Adhesin-like proteins have complex and diverse sequences with variable lengths. There are currently no comprehensive databases for the annotation of ALPs in methanogen genomes, and ALPs have, in general, not been defined. In this study, we applied an expanded search approach using annotated ALPs from 13 other methanogens (refer to materials and methods) to query the genomes of M. stadtmanae and M. smithii. We report 49 ALPs in the genome of M. smithii and 42 in the M. stadtmanae genome, which is more than that reported in earlier studies (Fricke et al., 2006; Samuel et al., 2007; Hansen et al., 2011). Some ALPs were previously annotated as hypothetical proteins and were therefore not reported at the time the genomes were published. The 91 ALPs in the two species reported here account for ~10% of the genome of M. smithii and M. stadtmanae each (Table 1). This fraction is significantly larger in comparison to bacteria, where only <1.5% of the proteome is dedicated to code for adhesins (Monzon et al., 2021). This difference is mainly due to the higher number of ALP genes in methanogens as compared to those in bacteria (Monzon et al., 2021). The lengths of ALPs ranged from 128 amino acids to 4,691 amino acids in M. smithii (average: 1,299 amino acids) and 251 amino acids to 3,356 amino acids in M. stadtmanae (average: 1,462 amino acids). Both the number and the length of ALPs contribute toward the large fraction of genome coding for ALPs in two species.

Functional annotations using reference-based tools available in InterPro and Pfam (Mistry et al., 2020; Blum et al., 2021; Paysan-Lafosse et al., 2022) indicate an absence of domains in a large part of ALP sequences or only limited similarity of domains with Pfam references. Only 30 M. smithii and 24 M. stadtmanae ALPs could be matched to at least one domain in Pfam, which is now part of the InterPro database. We assigned domains to ALPs based on sequence similarity to the nearest proteins in the AlphaFold database (Supplementary Table 1). It was found that archaeal ALPs from the two methanogens have mainly three to four different domain types: membrane-anchoring domain (s) (MAD), archaeal big domain (ABD), and right-handed beta-helical (RBH) domain, while other domains such as transglutaminase-like domain (TG-like) and carbohydrate-binding domains were also detectable in some ALPs (Figure 1). The frequency of occurrence of the different domains in ALPs is shown in Table 2. The detailed features of these domains are presented in the sections below.

Two types of MADs were recognized in ALPs of two organisms, transmembrane (TM) α-helices and amphipathic helices. TM helices were present in most ALPs of both organisms. In general, TM helices are present at the N-terminus of ALPs; in some cases, e.g., YP_001272624 of M. smithii, a duplication of the TMH at the N-terminus, can be observed. In 11 M. smithii ALPs, MADs were present at both termini, which may indicate further complexity of potential interaction with other microbes. In comparison, M. stadtmanae had a single N-terminal MAD in 39 out of 42 ALPs, and three ALPs had no MAD.

For some ALP sequences (10 in M. smithii, four in M. stadtmanae), we noticed the presence of N-or C-terminal helices with hydrophobic residues in the AlphaFold structure; however, TMHMM failed to assign them any transmembrane (TM) domain. These were < 20-amino-acid-long sequences of hydrophobic residues, which might not form a complete TM helix (Supplementary Table 2). The typical length of the TM helix was suggested to be 24.0 (±5.6) amino acids (Baeza-Delgado et al., 2013) though it depends on the amino acid sequence and hydrophobicity (Krishnakumar and London, 2007). We noticed that the hydrophobicity index values of short hydrophobic helices in ALPs typically were > 1, which was comparable to the values determined for full-length TM helices using the HeliQuest sequence analysis module (Gautier et al., 2008). These short helices were rich in long-chain hydrophobic residues such as leucine, isoleucine, and phenylalanine (Kyte and Doolittle, 1982). Together with TM helical domains, we marked these short helices as membrane-anchoring domains as they might also help anchor ALPs to the lipid membrane. Earlier studies have shown that the lipid bilayer can adapt to TM helices as short as 10–12 leucines (Baeza-Delgado et al., 2016) and can adjust to negative mismatches (Krishnakumar and London, 2007). It is proposed that the response of short helices to the surrounding lipid bilayer depends on the nature of lipids the helix is in contact with, its amino acid composition, and the distribution of amino acids along the helix. The lipid bilayer may respond to hydrophobic mismatch caused by short helices by compression and chain disordering (de Planque et al., 1998). Furthermore, the packing response of lipids around the single helix could be different than that for larger proteins (Weiss et al., 2003). Marginally hydrophobic α-helices have also been shown useful in membrane protein folding (De Marothy and Elofsson, 2015).

In addition to the short membrane-anchoring α-helices, we also observed the presence of amphipathic helices at the terminals in the structures of some ALPs, where in most cases a TM helix was missing. Using helical wheel diagrams (Supplementary Figure 1), we showed that such sequences in ALPs could fold into an amphipathic α-helix with a hydrophobic face and a hydrophobicity index >50%, as predicted by HeliQuest (Gautier et al., 2008). It is possible to assume that such helices could anchor ALPs to a lipid membrane by lying parallel to a lipid bilayer membrane with the hydrophilic surface interacting with charged lipid head groups while the hydrophobic side is exposed to the fatty acid chains of a membrane lipid. Almost all ALPs had at least one MAD except five ALPs (YP_001272625 and YP_001274282 from M. smithii and YP_447499, YP_447699, and YP_447953 from M. stadtmanae). This could be due to the partial annotations of these proteins from the genomic sequence, or it is also possible that the MAD could not be predicted with given algorithms.

Furthermore, ALPs are also found to be rich in asparagine/lysine/arginine residues at the N-terminus of the transmembrane domain. The cytoplasmic di-lysine motifs shown to be involved in the trafficking of protein to the ER and plasma membrane in earlier studies on eukaryotic cells (Chau et al., 2021; Jackson et al., 2012) were observed at the N-terminus of transmembrane helices in 22 ALPs, and 36 other ALPs had one of the lysine substituted by asparagine, while others also had arginine and aspartic acid (Supplementary Table 3). In ALPs, such motifs might be important for the protein to insert a MAD into the membrane in the required orientation. Interestingly, such motifs were also observed near the C-terminal TM helix, indicating their possible orientation to be on the cytoplasmic side (Supplementary Table 4). The presence of such signals adjacent to TM helices gives information about the possible orientation of helices in cell membranes such as M. smithii’s YP_001272624, which has two TM helices at the N-terminus. The di-lysine motif was found only before the second TM helix, indicating its possible orientation from cytoplasmic to the extracellular side. It is to be noted that such signal sequences were missing in most short helices, suggesting that small helices could help in insertion although not spanning the whole membrane bilayer.

Right-handed beta-helical is the third most represented domain found in ALPs of the two methanogens (Table 2). Methanobrevibacter smithii had 28 ALPs with 43 RBH domains, and M. stadtmanae has 34 ALPs with 52 RBH domains. Many ALPs had repeats of the RBH domain such as YP_001273761 of M. smithii, which had seven repeats.

The RBH domain was initially identified in Erwinia chrysanthemi, a plant pathogenic bacterium as a pectin-binding domain of pectate lyase C (Yoder et al., 1993), and subsequently has been discovered in many other enzymatic proteins such as those involved in hydrolyzing lectins and other carbohydrates (Jenkins and Pickersgill, 2001). The beta-helical rod of this parallel β-helical domain provides a larger groove on its surface for recognizing long carbohydrate molecules (Suits and Boraston, 2013; Villarreal et al., 2022). This is mediated through the chain of conserved asparagines in the loops, which has been suggested to be the most common amino acid in RBH domains (Iengar et al., 2006). Loops mainly have charged and polar amino acids, which explains their ability to bind to long polysaccharides. InterPro entry (IPR039448) indicates that the RBH domain is highly represented in bacteria as compared to other groups of organisms, while in Archaea, it is mostly found in Methanobacteriaceae and Methanosarcinaceae families. Particularly for the former, this could be due to their presence in ALPs, which form a large fraction of the archaeal proteome. Similar structures have also been observed in viruses for the purpose of host attachment and infection (Weigele et al., 2003).

The fold was suggested to have diverged from a common ancestor based on the presence of conserved alpha-helix capping the N-terminus of beta-helix. The cap motif also inhibits oligomeric interactions similar to those found in amyloid formations (Bryan et al., 2011). Furthermore, it is noted that RBH domains in archaeal ALPs are relatively large. The longest RBH domain (YP_447868 of M. stadtmanae) with a length of 1,236 amino acid folds into ~100 Å long RBH domain rod carrying 14 turns. Although rare in eukaryotes, the RBH domain is highly prevalent in surface proteins of bacteria and fungi, with many of them involved in pathogenesis (Bradley et al., 2001).

Archaeal big domains are the most abundant domain repeats in M. smithii and M. stadtmanae ALPs and are found in almost all the ALP sequences. Figure 2 depicts the phylogenetic tree constructed based on the alignment of 279 M. smithii ABDs and 222 M. stadtmanae ABDs with 80 bacterial stalk domains (Monzon et al., 2021). These ABDs seem to have diverged from bacterial stalk domains as most of the archaeal sequences cluster together in a phylogenetic tree and form groups distinct from the bacterial stalk domain sequences. The archaeal big domain (ABD) definitions are not included in the Pfam domain database (Mistry et al., 2020). Pfam either failed to assign domain family to a large part of archaeal ALPs or assigned Big3 (Pfam ID: PF16640) and DUF11 (Pfam ID: PF01345) domains in most cases, generally with high e-value. We also searched the ABD domains in “refseq_genomes (1,711 databases) at NCBI, excluding archaea (NCBI taxid: 2,157)”, and found no blast hit, indicating that these domains may be unique to archaeal species and are not ubiquitously found in other domains of life, suggesting their potential specific role in archaeal symbiosis.

We noticed broad clades of ABD domains in M. smithii and M. stadtmanae, as marked in Figure 2. Most clades contained no sequences of apparent bacterial origin; some ABD clades clustered with at least four bacterial stalk domain sequences (MucBP (A0A806LF85), LVIVD (A0A0S1YA82), and Trp_ring (F9N556) of non-ESET clan and Big6 (A0A150KJ36), Big3_5 (A0A2V7S5F5), Big3 (R5U8D9), and Big2 (A0A0E1X8Y2) of ESET clan) in the phylogenetic tree (Figure 2). These could be the precursors from which archaeal ABDs evolved and subsequently diverged to acquire unique features. Some ABDs were longer as compared to the most common ABD domains, and only a few ALPs had such domains (YP_001272746, YP_001272984, YP_001274106, and YP_001274107 of M. smithii) (Supplementary Figure 2).

We analyzed the frequency and patterns of amino acids present in ABDs and bacterial stalk domains (Figure 3). The comparison clearly shows that although archaeal ABDs have glycine residues conserved similar to bacterial stalk domains, they also acquired unique features with high conservation. The uniquely conserved residues of archaeal ABDs are marked in Figure 3.

An ABD folds into a typical β-sandwich in Greek-key topology with seven strands (Figure 4A). A β-sandwich domain of longer ABDs is formed by nine β-strands (Figure 4B). The conserved glycine residues present in loops are marked in the representative three-dimensional structure obtained from AlphaFold, as shown in Figure 4A. Notably, the conserved residues occur in loops, which may be important for interaction with other protein domains, while the core is conserved with hydrophobic residues. Furthermore, we noticed that some strands in ABD folds have conserved long-chain hydrophobic residues such as valine, isoleucine, phenylalanine, and leucine, while others show conservation of smaller residues such as glycine. The representative structure was searched in the Dali database to identify the closest structural homolog of ABD. It is interesting to note that the root mean squared deviation (r.m.s.d) of ABD of M. smithii (NCBI accession: YP_001272624) with the nearest structure (Big1 domain of bacterial invasion, PDB ID: 1CWV) was only 1.8 Å, while they shared only 10% sequence similarity. Similarly, the nearest structural homolog of another ABD domain of the same ALP belonging to a different clade in the phylogenetic tree (HLA Class-I Histocompatibility antigen, PDB ID: 1EWO, r.m.s.d.: 1.8 Å) was only 18% similar. This indicated sequence divergence from bacterial ancestral homologs while conserving the overall three-dimensional fold of seven β-strands.

Archaeal big domain is found in repeats on ALPs and May be important for extending the range to reach symbiotic microbes. ABD repeats of some ALPs are highly similar; for example, YP_447631 of M. stadtmanae has 27 ABDs belonging to two major clades in a phylogenetic tree, and 20 of them share 60.6% average similarity with each other, while some other ALPs have divergent ABD sequences belonging to other phylogenetic clades; for example, YP_447476 of M. stadtmanae has three ABDs, and all of them cluster with three different clades, indicating early divergence of ABDs from bacteria.

Less common domain types in the ALPs of M. smithii and M. stadtmanae are currently mostly grouped under ‘others’ as being different from the commonly observed MAD, RBH, and ABD domain types. While the overall occurrence of these domain types is low in ALPs of these two methanogens, it is observed that these domains vary and, in some ALPs, multiple ‘others’ can be detected. These can comprise domains that show limited structural similarity to domains such as transglutaminase, pseudomurein-binding protein, PQQ-like domain, and lectin-like domain. These are listed in Table 2, along with the number of occurrences in ALPs of two organisms.

Functional domain annotations using Pfam indicated that M. smithii and M. stadtmanae ALPs are less diverse as compared to bacteria. Only 11 domain families were identified from a sequence-based search in the Pfam database in both species together. On the other hand, in bacteria, altogether 109 types of stalk and adhesive domains are present (Monzon et al., 2021; Monzon and Bateman, 2022). The most common architecture in analyzed archaeal ALPs was RBH domain repeats at the N-terminus, followed by ABD domain repeats. A single transmembrane helix at the N-terminus in the majority of ALPs could act as a membrane-anchoring domain; however, in others, it was present at the C-terminus or at both ends. In addition to the above ALPs, there were 14 sequences in M. smithii and M. stadtmanae, respectively, that had missing RBH and ABD domains, although these were picked in our sequence-based search. These could be partial as they are short and could likely present incomplete domains. These sequences were discarded and not classified as ALPs in this study (Supplementary Tables 5A,B).

The alignment of ALPs with different repetitive structures of varying length may lead to misalignments, potentially introducing errors in phylogenetic inference. However, the detailed characterization of individual ALP domains allows to group different ALPs based on their specific domain architecture and independent of their sequence similarity (Figure 5). As an alternative approach, we have used density-based clustering of text strings that represent the domain architecture, which allows to bin ALPs into different distinct classes. Based on the clustering, we proposed five groups of ALPs in M. smithii and M. stadtmanae. A growing number of ALP groups might be expected in the future as more archaeal species are analyzed for divergent ALPs. The groups proposed here are based on the presence/absence and positions of ABD, RBH, transglutaminase, and other domains, as described further. In general, we observed that ALPs contain at least either the ABD or RBH domain. If the protein sequence did not have any of these domains, we did not classify it as ALP, although these proteins were picked up in our blast searches together with other ALPs. There were nine such sequences in M. smithii and five in M. stadtmanae (Supplementary Tables 5A,B). All ALPs of M. stadtmanae had a MAD only at the N-terminus (except three ALPs with no MAD), while 11 out of 49 M. smithii ALPs had MAD on both termini. Furthermore, five ALPs of M. smithii (NCBI accession: YP_001272625, YP_001272839, YP_001273878, YP_001274107, and YP_001274163) were not fully annotated for domains due to low sequence similarity with AlphaFold structures or partially predicted structures. Thus, they were tentatively assigned ALP groups based on current knowledge of their domains. The following three groups (excluding subgroups) and one currently uncategorized set of “others” ALPs are proposed in M. smithii and M. stadtmanae regardless of the position of the MAD, and they are listed in Supplementary Tables 6A,B.

First, a raw dataset of putative ALPs was constructed by extracting protein FASTA sequences from an annotation-based search using keywords “adhesin-like protein” and “Asn/Thr-rich large protein” in 13 complete Methanobacteriales proteomes (Supplementary Table 7). This search picked 516 putative ALPs. These 516 ALPs were taken to query proteomes specifically of M. smithii and M. stadtmanae with the aim of identifying distantly related ALP sequences using the BLASTP algorithm with an e-value cutoff of 0.0001 (Altschul et al., 1990). All the sequence hits thus obtained were combined in a single FASTA file, separately for both species, and the resulting FASTA file contained 58 and 47 unique sequences belonging to M. smithii and M. stadtmanae, respectively.

Each of the 105 ALPs of both organisms was queried against the AlphaFold database (Jumper et al., 2021; Varadi et al., 2021) via EBI’s “Sequence Similarity Search” tool¹ to identify the closest structure based on sequence homology. In the first round of searches, we used full-length ALP sequences. Overall, 43 of 58 M. smithii ALPs and 32 of 47 M. stadtmanae ALPs could be matched to >90% identity for most of their sequence length, while 11 M. smithii and 13 M. stadtmanae ALPs matched to 40–90% identity. Structures based on the highest matching score were downloaded from the AlphaFold database. The amino acid sequences of the downloaded structures were individually aligned with the corresponding ALP sequence by Clustalx version 2.0 (Larkin et al., 2007) in order to locate the putative structural domains on ALP protein sequences. The remaining ALPs (five of 58 ALPs in M. smithii and eight of 47 ALPs in M. stadtmanae) had sequence identity between 20 and 30%, and thus identified domains for these ALPs were less reliable based on sequence homology to the nearest structures (Supplementary Tables 5A,B).

Since AlphaFold structures have currently a limitation of 2,700 amino acids for proteins, many ALP sequences could be annotated only partially for domains. For the remaining part of ALP sequences (partially annotated sequences that could not be annotated in the first round), we did another iteration of identifying the closest structures by querying only the unannotated part of the ALP sequence and keeping the identity cutoff to 40%. This allowed identifying further domains in ALPs and filled annotation gaps. In few cases, if at least one ABD was identified by AlphaFold, we were able to manually identify other ABD repeats. For example, the two closest AlphaFold structures (AlphaFold IDs: B9ACY6 and R7PVK4) were downloaded for an ALP (Accession: YP_001272746, length: 2879) from M. smithii that matched to >90% sequence identity. Only one ABD was present in B9ACY6, while six ABDs were located based on R7PVK4. However, we could locate another 17 ABD repeat domains similar to the ones located with the help of the nearest AlphaFold structures and extend the domains further at the terminus. Similarly, an AlphaFold structure (AlphaFold ID: B9AFH2) was matched to M. smithii ALP (Accession: YP_001273761) at 78% identity. The structure had only two RBH domains; however, we could locate five more RBH domain repeats on this ALP sequence. We have also marked these manually identified domains in Figure 5 together with other domains.

In rare cases where domains were not identified by AlphaFold, we took annotations from Pfam to improve the domain identifications. InterProScan searches (Blum et al., 2021; Paysan-Lafosse et al., 2022) were carried out for all ALPs. The sequences of domains identified by Pfam were aligned with annotated domains in other ALPs, followed by manually extending the domain boundaries on both sides. Most Pfam predictions agreed with AlphaFold predictions except ABD, which was assigned bacterial Ig-like domains (Pfam IDs: PF16640, PF02369, and PF02368) by Pfam. This could be because the ABD domain definition is missing in the UniProt database. Pfam was also useful for assigning functional annotation for AlphaFold domains in ALPs, as listed in Table 2.

The transmembrane helices were identified by TMHMM² (Hallgren et al., 2022) and confirmed by the presence of helical structures in AlphaFold structures. HeliQuest was used to predict amphipathic helix with analysis window size taken as 18 amino acids (Gautier et al., 2008). In general, hydrophobicity values of approximately 0.5 indicate the possibility of forming an amphipathic helix. The PDB files were displayed using ChimeraX (Pettersen et al., 2021).

All multiple sequence alignments were created by Clustal version 2.0 (Larkin et al., 2007; Thompson et al., 1997) using default parameters. The phylogenetic tree was calculated using the neighbor-joining method (Saitou and Nei, 1987) as implemented in Clustal v2.0, based on the multiple sequence alignment and 1,000 bootstrap trials to confirm the robustness of branches (Felsenstein, 1985). The trees were displayed by iToL v6.8³ (Letunic and Bork, 2021). The sequence logo was created by WebLogo⁴ using a user-defined coloring scheme.

We also downloaded 82 bacterial stalk domain sequences from the Pfam database as per the accessions given in (Monzon et al., 2021) representing 3,542 bacterial fibrillar adhesins. These belong to the ESET clan and others. The pairwise sequence comparisons of 80 stalk domains of bacteria and major clades of ABD domains were carried out using Clustal Omega v1.2.4 (Sievers et al., 2011) in clustering mode. The pairwise distance matrix was calculated and converted into percent identities. The percent identity values were averaged across ABD clades and other groups and mentioned in Table 3 separately for two organisms.

Adhesin-like protein domains were abbreviated by single letter codes (TMD = t, RBH = r, ABD = a, pseudomurein-binding domain = b, transglutaminase = g, OTHER = o), and domain sequences of individual ALPs were transcribed into strings of letters (e.g., sequence TMD-RBH-ABD-ABD-ABD-ABD-ABD = traaaaa). The DBSCAN library in R was used to perform density-based clustering of the string. The analysis was run using default parameters. Clusters were defined by an epsilon (eps) of 0.5 and minimum number of points (minPts) as 1.