Donor strand complementation and calcium ion coordination drive the chaperone-free polymerization of archaeal cannulae

Abstract

Cannulae are structurally rigid tubular protein filaments that accumulate on the extracellular surface of archaea within the family Pyrodictiaceae during cell growth. These obligate anaerobes propagate under hyperthermophilic conditions in which cannulae form a biomatrix that interconnects and sustains cells. The persistence of cannulae in this environment suggests that these filaments display significant thermostability, which has attracted technological interest in their development as synthetic protein-based biomaterials. Here, we report cryoEM structural analyses of ex vivo and in vitro assembled recombinant cannulae. We demonstrate that the interactions between protomers in native and recombinant cannulae is based on donor strand complementation (DSC), a form of non-covalent polymerization previously observed for bacterial chaperone-usher pili. Unexpectedly, calcium ion coordination at the subunit interfaces reinforces the network of donor strand interactions in the cannulae. This study provides insight into the mechanism of assembly of cannulae and the structural origin of their high stability and rigidity.

Introduction

Filamentous protein assemblies are common in biological systems in which they are involved in a wide range of functions that are critical for the survival and propagation of the organism¹. Many of these functions, e.g., locomotion², adhesion^3,4, tunable mechanical response^3,4,5,6,7,8, multicellular organization^9,10,11,12, electrical conductivity^13,14,15,16, directional transport of substrates^{17,18,19,20,21}, regulation of enzymatic catalysis²², etc., would be desirable to emulate in synthetic protein-based nanomaterials that could be tailored for specific applications²³. However, this approach requires prior knowledge of the sequence-structure-function relationships that are responsible for the observed behavior of the corresponding biologically derived assemblies, as well as convenient methods for the ex vivo isolation or in vitro fabrication of the protein filaments. Structural analyses at near-atomic resolution have illuminated the physical principles that underlie the function of many biologically derived protein-based nanomaterials^24,25. This structural information provides a critical starting point for engineering synthetic analogues that mimic the function of the corresponding native protein assemblies²⁶. High-resolution structural information can also enable an understanding of the molecular-level interactions that underlie the mechanism of self-assembly, thereby facilitating the development of reliable methods for recombinant production of precursor proteins that can be fabricated into functional filaments that display the desired properties^27,28.

Here, we employ electron cryomicroscopy (cryoEM) to investigate the structure of cannulae, a structurally unique class of protein-based filaments expressed on the extracellular surface of the hyperthermophilic archaeon Pyrodictium abyssi and related organisms in the family Pyrodictiaceae^29,30,31,32. Cannulae form during the life cycle of P. abyssi, occultum, brockii strains under culture conditions that mimic the native growth environment^31,32,33. P. abyssi is an obligate anaerobe that propagates under hyperthermophilic, deep-sea conditions with optimal growth in the temperature range from 80 °C to 110 °C³⁴. The persistence of cannulae under these conditions suggested that the filaments display significant thermostability, which has attracted biotechnological interest for potential use as robust assemblies for the development of synthetic protein-based biomaterials³⁵. However, P. abyssi strains are challenging to culture at preparative scale and not genetically tractable at present³⁶.

The cannulae of P. abyssi have been the most thoroughly studied experimental system. Cell cultures of P. abyssi display robust growth and achieved high cell densities, albeit under restrictive culture conditions, e.g., high temperature and pressure and corrosive media³². Dark field microscopy of growing cultures of P. abyssi provided evidence that the flat cells divide through binary fission^36,37. During cell division, cannulae are synthesized on the cell surface such that the daughter cells remain connected after separation. These extracellular connections were maintained in subsequent rounds of cell division, which resulted in the formation of a dense network of cells that were linked through cannulae. In liquid culture, P. abyssi cells were never observed without cannulae, which have been postulated to serve an essential role in cellular propagation. Electron cryotomography (cryoET) studies of P. abyssi cultures³⁸ indicated that the cannulae passed through the S-layer and embedded in the pseudo-periplasmic space. While the biological function of cannulae remains a matter of speculation, cryoET analysis provided evidence for density within the lumen of cannula connecting P. abyssi cells, which led to a proposed role for cannulae as a primitive extracellular matrix through which cells could communicate or exchange material.

To gain insight into the structural principles underlying these unique physical properties, cryoEM three-dimensional reconstructions were performed using ex vivo and in vitro recombinant cannulae from P. abyssi as substrates^29,30,31,32. These structural analyses revealed that the interactions between protomers in ex vivo and in vitro filaments were based on donor strand complementation (DSC), a form of non-covalent polymerization in which a β-strand from one subunit is inserted into the acceptor groove of a β-sheet in a neighboring subunit. DSC has been observed in structures of bacterial chaperone-usher (CU) pili in which extracellular polymerization occurs through a distinct process^{3,4,5,6,7,9,10,11,39,40}. In cannulae, the donor strand-acceptor groove interaction is reinforced through calcium ion coordination between structurally adjacent protomers. Crystallographic analysis of a recombinant cannula-like protein suggested that calcium ion binding primed polymerization of the monomer into a filament. Bioinformatic analysis provided evidence that structurally similar cannula-like proteins were encoded within the genomes of other hyperthermophilic archaea and were encompassed within the TasA superfamily of biomatrix proteins. CryoEM structural analysis of tubular filaments derived from in vitro assembly of a recombinant cannula-like protein from an uncultured archaeon in phylum Thermoproteota, tentatively identified as a Hyperthermus species, revealed a common mode of assembly to the Pyrodictium cannulae, in which DSC and calcium ion binding stabilized axial and radial assembly. The facile in vitro assembly of cannula-mimetic tubules represents an attractive methodology for the fabrication of thermodynamically stable and biochemically addressable filamentous protein nanomaterials. In contrast to CU pili, in which the protein subunits undergo catalytic polymerization requiring a membrane-bound translocon, recombinant cannula proteins self-assemble spontaneously in the presence of calcium ions in aqueous solution under controlled conditions in vitro and displayed a similar structure to the ex vivo P. abyssi filament.

Results

CryoEM analysis of ex vivo P. abyssi extracellular filaments

We initiated a cryoEM study of ex vivo cannula filaments found on the extracellular surface of P. abyssi AV2, the type strain for which the genomic sequence was available (NZ_AP028907.1)⁴¹. From the raw micrographs, two main fiber populations could be discerned under the culture conditions (see “Methods”), namely cannulae and an unidentified class of pili (vide infra). Cannulae were present as long (> 5 µm) filaments with a diameter of ~26 nm that tended to associate laterally into large rafts (Supplementary Fig. 1a). High magnification cryoEM imaging (Fig. 1d and Supplementary Fig. 1b) revealed the presence of thinner, flexible (~2 nm diameter) fibrils that were either associated with the cannulae or freely floated in suspension. Although these 2 nm fibrils were mostly seen to be randomly bound to cannulae filaments, examples were found in which fibrils were helically wound inside the cannulae scaffold, seemingly following its helical pitch (Fig. 1d), which was suggestive of a specific interaction between both filament types.

Fig. 1: CryoEM structure of ex vivo cannulae from the hyperthermophilic archaeon Pyrodictium abyssi AV2.

a cryoEM volume revealing a two-start (C2) helical filament with a pitch p and diameter of 94.6 Å and 260 Å, and a rise and twist of 3.18 Å and 12.1°, respectively; b local interaction network between neighboring cannula protofilaments in which subunits interlock through donor strand complementation along the axial (e.g., i → j → k) direction and partake in further intermolecular contacts mediated by calcium coordination along the axial and radial (e.g., j-1→j → j + 1) direction; c stick representation of three types of calcium pockets: (i) axial stacking of subunits via type I and II calcium binding centers, (ii) radial contacts are mediated via type III Ca²⁺ binding sites; d raw cryoEM image of a cannulae filaments loaded with a helically winding thinner filament (2 nm; cargo). White arrows denote winding cargo and dangling cargo associated with the cannulae filaments; e surface electrostatics showing a predominantly negatively charged surface and a positively charged lumen for the cannula model. The representative image in this figure (d) is derived from a total of 2365 cryo-EM movies from two technical replicates.

After high-resolution data collection, standard helical refinement procedures led to a cryoEM reconstruction of the ex vivo cannulae at 2.3 Å global resolution (FSC 0.143 criterion; Fig. 1a, and Supplementary Fig. 1f, g), which revealed two strands having C2 symmetry, with a helical rise and twist along each strand of 3.18 Å and 12.1°, respectively, and a pitch of 94.6 Å. Given that the molecular identity of the ex vivo cannulae subunits was unknown, Modelangelo⁴² was used to build a de novo atomic model without providing an input protein sequence, followed by rounds of alternating manual and automated real-space refinement (see “Methods”). The final model revealed an intricate network of protein subunits that participated in polytypic radial and axial interactions (Fig. 1b and Supplementary Fig. 2). Specifically, the cannula protein subunits interlocked axially via N-terminal DSC (e.g., i → j → k; Fig. 1b), and axially as well as radially via intermolecular calcium coordination (Fig. 1c and Supplementary Fig. 1h). Three types of calcium binding sites were identified that could be distinguished based on interfacial interactions between subunits. Type I and II binding sites mediated axial interactions between protomers (e.g., protomers i and j), and type III binding sites were located at the interface between two structurally adjacent subunits in the radial plane (e.g., the protomer interface between j-1 and j). Each calcium-binding site type was unique and consisted of a combination of Ca²⁺ side-chain coordination via Asp or Glu residues, as well as main chain contacts with carbonyl groups and water molecules. We hypothesized that this elaborate network of calcium-binding sites contributed to (i) the remarkable stability of the cannulae (P. abyssi was cultured at 98 °C), but (ii) could also represent an environmental trigger for cannula protein assembly (vide infra).

To identify the major subunit of the ex vivo cannulae, we first performed a hidden Markov search using the previously investigated cannula protein CanA (UniParc UPI0024181436)^35,43 from P. abyssi TAG11^32,33 as query sequence against the P. abyssi AV2 genome (NZ _AP028907.1)⁴¹. This search led to 19 putative CanA-like candidate proteins, 15 of which aligned onto a common consensus sequence (Supplementary Fig. 3). Of those 15 sequences, 6 sequences were retained as plausible candidates based on sequence length arguments (i.e., primary mature sequence between 140 aa and 160 aa; Filter 1 in Supplementary Fig. 4). The final assessment of the major subunit sequence was based on sidechain identification through manual inspection of the cryoEM map (Filters 2 and 3 in Supplementary Fig. 4). WP_338251948.1 (annotated as “hypothetical protein”, InterProScan: no identified Pfam or PANTHER domains) was retained as the major subunit. Given the similarity (49.0% sequence identity) to previously identified CanA, and the fact that this sequence was derived from ex vivo analysis, we propose “CanX” as the naming convention. Interestingly, we found clear additional hexose-like density linked to residue Asn51 (Supplementary Fig. 4). N-glycosylation in Thermoproteota was found to comprise a eukaryote-like GlcNAcβ(1-4)GlcNAc core structure⁴⁴. The excess density at Asn51 and the adjacent hydrogen-bond network fit well with a β-linked N-acetylglucosamine (GlcNAc; albeit with some additional density at the 3′ position), including additional density for a partially resolved hexose bound to its 4′ position (Supplementary Fig. 4). We concluded that Asn51 is an N-linked glycosylation site, likely with GlcNAc core, which resulted in a band of glycans that helically wound around the cannula outer surface (Supplementary Fig. 1i).

Intrigued by the number of detected cannula-like proteins in the P. abyssi genome, we inspected the genomic loci and found that 11 out 15 sequences were grouped into two extensive gene clusters, with the remaining 4 scattered across the genome (i.e., orphan sequences; Supplementary Fig. 5). Interestingly, CanX (WP_338251948.1) was flanked by WP_338251946.1 (37.4% sequence identity) and WP_338251950.1 (29.5% sequence identity), which suggested that these three genes constituted an operon. Although we can only speculate on the function of the CanX homologues present in the genome at this point, we wondered if one or more of these CanX-like proteins could be sporadically integrated into cannulae (i.e., as minor subunits). The cryoEM volume shows no clear signs of subunit heterogeneity, but if one or more CanX-like proteins were randomly integrated into the cannula lattice at low occupancy, then their impact on and contribution to the final 3D reconstruction would be limited. After inspecting the AlphaFold2⁴⁵ structural predictions (Supplementary Fig. 5), numerous candidates were found with substantial domain insertions, and we reasoned that such domains could act as fiducials to guide a particle classification process. To test that hypothesis, we tuned the filament tracer algorithm of cryoSPARC to selectively pick the edges of the cannula filaments and extracted the particles using a smaller box size covering a length equivalent to 2.5 helical pitches to allow for greater sensitivity to local changes in downstream 2D classification jobs. Despite this approach, the resulting 2D class averages showed no signs of heterogeneity, nor could the presence of any additional domains be discerned as outward projections (Supplementary Fig. 1j). We recognize that this approach cannot provide definitive proof that the cannula filaments were solely composed of CanX, but it suggested that if any other CanX homologues were incorporated into the cannulae, it would most likely occur at low sub-stoichiometric levels.

Next, we set out to determine the identity of the 2 nm fibrils that were associated with the cannula filaments (Fig. 1d and Supplementary Fig. 1b). Given the clear interaction between the two types of filaments, we inferred that the 2 nm fibrils might be putative cannula-cargo. The reconstructed cryoEM volume showed no signs of the presence of the cargo, but that would be expected due to the helical reconstruction method that was employed, i.e., the cargo locally breaks helical symmetry, and, as a result, any cargo density would be averaged out in the reconstruction. To account for such a symmetry mismatch, we performed C1 reconstructions followed by 3D classification, but did not identify any 3D classes in which the cargo was resolved, which was most likely due to the absence of a specific cannula-cargo epitope (as suggested by the raw micrographs). In the absence of a cryoEM resolved cannula-cargo interaction, we hypothesized that the cargo could be nucleic acid and collected a second cryoEM dataset after extended DNase I treatment. The resulting raw micrographs showed no signs of the presence of any cargo after DNase I treatment. To formalize that observation, we adopted the following workflow. Starting from high-resolution helical reconstructions of both DNase I-treated and non-treated particles, we used the respective consensus volumes as input in particle subtraction jobs—the rationale being that particle alignment and classification would be dominated by the cannula signal for the non-subtracted particles. Next, the subtracted particles were 2D classified and inspected for the presence of fibrillar classes that would match the dimensions of the cargo in raw micrographs. For the non-treated cannulae dataset, such cargo 2D class averages were produced as evidenced from the apparent diameter (i.e., 2 nm) and the mapping of the particles-corresponding to the cargo 2D class averages-back onto the raw micrographs, which confirmed that these corresponded to boxed regions with cargo clearly present. In contrast, no cargo 2D class averages could be obtained following DNase I treatment, which allowed us to conclude that the cargo fibrils were DNA. Although the cryoEM images were suggestive of a luminal location, these did not unambiguously exclude the possibility that the cargo could be bound to the cannulae surface. However, indirect evidence further suggested a luminal interaction based on the surface electrostatics of the cannula filaments. Cannulae displayed a highly polar structure that has a predominantly negative outer surface and a highly positive lumen (Fig. 1e). The presence of a positively charged lumen for a negative cargo might seem contradictory from the perspective of DNA transport but could be rationalized through a mechanism of (i) counterion-mediated charge neutralization or (ii) the involvement of effector molecules bound to the DNA to facilitate transport through the cannulae fibers^17,18,19,46. Taken together, these data suggested a possible function for cannulae as molecular conduits between cells for the exchange of DNA, although we can’t exclude the possibility that other types of cargo exist as well.

In addition to the cannulae filaments, a second class of pili was also found to be present (Supplementary Fig. 1d). These pili existed as large super bundles with a local undulating character that was reminiscent of the structure of the archaeal bundling pili (ABP) of Pyrobaculum calidifontis (PDB: 7UEG)¹⁰. Although the 2D class averages of these putative P. abyssi bundling pili do not show homogenous packing of the filaments, the resulting power spectrum showed a pronounced maximum at 1/47 Å, which we presume corresponded to the axial rise of the subunits in the individual pili. In contrast, an axial rise of 33 Å was observed for individual filaments of P. calidifontis ABP, which were derived from DSC of the cellular protein AbpA (WP_011849593.1). Using HMMER⁴⁷ and Foldseek⁴⁸, we found no putative homologues of AbpA encoded in the genome of P. abyssi AV2. AbpA was distantly related to TasA (UniProt: P54507), the major biofilm matrix component of Gram-positive Bacillus subtilis, which was shown to have an axial rise of 48 Å (PDB: 8AUR)⁹. Using the TasA structure as a Foldseek⁴⁸ query, we identified the hypothetical protein WP_338249486.1 (AAA988_RS09245) in a structural similarity search (Supplementary Fig. 7). Although the sequence identity to TasA is low (16%), the AlphaFold3⁴⁹ prediction of a WP_338249486.1 dimer revealed notable structural similarities, i.e., WP_338249486.1 is predicted to form a donor strand complemented dimer with a monomer-to-monomer distance of 47 Å, which agreed with the experimentally determined axial rise of the putative bundling pili of P. abyssi AV2 (Supplementary Fig. 1d, e). Looking at the local genomic context of WP_338249486.1, we identified the protein WP_338249487.1 (AAA988_RS09250), presumed to form an operon with AAA988_RS09245 and predicted to encode a dedicated class I signal peptidase. Based on these data, we hypothesized that WP_338249486.1 formed the structural basis of the bundling pili found in the extracellular milieu of P. abyssi AV2. In analogy to CanX, we propose AbpX as the naming convention for WP_338249486.1. While never previously identified in EM analyses of Pyrodictium extracellular filaments, the AbpX filaments were more similar in structure to the taxonomically more distant TasA filaments than to previously characterized AbpA pili from P. calidifontis. Remarkably, two different precursor proteins, CanX and AbpX, undergo polymerization into structurally distinct extracellular filaments through a common process of DSC on the surface of P. abyssi.

Structural analysis of recombinant P. abyssi cannulae

While the cryoEM structural analysis of the CanX filament provided insight into the origin of the thermodynamic stability and mechanical rigidity of P. abyssi cannulae, the challenging culture conditions hinder preparation of cannulae in sufficient quantities to evaluate their potential in biomaterials applications. Previous experimental investigations demonstrated that recombinant expression of P. abyssi cannula-like proteins in heterologous bacterial hosts represented a viable preparative route to soluble precursors of structurally similar filaments^35,43. CanA, a cannula-like protein from P. abyssi strain TAG11, has been the most thoroughly studied of these recombinant precursors³². Previous investigators reported that recombinant CanA could assemble in vitro into tubular filaments in the presence of calcium ions. The resultant filaments displayed a similar diameter and morphology to ex vivo cannulae from P. abyssi TAG11 under low-resolution electron microscopy imaging³⁵. In addition, a recombinantly expressed truncation of CanA, K₁-CanA, corresponding to a deletion of the first ten amino acid residues of the putative donor strand of the mature protein, displayed no propensity for polymerization under similar conditions⁴³. The induction of self-assembly of recombinant CanA in the presence of calcium ions and its inhibition upon partial deletion of the N-terminal donor strand suggested that recombinant cannula-like proteins could polymerize in vitro through a mechanism that mimicked the assembly of the ex vivo filaments.

A structural determination of in vitro assemblies of recombinant CanA was undertaken to compare the corresponding structure to that of the ex vivo filament. A codon-optimized version of the coding sequence of the mature CanA protein, i.e., after deletion of N-terminal twenty-five amino acids of the putative signal peptide sequence⁵⁰, was designed for bacterial cell expression (Supplementary Figs. 8a and 9a). Previous research demonstrated that this CanA protein variant could be expressed and purified as a soluble protein in Escherichia coli host systems in good yield³⁵. Following the published procedure, recombinant CanA was obtained in an unoptimized yield of 31 mg of purified protein per L of culture from the cell lysate after induction of expression in E. coli strain BL21Gold(DE3) (see “Methods”). The purity and identity of the recombinant CanA protein were confirmed using SDS-PAGE analysis and electro-spray ionization (ESI) mass spectrometry, respectively (Supplementary Figs. 10a and 11a). Sedimentation velocity analytical ultracentrifugation indicated that CanA migrated as a monomer in low-salt buffer (50 mM Tris-HCl, pH 7.5, 80 mM NaCl, 0.1 mM, EDTA, 9% glycerol) at an apparent molecular mass that agreed with mass spectrometric analysis and the monoisotopic molecular mass calculated from the amino acid sequence of mature CanA (Supplementary Fig. 12a). Thermal treatment (50°–80 °C) of solutions of purified CanA in a low-salt buffer led to rapid polymerization in the presence of CaCl₂ (20 mM)³⁵. The experimental conditions for CanA assembly differed significantly from the native growth conditions of P. abyssi (seawater, 80–110 °C, 20–35 MPa hydrostatic pressure). The facile in vitro assembly of CanA under non-physiological conditions for P. abyssi growth suggests that cannula biosynthesis under native conditions may be regulated to prevent premature assembly. Negative-stain TEM images of the in vitro assemblies revealed the presence of a structurally uniform population of high aspect-ratio tubular assemblies of ~25 nm in diameter (Supplementary Fig. 13a), which agreed with previous reports from EM imaging of ex vivo P. abyssi TAG11 cannulae and recombinant polymers of CanA^32,35,38. The cannula-like filaments of CanA laterally associated into large rafts like those observed in ex vivo preparations of P. abyssi AV2 cannulae (vide supra). Similar behavior was observed for cannulae under native growth conditions in cultures of P. abyssi TAG11, in which fast freeze/deep etch electron microscopy imaging revealed the presence of bundles containing up to a hundred cannulae at the cell surface³².

CryoEM analysis was employed to determine the structure of the recombinant CanA assemblies (Fig. 2). Despite extensive lateral association between filaments observed in the raw micrographs, a sufficient number of isolated particles could be classified as the basis for helical reconstruction (Fig. 2a). The 2D class averages confirmed that the tubular filaments were uniform in inner and outer diameter, ~18 nm and ~27 nm, respectively, despite in vitro assembly conditions that differed significantly from the native growth environment of P. abyssi. A 3D volume was reconstructed at 2.6 Å global resolution (FSC 0.143 criterion) from the 2D projection images using iterative helical real-space reconstruction^51,52 after determination of helical symmetry (Supplementary Fig. 14a)²⁵. An unambiguous atomic model could be built into the reconstructed density (Fig. 2b and Supplementary Fig. 14b, c). In contrast to the C2 symmetry observed for the ex vivo cannulae, the CanA filament was a C1 assembly of protomers along a left-handed 1-start helix with a rise of 1.59 Å and a twist of −173.75°. The most prominent structural feature of the CanA filament, as for the CanX filament, was the presence of a pair of right-handed protofilaments with a ~ 4.6–4.7 nm periodicity (Figs. 1a and 2b, c), i.e., half the pitch of the 2-start helices, which was consistent with previously reported TEM images of native cannulae and recombinant cannula-like tubules^32,35,38.

Fig. 2: CryoEM structure of in vitro cannula-like tubules formed by the self-assembly of recombinant CanA.

a representative raw cryoEM micrograph and 2D-class average (inset) of the recombinant CanA filament; b cryoEM volume depicting right-handed 2-start helical packing of the helical reconstruction (rise and twist of 1.59 Å and −173.75°, respectively); c helical model of CanA filament; d local interaction network between neighboring CanA subunits with three Ca²⁺ ions bridging protomers in a 12-mer segment of the cannulae; e close-up view of three Ca²⁺ ions binding sites in stick representation. The representative image in this figure (a) is derived from a total of 11143 cryoEM micrographs from a single technical replicate.

As observed for the ex vivo filament, the atomic model of the CanA tubules suggested that DSC was the primary mechanism that drove polymerization of CanA into cannula-like filaments and stabilized the resultant assemblies. The relative orientation of protomers in the polymer, i.e., along the h → i → j → k direction (Fig. 2d) was defined by a rotation of 1.25° and an axial rise of 46 Å (1.5° and 48 Å, respectively, for the corresponding interaction in CanX). Successive protomers were aligned nearly parallel to the central axis of the helical assembly (Figs. 1b and 2d). Donor-strand complementation linked protomers located on structurally adjacent 2-start helices (Fig. 2b, c) and reinforced the structural interfaces within the CanA filament. The structure of the CanA cannulae displayed an extensive network of inter-protomer contacts that involved subunits located as far away as j ± 60 on either side of a central protomer (Supplementary Fig. 14d). Protein, Interfaces, Surfaces, and Assemblies analysis⁵³ indicated that, while DSC between protomers buried the greatest amount of surface area, significant interactions also occurred between subunits located within the same protofilament (e.g., subunits j-1 and j) along a 2-start helix (Fig. 2d and Supplementary Fig. 14). In addition, as observed for the ex vivo filament, three calcium ions could be fit into the cryoEM density map of the in vitro filament (Fig. 2d). The calcium ion binding sites were structurally homologous to those of the ex vivo filament, mediating similar polytypic radial and axial interactions between structurally adjacent protomers (Fig. 2e). Other than the observed helical symmetry, the most significant difference between the ex vivo and in vitro cannula filaments was the structure of the protomers. While the core structures of CanX and CanA were based on homologous jelly-roll folds (49.0% sequence identity), extended loops were observed in the CanA protomer and the AlphaFold3⁴⁹ predicted monomer structure, which were not present in CanX (Supplementary Fig. 15a). In the recombinant CanA structure, these loops projected from the outer surface of the filament, which resulted in a pair of ridges that coincided with the 2-start helices (Fig. 2b, c). The CanX protomer lacked these loop insertions, although a homologous protein (WP_338251966.1, 51.1% sequence identity to CanA) was identified in the P. abyssi AV2 proteome that presented loop protrusions from the core fold (Supplementary Fig. 5). However, outer surface projections that could account for the presence of this CanA-like protein were not detected in the ex vivo filament (vide supra).

Potential role of calcium ion binding in assembly of cannulae

The conservation of the calcium-binding sites in the structures of the ex vivo and in vitro cannula filaments, as well as the calcium ion induction of CanA assembly, raised a question regarding the relationship between calcium ion binding and DSC. The polymerization of the cannula-like proteins resulted from progressive DSC in combination with calcium ion binding. However, for CanA, calcium ion binding was a necessary but not sufficient criterion for polymerization under the conditions that we investigated. While the monomer was competent for polymerization in the presence of calcium ion⁵⁴, the polymerization process was kinetically slow at ambient temperature and required heating to promote more rapid assembly. Under native growth conditions for P. abyssi (~100 °C), the temperature would be sufficient to initiate polymerization of cannula-like proteins at the extracellular surface since the calcium concentration in seawater (400 ppm, ~10 mM) would be within the permissible concentration range. Hypothetically, other divalent metals ions present in seawater could promote self-assembly of cannula-like proteins into filaments at the cellular surface of P. abyssi. The Mg²⁺ ion is the most abundant divalent cation in seawater (1300 ppm, ~54 mM). However, under the assembly conditions that were employed (50–100 μM protein, 20 mM divalent ion, 50–80 °C, low-salt buffer, pH 7.5), filamentation of recombinant CanA was not observed in vitro, although a recent report suggested that polymerization occurred at higher protein concentration (1–2 mM)⁵⁴. The difference in ionic radius between Mg²⁺ and Ca²⁺, 72 pm versus 100 pm, can lead to different binding affinities between the two divalent metal ions within coordination sites in metalloproteins. Notably, the addition of terbium ion (10 mM), which is often employed as a calcium ion surrogate in biophysical studies of calcium-binding proteins (Tb³⁺ ionic radius ~106 pm)⁵⁵, rapidly initiated polymerization of recombinant CanA into cannula-like filaments at ambient temperature. EDTA (10 mM), a strong chelator of calcium ion (K_b’ ~10⁸ M⁻¹ in 10 mM Tris-HCl, pH 7.5)⁵⁶, inhibited the Ca²⁺ induced polymerization of CanA into cannula-like filaments. However, treatment of pre-formed cannulae with EDTA did not induce dis-assembly unless the solution was heated to ≥ 60 °C. SV-AUC measurements indicated that CanA remained a monomer in solution over the course of the experiments in the presence of sub-critical concentrations of Ca²⁺ ion below the polymerization temperature (Supplementary Fig. 12a).

To gain insight into the potential role of calcium ions in driving filament formation, single-crystal X-ray diffraction was employed to determine the structure of the monomeric form of CanA in the absence of calcium ions (Fig. 3a). A comparison between the crystal structure of the CanA monomer and the corresponding CanA protomer in the assembly revealed significant conformational differences between the two structures that localized to protein segments associated with subunit-subunit interfaces in the filament. The most obvious difference between the two structures lies in the conformation of the N-terminal nineteen amino acids in CanA. These residues were not observed in the crystal structure of the CanA monomer, which suggested that the N-terminal segment was disordered and did not contribute to the diffraction. In contrast, these N-terminal amino acids adopted a β-strand conformation (strand A) in the cryoEM structure of the CanA protomer, which mediated the donor strand-acceptor groove interaction between axially adjacent protomers (e.g., i,j) in the filament (Fig. 2b, c). Deletion of the first 10 amino acids of mature CanA resulted in a construct, K₁-CanA⁴³, that was incapable of polymerization, which further argued for the importance of an intact donor strand in cannulae formation.

Fig. 3: Crystal structure analysis of the CanA monomer.

a crystal structure of the CanA monomer (cyan) depicted as a ribbon diagram. The arrow indicates a rendering of the disordered loop region 34–40; b close-up of the superposed CanA monomer from the crystal structure (blue) with the cryoEM CanA protomer (tan) depicting the conformational change within the loop region 161–173; c close-up of the RMSD rendering from a Matchmaker backbone alignment between the CanA protomer and monomer that highlights the difference in loop region 161–173; d alignment of the CanA monomer from the crystal structure (blue) to an interacting axial pair of CanA protomers from the cryoEM structure. The N-terminal donor strand from protomer i (violet) inserts into the acceptor groove of protomer j (tan); e, f zoomed-in views of loop region 161–173 in the CanA cryoEM structure (e) and the superposed CanA monomer (f). The latter view revealed a potential steric clash (arrow) between this region of the CanA crystal structure with donor strand residues Tyr11 to Ser20 that potentially precludes CanA oligomerization in the absence of calcium ion binding.

Significant structural differences were also observed in the conformation of two loops, corresponding to residues 34–40 and 161–173, which contributed to the axial Ca²⁺ binding sites (Fig. 3b) within the cannula filament. In the crystal structure, the electron density corresponding to residues 35–39 was absent, which implied that the loop conformation was dynamic in the absence of calcium ions. In addition, a backbone alignment of the CanA monomer crystal structure and the CanA cryoEM protomer structure indicated that the loop region corresponding to residues 161–173 underwent a significant displacement upon formation of the filament (Fig. 3b, c). We attributed the observed conformational changes in these two loops to the influence of calcium ion binding. Alignment of the monomeric CanA structure to an axially interacting pair of protomers (i,j) in the filament suggested a mechanism whereby calcium ion binding could promote DSC (Fig. 3d). In the cryoEM structure of the CanA filament (Fig. 3e), calcium ion binding stabilized the conformation of the two loop regions and promoted a local interaction that exposed the acceptor groove of the monomer. In the crystal structure of the monomer, loop region 161–173 adopted a conformation that occluded the acceptor groove, which prevented it from accepting the donor strand of another monomer and precluded DSC potentially preventing filament formation (Fig. 3f). We propose that calcium ion binding to the monomer triggers a structural rearrangement that renders the monomer competent for subsequent donor-strand complementation and polymerization into a cannula-like filament. However, the rate at which polymerization occurred was observed to depend on temperature, perhaps mediated through a thermally activated process that was not apparent from the available biophysical data. The crystal structure of CanA showed that N-terminal strand A was not self-complemented in the monomeric form and, consequently, available to interact with another subunit upon calcium coordination through donor-strand complementation, thereby initiating polymerization into a filament. Structural comparison between the CanA protomer in the cryoEM structure and a recently reported NMR structure of N-terminal deletion K₁-CanA (PDB: 9HDI) similarly indicated that structural differences between monomer and protomer were localized to the N-terminal donor strand and loop regions associated with divalent ion binding⁵⁴. The most pronounced structural difference between the two structures, outside of the donor strand conformation, occurred within the loop region 34–40, which undergoes a significant conformational rearrangement to accommodate calcium ion coordination at the axial binding sites (Supplementary Fig. 16).

CryoEM of recombinant cannula-like protein homologues

To identify other potential cannula-mimetic proteins in the genomes of related archaeal species, Foldseek was employed using the protomer structure of CanA (PDB: 7UII) as a query⁴⁸. Several putative cannula-like sequences were identified within the proteomes of species within the family Pyrodictiaceae, including five proteins from the related species Pyrodictium occultum and four proteins derived from metagenome-assembled genomic (MAG) data reported for an uncultured archaeon in phylum Thermoproteota, tentatively assigned as a species within the genus Hyperthermus (Supplementary Table 1). AlphaFold2 structural predictions were available for the respective proteins through the UniProt webserver (www.uniprot.org)⁵⁷. In each case, the presence of a jelly-roll fold was predicted for the respective monomeric structures with high pLDDT and pTM, although poorly predicted insertions and N-terminal extensions were also observed for several of the larger proteins as previously observed for putative cannula proteins encoded within P. abyssi AV2 genome (Supplementary Fig. 5). In most cases, AlphaFold2 predicted an unstructured N-terminal domain for the mature protein sequences. As determined for CanA and CanX, most of these cannula-mimetic protein sequences contained potential N-glycosylation sequons (NxS/T) that localized to surface loops in the AlphaFold2 structural predictions.

Two of these cannula-like proteins, Hyper1 (hypothetical protein DSY37_00495, UniProt: A0A432RA20) and Hyper2 (hypothetical protein DSY37_03180, UniProt: A0A432R7L7), were chosen as substrates for a more detailed analysis. The mature sequences⁵⁰ of these two proteins were similar in length to CanX and shared a pairwise sequence identity of 38.9% to each other and of 36.4% and 26.2%, respectively, to CanX. Codon-optimized coding sequences (Supplementary Fig. 9) of the putative mature proteins were cloned into the plasmid pD451-SR and transformed into E. coli strain BL21Gold(DE3). Expression cultures were screened for the presence of the respective proteins under IPTG induction using the experimental procedure optimized for recombinant CanA (see “Methods”). While SDS-PAGE analysis indicated that both proteins were successfully expressed, the Hyper1 protein accumulated in inclusion bodies that could not be successfully refolded after denaturation and was not further studied. The Hyper2 protein accumulated in the soluble fraction of the cell lysate and could be purified using the procedure developed for recombinant CanA. Recombinant Hyper2 was obtained in an unoptimized yield of 80 mg of purified protein per L of culture from the cell lysate of E. coli expression cultures. The purity and identity of the Hyper2 protein were confirmed using SDS-PAGE analysis and ESI mass spectrometry (Supplementary Figs. 10b and 11b). As observed for CanA, sedimentation velocity AUC analysis in low-salt buffer indicated that Hyper2 behaved as a monomer in the absence of calcium ion (Supplementary Fig. 12b). The apparent molecular mass of the Hyper2 protein from SV-AUC analysis agreed with that observed from mass spectrometry and the monoisotopic molecular mass calculated from the amino acid sequence of the recombinant Hyper2 protein.

In contrast to CanA, Hyper2 assembled rapidly at ambient temperature in the presence of calcium ion (≥ 10 mM) without heat treatment. However, negative-stain TEM analysis (Supplementary Fig. 13b) indicated a range of filamentous assemblies, including single-walled tubular filaments of different diameters and multi-walled tubes. The structural polymorphism of the Hyper2 assemblies was confirmed in the raw cryoEM micrographs and 2D class averages (Fig. 4a and Supplementary Fig. 17a, b). Helical reconstruction was employed to structurally analyze two of the single-walled tubes corresponding to diameters of ~240 Å and ~270 Å. After independent 2D classification (Supplementary Fig. 17b) and manual assignment of helical symmetry (Supplementary Fig. 18a), standard helical refinement procedures led to cryoEM reconstructions (Fig. 4b and Supplementary Fig. 17c) at 3.5 Å global resolution for both classes of filaments (FSC 0.143 criterion; Supplementary Fig. 18b, c and Supplementary Table 2). Atomic models could be reliably built into the respective EM density maps for the two classes of Hyper2 filaments, which revealed that they were structurally related but based on different helical symmetries. The structure of the thicker tube displayed C8 symmetry with a helical rise of 11.16 Å and twist of 9.59°, while the structure of the thinner tube revealed C1 symmetry with a helical rise of 1.42 Å and a twist of 46.23° (Fig. 4b, c and Supplementary Fig. 17c). As observed for the Pyrodictium cannulae reconstructions, the protomers in the Hyper2 filaments were integrated through a combination of DSC and calcium ion binding interactions (Fig. 4d, e and Supplementary Fig. 17d).

Fig. 4: CryoEM structure of cannula-like tubules (Hyper2 thick tube) formed by the in vitro self-assembly of recombinant Hyper2.

a representative cryoEM micrograph and 2D-class average of recombinant Hyper2 tubes; b cryoEM map showing right-handed 8-start helical packing with C8 rotational symmetry (rise and twist of 11.16 Å and 9.59°, respectively); c helical model of Hyper2 thick tube; d local interaction network between neighboring Hyper2 subunits with three Ca²⁺ ions bridging protomers in a 12-mer segment of the cannulae; e close-up view of the three Ca²⁺ ions binding sites in stick representation. The representative image in this figure (a) is derived from a total of 5502 cryoEM micrographs from a single technical replicate.

The most significant structural difference observed between the Pyrodictium and Hyperthermus cannulae lay in the number and geometrical arrangement of the protofilaments. For the putative Hyperthermus cannulae, eight protofilaments propagated along right-handed 8-start helices having a periodicity of ~5.2 nm (one-eighth of the 8-start helical pitch) and twists of 9.59° (C8) and 9.84° (C1), respectively (Fig. 4b and Supplementary Fig. 17c). For the Pyrodictium cannulae, two protofilaments propagated coincidently with a pair of right-handed 2-start helices having a periodicity of 4.6–4.7 nm (Figs. 1a and 2b, c). However, the distance between the protofilaments along the direction of donor strand polymerization was comparable in the structures of the Hyperthermus cannulae (~4.6–4.7 nm) and the Pyrodictium cannulae (~4.6–4.8 nm). In all cases, DSC occurred between structurally adjacent protofilaments and reinforced the supramolecular structure of the respective cannulae. The donor strand-acceptor groove interaction occurred along a series of left-handed 31-start (C1) or 32-start (C8) helices for Hyperthermus cannula-like filaments (Fig. 4d and Supplementary Fig. 17c). In contrast, the donor strand polymerization in the Pyrodictium cannulae occurred along a set of right-handed 30-start helices for CanX (Fig. 1c) or 29-start helices for CanA (Fig. 2d). The propagation axis of the DSC interaction was nearly parallel to the central helical axis for the Pyrodictium cannulae (~ 4° tilt), whereas donor strand propagation was tilted at an angle (~ 18°) from the central axis in the Hyper2 assemblies.

Despite these differences in helical symmetry, the protomer structure and local structural interfaces were remarkably conserved between the putative Hyperthermus and Pyrodictium cannulae. The core structure of the protomer was based on a donor-strand complemented jelly-roll fold in which calcium ions were observed to bind at the radial and axial interfaces between protomers. Structural alignment of the Cα backbone atoms of the CanX, CanA, and Hyper2 protomers suggested a close correspondence between the respective jelly-roll folds despite the presence of two large loops in the CanA structure (Supplementary Fig. 15a). Two distinct dimeric structural interfaces were identified in the local interaction network of the respective filaments (Supplementary Fig. 2), which were associated with radial interactions between laterally adjacent protomers within the same protofilament (e.g., subunits j−1,j,j + 1) and axial interactions between protomers located in different protofilaments (e.g., subunits i,j,k). Pairwise backbone alignments of the dimeric radial and axial interfaces between protomers in the different cannulae revealed that the structures were strongly conserved (RMSD ≈ 0.8–1.2 Å, Supplementary Fig. 15b). This strong conservation of interfacial interactions among the cannulae can be contrasted to the weaker sequence identity between the Pyrodictium and Hyperthermus cannula-like proteins (Supplementary Fig. 15c).

As in the ex vivo and in vitro P. abyssi cannulae, three distinct calcium ion binding sites were observed in the Hyper2 filaments at structurally homologous positions. A multiple sequence alignment (MSA) of the mature CanX, CanA, and Hyper2 proteins indicated that the residues in the calcium ion binding sites were conserved (Fig. 5a), which suggested a similar role for calcium ions in inducing assembly and maintaining the structural integrity of the two classes of cannulae (Figs. 1b, c, 2d, e, and 4d, e). Likewise, the structural conservation of the donor strand-acceptor groove interaction was observed between the Pyrodictium and Hyperthermus cannulae. In each case, donor strand A of subunit j forms a series of hydrogen-bonded interactions with strands B and K of the acceptor groove on an axially adjacent subunit i (shown in Fig. 5b for the CanX filament). Strands B and K define a hydrophobic groove consisting of five pockets, P1–P5, which accommodate residues on one face of the donor strand (Fig. 5c–e). The donor strands of the cannula-like proteins displayed an amphipathic sequence pattern in which hydrophobic residues alternated with hydrophilic residues. The sidechains of the former residues were buried in the acceptor groove, while the polar sidechains of the latter residues were exposed on the filament surface. The donor strands of the cannula-like proteins displayed significant sequence homology, in which an aromatic residue (F/Y) occurred at the site that occupied the P1 pocket, while smaller sidechain hydrophobic residues (G/A/V) occupied the sites on strand A that interacted with P2–P5 pockets (Fig. 5c–e). Notably, a central Gly residue (cyan, Fig. 5c–e) was conserved in all three proteins, as well as for all putative cannula-like proteins tentatively identified in Pyrodictium and Hyperthermus species. We hypothesize that the conserved glycine residue may guide the alignment between the donor strand and acceptor groove, ensuring the donor strand adopts the registry that promotes structural complementation. To test the importance of the glycine residue, a recombinant protein corresponding to the G13A mutant of Hyper2 was prepared and tested for its ability to undergo polymerization in the presence of calcium ions. Under identical conditions to those successfully employed for the wild-type Hyper2 protein, the Hyper2G13A mutant was severely compromised in its ability to form filaments, which implied that donor-strand complementation was partially to completely inhibited through this conservative substitution. In bacterial CU pili, a similar process of molecular recognition between the side chains of residues on the donor strand and pockets in the acceptor groove promotes correct registry during donor-strand complementation and governs the order of incorporation of different pilins into the filament during biosynthesis^3,4,8,58,59.

Fig. 5: Conserved donor strand complementation across cannulae forming subunits.

a multiple sequence alignment (MSA) of mature CanA, CanX, and Hyper2 sequences (Jalview, Clustal coloring). Red stars denote regions where CanA loop/domain insertions were removed for visual clarity. Red arrows highlight conserved residues that interact with calcium in type I, II, and III coordination centers. Arrows below the MSA denote secondary structure elements color coded according to the monomer cartoon representation in (b); b Illustration of the cannula donor strand complementation mechanism shown here for CanX: the N-terminal strand A of subunit j docks into the surface exposed hydrophobic groove defined by the region between strands B and K of subunit i; c–e zoomed-in renditions of the donor strand (stick representation) docking into the respective B/K grooves for CanX, CanA and Hyper2, respectively. Donor sequences above (c–e): hydrophobic residues docked into complementary pockets highlighted in bold; residues forming sidechain-mediated intermolecular hydrogen bonds shown in red; proline directing the N-terminus towards the “lattice contact” underlined.

Distribution of cannula-like proteins in archaea

This similarity in donor-strand/acceptor groove recognition prompted us to systematically evaluate the evolutionary relationships of CanX and AbpX with other bacterial and archaeal cell surface filaments formed through donor-strand complementation^{3,4,5,6,7,9,10,11,39}. Based on HMMER searches and signal peptide prediction, all CanX homologs are exclusively found in the Pyrodictiaceae family and possess a Sec/SPI signal sequence, which suggested that they undergo transport through the Sec translocon and are subsequently cleaved by Signal Peptidase I, similar to other characterized donor strand complemented filaments. However, the genomic neighborhood of these putative cannula-like proteins lacked genes encoding putative chaperones or dedicated signal peptidases, supporting the hypothesis that CanX could assemble in the presence of calcium without external assistance. Unlike CanX, HMMER searches revealed that AbpX homologs were more widely distributed across archaea¹⁰, including in Pyrodictiaceae that lacked CanX homologs, such as P. delaneyi Su06³⁴ (WP_055409108.1). The sequence homology searches with AbpX and its archaeal homologs also identified sequence matches to bacterial TasA family proteins, with pairwise sequence identity levels of 30%–35%, which supported the classification of AbpX within the TasA superfamily (Fig. 6a, b). Like bacterial TasA proteins⁹, AbpX homologs in archaea were frequently found adjacent to genes encoding a signal peptidase I or other AbpX-/TasA-like proteins, underscoring a potential evolutionary and functional connection between these systems.

Fig. 6: CanX and AbpX belong to the TasA superfamily.

a pairwise profile HMM comparison of representative archaeal and bacterial cell-surface filament subunits. The heatmap shows HHsearch probabilities (%) for each pairwise comparison. The proteins are grouped into five categories: archaeal CanX proteins (arc CanX), archaeal and bacterial TasA-like proteins (arc & bac TasA-like), bacterial type I pilins (bac T1P), archaeal type IV pilins and archaellins (arc T4P & archaellins), and bacterial type IV pilins (bac T4P), all of which are β-sandwich/jelly-roll fold proteins that assemble into filaments. The NCBI/UniProt accession numbers of the proteins are provided in the “Methods” section; b distribution of the TasA superfamily across archaea, visualized on a species-level archaeal tree from the Genome Taxonomy Database. Colored dots indicate the presence of specific protein families: TasA/AbpX (blue), CanX (red), AbpA (green), and S. acidocaldarius 0406 thread (magenta). Phyla are indicated on the outer ring of the tree. Asgard Asgardarchaeota, Methano Methanobacteriota, Met B Methanobacteriota_B, Iain Iainarchaeota, and Aenigm Aenigmatarchaeota; c, d structural representations of representative TasA superfamily proteins. AlphaFold3 models are shown for selected proteins, while cryoEM structures are displayed for P. calidifontis AbpA (PDB: 7UEG) and the S. acidocaldarius 0406 thread subunit (PDB: 7PNB). β-helices are colored red, and β-strands are colored yellow.

To uncover potentially distant CanX homologs that HMMER searches might have missed, sensitive profile HMM-based sequence searches were employed using HHsearch. We searched the profile HMM databases of proteomes from several representative archaea for homologs of CanX, which identified three distant homologs in Saccharolobus solfataricus (WP_009990924.1, WP_009991716.1, WP_029552507.1) with HHsearch probability values of greater than 90% (Supplementary Fig. 19a). Despite sharing only ~18% pairwise sequence identity with P. abyssi CanX, the structures of these putative proteins, predicted by AlphaFold2, retained the jelly-roll fold, showing a striking structural resemblance to CanX (Fig. 6c). Homo-oligomers of these proteins, as predicted by AlphaFold2-multimer⁶⁰, suggested their potential to polymerize through DSC (Supplementary Fig. 19b). These proteins were employed in subsequent HMMER searches that identified over 100 CanX-like proteins in archaea, mostly within the phylum Thermoproteota, specifically in the Sulfolobaceae and Thermocladiaceae families (Fig. 6b). For instance, we identified a homolog each in Vulcanisaeta distributa DSM 14429 (WP_013336327.1) and Caldivirga maquilingensis IC 167 (WP_156769888.1), two in Sulfolobus acidocaldarius DG1 (WP_011279025.1, WP_011279026.1), and four in Metallosphaera yellowstonensis MK1 (WP_009071828.1, WP_009072785.1, WP_009072754.1, WP_009071324.1). AlphaFold3 models for these homologs suggested that they could also adopt the jelly-roll fold (Supplementary Fig. 19b) and could potentially undergo DSC. However, experimental validation will be required to confirm whether these predicted homologs indeed assemble into extracellular filaments that have physiological function in the corresponding organisms in their native environment.

The jelly-roll fold of cannula-like proteins was a structural feature that was shared with the bacterial TasA protein from B. subtilis and its recently identified archaeal homologs, AbpA, from P. calidifontis and 0406 thread subunit from S. acidocaldarius. To explore the evolutionary relationships among these proteins, HHsearch was employed for pairwise sequence comparisons and searches against the PDB and Pfam profile HMM databases. These searches identified statistically significant matches (HHsearch probability values between 80% and 95%) between CanX-like proteins and AbpA (PDB: 7UEG), the S. acidocaldarius 0406 thread subunit (PDB: 7PNB), as well as the Pfam Peptidase_M73 family (PF12389), which encompassed bacterial TasA homologs. The pairwise sequence identity across these matches was low (below 20%), indicating that CanX and CanX-like proteins form a highly divergent family within the TasA superfamily. The S. acidocaldarius thread subunit, which was recently described as representative of a newly discovered class of archaeal surface filaments, also showed statistically significant matches to CanX-like proteins, AbpA, and TasA family proteins in our HHsearch searches. However, the low sequence identity among these proteins (below 20%) suggested that both the thread subunit and AbpA constituted distinct families within the TasA superfamily (Fig. 6b). Additionally, our searches uncovered subdomain-level matches to archaeal flagellin C-terminal globular subunits and bacterial type I filament-forming proteins, such as CupE1, FimA, and PapA (Fig. 6a), all of which adopted the immunoglobulin-like (Ig) fold. Despite the low sequence similarity, the core structures of these proteins displayed a remarkably conserved topology.

While type IV filament (T4F) superfamily assemblies, including the archaeal flagellins, are widespread across archaea⁶¹, the distribution of TasA-like proteins in archaea remains less well understood¹⁰. To address this, we used HMMER to search for homologs of bacterial TasA, AbpX, AbpA, CanX, and the 0406 thread subunit in representative archaeal genomes from the Genome Taxonomy Database (GTDB). Our searches revealed that homologs of AbpA, CanX, and the 0406 filament are primarily restricted to genomes in the Thermoproteota phylum, whereas putative homologs of AbpX and bacterial TasA are found more extensively across the genomes of several archaeal phyla (Fig. 6b). Interestingly, the genomes of many archaeal species encode two or more of these putative TasA-like proteins. For example, the V. distributa DSM 14429 genome encodes potential homologs of CanX (WP_013336327.1), AbpA (WP_148678262.1), and the thread subunit (WP_013336193.1, WP_013336196.1), highlighting the potential diversity and functional versatility of these putative filament-forming proteins. These findings demonstrate that genes encoding TasA-like proteins may be far more widespread in the genomes of archaea than previously recognized. The observation of bundling pili in P. calidifontis and S. acidocaldarius, as well as the presence of the AbpX filaments in the ex vivo P. abyssi culture, suggest that many archaeal species may have the potential capacity to form biofilms under conditions in which these TasA-like proteins are expressed as filaments at the extracellular surface.

Discussion

DSC underlies the polymerization of several classes of bacterial^{3,4,5,6,7,9,39} and archaeal^10,11 pili that are involved in cellular adhesion and biofilm formation. Proper assembly of bacterial pili often requires the involvement of specialized cellular machinery such as the CU secretion system^62,63. In absence of the usher catalyst, in vitro polymerization of recombinant CU-type pilin monomers is kinetically slow and results in polymers that have shorter lengths and lower stability than the native pili, which has been attributed to the introduction of structural defects in the assemblies^63,64. In contrast, in vitro polymerization of recombinant cannula-like protein monomers afforded filaments that mimicked the quaternary structure of ex vivo cannulae and preserved the critical interfacial interactions between protomers. Accordingly, unlike most bacterial protein filaments derived from DSC, the polymerization of the cannulae does not require the presence of a dedicated membrane-bound translocation/assembly platform⁶⁵ or accessory proteins such as chaperones⁶² or proteases^39,66 to mediate polymerization. In agreement with this hypothesis, neither dedicated signal proteases nor putative chaperone proteins were identified within the genetic loci encoding cannula-like proteins. The presence of calcium ion binding sites distinguished cannulae from other extracellular filaments that arose from DSC. Our results suggested that calcium ion coordination primed DSC and promoted the in vitro polymerization of recombinant cannula proteins into cannula-like filaments (Fig. 3). In addition, calcium ion coordination stabilized the quaternary structure of the filaments through formation of a network of axial and radial interactions between structural subunits in both the ex vivo and in vitro cannulae.

Coordination of calcium has been previously reported to induce structural transitions in proteins secreted at prokaryotic cell surfaces. These transitions have been attributed to the structural influence of calcium ion coordination, which arises from the higher Ca²⁺ ion concentration between the extracellular and intracellular environment ([Ca]_extra/[Ca]_intra ~10⁴)⁶⁷. Calcium-dependent structural transitions have been studied extensively in extracellular filaments such as the repeats-in-toxin adhesion proteins (RTX adhesins)⁶⁸, which are megadalton proteins that are translocated through the outer membrane (OM) of Gram-negative bacteria via the Type I secretion system (T1SS)⁶⁹. RTX adhesins can consist of >100 copies of tandemly repeated bacterial Ig-like (BIg) protein domains capped with a C-terminal RTX β-roll domain. Since the secretion signal is located at the C-terminus of the protein, translocation cannot proceed until completion of protein synthesis, which requires that the nascent polypeptide chains are maintained in an unfolded state that is permissive for translocation through the T1SS translocon. Biophysical analyses of pathogenic⁷⁰ and non-pathogenic^71,72 RTX adhesins have demonstrated that calcium ion binding plays several critical roles for this class of proteins, including facilitation of transport across the OM and induction of folding in the extracellular environment^73,74. RTX adhesin proteins are proposed to be maintained in a fully or partially unfolded state at the low intracellular calcium ion concentration prior to secretion through the translocon^74,75,76. In the extracellular environment, the higher calcium ion concentration initiates folding of the BIg protein domains, essentially serving as a “chemical foldase”⁶⁹. Crystallographic analyses of single or tandemly repeated RTX BIg domains provided evidence that calcium binding not only stabilized the internal structure of individual proteins but also structured flexible loops between domains, causing a rigidification of the structure that enabled the filament to extend outward from the cellular surface^{70,72,74,75,77}. We envision a similar dual functional role for calcium ion coordination in the assembly of cannulae. At the low intracellular calcium ion concentration, cannula precursor proteins remain in a monomeric or low oligomeric state. Upon secretion into the extracellular environment, the higher calcium ion concentration triggers self-assembly of the monomers into cannulae by facilitating DSC. Calcium ions coordinate at binding sites at the axial and radial interfaces between protomers, further rigidifying the structure of the cannulae. Likewise, a calcium ion gradient has been demonstrated to induce self-assembly of bacterial S-layers^78,79,80. Binding of calcium ion to interfacial loops of these cell surface structures induced a structural transition into their assembly-competent conformation and promoted adoption of the correct quaternary structure. Like the RTX adhesins and cannulae, calcium-induced structural transitions were proposed to provide a control mechanism to prevent premature assembly of nascent subunits in the cytoplasmic environment, where calcium is held at low nanomolar concentrations^78,79.

The large lumen (~18 nm) of the cannula filaments provides an attractive target for the development of functional nanomaterials that can encapsulate and release large substrates, but also begs the question of the native function of cannulae. CryoET analysis of P. abyssi TAG11 cannulae provided evidence for the presence of density within the lumen, which led to a proposed role in substrate transport³⁸. Our structural analysis of the ex vivo cannulae indicated that they could accommodate filamentous cargo within the lumen, which in several instances was observed to be helically wound at the interior surface (Fig. 1d). The thin diameter of the filamentous cargo (~2 nm) and its susceptibility to DNase I treatment suggested the presence of DNA in the lumen, although the structural analysis was inconclusive regarding the cargo identity. Extracellular protein filaments have been previously demonstrated to mediate horizontal gene transfer (HGT) in prokaryotes through transport of DNA between cells. The proposed transport mechanism in archaea involves the synthesis of a thin conjugative pilus (~7–9 nm outer diameter) through which DNA uptake can occur from a donor cell. We wondered if cannulae could serve as an alternative conduit that could mediate HGT through cross-cellular DNA transport. Conjugative DNA uptake in archaea involves the Ced/Ted DNA import systems^19,81, which are encoded on the archaeal chromosome and are homologous to the prokaryotic type IV secretion system (T4SS)⁸². Could cannulae replace the more widespread conjugative DNA uptake process in Pyrodictiaceae? To answer this question, we searched for the presence of genes encoding Ced/Ted-related proteins within the genomes of organisms in Pyrodictiaceae family. Pilin protein sequences Aeropyrum pernix CedA1 (WP_010865579) and P. calidifontis TedF (WP_011849449) were employed to query the genomes of the family Pyrodictiaceae in a PSI-BLAST search. The TedF query resulted in no positive hits, which was not surprising given that the two families reside in different orders. In contrast, the A. pernix CedA1sequence, belonging to the same order Desulfurococcales, returned multiple hits at E-values (10⁻²⁴–10⁻²⁷) that were suggestive of high sequence identity. The recovered CedA1 homologs encompassed all members of the family Pyrodictiaceae for which genomic information was available as well as MAG data from uncultured Pyrodictium and Hyperthermus species. While cannulae may play a role in HGT through DNA transport, the conservation of an endogenous Ced DNA import machinery suggested that a putative cannula-mediated mechanism must be complementary to and independent from HGT through the Ced conjugative uptake pathway.

The narrow distribution of putative cannula-like proteins in archaeal genomes raises a question regarding the evolutionary and structural relationship between cannulae and the more widely distributed ABP of the TasA superfamily. The co-expression of structurally distinct AbpX and CanX extracellular filaments in P. abyssi cultures implies different functional roles despite similar jelly-roll folds and a common process of polymerization through donor-strand complementation. A structural homology search of the proteome of Pyrodictium occultum PL-19⁸³, a shallow-sea hyperthermophilic archaeon in which cannulae have been physically observed and for which a genome sequence (NZ_LNTB01000001.1) is available³¹, identified a gene encoding a putative ABP (WP_058370724.1). Previous EM analyses of ex vivo samples in Pyrodictium occultum cultures did not detect Abp-like structures^29,31, although these thin filaments would have been difficult to detect under the experimental imaging conditions. For comparison, a gene encoding a putative AbpX-like protein (WP_055409108.1, Uniprot: A0A0N7JD56) was identified within the genome of Pyrodictium delaneyi Su06 (NZ_CP013011.1), a motile deep-sea archaeon phylogenetically most closely related to P. abyssi AV2^34,41. Neither sequence nor structural homology searches identified genes encoding putative cannula-like proteins in the genome of P. delaneyi or in the complete genomes available for other archaea within the Pyrodictiaceae family.

We speculate that cannulae may have evolved from the structurally similar and more widely distributed ABP as a unique evolutionary adaptation within this sub-class of archaea to accommodate a sessile life cycle under hyperthermophilic environmental conditions. These Pyrodictium species, thriving at elevated temperature (~100 °C) in deep-sea hydrothermal vents and submarine solfataric fields, form intricate colonies with individual non-motile, disk-shaped cells linked through cannulae. Biofilm integrity may be further supported by the presence of filamentous networks derived from assembly of other matrix proteins, e.g., AbpX, that independently undergo polymerization through donor-strand complementation. These cellular colonies may display a clonal form of multicellular organization that is distinct from that observed in most microbial communities. Cannulae can potentially enable physical and chemical communication (including possible DNA exchange) between cells, creating a structure reminiscent of biological tissues. Interconnected cells form a mesh-like network that, while not truly multicellular in the sense of differentiated cells in animals or plants, represents a cooperative and coordinated community. This network potentially enhances resilience under extreme conditions, allowing for resource sharing (e.g., DNA repair) and robustness to environmental stress—characteristics often associated with the early stages of multicellularity.

Methods

Materials

Luria-Bertani (LB) medium was purchased from IBI Scientific (Dubuque, IA). Isopropyl β-D-1-thiogalactopyranoside (IPTG) was purchased from GoldBio (St. Louis, MO). Chicken egg white lysozyme was purchased from Research Products International (Prospect, IL). Benzonase nuclease was purchased from Merck KGaA (Darmstadt, Germany). Protease Inhibitor Cocktail Set V was purchased from Calbiochem (San Diego, CA). Competent cells of E. coli strain BL21Gold(DE3) were purchased from Agilent Technologies (Cedar Creek, TX). Precision Plus Protein™ Kaleidoscope™ pre-stained protein standards were purchased from Bio-Rad (Hercules, CA). Custom gene synthesis was performed by ATUM (Newark, CA). Carbon-coated copper grids (200 mesh) and C-flat cryoEM grids were purchased from Electron Microscopy Sciences (Hatfield, PA). Lacey carbon cryo grids and 2% (w/w) solutions of methylamine vanadate (Nano-Van) and methylamine tungstate (Nano-W) negative stains were purchased from Ted Pella (Redding, CA). All other chemical reagents were purchased from either VWR (Radnor, PA) or Sigma-Aldrich Chemical Co. (St. Louis, MO) and used without further purification.

Protein expression and purification

Codon-optimized genes corresponding to the expression cassettes of CanA, Hyper1, and Hyper2 (Fig. S9) were purchased from ATUM (Newark, CA) as sequence-verified clones in plasmid pD451-SR (Kan^R). The synthetic genes were cloned as transcriptional fusions under the control of a T7 promoter with a strong ribosome binding site in the high copy number plasmid pD451-SR (pUC ori). Purified plasmids encoding the respective proteins were transformed into E. coli strain BL21Gold(DE3). Single colonies of the respective transformants were inoculated into sterile LB medium (4 mL) supplemented with kanamycin (100 g/mL) and incubated at 37 °C for 12 h. The pre-incubated cultures were used to inoculate sterile LB media (200 mL) in a shake flask culture supplemented with kanamycin to a final concentration of 100 μg/mL. Bacterial cultures were incubated on a rotary shaker at 37 °C until the OD₆₀₀ reached a value of ca. 0.4 absorption units (AU). Protein expression was induced through addition of isopropyl β-D-1-thiogalactopyranoside (IPTG) to a final concentration of 1 mM and the bacterial cultures were incubated with shaking for an additional 4 h at 30 °C. Cells were isolated from the expression culture through centrifugation at 9100 × g for 3 min and resuspended in 2 mL of low-salt buffer (50 mM Tris-HCl pH 7.5, 80 mM NaCl, 9% glycerol). The following reagents were added to the cell lysate: hen egg-white lysozyme (final concentration of 1.25 mg/mL), Benzonase nuclease (250 units), MgCl₂ (final concentration of 1 mM), and 100× protease inhibitor cocktail (to 1×). The cell lysate was incubated with shaking at 30 °C for 1 h. The clarified cell lysate was dialyzed against low-salt buffer containing EDTA (final concentration of 5 mM). After several changes in this high EDTA buffer, the protein was dialyzed in a low-EDTA version of the same buffer (0.01 mM EDTA) with several changes. The dialysate was heated to 80 °C for 20 min to denature the heat-labile proteins of the host bacterium and was subsequently cooled to ambient temperature. The resultant solution was centrifuged at 14,000 × g for 10 min at ambient temperature. The clarified supernatant fraction was dialyzed overnight against low-salt buffer in a membrane having MWCO of 10,000 Da for CanA and 7000 Da for Hyper 2. The protein concentration of the dialysate was determined spectrophotometrically using a Nanodrop spectrophotometer. Purified protein samples (4 mL) in low-salt buffer achieved a concentration of ca. 400 µM after purification. Protein purity was assessed using SDS-PAGE gel electrophoresis and electrospray-ionization (ESI) mass spectrometry.

Mass spectrometry

Recombinant protein samples were dialyzed extensively against deionized water to remove salt prior to mass spectrometric analysis. Mass spectra were acquired on a Thermo Exactive Plus using a nanoflex source with off-line adapter. Solutions of peptide (10 μL) were deposited into a New Objective econo picotip and placed in the nanosource using the off-line adapter. A backing pressure was applied using an off-line syringe to assure flow to the tip. A voltage of 1.0–2.5 kV was applied to the picotip. Typical settings employed for the analysis were a capillary temperature of 320 °C and an S-lens RF level between 30 and 80 with an AGC setting of 1 × 10⁶. The maximum injection time was set to 50 ms. Spectra were taken at 140,000-resolution at a mass/charge ratio (m/z) of 200 using Tune software and analyzed with Thermo Freestyle software. Representative mass spectra for the protein samples were acquired from analysis involving two technical replicates.

Analytical ultracentrifugation

Sedimentation velocity experiments were performed on a Beckman Coulter Optima AUC at the Canadian Center for Hydrodynamics at the University of Lethbridge. Recombinant CanA (8.3 and 35 μM) and Hyper 2 (10.0 and 34 μM) samples were measured in low-salt buffer (50 mM Tris-HCl pH 7.5, 80 mM NaCl, 0.01 mM EDTA, 9% glycerol); either alone or supplemented with aqueous CaCl₂ to a final concentration of 1–20 mM. All measurements were conducted in epon-charcoal centerpieces fitted with quartz windows and measured in intensity mode at 45,000 RPM and 20 °C. Data were collected at 220 nm for the low concentration protein samples and 295 nm for the high concentration samples. All data were analyzed with UltraScan III version 4.0⁸⁴. The data were fitted with an iterative two-dimensional spectrum analysis⁸⁵ to fit time- and radially invariant noise and meniscus positions. Sample analysis was further refined using Monte Carlo analysis⁸⁶, and diffusion-corrected integral sedimentation coefficient distributions were generated using the enhanced van Holde–Weischet analysis methods⁸⁷. UltraScan calculated the buffer density and viscosity to be 1.023810 g/cm³ and 1.27097 cP, respectively. Representative sedimentation coefficient distribution curves were acquired from datasets derived from two independent technical replicates.

In vitro cannulae assembly

Aliquots (100 μL) of the recombinant CanA and Hyper2 proteins (50–100 μM) in low-salt buffer were incubated in the presence of different concentrations of CaCl₂ and MgCl₂ (final concentration from 5 to 20 mM in the respective divalent ions). The temperature of the polymerization mixture for CanA was heated in the temperature range from 50 to 80 °C for 20 min and cooled to ambient temperature. The polymerization mixture of Hyper2 was incubated for 24 h at ambient temperature after addition of Ca²⁺ ion to the desired concentration. Polymerizations samples were screened using negative-stain transmission electron microscopy (nsTEM) for the presence of fibrillar assemblies.

Negative-stain transmission electron microscopy

Aliquots (5 µL) of the polymerization mixtures under different assembly conditions were deposited onto 200-mesh carbon-coated copper grids. After 1 min of incubation, excess solution was removed by wicking the sample. An aliquot (5 µL) of negative stain (1% mixture of 50% NanoVan/50% NanoW) was placed on the grid and incubated for 1 min. Excess stain was wicked away and the grids were air-dried. TEM images were collected with a JEOL JEM-1400 transmission electron microscope at an accelerating voltage of 80 kV.

CryoEM structural analysis of ex vivo cannulae

An active culture of P. abyssi strain AV2 (DSMZ 6158) was purchased from DSMZ-German Collection of Microorganisms and Cell Cultures GmbH. P. abyssi was cultured in medium 283 (Pyrodictium medium) under anaerobic conditions at 98 °C in a pressurized vessel with 2 bar over-pressure of sterile 80% H₂ and 20% CO₂ gas mixture. Aliquots (100 µl) of pressurized culture were taken using a 1 mL Injekt-F syringe (Braun) fitted with a 23-gauge sterile needle (BD Microlance) by puncturing the rubber fitting. Harvested aliquots were transferred to 1.5 mL Eppendorf tubes, effectively breaking the anaerobe storage conditions. To minimize the effects of exposure to oxygen, care was taken to prepare cryoEM grids shortly after sample harvesting. For that, aliquots (3 µl) of the culture were used as is during grid preparation with no further processing to avoid sample processing artefacts.

A high-resolution cryoEM dataset was collected using Quantifoil™ R2/1 300 copper mesh holey carbon grids. Grids were glow-discharged at 5 mA plasma current for 30 s in an ELMO (Agar Scientific) glow-discharger. A Gatan CP3 cryo-plunger set at −176 °C and relative humidity of 90% was used to prepare the cryo-samples. A sample (3 μL) of the culture was applied on the holey grid and incubated for 60 s. The sample was back-blotted using Whatman type 2 paper for 3 s and plunge-frozen into precooled liquid ethane at −176 °C. High-resolution movies were recorded at the VIB-VUB Facility for Bio Electron Cryogenic Microscopy (BECM) at 300 kV on a JEOL Cryoarm300 microscope equipped with an in-column Ω energy filter (operated at slit width of 14 eV), automated with SerialEM v4.1.8. The movies were captured with a K3 direct electron detector run in counting mode at a magnification of 60 K with a calibrated pixel size of 0.71 Å/pixel, and a total exposure of 60 e/Ų over 60 frames. A total of 2365 movies were collected and imported into cryoSPARC v4.5.3⁸⁸ for further processing. Movies were motion-corrected using Patch Motion Correction and defocus values were determined using Patch CTF. Exposures were curated and segments were picked using the filament tracer and extracted (750k particles) with a box size of 300 × 300 pixels at a pixel resolution of 1.42 Å/pixel. After several rounds of 2D classification (310 k particles retained), Fourier-Bessel indexing was used to calculate initial estimates of the helical rise and twist values, which were used in a subsequent helical refinement job. Next, particles were re-extracted at 0.71 Å/pixel (600 × 600 pixel) and used as input for a second round of helical refinement, followed by local and global CTF refinement, and helical refinement. The resulting high-resolution volume and particle stack were used for reference-based motion correction after particle duplicate removal (310k particles), followed by a final helical refinement (2.30 Å global resolution, FSC 0.143 criterion) after mask application in cryoSPARC. An initial atomic model was built using ModelAngelo⁴² v1.0.12 without providing an input sequence (build_no_seq), after which the model was manually adjusted to the CanX sequence (WP_338251948.1) and real-space refined in Coot⁸⁹. A final round of refinement was performed using Phenix^90,91) and figures were created using ChimeraX⁹². Map and model statistics are found in Supplementary Table 2.

CryoEM imaging of in vitro assembled cannulae

The CanA and Hyper2 samples (ca. 3–3.5 μL) were applied to glow-discharged lacey carbon grids and plunge frozen using an EM GP Plunge Freezer (Leica). Grids were positively glow-discharged using amylamine for the Hyper2 sample. The cryoEM movies were collected on a 300 keV Titan Krios with a K3 camera (University of Virginia) at 1.08 Å/pixel and a total dose of 50 e/Ų using software EPU v2.5.0.4799REL. First, motion corrections and CTF estimations were done in cryoSPARC^88,93,94 v3.2 (CanA) or v3.3.2 (Hyper2). Next, particles were automatically picked by “Filament Tracer” with a shift of 13 and 24 pixels for CanA and Hyper2, respectively. All auto-picked particles were subsequently 2D classified through multiple rounds of selection, and all particles in bad 2D averages were discarded. After initial processing, 659,059 particles remained for CanA. In contrast to the CanA tubes, the Hyper2 assemblies displayed significant polymorphism, including the presence of single-walled and multi-walled tubes. The single-walled tubes constituted populations of tubes that displayed different diameters. 2D classification of the single-walled tubes gave enough particles for 3D reconstruction for two distinguishable classes (95,418 and 140,207 particles) representing tubes having diameters of 240 Å (Hyper2 thin tube) and 270 Å (Hyper2 thick tube), respectively. A range of possible helical symmetries was calculated from the respective averaged power spectra derived from aligned raw particles. To speed up the reconstruction, a smaller subset of particles was first used to test all possible helical symmetries by trial and error until amino acid side chains, and the hand of a short α-helix was seen, which was followed by final reconstructions with all good particles^51,95. The final volumes were then sharpened with a negative B-factor, −85, −88, and −89 Ų for CanA, Hyper2 thick tube, and Hyper2 thin tube, respectively, which was automatically estimated in cryoSPARC. After final helical refinement, resolution was estimated from GS-FSC calculations (0.143 criterion)⁹⁶ after mask autogeneration in cryoSPARC. The resolutions of the reconstructions are reported in Supplementary Figs. 14 and 17 and Supplementary Table 2. EMReady was employed to perform post-processing on raw non-sharpened maps⁹⁷.

Model building of recombinant cannulae

The absolute helical hand of the maps of CanA and Hyper2 tubes was determined from the hand of an α-helix. The hand assignment also agreed with the AlphaFold2⁴⁵ prediction of CanA and Hyper2. The cryoEM map corresponding to a single subunit of CanA and Hyper2 was segmented in Chimera⁹⁸. The AlphaFold2 predictions of CanA and Hyper2 were used as the starting model and docked into the cryoEM maps. The docked models were refined by iterative cycles of manual refinement using Coot⁸⁹ and automatic refinement using PHENIX⁹⁹ and ISOLDE¹⁰⁰ as needed to reach the reasonable map-to-model agreement and model geometry. Densities for eight (1-8) N-terminal residues and four (149–152) C-terminal residues of Hyper2 were not resolved in the map and thus these residues were left unmodeled. MolProbity¹⁰¹ was used to evaluate the quality of the filament model. The refinement statistics are shown in Supplementary Table 2.

CanA crystallization and data collection

The recombinant CanA protein was concentrated to 10 mg/ml and put into crystallization trials. All crystals were grown by sitting drop vapor diffusion at 20 °C using a protein to reservoir volume ratio of 1:1 with total drop volumes of 0.2 μL. Crystals of CanA were grown using a crystallization solution containing 100 mM MgCl₂, 30% PEG 4 K, 3% Xylitol, and 100 mM Tris pH 8.5. All crystals were flash frozen in liquid nitrogen after a short soak in the appropriate crystallization buffers supplemented with 20–25% ethylene glycol. Data for CanA were collected at the beamline 19-ID NYX at the National Synchrotron Light source II (NSLS II), Brookhaven National Labs. All data was indexed, merged, and scaled using HKL2000, then converted to structure factors using CCP4.

Crystal structure determination and refinement

The CanA crystal structure was solved by molecular replacement using the program Phaser¹⁰². Molecular replacement calculations were performed using the coordinates of the CanA monomeric subunit from the CanA cryoEM structure (PDB: 7UII as the search model. The resulting phase information from molecular replacement was used for manual model fitting of the CanA structure using the graphics program COOT⁸⁹. Structural refinement of the CanA coordinates was performed using the PHENIX package⁹⁰. During refinement, a cross-validation test set was created from a random 5% of the reflections. Data collection and refinement statistics are listed in Supplementary Table 3. Molecular graphics of the CanA crystal structure were prepared using the PyMOL Molecular Graphics System (Schrodinger) (DeLano Scientific LLC, Palo Alto, CA) or ChimeraX⁹². A representative image of a portion of the CanA electron density map is provided in Supplementary Fig. 20.

Bioinformatic analyses of CanX and AbpX

All sequence similarity searches were performed using the HMMER webserver⁴⁷ against the UniProtKB database and a local installation of HMMER v3.4¹⁰³ against the non-redundant (nr) protein sequence database. Additionally, HHsearch within the MPI Bioinformatics Toolkit¹⁰⁴ was utilized to detect distant homologs. For pairwise sequence comparisons of representative archaeal and bacterial cell-surface filament-forming proteins (Fig. 6a), we constructed MSAs by running three iterations of HHblits¹⁰⁵ against the UniClust30 database¹⁰⁶. Secondary structure information was incorporated into these alignments using the addss.pl script from HH-suite3. Profile hidden Markov models (HMMs) were then generated using hhmake, and context-specific transitions were calculated using cstranslate. A custom-built profile HMM database was built as outlined in the HH-suite3 user guide. To obtain pairwise HHsearch probabilities, individual alignments were searched against this custom-built profile HMM database using default parameters.

The query sequences correspond to previously characterized archaeal and bacterial surface proteins that assemble into filaments based on β-sandwich/jelly-roll folds, employing either donor-strand complementation (type I pili, TasA-like proteins) or helical packing of N-terminal α-helices (type IV pili and archaeal flagellins/archaellins). For clarity, the proteins analyzed in Fig. 6a were grouped into five categories: archaeal CanX proteins, archaeal and bacterial TasA-like proteins, bacterial type I pilins, archaeal type IV pilins and archaellins, and bacterial type IV pilins. Figure 6a includes the following proteins: CanX from P. abyssi (WP_338251948.1) and Hyperthermus sp. (RUM47266.1); CanX-like from S. solfataricus (WP_009990924.1, WP_009991716.1, WP_029552507.1); AbpX from P. abyssi (WP_338249486.1); AbpA from P. calidifontis (A3MUL8); 0406 thread subunit from S. acidocaldarius (Q4JBK8); TasA from Archaeoglobus fulgidus (A0A075WE38), Methanosarcina barkeri (A0A0E3QZX2), and B. subtilis (P54507); CalY1 from B. subtilis (AKB91166); bacterial type I pilins from E. coli (PapA—P04127, FimA1—P04128), Acinetobacter baumannii (CsuA/B—A0A6F8TDQ5), and Pseudomonas aeruginosa (CupE1—Q9HVE4); archaeal flagellins from Methanospirillum hungatei (Q2FUM4), Methanococcus maripaludis (Q6LWP3), and Pyrococcus furiosus (A0A0B4ZYM1); archaeal type IV pilins from S. solfataricus (A0A157T322, UpsA—Q7LXX9), Sulfolobus islandicus (M9UD72), and S. acidocaldarius (AapA—Q4J6I2); and bacterial type IV pilins from Thermus thermophilus (PilA5—Q72GL2), P. aeruginosa (PilA—P02973), and Neisseria gonorrhoeae (PilE1—P02974).

To investigate the prevalence of TasA-like filaments in archaea (Fig. 6b), we searched for homologs of TasA, AbpX, AbpA, the 0406 thread subunit, and CanX-like proteins in archaeal proteomes within the GTDB (release R220)¹⁰⁷ using HMMER. First, we built MSAs by running three iterations of jackhmmer against the nr70_arc database, a filtered version of the nr protein sequence database containing only archaeal sequences and with a maximum pairwise sequence identity of 70%. For TasA, we used the pre-existing alignment (PF12389) from the Pfam database¹⁰⁸. The resulting alignments were then converted into profile HMMs using hmmbuild. To detect homologs of each query protein in the GTDB archaeal proteomes, we performed searches with hmmsearch using a stringent E-value threshold (-E) of 1e-10. We visualized the presence of these homologs in a GTDB archaeal phylogenetic tree using iTOL¹⁰⁹. For this, we converted the GTDB archaeal trees into iTOL-compatible format using the convert_to_itol method from GTDB-tk v2.4.0+¹¹⁰. For signal peptide prediction, we used DeepTMHMM¹¹¹, and for obtaining structural models, we used a local installation of AlphaFold2⁴⁵, the AlphaFold3 webserver⁴⁹, as well as the AlphaFold Protein Structure Database¹¹².

Reporting summary

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