To identify new non-opisthokont septin sequences, we utilized both the Joint Genome Institute Phycocosm webpage (https://phycocosm.jgi.doe.gov/) and the NCBI Genome database (https://blast.ncbi.nlm.nih.gov/). We used the initial set of queries consisting of Chlamydomonas, Symbiodinium, and Paramecium septins. These searches identified several septins in the phyla in which they have not been reported. To enhance the chance of finding new sequences in these and other divergent branches, we added Porphyra, Ectocarpus, and Cyanophora to the list of queries and performed additional searches (Table 1; Supplementary Material S1). BLASTP searches were performed on 14 November 2021 using a BLOSUM62 matrix, E-value cutoff of 1 × 10⁻⁵, word size of 3, and filtered low complexity regions. The JGI database searches used proteomes from Excavata, Archeaplastida, Rhizaria, Heterokonta, and Alveolata (Supplementary Material S2). Due to the limited availability of information for ciliate species on JGI, additional searches were performed using the NCBI database, specifically focusing on Alveolata (taxid:33,630) (Supplementary Material S2). Identified sequences were further examined manually for the presence of G-motifs (G1, G3, and G4) and S-motifs (S1-S4) to confirm that they are bona fide septins. Opisthokont septins were selected from (Auxier et al., 2019).

Phylogenetic trees were constructed following the methodology described by (Auxier et al., 2019). A total of 131 opisthokont and 123 non-opisthokont septins were used; as an outgroup, several prokaryotic YihA proteins were also included (Supplementary Material S3). Sequences were first aligned using the constraint-based alignment tool (COBALT) (Papadopoulos and Agarwala, 2007), which incorporates information about protein domains in a progressive multiple alignment. This tool biases the alignment within the septin GTPase domain. To remove regions of randomly similar sequences from the alignment, we employed ALISCORE and ALICUT (Misof and Misof, 2009; Kück et al., 2010; Kueck, 2017). ALISCORE identifies regions of ambiguous alignment, which were subsequently removed using ALICUT. This process resulted in a reduced MSA file containing highly conserved regions within the GTPase domain (Supplementary Material S4), which was then used to generate the phylogenetic tree.

Tree generation was performed using the CIPRES gateway (Miller et al., 2010), employing RAxML-HPC v.8 on XSEDE with the PROTCAT substitution model and the LG protein matrix and a rapid 1,000 bootstrap analysis. The generated trees were visualized using the Rstudio package “ggtree.” Bootstrap values displayed on the trees have been limited to values greater than 25.

For ancestral sequence reconstruction (ASR), we utilized the FASTML server for maximum-likelihood computing of the ancestral states (Ashkenazy et al., 2012). Due to limitations with the FASTML server, we reduced our list of septin sequences from 254 to 200 by removing some sequences from some fungal species and all sequences from the genus Paramecium except for the species tetraurelia. The resulting 200 sequences (Supplementary Material S5) were aligned using COBALT alignment. As ASR provides meaningful interpretation when the entire protein sequence is provided, we did not utilize ALISCORE and ALICUT processing. To generate a new phylogenetic tree, we used the IQTree webserver (http://iqtree.cibiv.univie.ac.at/) with an automatic amino acid replacement matrix, 1,000 ultrafast bootstraps, and all other default parameters (Trifinopoulos et al., 2016; Minh et al., 2020). This tree reproduced the same phylogenetic groupings and general branching patterns as our more rigorous ALISCORE and ALICUT processed tree. Nodes of interest, including parental nodes for the septin phylogenetic groups, opisthokont and protist divide, and the last eukaryotic common ancestor (LECA) node, were defined based on the joint reconstruction output file and labeled in Supplementary Material S5. The protein sequences at these nodes were extracted and referred to as the ancestral septins.

AlphaFold predictions were executed using the Colabfold Google notebook v1.3.0. The specific parameters can be found within the “config.json” file in each respective folder. Due to computational limitations of AlphaFold with extremely long sequences, some sequences required trimming. The objective of trimming was to preserve the entire GTPase domain and the CTE while reducing the sequence length to a manageable size (approximately 800 amino acids). Generally, the protein sequence was truncated from the N-terminal end. Predictions primarily used an MMseqs2 MSA. Five models with three recycles each were generated and the highest-ranking model was selected (Supplementary Material S6). The resulting 3D structures were visualized using ChimeraX. Topology diagrams were drawn in Adobe Illustrator, following the convention used in (Cavini et al., 2021). For AlphaFold predictions of Klebsormidium flaccidum and Ichthyophthirius multifiliis septins, we used version 1.5.2 of the ColabFold notebook. The structures were visualized using ChimeraX and colored according to AlphaFold confidence.

To search for potential polybasic domains in the NTE of our reconstructed ancestral sequences, we developed a Python script that uses a sliding 10-amino-acid window to calculate the local average isoelectric point and plots this value against the first amino acid position across the entire protein length. To focus solely on the NTE, which is where PB1 in extant septins is primarily located, we aligned the ancestral septins to the GTPase domain of S. cerevisiae Cdc3 using CLUSTALω. Only residues before the start of the GTPase domain were plotted. To visualize the multiple sequence alignment (MSA) of the ancestral septins, a CLUSTALω alignment was performed without the Cdc3 GTPase domain to compare the amino acid composition between GTPase domain-adjacent polybasic domains. The MSA was visualized using the R package “ggmsa,” and the amino acids were colored according to their properties.

For high-throughput prediction of amphipathic helices, we developed a Python script that consists of two steps of analysis: (1) secondary structure prediction by s4pred (Moffat and Jones, 2021) followed by (2) amphipathicity assessment of α-helices. In (1), secondary structure prediction was performed for the amino acid sequence of a given septin protein using the run_model.py script provided in https://github.com/psipred/s4pred. In (2), either a “fully-helical” or “partially-helical” segment of an amino-acid sequence was extracted by a sliding 18 amino-acid window. In a “partially-helical” segment, at least six amino acids at both ends of the 18 amino acid window must be fully helical. For example, while a segment with a prediction “HHHHHHCCCCCCHHHHHH” (6x H–6x C–6x H) was permitted, those with “HHHHHCCCCCCHHHHHHH” (5x H–6x C–7x H) were not. We included “partially-helical” segments for further assessment because some membrane-bound Ahs could be predicted as “partially helical,” where two helices are broken apart by non-helical sequence (e.g., Sun2 AH: Lee et al., 2023). For each helical segment, the amphipathicity was calculated and assessed similarly to HeliQuest software (Gautier et al., 2008), but with modifications. First, the mean hydrophobic moment value <µH> was calculated as previously described (Eisenberg et al., 1982) using the hydrophobicity scale values (Fauchere and Pliska, 1983) based on an assumption that all helices rotate with a 100° step. Then, the discriminant factor D = 0.944 x <µH> + 0.33 x z (where z is the net charge) was calculated accordingly to HeliQuest. Finally, the helical segment was considered amphipathic if all of the criteria below were satisfied: i) D > 0.68 OR (<µH> > 0.4 AND z = 0); ii) The hydrophobic face contains at least three consecutive bulky hydrophobic residues (L, V, F, I, W, M, Y) (e.g., a hydrophobic face “SYALLVT” is satisfactory); iii) “Core” of the hydrophobic face does NOT contain any charged residue (“core”: the area of 90° centered around the pole). This search resulted in the identification of 4,809 possible AH domains, with the vast majority showing overlap with one another (Supplementary Material S7).

We then filtered the data to exclude AHs that are positioned inside of an septin GTPase domain. The GTPase domain of Cdc3 from S. cerevisiae was used as a reference to define the start and end residues for the GTPase domain of the other 254 extant sequences. The list of possible AHs of 18 amino acids in length was then screened by excluding those that overlapped with the GTPase domain. Sequences satisfying these criteria were considered to possess an AH (Supplementary Material S8) and were highlighted in a cladogram generated using the R package “ggtree.” To generate helical wheel diagrams, individual AH sequences from the dataset were used as input to run the HeliQuest program (Gautier et al., 2008).

To identify septins with coiled-coil domain and/or putative transmembrane domains in the set of 254 extant septins, we used the existing annotations on the UniProt database (The UniProt Consortium, 2023) release 2023_04. A BLASTP search using our list of 254 septins as query against the UniprotKB database retrieved 206 hits, for which “Coiled coil” and “Transmembrane” annotations were downloaded from the database. According to the UniProt documentation, these annotations are based on the COILS program (Lupas et al., 1991) with a minimum size of 28 amino acids for coiled-coil domains, and TMHMM and Phobius predictions (Krogh et al., 2001; Käll et al., 2004) for transmembrane domains. For the remaining 48 sequences, manual searches for coiled-coil and transmembrane domains were performed using Cocopred (Feng et al., 2022) and Phobius. These predictions are conservative and unlikely to identify all possible coiled-coil and transmembrane domains; for example, the present analysis identified fewer coiled-coil-containing septins than Auxier et al. (2019), which used the hidden-Markov-model-based Marcoil program. Results of these searches are summarized in Supplementary Material S8.

To search for septin sequences outside of opisthokonts, we compiled a small query list of previously identified septin sequences from algal and protist species (Table 1). These sequences were selected based on their evolutionary diversity, aiming to enhance the chance of identifying septins from various taxa. We conducted BLASTP searches using the BLOSUM62 matrix and an E-value cutoff of 1 × 10⁻⁵, utilizing the protein databases available on the Joint Genome Institute’s (JGI) Phycocosm webpage and the Alveolata database on the NCBI BLAST website (see Materials and Methods). These searches revealed previously unreported sequences in multiple taxa under the supergroups Archaeplastida and Chromista (Figure 1). Our searches also reproduced a previous failure to identify any septin sequences in the entire supergroups of Amoebozoa and Excavata (Figure 1; Onishi and Pringle, 2016). At lower phylogenetic levels, septins were also not detected in Viridiplantae (land plants) (Figure 1).

The discovery of new septin sequences in distant branches of eukaryotes raised questions about their phylogenetic relationship with other septins. Previous studies have classified septins into five groups, but these groupings were defined predominantly based on septin sequences within the opisthokont lineage. We thus combined these new non-opisthokont septin sequences with a preexisting list of opisthokont septins (Auxier et al., 2019) and used the resulting 254 sequences to generate a consensus RAxML tree (Supplementary Figure S1) and a simplified cladogram (Figure 2A). Briefly, the 254 sequences and four prokaryotic YihA NTPases (used here as an outgroup; Weirich et al., 2008) were aligned using NCBI’s COBALT alignment tool and processed using ALISCORE and ALICUT to remove ambiguous regions of alignment.

Consistent with results from previous reports (Momany et al., 2001; Kinoshita, 2003a; Pan et al., 2007; Shuman and Momany, 2021), our phylogenetic analysis grouped the opisthokont septins into five distinct clades (Figure 2A; Supplementary Figure S1): Groups 1 and 2 include septins from both animals and fungi, while Groups 3, 4, and 5 represent fungi-specific clades. Although limited sampling of non-opisthokont septins has previously placed some of them in Group 5 (Onishi and Pringle, 2016; Shuman and Momany, 2021), it is now clear that Group 5 septins are distinct from non-opisthokont septins, consistent with the proposal by Yamazaki et al. (2013).

The non-opisthokont septins themselves form three new groups (Groups 6–8) (Figure 2B; Supplementary Figure S2). Group 6 is a monophyletic group of green algal species divided into two subgroups: Group 6A includes some septins that are encoded as a single gene in the genome, in species such as C. reinhardtii and N. bacillaris (Versele and Thorner, 2005; Yamazaki et al., 2013). Group 6B, in contrast, exclusively represents septins that appear to have emerged through gene duplication. For example, of five septins in the green alga Gonium pectorale, only one belongs to Group 6A while the remaining four belong to Group 6B (Figure 2B; Supplementary Figure S2). The genes for these four septins form a cluster in the assembled G. pectorale genome. (Scaffold_65:140,824–165,695), suggesting a very recent gene duplication event. Similarly, of the seven septins in Desmodesmus armatus, five belong to Group 6B (Figure 2B; Supplementary Figure S2). Group 7 is a paraphyletic group composed of septins from various groups of algae, such as additional green algae (e.g., Symbiochloris reticulata), heterokonts (Ectocarpus siliculosus), haptophytes (Chrysochromulina Phaeocystis antarctica), cryptophytes (Crytophyceae sp. CCMP2293), chlorarachniophytes (Bigelowiella natans), and rhodophytes (P. umbilicalis) (Figure 2B; Supplementary Figure S2). Finally, Group 8 is a monophyletic group comprised exclusively of septins from ciliates, except for one highly divergent sequence from the unicellular opisthokont Capsaspora owczarzaki. Within Group 8, septins from Paramecium and Stentor coeruleus formed genus-specific clades, suggesting recent expansion events of septin genes within their lineages (Figure 2B; Supplementary Figure S2).

Several non-opisthokont sequences are currently not classified in Groups 6–8 because their phylogenetic positioning was sensitive to the programs and parameters used (Figure 2A; Supplementary Figure S1). These include sequences from glaucophytes (C. paradoxa), dinoflagellates (S. minutum, Pseudonitzschia multistrata), and coccolithophores and related haptophytes (Emiliania huxleyi, Phaeocystis globosa, Chrysochromulina tobinii, Diacronema lutheri). Curiously, a septin from Fonticula alba, an opisthokont cellular slime mold, also belonged to this orphan group. Additional sampling of sequences from these and related species will likely help improve the confidence in their phylogenetic positioning.

In previous studies, septins from Groups 1–5 were found to have several highly conserved regions in their GTPase domains (Figure 3A) that participate in inter-subunit contacts across the G- and NC-interfaces (Figure 3B, C; Pan et al., 2007; Pan et al., 2007; Auxier et al., 2019; Rosa et al., 2020; Shuman and Momany, 2021). To gain insights into the evolution of these interfaces in septins across the eukaryotic tree, we expanded the alignment to all 254 septins and generated a Weblogo representation for each septin group (Figure 3D). In general, the GTPase-specific motifs (G1, G3, G4), septin-specific motifs (S2, S3, S4) except for the S1 motif (Pan et al., 2007; Nishihama et al., 2011; Onishi and Pringle, 2016; Auxier et al., 2019), and some key residues in the septin-unique element are all well conserved. More specifically, most of the key residues in the five G-interfaces (Gig1-Gig5) are all conserved, except for Gig2 which appears to be variable in Group 8 (Figure 3D). In contrast, key residues in the four NC-interfaces (NCig1-4) are poorly conserved in Groups 6B, 7, and 8. These results suggest that non-opisthokont septins may primarily form homo- or hetero-dimers through the G-interface, and further addition of subunits through NC-interfaces may be limited to Group 6A. In support of this speculation, we found a unique arginine residue that is highly conserved in many Group 6–8 septins but not in Groups 1–5 (Figure 3D); similar “arginine (R-) fingers” are found in other GTPases that form G-dimers (Koenig et al., 2008; Schwefel et al., 2013; see below).

To delve deeper into the evolution of the structural motifs within the septin GTPase domain, we used ancestral sequence reconstruction (ASR) (Ashkenazy et al., 2012) to resurrect ancestral septins. Due to the limitations of the program used, we reconstructed an IQTree of 200 of the 254 septins (Figure 4A; Supplementary Figure S3). The grouping of septin clades and the overall topology of the tree were largely consistent with the RAxML tree (Figure 2). Using this IQTree, ASR prediction was made for several key nodes representing Groups 1–8 and their parental nodes, and then AlphaFold2 (Jumper et al., 2021) was used to predict their 3D structures for the GTPase domain and the C-terminal extension (see Materials and Methods). Perhaps unsurprisingly given the conservation of the extant sequences (Figure 3D), the tertiary structures of the ancestral sequences all appeared similar among themselves and with experimentally determined septin structures (Supplementary Figure S4).

To highlight gains and losses of sub-domain motifs during the evolution of ancestral septins, interpretive topology diagrams of the GTPase domains were generated based on the AlphaFold predictions (Figures 4B, C). This analysis revealed a largely consistent core structure of the GTPase domains consisting of six α-helices (α1-α6) and nine β-sheets (β1-β6 and βa-βc), as well as a few variable α-helices that emerged or were lost at specific ancestral nodes (see below). In addition to the helices and sheets, we identified an arginine residue positioned in the S3 motif of AncGroup 6–8 and LECA septins (Figures 3B, D, 4C). Although this residue is not found in the reconstructed AncGroup 1–5 septins (Figure 4C), some extant Group 5 septins, such as Aspergillus nidulans AspE, appear to have it (see below). Thus, this “R-finger” arginine is an ancestral feature of septin family proteins that has been lost in most opisthokonts. Intriguingly, it has been reported that this R-finger in the single septin of C. reinhardtii is required for its homo-dimerization across the G-interface (Pinto et al., 2017), where it reaches into the GTP-binding pocket of the opposite subunit to accelerate GTP hydrolysis (see Figure 3C, G-interface). Thus, we suspected that the R-finger would invariably be conserved in single septins found in other species. This prediction was partially confirmed: 20 of the 23 single septins that were included in our analysis have an R-finger at the expected position (Figure 5A), suggesting that the dimerization mechanism observed in C. reinhardtii may be ancestral and conserved in many algae and protists. Of the other three that lacked an R-finger, the sequence from the dinoflagellate S. minutum is an extremely large 4484-aa protein, with a septin-like domain near the N-terminus and some additional domains (e.g., SMC domain, HSP70) that are not found in other septins. The other two (from the ciliates Halteria grandinella and Stylonychia lemnae) have the arginine replaced by a histidine residue. It is unknown whether these single septins still form a G-dimer without an R-finger or have taken unique evolutionary paths to function without dimerizing through the G-interface.

Interestingly, in many algae and protists with multiple septin genes, a loss of the R-finger is observed in some of the duplicated genes (Figure 5A). For example, the ciliate I. multifiliis possesses two septins: XP004037107 with an R-finger and XP004027529 without (Figure 5B). Similarly, the filamentous charophyte green alga K. flaccidum has two proteins with and without an R-finger (GAQ92127 and GAQ78635, respectively; Figure 5B). Given the apparent selective pressure against the loss of R-finger in single septins as well as the loss of R-finger in most opisthokont septins that are invariably encoded as multiple copies in a genome (see below), it is tempting to speculate that these septins may have lost their R-finger because of evolution to form hetero-oligomers. Biochemical characterization of these septins is needed to address this possibility.

Unlike the non-opisthokont counterparts, the vast majority of opisthokont septins do not possess an R-finger between the S2-S3 motifs (Figures 3D; Figure 4C). In Group 1–4 septins, the arginine residue is replaced by small uncharged amino acids such as serine, glycine, or alanine. Although there is an invariant histidine residue in the adjacent position (Figure 3D) that could potentially be involved in GTP hydrolysis (Weirich et al., 2008), a mutation to this amino acid in human SEPT2 did not affect its GTPase activity (Sirajuddin et al., 2009). Thus, it is unlikely that the Group 1–4 opisthokont septins employ an R-finger-like molecular mechanism to interact through their G-interfaces. The R-finger is also absent in most filamentous-fungus-specific Group 5 septins (Figures 3D; Figure 4C), consistent with the previous observation that the S1-S4 motifs in septins in these groups are highly variable (Shuman and Momany, 2021). However, some septins, such as A. nidulans AspE (Figure 5A), have an arginine residue located between the divergent S2-S3 motifs. Available data suggest that AspE is not incorporated into canonical septin complexes, although it interacts with them in a developmental-stage-specific manner (Hernandez-Rodriguez et al., 2014). It is interesting to speculate that AspE-type Group 5 septins have retained the ancestral trait to form a homomeric G-dimer using their R-fingers.

Some opisthokont septins that lack the R-finger have lost their activity to hydrolyze GTP by losing a catalytically active threonine (or serine) within the switch I region, making them GTP-bound subunits (Supplementary Figure S5 AB, yeast Cdc3, Cdc11, and human SEPT6; Rosa et al., 2020). We examined some representative non-opisthokont septins to ask if this residue is conserved. In septins with R-finger from C. reinhardtii, V. carteri, and G. pectorale (all Group 6A), this threonine is invariably conserved, consistent with the idea that these septins are active GTPases (Supplementary Figure S5A; Pinto et al., 2017). Interestingly, the other four septins from G. pectorale (Group 6B) all lack both the R-finger and the threonine (Supplementary Figure S5A), suggesting that they may have lost their GTPase activity. Similar concomitant loss of R-finger and catalytic threonine was observed in pairs of septins from K. flaccidum (Group 7) and I. multifiliis (Group 8). In contrast, all septins in P. tetraurelia (Group 8) contain the catalytic threonine, regardless of the presence or absence of their R-finger. These results suggest that sequential loss of R-finger (reduction of GTPase activity) and catalytic threonine in switch I (loss of GTPase activity) may have occurred independently in many (but not all) lineages during septin evolution, and that some non-opisthokonot species may form a septin complex consisting of a mix of GTP- and GDP-bound subunits, like their animal and fungal counterparts. We also observed that in some cases, such as K. flaccidum and I. multfiliis, where the catalytic threonine was lost, AlphaFold prediction positions the Switch-I loop away from the G-interface (Supplementary Figure S5B). It is interesting to speculate that a potential rearrangement of Switch-I may have destabilized G-interface interactions in support of an emergent NC-interface interaction motif.

In addition to the core helices and sheets, AlphaFold predictions of AncGroup 1–5 (opisthokont) septins displayed two additional invariant α-helices, both positioned in the NC-interface: α0 at the junction between the N-terminal extension and the GTPase domain, and α5′ that is positioned in-between α4 and β6 (Figure 4C). Interestingly, however, these helices are not predicted by AlphaFold in AncGroup 6–8 septins (Figure 4C). In the human SEPT2/6/7 complex (and plausibly in many other opisthokont septins complexes), the α0 helix is an integral part of the NC interface where it forms an electrostatic inter-subunit interaction (Cavini et al., 2021). In addition, the α5′-helix contains a polyacidic region that is known to interact with the polybasic region 1 (PB1) within the α0 helix of a neighboring subunit across the NC interface (Figure 3C; Cavini et al., 2021). Thus, it is conceivable that the α0 and α5′ helices evolved together in the opisthokont lineage as the positioning of PB1 was fixed in the former (see below).

The PB1 domain in α0 helix binds to phospholipids such as phosphatidylinositol 4-phosphate, 4,5-bisphosphate, and 3,4,5-triphosphate (Zhang et al., 1999; Casamayor and Snyder, 2003; Bertin et al., 2010; Onishi et al., 2010; Krokowski et al., 2018). The PB1 domain has been observed in some septins in non-opisthokont species such as in C. reinhardtii (Wloga et al., 2008; Nishihama et al., 2011; Pinto et al., 2017) despite the lack of α0 in the same proteins (Figures 3D, 4B), raising the possibility that the emergence of PB1 precedes that of α0. To test this, we examined the NTEs of the reconstructed ASR sequences for the presence of PB1 by developing a Python script that calculates the isoelectric point of a 10 amino-acid window moving along protein sequences. We observed a basic region proximal to the beginning of the GTPase domain in AncGroup 1–5 septins (including in the very short NTE of AncGroup3 septin) (Figures 6A, B), consistent with the presence of PB1 in the majority of extant opisthokont septins (Nishihama et al., 2011; Shuman and Momany, 2021). Similarly, the regions immediately upstream of the G1 motif in AncGroup 6 and 6/7 septins are also highly basic (Figure 6A). In contrast, the NTE of AncGroup8 is overall acidic (Figure 6A), and a few basic residues found in this region are interdigitated by acidic residues (Figure 6B), consistent with the reported ambiguity about the presence of polybasic regions in septins in T. thermophila and P. tetraurelia (Wloga et al., 2008). Interestingly, CLUSTALω alignment identified additional polybasic domains in AncGroup 6B and 6/7 septins at positions 339 and 214 aa upstream of the G1 motif, respectively, which exhibited greater similarity to the proximal PB1 observed in AncGroup 1–5 septins (Figure 6B), and the G1-proximal sequences (PB1′) are non-opisthokont-specific (Figure 6B). Given the low overall sequence conservation of these regions in AncGroup 8 (Figure 6B), it is not clear whether PB1’ is an ancestral feature that has been lost in opisthokont septins, or it was newly inserted adjacent to the G1 motif in the lineage leading to Group 6 and 7 septins. Overall, however, the presence of a polybasic region in the NTE appears to be an ancestral feature that predates the emergence of opisthokont-specific α0.

Some opisthokont septins have the ability to recognize micron-scale membrane curvature through an amphipathic helix (AH) (Bridges et al., 2016; Cannon et al., 2019). Perturbation of these AHs can lead to abnormal subcellular localization of septin proteins (Cannon et al., 2019). To ask if putative membrane-binding AHs are found outside of opisthokonts and therefore can be an ancestral feature of septins, we developed a high-throughput pipeline to identify AH domains in a large number of polypeptide sequences by predicting alpha helices and then calculating their amphipathicity (see Materials and Methods), and applied it to the NTE and C-terminal extension (CTE) of our eukaryotic septin collection. This pipeline precisely identified previously reported AH domains in fungal and animal septins (Cannon et al., 2019; Lobato-Márquez et al., 2021; Woods et al., 2021), such as Cdc12 and Shs1 in S. cerevisiae and Ashbya gossypii, human SEPT6, Caenorhabditis elegans UNC-61, and Drosophila melanogaster Sep1 (Supplementary Figure S6). In some cases, multiple AHs were found in a single septin. These additional AHs could potentially be a result of the inherent amphipathicity of coiled-coil domains (where many AH domains reside). We limited our downstream analysis to hits with the largest calculated D-factor to focus on putative membrane-binding AHs. Our analysis revealed the presence of predicted AHs in septin sequences spanning all Groups (Figure 7A; Table 2) with varying levels of conservation. In opisthokonts, for instance, predicted AHs were detected in 68% of Group 2 and Group 4 sequences, while only 13% of Group 3 sequences exhibited AHs. In Group 1, there is a striking difference between the two subclades: a predicted AH is completely absent in 1A (animals and fungi), while it is found in 75% of septins in 1B (animal-specific). This suggests a potential connection between the evolution of AHs and the positioning of subunits within a canonical octameric protomer, in which 1A subunits occupy the central dimer. Like Group 3, only a small fraction of Group 5 septins (22%) have predicted AHs; unlike Group 1, there is no specific subgroup in which AHs are conserved, suggesting sporadic loss/gain of the domain within this group (Figure 7A; Table 2). In general, the AHs in Groups 1–5 displayed features consistent with stereotypical amphipathicity, with a large hydrophobic window and a hydrophilic face composed of both positively and negatively charged residues (Supplementary Figure S6).

The wide distribution of AHs is also observed in all non-opisthokont groups (Figure 7A; Table 2). Group 6A, consisting largely of single septins, has the highest rate of AH domains at 68%. In Group 6B, septins with predicted AHs were found in most subclades, with a total preservation rate of 50%. In Groups 7 and 8, septins with predicted AHs were found in 38% and 19%, respectively. In the Heliquest visualization, both AHs present in Group 6B and Group 7 exhibited hydrophilic faces primarily composed of positively charged residues interspersed with small polar residues such as serines and threonines (Supplementary Figure S6). In some instances, weaker amphipathic helices were observed, as exemplified by P. umbilicalis 6581, which lacked a strongly pronounced hydrophilic face but still fulfilled the criteria of our search because of their high net charges that raised the D-factor (Figure 7E; Supplementary Figure S6). Some Group 6A and Group 8 septins have predicted AHs similar to those observed in Groups 1–5 with a large hydrophobic window opposite the cluster of both positively and negatively charged residues.

Many animal and fungal septins contain a coiled-coil (CC) motif in the CTE which is thought to be involved in polymer stabilization and the formation of bundles and filament pairs (Sirajuddin et al., 2007; Bertin et al., 2010; Cavini et al., 2021). We utilized the existing annotation of CC domains in the Uniprot database to identify them in our list of 254 extant septins. Interestingly, we observed the presence of CCs in Groups 1B, 3, 4, and 6B (Figure 7B; Table 2). The majority of these sequences were also positive for AH domains (Figure 7D), with AH domains residing within CC domains in many cases, such as in S. cerevisiae Cdc12 (Figure 7E; Cannon et al., 2019). Interestingly, CC domains were almost entirely excluded from non-opisthokont Groups 6A, 7, and 8 (Figure 7B; Table 2), suggesting that the CC domains observed in Group 6B were a result of convergent molecular evolution. It is interesting to speculate that septin gene duplication in some green algae (Figure 5A) and the formation of heterooligomeric complexes may have led to the emergence of lateral pairing between septin subunits.

Lastly, it has previously been reported that some non-opisthokont septins possess putative transmembrane (TM) domains or short hydrophobic patches (Wloga et al., 2008; Nishihama et al., 2011). Thus, we searched for the presence of potential TM domains in our list of 254 extant septin sequences. Except for one sequence from the parasitic fungus Catenaria anguillulae (A0A1Y2I4M7, Group 2A, 46% identical to S. cerevisiae Cdc3) that has a unique N-terminal TM domain, all septins with a TM domain were found in the non-opisthokont lineages, with notable enrichment in Groups 6A and 7 (Figures 7C, E; Table 2). This distribution of TM domains in our dataset seems to suggest that they emerged early in the non-opisthokont branch after its split with opisthokonts and were subsequently lost in many species in Group 6B and 8. [See, however, Discussion for another possibility given a recent report by (Perry et al., 2023).] It is interesting to note that there is little overlap between the distributions of CC and TM domains in Group 6 septins (Figure 7D), perhaps suggesting that the evolution of septin-septin interactions through CC domains necessitated a concomitant loss of TM that would otherwise restrict the accessibility of CTE.

In summary, our searches for α-helix-based structures that are often associated with septin CTE suggest that the AH and TM domains may have ancient origins in septin evolution, while the CC domain may have evolved independently in multiple lineages.