotile, its genome encodes a complete archaellum cluster and a cheY homologue (Fig. 2a), along with pilO and pilN homologues nearby (Fig. 2 and Supplementary Fig. 3). Cells grown in marine broth for 5 days formed long multicellular filaments with distinct cell segments. Shorter filaments, similar to those of Chloroflexus islandicus, were also observed25 (Fig. 4). NileRed and SYTO13 staining of membranes and DNA, respectively, confirmed that both filament types consisted of multiple cells. Electron microscopy showed segmented filaments but no polar archaella (Fig. 4, upper panel).

Fig. 4: Light, fluorescence and transmission electron microscopy of the archaellated bacterium L. aerophila.
figure 4

a, Light and fluorescence microscopy of L. aerophila after labelling with NileRed and SYTO13 from liquid cultures and motility plates. Cell filaments shorten after growth on semi-solid agar plates. b, Transmission electron micrographs of cells. Cells isolated from semi-solid agar plates show distinct polar-located cell surface filaments reminiscent of archaellum filaments indicated by red arrows. Archaella were observed in 0 out of 86 cells from liquid culture (n = 86), 19 out of 108 cells from cells isolated from midpoint (n = 108) and 53 out of 95 cells from the rim of motility halo (n = 95). Scale bars are as indicated.

As archaellation is often not constitutive in Archaea22, we grew L. aerophila on semi-solid agar. After 5 days, a thin halo appeared. Cells from the rim and centre were imaged via light and fluorescence microscopy. These cells were mostly shorter (4–10 µm) but remained multicellular (Fig. 4). Electron microscopy revealed 11–12-nm polar surface appendages on cells from both halo regions (Fig. 4b). qRT-PCR showed increased expression of arlB and other archaellum genes in cells from motility plates compared with those in liquid culture (Supplementary Fig. 5).

Time-lapse microscopy revealed that L. aerophila cells showed active swimming with an average speed of 10.46 ± 6.68 µm s−1 (Supplementary Videos 17). Predominantly shorter cells moved directionally with occasional switching, suggesting regulatory switching mechanism. Rotation of the cell pole around the longitudinal axis indicated torque-driven motility, similar to some motile Archaea26,27,28. Slow body rotation, as seen in Halobacterium salinarum29, was also observed (Supplementary Videos 3 and 7). This behaviour contrasts sharply with the gliding motility seen in Chloroflexus aggregans, in which cells move along surfaces with jerky body movements30. In summary, L. aerophila expresses an archaellum and shows directed swimming behaviour characteristic of model motile Archaea. This identifies a motile member of the Chloroflexota phylum and provides evidence of a bacterium using an archaellum for motility.

CryoEM single-particle analysis of the bacterial archaellum filament

To verify that the identified surface structure is encoded by the archaellum cluster, filaments were purified. Mass spectrometry confirmed ArlB as the main filament protein. CryoEM single-particle analysis revealed numerous filaments, which were picked using CryoSPARC’s filament tracer (Supplementary Fig. 6). Initial helical refinement used parameters from the Methanocaldococcus villosus archaellum (twist, 108°; rise, 5.57 Å), refined to a twist of 108.14° and a rise of 5.64 Å (ref. 31). This yielded a 3.67-Å map, further improved to 2.7 Å after contrast transfer function (CTF) refinement and motion correction and corroborated without applying helical parameters (Supplementary Fig. 7 and Supplementary Table 1). The final map showed a helical filament with an alpha-helical core and globular domains. ModelAngelo confirmed ArlB identity and the structure lacked the first 28 amino acids, starting at Ile29 (refs. 32,33). ArlB carries a class III signal peptide, and SignalP6 (ref. 34) predicted a cleavage site at I29 (Supplementary Fig. 8). Each ArlB subunit consists of an N-terminal alpha-helical tail and a C-terminal globular domain (Fig. 5 and Supplementary Fig. 9). Multiple sequence alignment showed conservation of the cleavage site and hydrophobic signal peptide region (Supplementary Figs. 8 and 10). As in Archaea, the N-termini contribute to the filament core via hydrophobic interactions (Fig. 5).

Fig. 5: CryoEM structure of the L. aerophila archaellum.
figure 5

a, CryoEM-derived map at 2.7 Å resolution in surface view with fitted-chain representation. Representation of the archaellum structure in side, top and end-on views showing the core of alpha helices and globular domain facing outwards. Three left-handed strands are coloured accordingly. Scale bars, 100 Å. b, CryoEM-derived map at 3.4 Å of asymmetric filament reconstruction. The EM map shows a slight tilt and is coloured by fitted chains. Representation of the tilted archaellum filament in side view, top view and end-on view. Scale bars, 100 Å. c, Additional dead-end protrusions corresponded to the residues Thr145 and Thr66, likely to be O-glycosylations. d, Each subunit harbours a metal-binding site coordinated by residues D121, D127, E131 and E134.

To examine supercoiling, particles were re-extracted with a larger box size and asymmetrically reconstructed, revealing a bent filament model (Fig. 5 and Supplementary Figs. 11 and 12). Using ModelAngelo, 52 chains were identified in the map arranged in a left-handed three-start helix. Cross-sectional views of straight and bent filaments revealed ten protofilaments. While straight filaments showed low conformational variability, supercoiled filaments had greater shifts in the N-terminal α-helices. Alignment showed a shift of 1.02 Å in straight filaments and 2.8 Å in supercoiled ones (Supplementary Fig. 12). Comparison of subunits from the inner seam, outer seam and straight filament revealed root mean square deviation (RMSD) values of 3.3 Å and 2.4 Å along the N-terminal helix (Supplementary Fig. 13). In summary, these findings show that the bacterial archaellum is capable of supercoiling and exhibits slight tilting accommodated through conformational variability, similar to what has been reported for Saccharolobus islandicus35.

The bacterial archaellum is remarkably similar to archaeal archaellum filaments

The architecture of the bacterial archaellum filament was compared with solved structures of archaeal archaella from P. furiosus, Methanospirillum hungatii, Methanococcus voltae and S. islandicus REY15A31,35,36,37. Despite differences in helical parameters, the bacterial archaellum shows strong structural similarity (Supplementary Fig. 9). All filaments form a left-handed three-start helix with each subunit contributing to filament architecture through tight hydrophobic interactions of the N-terminal α-helix (Fig. 5 and Supplementary Fig. 9). The C-terminal globular domain shows a β-sandwich fold characteristic of archaellins (Fig. 5 and Supplementary Fig. 9). Structural alignment revealed high similarity to euryarchaeal archaellins lacking the extended C-terminal domain found in S. islandicus ArlB, with RMSD values under 2 Å (Supplementary Fig. 9a,c). Conservation was especially strong in the N-terminal α-helix, while the globular domain showed more sequence variability, particularly on its outward-facing side, a feature commonly seen in archaeal archaellins (Supplementary Fig. 9a,b). Two unassigned densities near residues T66 and T145 may represent O-glycosylation (Fig. 5c,d). While all archaella filaments are highly N-glycosylated, O-glycosylation was only known from M. hungatii and S. acidocaldarius archaella37,38 and was commonly found in bacterial surface filaments such as Neisseria meningitidis pili and Campylobacter jejuni flagella39.

Additional unassigned density was observed in the C-terminal domain, coordinated by D121, D127, E134 and E131, suggesting a conserved metal-binding site. Sequence alignments confirmed conservation of D121 and E134 (Supplementary Fig. 15c), and structural comparison with M. jannaschii archaellin showed a similar metal coordination. This feature is also shared with the type II secretion system (T2SS) pseudopilus (Fig. 5 and Supplementary Fig. 15b)36,40,41. ConSurf analysis showed conservation of the coordinating residues across bacterial ArlBs, indicating a conserved metal-binding motif similar to that in euryarchaeal archaella and T2SS pseudopilus (Supplementary Fig. 15).

Horizontal gene transfer of the archaellum machinery to Chloroflexota

To unravel the origin of the archaellum cluster in Chloroflexota, a concatenation of the core machinery proteins (ArlIJ) present in one copy in the cluster was used to infer a phylogenetic tree of archaeal T4P loci in Archaea and Chloroflexota. By mapping the different subtypes of archaeal T4P according to a previous study3 on the resulting phylogeny, we identified two well separated clades (ultra fast bootstrap support (UFB) = 100%; source data). While the first clade predominantly contains archaeal pili comprising subtypes such as UV pilus (subclade 4I), adhesion pilus (subclade4b), bindosome (subclade 4I) and Epd pilus (clade 1), the second clade comprises solely archaellum clusters (source data). Chloroflexota members are well nested within both archaeal clades, indicating two separate horizontal gene transfer events. The first clade contains two members of Chloroflexota (Anaerolineales order). However, as they lack a pilin subunit, these systems are probably incomplete and likely non-functional. The second clade contains archaellum loci belonging to Dehalococcoidia and Anaerolineae classes (Fig. 6 and source data). The completeness of the archaellum system within these genomes is striking and indicates a complete transfer of the archaellum cluster to Bacteria. This transfer probably occurred from euryarchaeal members belonging to Methanotecta (UFB = 100%) that encode the arl2-like archaellum cluster to Chloroflexota, as these two groups form strongly supported sister lineages in the tree (Fig. 6 and source data). Gene transfers between the two domains of life are frequent but highly asymmetric with more frequent gene transfers from Bacteria to Archaea than vice versa42. Indeed, previous analysis of TFF systems in Bacteria and Archaea revealed a horizontal gene transfer of an ancient archaeal pilus to Bacteria that led to the bacterial Tad system2. The presence of the archaellum machinery in Chloroflexota depicts the second horizontal gene transfer event of a functional cell surface machinery to Bacteria.

Fig. 6: Horizontal gene transfer of the archaellum machinery to Bacteria.
figure 6

Phylogenetic tree of concatenated core archaellum machinery subunits. The maximum likelihood tree was inferred from a supermatrix of 431 sequences and 987 amino acid positions with the model LG + R10 + C60. The archaellum clade of Chloroflexota is well nested in Archaea with Methanomada as the closest relative. Archaellum gene clusters of representative archaellated species are shown coloured as in Fig. 2 indicating the presence of an arl2 cluster to Chloroflexota. Scale bar as indicated.

Discussion

Diversification of the TFF superfamily in prokaryotes is driven by adaptation of existing macromolecular machineries for specialized functions. In Archaea, an ancient TFF system evolved into the rotating archaellum2. While Archaea typically use archaella for motility and Bacteria use flagella1, archaeal archaellum genes were identified in Chloroflexota metagenomes19 and archaellum loci were reported in cultivated SAR202 members18. The archaellum clusters were restricted to some members of the Chloroflexota and notably in these flagellar components were absent (Fig. 1a,b and source data). The only known flagellated Chloroflexota, Tepidiforma thermophila, was identified recently43. Phylogenetic analyses suggest that flagella were ancestral in this phylum but were repeatedly lost, probably owing to genome reduction43. It is plausible that the simpler archaellum was later acquired through horizontal gene transfer after flagellar loss. This parallels the emergence of the bacterial Tad pilus from an archaeal ancestor, involving structural additions as a secretin to accommodate diderm envelopes2.

Our phylogenomic analysis of the archaeal T4P cluster in Archaea and Bacteria revealed two separate horizontal gene transfer events that led to the co-option of the archaellum machinery in Chloroflexota. The second involves a complete archaellum locus found in the ancestor of Chloroflexota classes Dehalococcoidia and Anaerolineae (Fig. 6 and source data). Although extremophiles frequently exchange genes via horizontal gene transfer, archaellum-encoding Chloroflexota are found in diverse environments (marine, anaerobic, thermophilic), suggesting that environmental pressure influenced but did not drive archaellum acquisition. Additional archaeal genes in these bacteria for example, Mbh hydrogenase, malate dehydrogenase and archaeal-type ATP synthase19 support broader archaeal gene acquisition. The ATP synthase may be an adaptation to meet the high energy demands of the archaellum machinery.

The presence of archaellum-related genes in Chloroflexota members is indicative of a conserved mechanism of assembly and function in the cell envelope (Figs. 2 and 3), and indeed, swimming motility is observed (Supplementary Videos 17). In Sulfolobus islandicus, S-layer interaction is essential for archaellum function: deletion of slaAB leads to immotility despite intact archaella12. Yet, some Archaea lacking an S-layer (for example, Oxyplasma meridianum) or with complex envelopes (M. hungatii) remain motile37,44, suggesting that the archaellum machinery can easily adapt to various cell envelopes. Similarly, Chloroflexota show a range of envelope architectures, including S-layer-containing monoderms45. This simplicity may have facilitated archaellum acquisition and restricted its spread to this phylum.

The cultivated and archaellated bacterium L. aerophila possesses an outer layer resembling the archaeal S-layer and also encodes peptidoglycan synthesis genes (Fig. 3 and Supplementary Fig. 16). Genes upstream of the archaellum locus encode proteins with partial PilO homology and C-terminal LysM domains that bind peptidoglycan, as well as a PilN homologue—proteins known to aid pilus anchoring in bacterial cell walls46 (Fig. 2 and Supplementary Fig. 3). qRT-PCR reveals low expression of these genes in L. aerophila, suggesting accessory roles in anchoring the archaellum to the envelope (Supplementary Fig. 5). Thus, co-option of such proteins may have enabled archaellum integration into peptidoglycan-containing bacteria.

Chloroflexota archaellum machinery and filament architecture closely resemble those of Archaea, suggesting a conserved rotation mechanism. Unlike in Archaea, the archaellum filaments in Bacteria are modified by O-glycosylation (Fig. 5). Archaea primarily use N-glycosylation, which is essential for proper motility, and defects in glycan attachment or the presence of truncated glycans can lead to severe motility impairment due to filament aggregation47. Bacteria, in contrast, use both N- and O-linked glycosylation systems, but commonly apply O-glycosylation to pilins and flagellins. It is likely that this existing bacterial glycosylation machinery was repurposed to modify the archaellum filament39.

Strikingly, L. aerophila seems to link archaellation to alteration of its cell shape. Chains of cells become shorter in later growth phases and are then archaellated. The multicellular filamentous cyanobacterium Nostoc punctiforme forms hormogonia, which are differentiated shorter filaments that show gliding motility through type IV pili and secretion of polysaccharides48. These observations are in concordance with what was observed in the well-studied archaeon H. volcanii, in which motility is restricted to early, rod-shaped stages and depends on polar localization of the motility and chemotaxis systems, governed by the MinD4 protein49. L. aerophila encodes CheY and two MinD-like proteins with strong homology to their archaeal and bacterial counterparts (Supplementary Fig. 3 and source data). While Archaea rely on the CheF adaptor to link CheY signalling to the archaellum motor50, this component is absent in Chloroflexota (source data). Analogously to archaea, Chloroflexota might have evolved another protein component to enable signal transfer from the chemotaxis system to the bacterial archaellum motor complex. Further studies are needed to clarify whether a functional chemotaxis system and MinD have a related role in L. aerophila and its link to cell shape alteration and positioning of the archaellum machinery. Ultimately, our study highlights a striking case of horizontal transfer of the archaellum machinery to the Chloroflexota. There, it assembles into a functional filament as observed in Archaea, underscoring the exaptation of what was previously considered to be an archaea-specific machinery, repurposed for identical functional roles.

Methods

If not stated otherwise, all chemicals were purchased from either Roth or Sigma.

Strain and growth conditions

L. aerophila ATCC BAA-2444, DSM25763 (DSMZ), was grown in Difco marine broth 2216 medium at 55 °C, at 90 rpm shaking with ambient light in 5-ml plastic tubes or on marine broth agar plates solidified with 1.5% Bacto Agar (BD).

RNA isolation and quantitative real-time PCR

For RNA isolation, cells from motility plates or 5 ml of liquid cultures grown for 5 days were collected and briefly washed with phosphate-buffered saline (PBS). RNA was isolated using TRIzol reagent followed by phenol–chloroform extraction. Residual DNA was removed by DNase treatment, and the removal of DNA was checked with PCR. cDNA synthesis was performed using the Thermo Fisher cDNA synthesis kit (Thermo Fisher). Relative qPCR was performed using qPCRBIO SyGreen Mix (PCRBioSystems) using cDNA as a template. DNase-treated RNA was used as a non-template control. Fold changes were calculated using the Livak method with rpoB as a normalizer51.

Motility plates

Into 400 ml of marine broth medium was dissolved 0.5% Bacto agar (BD). Subsequently, 5 µl of a 1-day-old culture of L. aerophila was spotted on the plates. The plates were incubated at 55 °C for 5 days in a sealed plastic box.

Cell surface filament isolation

The cells of the motility plates were collected and resuspended in 1× PBS with 2% NaCl. Isolation was done as previously described52. Cell surface filaments were sheared using a blender (Russell Hobbs) or a peristaltic pump (Gilson Minipuls). Cell debris was pelleted by centrifugation at 12,000 × g for 25 min. The cells were pelleted from the supernatant by ultracentrifugation at 200,000 × g for 1 h 10 min. The resulting pellet was resuspended in 500 µl 1 × PBS with 2% NaCl. This was applied to 4.5 ml of CsCl2 (0.5 g ml−1) dissolved in PBS with 2% NaCl for density gradient centrifugation at 250,000 × g for 16 h 30 min. A white band in the upper third was recovered and diluted in 8 ml buffer (1× PBS in 2% NaCl). This fraction was centrifuged at 250,000 × g for 1 h. The resulting pellet contained purified cell surface filaments and was resuspended in 100 µl and stored at −20 °C.

Negative-stain electron microscopy

Around 5 µl of cells or purified archaella filaments was applied on a freshly glow-discharged 300-mesh carbon–formvar-coated copper grid (Plano). This was incubated for 30 s and excess liquid was blotted away. The grid was washed three times with ddH2O and stained with droplets of 2% uranyl acetate. Imaging was done with a Hitachi HT7800 operated at 100 kV, equipped with an EMSIS Xarosa 20-megapixel CMOS camera.

Cryo-electron microscopy

Around 3.5 µl of isolated archaella filaments was vitrified on freshly glow-discharged Quantifoil R2/2 grids using a Mark IV Vitrobot (Thermo Fisher). The dataset was collected using a Titan Krios equipped with a Falcon4i and a Selectris energy filter (Gatan). The detector was operated in counting mode at a calibrated pixel size of 0.94 Å, corresponding to a magnification of ×130,000. With the use of EPU 3.6 (Thermo Scientific), 40-fraction videos were recorded with an exposure time of 1.95 s and a total electron dose of 40 e Å2 at a defocus range of −0.5 µm to −2 µm. A total of 6,110 videos were collected and processed in CryoSPARC v4.6 (Supplementary Table 1)32. Briefly, videos were motion corrected, and CTF was estimated using patch.motion and patch.ctf within CryoSPARC. Helical segments were picked with the filament tracer job using 150 Å in diameter and a separation distance of 0.4 diameter between segments. The picked particles were extracted with a Fourier-cropped box size of 200 px, and two rounds of two-dimensional (2D) class averaging were done to remove junk particles. Particles from filament-indicating classes were extracted at an initial box size of 256 px and subjected to 2D class averaging. Selected classes were used to determine the helical parameters using CryoSPARC’s helix refine job. Initial helical parameters of M. villosus archaellum were applied31 and refined to a twist of 108.09° and a rise of 5.63 Å. The obtained map was improved with global CTF and local CTF refinement as well as reference-based motion correction. After a final round of helical refinement, the parameters were determined to be a twist of 108.14° and a rise of 5.64 Å. Local resolution and Fourier shell correlation (FSC) estimation were performed, and resolution was determined to be 2.71 Å.

Helical parameters were further corroborated by running helix refine without implying any helical parameters. Using a symmetry search job within CryoSPARC with a range of 107–109° twist and 4–6 Å rise, 4 possible helical parameters were determined. These parameters were used for a subsequent helix refine job. The obtained map was improved with global CTF and local CTF refinement, as well as reference-based motion correction with final helical parameters closely matching the initially obtained ones (Supplementary Fig. 7).

Asymmetric reconstruction

Supercoiling properties of the purified archaellum filament were determined through an asymmetric reconstruction protocol based on a previous study35 (Supplementary Table 1). Using CryoSPARC, particles were picked using the filament tracer mode. The particles were extracted at 512 px box size and Fourier cropped to 64 px. After two rounds of 2D classification removing junk particles, the filament particles were re-extracted at the initial box size and subjected to homogenous refinement with a low-resolution helical volume (20–50 Å) as an initial volume. The particles and map were subjected to three-dimensional variability analysis within cryoSPARC and reconstructed into three-dimensional (3D) variability clusters. The best volumes were chosen and further processed with a local refinement with shift constraints from 5 Å to 20 Å and rotational searches from 5° to 20° yielding a map of a slightly curved filament. This was subjected to a global and local CTF refinement followed by a second local refinement and sharpening yielding a map with 3.4 Å resolution (Supplementary Fig. 11).

Model building and validation

Model building was done using the ModelAngelo build command, and the protein sequence of ArlB1 (WP_141610922.1) from L. aerophila was used33. ModelAngelo fitted the sequence in all well-resolved chains of the archaellum filament map. The model was manually curated and adjusted in Coot53. The model was further iteratively refined using phenix.realspace.refine and manually correcting for outliers. Phenix validation with phenix.validation, including phenix.molprobity and phenix.mtriage, was run to determine any rotamers and Ramachandran outliers54,55. Molecular models and graphs were generated with ChimeraX56. Final validation parameters can be found in Supplementary Table 1.

Light and fluorescence microscopy

Cells isolated from the rim and middle of the motility halos from the motility plate or grown in liquid medium were collected and resuspended in 500 µl PBS. Cells were washed with PBS and stained with 5 µl NileRed (5 mg ml−1 in DMSO; Thermo Fisher) and 1 µl SYTO13 (Thermo Fisher). The samples were observed on an agarose pad (1% in PBS) using an inverted Zeiss Axio Observer Z1 phase contrast microscope equipped with a Plan Apochromat 100 × 1.4 Oil Ph3 M25 objective controlled via Zeiss Blue v.3.3.89. Image analysis was performed using ImageJ.

Swimming videos and automated cell tracking

For swimming videos, cells from the motility plate were inoculated in filtered marine broth medium and incubated for 90 min at 55 °C while shaking. Then, 1 ml of cells was diluted with 1 ml prewarmed filtered marine broth medium, and 1.5 ml was transferred into a round 0.17-mm Bioptechs Delta TPG microscopy dish. Imaging was done using an inverted Zeiss Axio Observer Z1 phase contrast microscope equipped with a Plan Apochromat 100 × 1.4 Oil Ph3 M25 objective controlled via Zeiss Blue v.3.3.89 preheated at 55 °C. Videos of swimming cells were recorded for 10 min in camera streaming mode. Videos were opened in Fiji57 with a time interval of 0.051 s. The first five frames of a representative video were used to train the Weka detector58 in Fiji57 to detect cells and background signals. This trained Weka model was used to automatically track swimming cells in the full video using TrackMate7 (ref. 59). Low-quality tracks and tracks of multiple cells or background were removed by adjusting the detection threshold in TrackMate7. Tracks were refined through visual inspection and manual bridging of gaps within tracks. Swimming videos were exported in MP4 or AVI format with 30 fps.

Databases

Three different databases were used in this study:

Bacteria

This database was previously described and used in a previous study60. Briefly, Bacteria assemblies were retrieved from the National Center for Biotechnology Information (NCBI) as of April 2020, and three species were collected within each order, following NCBI taxonomy. Genomes from reference species and the most complete assemblies were preferably selected. This resulted in a database of approximately 1,048 genomes covering the available bacterial diversity.

Archaea

This database was assembled based on a previous study61. Briefly, 8,415 archaeal genomes were retrieved from the NCBI in 2022, to which 3,038 metagenome-assembled genomes from a genomic catalogue of earth microbiome were added62. All downloaded assemblies were functionally and taxonomically annotated using rapid prokaryotic genome annotation (PROKKA)63 and Genome Database Taxonomy (GTDB-tk)64, respectively. A preliminary tree was inferred using FastTree265 and a supermatrix based on the concatenation of four marker proteins (RpoB, IF2, Ul10 and Us9) retrieved from all assemblies. TreeCluster66 was next used with a threshold of 0.05 to cluster similar genomes within the preliminary phylogeny. One representative genome was selected from each cluster based on assembly quality, evaluated using CheckM resulting in 3,702 archaeal genomes67.

Chloroflexota

A new Chloroflexota database was assembled specifically for this study. We retrieved 9,868 genomes annotated as Chloroflexota and available in NCBI as of June 2024. All assemblies were functionally and taxonomically annotated using PROKKA63 and GTDB-Tk64, respectively, and dereplicated using dRep68 with a threshold of 96% to retain representatives of each species. This resulted in 3,780 clusters of Chloroflexota genomes, from which one representative based on assembly quality, as evaluated by CheckM67.

Chloroflexota reference phylogeny

To infer a reference species phylogeny of Chloroflexota, we first searched for three conserved and universal markers (IF-2, RpoB and RpoC) in the Chloroflexota database. We used hmmsearch from the HMMER-3.1b2 package69, and from the Pfam database70, the domains PF04997 and PF04998 for RpoB; PF04563 and PF04562 for RpoC; and PF11987 for IF-2 and extracted the markers from all Chloroflexota assemblies. We discarded the assemblies with only one marker out of the three and concatenated the markers for the remaining assemblies into a supermatrix of 3,543 sequences and 2,131 amino acid positions. We inferred a preliminary phylogeny using FastTree2 (ref. 65), which was used to sample taxon according to patristic distances using TreeCluster66. After two rounds of TreeCluster using length clade= 0.1 and max clade= 0.3 in the first and second round respectively, we selected 443 assemblies, representative of all the taxonomic and phylogenetic diversity of Chloroflexota, and a supermatrix with 443 sequences and 2,456 amino acid positions was generated. Finally, a maximum likelihood phylogeny was inferred using IQ-TREE2 (v2.3.4)71 and the evolutionary model LG+C60+I+R10 selected by ModelFinder72, and ultrafast bootstrap values for branch supports73.

Bacterial protein secretion systems and archaeal TFF detection

To search for TFF and flagellar macromolecular systems, we downloaded the 20 models implemented in the package TXSScan (v1.1.3) dedicated to the genomic detection of bacterial secretion systems and related appendages, including the archaeal pili and archaellum2. Next, we used MacSyFinder2 (ref. 74) and scanned the 3,780 genomes of Chloroflexota, as well as 3,702 genomes of Archaea and 1,048 genomes of Bacteria from our locally maintained databases60,61 for the presence of each system. Finally, all the hits identified as part of best solutions by MacSyFinder2 were retrieved and analysed. MacSyFinder2 was run with the generic archaeal T4P model, and candidate loci were manually curated to distinguish archaella from other archaeal T4P systems based on gene content, synteny and supporting arCOG annotations. Archaellum loci were identified based on the co-localization of core archaellum genes (arlC–J) and the filament gene arlB within a single gene locus. Only loci with a near-complete set of archaellum-specific genes and no signature genes of other T4P types were considered true archaella. In parallel, we carried out additional homology searches for FliCD, MotAB and FlgEGKL proteins as they are not included in the flagellum model of TXSScan. For this, we searched the domains PF12445 (FliC), PF07195 (FliD), PF20560 (MotA), PF13677 (MotB), PF06429 (FlgEG), TIGR02492 (FlgK) and TIGR02550 (FlgL) against the Chloroflexota database using hmmsearch from the HMMER-3.1b2 suite69 and the option cut_nc.

Phylogenetic analyses

Sequences corresponding to the most conserved components of the archaeal T4P were extracted from Chloroflexota and Archaea databases, aligned using MAFFT75 with the linsi option and trimmed using trimAl76 with the option gappyout, and single gene maximum likelihood phylogenies were generated using IQ-TREE2 (v 2.32.4) with the best-fit evolutionary model selected by ModelFinder72 using the BIC criteria72 and ultrafast bootstrap values for branch supports73. To infer the origin of the archaellum in Chloroflexota, we extracted the sequences corresponding to the core machinery ArlIJ from the archaellum loci comprising only one copy of each gene. Single gene alignments were generated using MAFFT (with the linsi option), trimmed using trimAl (with the option -gt 0.5) and concatenated into a supermatrix of 431 sequences and 987 amino acid positions. Finally, a maximum likelihood phylogeny was generated using IQ-TREE2 and the evolutionary model LG+R10+C60 and ultrafast bootstrap values for branch supports. All the trees were annotated using IToL77.

Structure prediction and analysis

Structure prediction was done with Alphafold3server78 (https://alphafoldserver.com/) and visualized through ChimeraX56. Structural conservation was analysed through ConSurf WebServer79. Gene loci were depicted using GeneGraphics80. Sequence logos of bacterial ArlBs were created using WeblogoServer81. The peptidoglycan synthesis pathway was mapped used KEGG-mapper82.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.