With the first description of a septin structure (Sirajuddin et al., 2007), the critical role of the G-interface for the molecular assembly of the core particle became evident. As a result of the dozens of structures determined subsequently, the nature of the interactions involved at the interface has become clear and the exquisite manner by which stability, specificity and exchangeability arise have begun to be understood (Brognara et al., 2019; Rosa et al., 2020). In general, the G-interface is extremely well conserved in different structures, preserving almost perfectly the relative positions of the two monomers involved. Furthermore, no crystal structure presenting an intact G-interface has been observed in the absence of bound nucleotide. This suggests the ligand to be an integral component of the interface and/or that substrate hydrolysis through a GAP-like mechanism depends on dimer formation (Gasper et al., 2009; Sirajuddin et al., 2009; Zent and Wittinghofer, 2014). This section will describe the surface regions on the monomers which participate in the G-interface, highlighting the structural differences between groups and specifying the details which contribute to correct filament assembly.

The standard model for the human hexamer implies the existence of G-interfaces between SEPT2 and SEPT6 and between two copies of SEPT7 (Figure 1B). In the octamer, the latter is replaced by a SEPT7-SEPT3 interface. At these interfaces, up to seven contact regions within the G-domain of each monomer form a diamond-like shape (Figure 6). At the top of the diamond, the two switch II regions interact, as shown in Figure 4. The central region, where the contact surface is widest, involves Tr1, the P-loop and G4. This also includes the switch I region in the case of SEPT2-and SEPT6-group member interfaces. At the bottom, the β-meander participates in interactions with Tr2 of the corresponding partner. Even at promiscuous interfaces, many of these features are preserved, including those involved in nucleotide binding. Although some subtle variation has been observed within the interface core (e.g., mutation to Glu(P-loop) in the SEPT3 group and in Drosophila Pnut/SEPT7 or to the β-meander in the SEPT3 group), it is the region at the rim of the interface, particularly switch I, which presents the greatest variability between septin groups (Rosa et al., 2020).

How do the correct pairings of septins arise at the different G-interfaces of the core particle during spontaneous assembly? This is a question relating to the interactions involved in molecular recognition. Switch I makes a significant contribution to the contact area at the G-interface between septins of the SEPT2 and SEPT6 groups (but not those between SEPT7 and SEPT3) (Rosa et al., 2020). This raises the possibility that the selectivity that drives members of these two groups together may be related to selective interactions involving group-specific residues found within the switch I regions. Such interactions are shown in Figure 7 and Supplementary Figure S2.

An interaction network involving group-specific (or characteristic) residues is observed between the Tr1 of SEPT2 and the P-loop/Switch I of SEPT6/8/11 (Figure 7). The hydroxyl of a characteristic threonine residue in the SEPT6 group, Thr(P-loop/SEPT6), interacts with the γ-phosphate of GTP via a water molecule, thereby orienting its methyl group towards His(Sw1/SEPT6). This allows the correct orientation of the imidazole ring of His(Sw1/SEPT6) to accept a hydrogen bond from the main chain of SEPT2. His(Sw1/SEPT6) fits snugly into a pocket formed by Tr1 of SEPT2, whose conformation is determined by the characteristic residues Phe(Tr1/SEPT2) and GlyII(Tr1/SEPT2). Compared with septins from other groups, the trans-loop 1 of SEPT2 adopts a more extended conformation (Supplementary Figure S2A; Rosa et al., 2020). Both the ordering of switch I in SEPT6/8/11 and the conformation of trans-loop 1 in SEPT2 are necessary for correct partner pairing at the interface. No other group combination would provide the appropriate structural features for the fit schematically represented in Figure 7. This can be verified by observing the disordered and/or incomplete switch I regions, present at promiscuous G-interfaces such as the homodimer of SEPT2, SmSEPT10 and SEPT3 (Sirajuddin et al., 2007; Zent et al., 2011; Macedo et al., 2013; Zeraik et al., 2014).

The involvement of GTP in the interaction network (Figure 7) suggests a functional significance for the lack of catalytic activity in the SEPT6 group. The persisting γ-phosphate should be considered a characteristic feature of the SEPT6 group as it aids in correctly orienting Thr(P-loop/SEPT6) towards His(Sw1/SEPT6). In the remaining groups of septins, this threonine is replaced by a serine, incapable of forming the hydrophobic contact with His(Sw1/SEPT6).

On the other side of the G-interface, in the switch I region of SEPT2, a characteristic amino acid Ala(Sw1/SEPT2), fits into a cavity in SEPT6/8/11 (Figure 7 and Supplementary Figure S2B). The presence of amino acids with longer side chains at this position in the SEPT3 group, or a proline in SEPT7 or even its absence altogether in the SEPT6 group (Supplementary Table S1) suggests that this interaction would only be viable in the case of the SEPT2 group (Rosa et al., 2020). Ala(Sw1/SEPT2) interacts via Van der Waals contacts with Ser(Tr1/SEPT6) which forms a hydrogen bond with the main chain of an aspartic acid in SEPT2. This interaction is only possible for the SEPT6 group where Ser(Tr1/SEPT6) is always either serine or threonine. This allows for the correct orientation of switch I of SEPT2, permitting Ala(Sw1/SEPT2) to fit into its complementary pocket in SEPT6.

In summary, the presence of characteristic residues converges on the correct structural organization of switch I and the trans-loop 1 of the SEPT6 and SEPT2 groups, favoring the formation of the specific G-interface. Since the interactions involved, by definition, can be generated by any member of the groups involved, this provides a molecular basis for the understanding of Kinoshita’s rule (substitutability between members within a group). As a “side effect”, we begin to understand the reason why the SEPT6 group lacks catalytic activity. By retaining the γ-phosphate, a network of interactions can form, guaranteeing the correct pairing of members of these two groups at the G-interface.

The crystal structures of homodimers and heterodimers have made it possible to demonstrate that the complete ordering of switch II is related to the formation of physiological G-interfaces (Zent et al., 2011; Zeraik et al., 2014; Brognara et al., 2019; Castro et al., 2020; Rosa et al., 2020). The structure which results from the pairing of the two switch II regions is shown in Figure 4. In general, this pairing does not arise at promiscuous G-interfaces, which frequently appear in crystal structures of isolated G-domains of a single septin (Supplementary Figure S3).

By comparing the GTP- and GDP-bound forms of the Schistosoma septin, SmSEPT10, a mechanism, controlled by nucleotide hydrolysis, was described for β-strand slippage (Zeraik et al., 2014). When bound to GTP, the β3-strand is in the “non-slipped” state and switch II is partially ordered (forming the type II β-turns described in Section 5.1.2). However, in the GDP-bound state, β3 is slipped towards the G-interface by two residues causing a dramatic rearrangement of the hydrogen bonding within the sheet. As a consequence of slippage, switch II becomes disordered and the G-interface is partially destroyed (Figure 8A). This phenomenon was also described for SEPT2 (Valadares et al., 2017) and, in both septins, the homotypic G-interfaces are promiscuous (i.e., not predicted by the canonical model shown in Figure 1B). β-strand slippage was initially proposed as a mechanism for transmitting information from the G-interface to the NC-interface, forcing the α0 helix to change conformation (see Section 9 and Section 10.1) (Valadares et al., 2017).

The accumulation of many high-resolution structures for known physiological G-interfaces, including SEPT2-SEPT6/8/11, SEPT7-SEPT3_T282Y and SEPT7 alone (Brognara et al., 2019; Rosa et al., 2020), has forced this idea to be revised. In all these cases (as well as the homotypic interfaces formed by SEPT3-group members) the two switches are very well ordered (Supplementary Figure S3), forming a conserved wide-type β-bridge (Figure 4, Figure 8B) and the β3-strand remains unslipped. The available data suggests that the presence of the β-bridge is a consistent and necessary feature of a physiological G-interface and that slippage only occurs at promiscuous interfaces. Indeed, it has been suggested that slippage may be the result of “negative design” during evolution to disfavour promiscuous interfaces from forming in vivo (Brognara et al., 2019). In this sense, it is interesting that the SEPT3 group also presents a well ordered β-bridge and may be indicative of alternative subunit arrangements including homopolymers (Nakos et al., 2019b). On the other hand, the absence of β-strand slippage in septins of the SEPT3 group may instead be an artefact related to the presence of Mg²⁺ in the structure bound to GDP, which may aid in holding the strand in the non-slipped position (Castro et al., 2020).

The intermolecular β-bridge is stabilized by main chain hydrogen bonds, which place asparagine or aspartic acid residues (Asx(Sw2)) paired across the interface (Figure 8B). The main chain torsion angles observed for Asx(Sw2) are unusual and rarely adopted by other amino acids (Hovmöller et al., 2002), which suggests the β-bridge to be a unique structural motif in septins, differentiating them from other small GTPases. β-turns before and after the β-bridge aid in its correct orientation (Brognara et al., 2019; Rosa et al., 2020). SEPT3 and SEPT7 have an asparagine (Asn(Sw2)) immediately after Asx(Sw2), whose side chain forms hydrogen bonds in the homo- (SEPT3-3, SEPT7-7) and heterodimers (SEPT3-7), further stabilizing the structure (Figure 8B). Heterodimers of SEPT2 with SEPT6/8/11, on the other hand, present a cysteine (partially conserved for the SEPT2 group) or a lysine (conserved in the SEPT6 group) at this position (Figure 8B). This consequently eliminates the interactions made by the asparagines and suggests an alternative mechanism for stabilizing the structure in these cases. Rosa et al. (2020) have speculated that the side chains of the lysine and the cysteine, which face one another under the β-bridge, could potentially form a rare Lys-Cys covalent bond described for the first time only recently (Ruszkowski and Dauter, 2016; Wang, 2019; Wensien et al., 2021). Further investigation of such a bond forming in vivo and its potential relevance in stabilizing septin core particles is clearly necessary. Particularly interesting is the fact that the cysteine residue is not conserved in all human SEPT2-group members (only in SEPT2 itself and SEPT1) and this may provide a means to fine-tune inter-subunit affinities even within the paradigm imposed by Kinoshita’s rule.

The septin N-terminal domain is the least studied in terms of structure and the most variable region amongst all septins. This domain contains a structured α-helix (α0, preceding the G-domain), which often contains a polybasic basic region (PB1) believed to be crucial for membrane interaction (Zhang et al., 1999). However, the greater part of the septin N-domain, upstream to this helix, is intrinsically unstructured (Garcia et al., 2006). Some septins with long N-terminal prolongations (such as human SEPT4 and SEPT9) can be expressed as numerous alternatively spliced isoforms, giving rise to multiple possibilities and functional variation in the different tissues where they are present. Amongst mammalian septins, for example, SEPT9 possesses more than 30 different isoforms (Connolly et al., 2014; Zuvanov et al., 2019), and many functions have been shown to be isoform-specific (Estey et al., 2010; Connolly et al., 2011; Kim et al., 2011).

A description of the N-domain can be made with reference to the characteristics of the conserved regions observed in mammalian septin sequences and highlighted in the alignment given in Figure 9A. Conveniently, this can be divided into two modules. Firstly, the unstructured (variable) region (IDR), including the prolongations observed for some SEPT4 and SEPT9 isoforms, whose long N-termini may have analogues in other species (Cdc3 in yeast or Pnut in Drosophila), along with some modulatory motifs within the downstream proline-rich region. Secondly, the structured (conserved) region consists of two components: 1) a domain-swapped loop (the “hook-loop”) important for NC-interface stabilization and 2) the α0 helix including PB1, the most conserved, structured and functionally characterized region of the N-domain.

The N-domains observed in the longest isoforms of SEPT9 and SEPT4 possess specific motifs attributed to interacting with cytoskeletal proteins such as actin and microtubules (Bai et al., 2013; Smith et al., 2015; Verdier-Pinard et al., 2017). In particular, attempts have been made to divide the long domain found in SEPT9 into two distinct regions based on amino acid content. The first half is a basic domain containing a cytoskeletal binding region (CBR) involved in cytoskeletal protein recognition. The second half is more acidic and includes a proline-rich motif together with the structured (conserved) region (Supplementary Figure S4). The prolongations observed in these septins have also been shown to directly mediate interactions with proteins associated with other functions, such as vesicle trafficing (dynactin) (Kesisova et al., 2021) and signalling pathways (CIF15, SA-RhoGEF) (Nagata and Inagaki, 2005; Diesenberg et al., 2015). For a more detailed review on interactions involving the N-domain see Spiliotis and Nakos (2021).

The proline-rich region, which follows the basic CBR, has also been credited with binding to different protein partners, thereby modulating protein interactions and functions. Some septin isoforms lacking the CBR present enhanced affinity towards signaling factors containing SH3 domains known to recognize proline-rich motifs. Many modulatory motifs and PTM sites have also been described within this region (such as acetylation, phosphorylation and SUMOylation motifs) (Van Damme et al., 2012; Zhou et al., 2013; Ribet et al., 2017). These modules could act as switches controlling and altering the effects of their flanking regions such as the CBR itself.

Shorter isoforms lacking the unstructured region were also shown to lose interaction specificities, and to increase binding to non-canonical paralogs and/or promiscuous partners, thereby, expanding their ordinary interactome (Devlin et al., 2021). It has been suggested that the presence of these proline-rich motifs (together with charged residues) (Supplementary Figure S4) might fine-tune NC-interface interactions and further restrict abnormal contacts and unusual filament assembly (Kim et al., 2012; Weems and McMurray, 2017; Jiao et al., 2020; Soroor et al., 2021).

Coiled-coil (CC) sequences are found in the C-terminal domain of most septins, with some exceptions (e.g., members of the human SEPT3 group and the yeast septin Cdc10). In humans, the length of the coiled coils varies among the different groups: in the SEPT2 group, it comprises around 30 residues, whilst in the SEPT6 and SEPT7 groups, it is approximately twice as long (Figure 10A; Leonardo et al., 2021). In yeast, there is a lack of direct structural information available and the output from prediction tools together with the information/assumptions adopted by different authors (Versele et al., 2004; Barth et al., 2008; Meseroll et al., 2013; Finnigan et al., 2015; Mela and Momany, 2019; Taveneau et al., 2020) means that it is unclear whether yeast septins share similar coiled-coil lengths or not (Supplementary Figure S5).

Apart from the coiled coil itself, there are potentially two flanking regions in the C-domain: C_N, the region between the final helix of the G-domain (α6) and the coiled coil, and C_C, the region subsequent to the coiled coil (Figure 10A). The C_N region is highly variable and believed to be quite flexible. It appears to act as a hinge, allowing the coiled coil to move with respect to the G-domain (Sirajuddin et al., 2007; Mendonça et al., 2021). This region in Shs1 (more specifically residues 350–445, Supplementary Figure S5) has been implicated in stabilizing octamers (Taveneau et al., 2020). Similarly, the region flanking the C-terminus of the coiled-coil (C_C) is believed to be mostly unstructured. In the SEPT6 and SEPT7 groups, at the very end of the C-terminal domain, a polybasic sequence is present (K/RK/RDKxK/RKN/K and EKNKKKGK, respectively). This motif, together with other polybasic domains in septins (PB1 and PB2), could assist in membrane interaction. Although still under debate, some studies indicate that specific C-domains (or parts of them) do indeed have a role in membrane association (Zeraik et al., 2016; Cannon et al., 2019; Jiao et al., 2020; Woods et al., 2021).

Coiled coils are present in a wide variety of proteins and can be described as super-helical assemblies of two or more α-helices coiled together (Lupas and Bassler, 2017; Woolfson, 2017). One of the main roles of coiled coils is to promote protein oligomerization. The hallmark of dimeric coiled-coil sequences is the presence of heptads, repetitions of seven amino acid residues dubbed a-b-c-d-e-f-g. Positions a and d are occupied mostly by hydrophobic residues, frequently leucine or isoleucine. These two positions form the hydrophobic core at the interface of the coiled coil and they interact with neighboring residues by “knobs-into-holes” contacts (Crick, 1953). In parallel dimeric coiled coils, the core interactions are a-a and d-d, creating mixed strips including both a and d side chains on both sides of the coiled-coil interface (Figure 10C and Supplementary Figure S6A). In antiparallel dimeric coiled coils, however, the contacts are a-d (i.e., an a residue from one helix paired with a d residue from the other). This places all side chains from the a residues on one side of the coiled-coil interface (and therefore the side chains from d residues on the other) (Figure 10D and Supplementary Figure S6B).

Other positions (b, c, e, f, g) are more exposed and are usually occupied by hydrophilic residues. Since two turns of a standard helix do not exactly match the heptad length, the helices pack into a left-handed super-coiled structure. However, insertion of non-canonical repeats with lengths other than seven may modify the packing angle between the helices (Brown et al., 1996; Gruber and Lupas, 2003). An insertion of four residues (called a stutter), for example, creates a block of 11 residues (7 + 4) and leads to the unwinding of the left-handed supercoil (Brown et al., 1996; Gruber and Lupas, 2003).

Not visible in the structures of human septin oligomers due to the flexibility of C_N (Sirajuddin et al., 2007; Mendonça et al., 2021), the C-domains are expected to participate in two different types of coiled coil along the filament: the SEPT2 homodimer and the SEPT6-SEPT7 heterodimer (in yeast, the Cdc11 homodimer and the Cdc12-Cdc3 heterodimer, Figure 1B), both at NC-interfaces. Given the directions of the final helices (α6) of the respective G-domains, and how they project perpendicular to the main filament axis, septin coiled coils were inferred to be parallel. FRET experiments show that the C-domains of SEPT6 and SEPT7 do indeed form heterodimeric parallel coiled coils (Low and Macara, 2006), even though they are capable of also assembling into homodimers (Almeida Marques et al., 2012; Sala et al., 2016). Circular dichroism data showed that the heterodimer is more stable than the homodimers (Almeida Marques et al., 2012; Sala et al., 2016), presumably due to unfavorable like-charge repulsion at a positions in the latter (Sala et al., 2016). No detailed structural evidence is currently available for the heterodimeric coiled coil, but it is expected that it would aid in guiding the correct assembly of the core particles (Almeida Marques et al., 2012; Meseroll et al., 2013; Sala et al., 2016). Concerning the C-terminal domain of SEPT2, it has been shown that its cleavage by Zika virus NS2B-NS3 protease is associated with mitotic defects in neural progenitor cells (Li et al., 2019), emphasizing the importance of the domain presumably in correct filament assembly.

Recently, the first structures of septin coiled coils were solved by X-ray diffraction. These were five homodimeric coiled coils of human SEPT1, SEPT4, SEPT5 (from the SEPT2 group), and SEPT6 and SEPT8 (from the SEPT6 group) (Leonardo et al., 2021). In the SEPT2-group structures, while SEPT5CC is a conventional parallel coiled coil with hydrophobic residues in a and d (Figures 10C,E), SEPT1CC and SEPT4CC (Figure 10F) are antiparallel (Figure 10D) and use a different contact interface which only partially overlaps with that observed for SEPT5CC. It has been suggested that this implies that the sequences are orientationally ambiguous (Leonardo et al., 2021). The residues which are common to the interface in both arrangements are shown by an arrow in Figure 10B and a similar pattern appears to be present in yeast septins. Additionally, the residues in a, which occupy e positions in the parallel form, are all hydrophilic (Figure 10B, note that the sequence register must be altered in order to preserve the standard definitions for the heptad positions). These establish hydrophilic contacts down the a-side of the interface, forming a chain of hydrogen bonds which interleaves acidic residues with glutamines thereby avoiding like-charge repulsion (Figure 10F) (Leonardo et al., 2021).

The two coiled-coil structures of the SEPT6 group (SEPT6, Figure 10G, and SEPT8) are also antiparallel. Both structures are very similar and, essentially, the same region forms the coiled coil, which confirms the intrinsic disorder of the flanking C_N and C_C regions since, for the SEPT8 construct, the entire C-domain was crystallized (Leonardo et al., 2021). Here, a positions are populated by lysines and glutamates. Since these residues form a strip down the a-side of the interface, this results in a chain of potential inter-helical salt bridges which would stabilize the dimer (Figure 10G). One particularity of these two structures (compared to the antiparallel structures of SEPT1CC and SEPT4CC) is the presence of a conserved stutter in the sequences of all SEPT6-group members (FE/DxL↓KxxH/Q, where the arrow indicates the break in the heptad register). The stutter decreases the supercoiling, leading to a structure in which the helices effectively lie side-by-side. This lack of supercoiling effectively maintains the side chains of equivalent register positions (a or d) on the same side of the interface along the entire length of the coiled coil (Figure 10G).

It has been suggested that the two orientations for the coiled coils could be metastable structures (Leonardo et al., 2021) and modulated by the chemical microenvironment in which the coiled coil is embedded. NMR studies with coiled-coil peptides from the SEPT2 group in aqueous solution show that they have a tendency to be parallel with a conventional hydrophobic interface (Leonardo et al., 2021). On the other hand, when antiparallel, the septin coiled coil buries hydrophilic residues inside the interface. This would only be expected to happen in an environment of low dielectric constant where solvent has been largely excluded, similar to that found inside crystals.

Although apparently important for the stabilization of the NC-interface in the parallel orientation, the C-domains are not required for polymerization (Sirajuddin et al., 2007; Szuba et al., 2021), indicative of an additional role. They have also been associated with the formation of paired filaments, gauzes and stacked filament structures (Bertin et al., 2008; Bertin et al., 2010; Jiao et al., 2020; Szuba et al., 2021). Two kinds of spacing between filaments have been reported in vitro both in yeast and in mammalian septins: tight (∼5 nm) (Bertin et al., 2010; Jiao et al., 2020; Leonardo et al., 2021; Szuba et al., 2021) and loose (15–20 nm) (Frazier et al., 1998; Versele et al., 2004; Bertin et al., 2008, 2010; Leonardo et al., 2021). The former resembles the lengths of the coiled coils of the SEPT2-group members (4–5 nm). The wider spacing is compatible with the coiled coils of the SEPT6-and SEPT7-group members (8–11 nm) when taking into account contributions from unstructured parts of the C-domain (C_N and C_C) (Leonardo et al., 2021). The former has also been proposed to arise from lateral contacts between G-domains (Szuba et al., 2021). However, in yeast, deleting the C-domain of Cdc11 or Cdc3/Cdc12 eliminated tighly- and loosely-paired filaments, respectively, supporting the involvement of specific C-domains in maintaining each type of spacing (Bertin et al., 2010).

A recent model attempts to relate the structural data on the predominantly antiparallel coiled coils to their potential role in mediating filament cross-bridging (Figures 11A,B; Leonardo et al., 2021). Previously, based on experiments with yeast septins, the formation of four-helix-bundles interconnecting filaments had been proposed (Bertin et al., 2008). However, evidence for tetrameric coiled coils has not been forthcoming (Almeida Marques et al., 2012; Leonardo et al., 2021). Additionally, the use of equimolar mixtures of the coiled-coil peptides from SEPT6 and SEPT7 in crystallization assays—expecting the crystallization of the heterodimer—yielded only antiparallel SEPT6 homodimers (Leonardo et al., 2021). The model shown in Figure 11 is a proposal for how both tight and loose spacings could arise by simple antiparallel pairings, which does not require the appearance of four-helix bundles. Although the details remain unknown, it is likely that the hydrophilic face (a positions) of these coiled coils would have to be solvent-protected to be stable, for instance by facing the membrane, where the dielectric constant is known to be lower (Figure 11C).

Although the NC-interface between SEPT6-7 was originally reported in the crystal structure deposited in 2007 (PDB:2QAG; Sirajuddin et al., 2007), the low resolution of the data at the time precluded its full description. The cryo-EM structure (PDB:7M6J; Mendonça et al., 2021), taken together with high-resolution crystal structures of its components, reveals that the NC-interface can be divided into two regions: the upper part, where salt bridges are formed by residues from the α6 helix and the loop following α2, and the lower part, formed mainly by contacts made by helix α0. The two regions are connected by one face of helix α6 (Figure 12). The upper part of the interface has been extensively described previously (Valadares et al., 2017; Castro et al., 2020). It involves inter-subunit electrostatic interactions made between, Glu(α2) and Arg(PB2), Glu(α6) and Arg(PB2) and Arg(α6) with the helical dipole of helix α2, according to our simplified nomenclature (Figure 12; Sirajuddin et al., 2007; Macedo et al., 2013; Valadares et al., 2017).

Helix α0 of SEPT7 contains a genuine polybasic region (PB1), composed of seven residues, including four positive charges. The same region in SEPT6 contains only one (Figure 9A). In SEPT7, PB1 can conveniently be divided into a proximal region and a distal region, each containing two basic residues. Both regions form contacts with SEPT6 across the interface; the proximal region with helices α5 and α6 (including the elbow which connects them) and the distal region with the PAR.

One feature revealed by the cryo-EM structure is a large cavity at the SEPT6-7 NC-interface (Figure 12), whose perimeter is defined by the upper and lower contact regions together with α6. The bottom of the cavity is limited by a platform formed by the two α0 helices, held in place by the interactions described above, together with a phenylalanine anchor (Phe(α0)) at the boundary between α0 and the G-domain. It is interesting to note that the α0 helix is therefore anchored by two well-conserved phenylalanines at either end; Phe(HL) from the hook-loop (Section 7.4.1) and Phe(α0) at the interface with the G-domain. The cavity has no known function but is necessary for the monomers to be able to slide with respect to one another, although it is still unknown if this shifting is a general phenomenon which applies to all NC-interfaces or if it is restricted to the SEPT3 group alone (see Section 9.3 and Section 10.3). Other possible roles for the cavity cannot be eliminated, for example, in lipid binding during membrane association.

The homotypic SEPT2-SEPT2 NC-interface was partially described by the 3.4 Å structure of SEPT2 lacking the C-domain, reported in 2007 (PDB:2QA5; Sirajuddin et al., 2007). It is the main determinant of septin filament polymerization and its exposure at the ends of the oligomers permits the formation of filaments including a mixture of octameric and hexameric core particles. In this crystal structure, the upper part of the interface is similar to that described above. Indeed, the network of salt bridges involving charged residues from the α6 helix and the loop following α2 is a general feature of homologues in general, suggesting this to be a constant feature of all NC-interfaces.

In order to better understand the totality of the interface, we generated a SEPT2α0 model based on the highest resolution structure available (PDB:6UPQ) to which the α0 helix from PDB:2QA5 had been grafted. The SEPT2α0 NC-interface is in the open conformation in this model with α0 buried, as described for the case of SEPT6-7.

Like SEPT7, SEPT2 possesses a genuine PB1 with four positively charged residues divided into proximal and distal parts. Based on these observations, it has been suggested that similar interactions to those described for SEPT6-7 would be expected to participate across the interface (Mendonça et al., 2021). It appears clear from the model that interactions between PB1 and the PAR are to be anticipated. However, the SEPT2 NC-interface is now known to be the weakest link along the filament (at least in terms of its susceptibility to salt concentration) and molecular simulations appear to justify its fragility (Mendonça et al., 2019). In contrast, in terms of contact area and estimated free energy, the SEPT6-7 NC-interface (PDB: 7M6J) is estimated by PISA (Krissinel and Henrick, 2007) to be more stable than the SEPT2-2 NC-interface (2,149.8 Å² and -16.6 kcal/mol, respectively, compared to 1,636.3 Å² and −6.5 kcal/mol). Taken together, it is therefore possible to rationalize the rupture of filaments preferentially at homotypic SEPT2 interfaces at high ionic strength.

In all crystal structures presenting an NC-interface, the same canonical “open” conformation is observed, except for the SEPT3 group (Sirajuddin et al., 2007; Macedo et al., 2013; Castro et al., 2020). In these, the homotypic NC-interface is physiological and occupies the center of the octamer (Mendonça et al., 2019; Soroor et al., 2021). Its plasticity results in at least three different conformations: open, closed, and shifted (Macedo et al., 2013; Castro et al., 2020). The crystal structures of the SEPT3 G-domain (PDB:4Z51 and PDB:4Z54) complexed to either nucleotide are found to be in the closed form. For the G-domain of SEPT12 (PDB:6MQ9 and PDB:6MQK), both types of interface (open and closed) are found within the same filament, independent of the nucleotide, and the closed interface presents a displacement that breaks the twofold symmetry (the “shifted” conformation). In the case of SEPT9 (PDB:5CYP and PDB:5CYO), the conformation depends on the bound nucleotide, being open when bound to GDP and closed when bound to GTPγS (Macedo et al., 2013; Castro et al., 2020).

Since the closed conformation has thus far only been observed for SEPT3/9/12, it is tempting to believe that this is a unique property of the group. As mentioned above (Section 7.4), the N-terminal region in these septins presents some unique properties, including the charge distribution along α0 and the absence of one of the anchoring phenylalanines (Phe(HL)). These features could facilitate its exit from the cavity, allowing for the closure of the NC-interface, which would otherwise be impossible due to steric hindrance.

The canonical open conformation of the NC-interface is characterized by a ∼20 Å separation between the two α6 helices, which is reduced to approximately 12 Å in the SEPT3 group when the interface closes. This conformational change results from a rearrangement of the salt bridges at the upper part of the interface (Macedo et al., 2013; Castro et al., 2020). With the closure of the interface, the PB2 region, which follow α2, wraps around α6 of its neighbor and approaches the polyacidic region (PAR). Specifically, Arg(PB2) dips down into the interface forming a new salt bridge with a glutamic acid from the PAR (Macedo et al., 2013; Castro et al., 2020). In this new conformation, Glu(α2) interacts directly with Arg(α6), which no longer interacts with the α2 helix dipole (Macedo et al., 2013; Castro et al., 2020). As a consequence, the α6 helices approach one another while the α2 helices move apart.

With the closure of the interface, the α0 helix is displaced out of the NC-interface, thereby gaining conformational freedom (Castro et al., 2020). This is not only suggested by the steric hindrance, which would result if α0 were to remain within the interface but has actually been observed experimentally in one of the crystal structures of SEPT3 (PDB:4Z54). Different to when hidden within the interface, where its positive charges are inward-pointing and occupied in stabilizing the interface itself, once liberated, PB1 would be free to interact with membranes (Zhang et al., 1999; Bertin et al., 2010).

Figure 13 schematizes the differences between the open and closed states. Under this proposal, the PAR plays two important roles: 1) harboring α0 and its PB1 when buried in the open state and 2) stabilizing the closed conformation by interacting with PB2. As such, the open conformation appears incompatible with membrane association by PB1, but rather is necessary for its safe storage when this is not required (Castro et al., 2020). This mechanism is also compatible with a role for PB2 in membrane association on interface closure. As this occurs, and PB2 wraps around helix α6 of the neighboring subunit to interact with the PAR, it becomes more exposed on the filament side where it could act in concert with PB1 in membrane association. Indeed, its role as such has already been proposed (Omrane et al., 2019) and is compatible with the sense of bending of the hexamer observed by cryo-EM (PDB:7M6J) (see Section 10.3).

In all protein structures from the SEPT3 group, the α5′ helix is oriented differently from that observed for other septins, lying roughly parallel to the filament axis. This is due to the presence of characteristic residues such as cisPro(α4-α5′/SEPT3), which result in the N-terminus of α5′ coming closer to α6 (Castro et al., 2020). This displaces the PAR upwards in a way which would favor interacting with PB2 in the closed conformation. The combined differences observed in the SEPT3 group, including those associated with α0, α5′ and the lack of a C-terminal coiled coil, may sum to give this NC-interface its apparently unique plasticity.

The closed structure of the G-domain of SEPT9 (PDB:5CYP) was obtained by soaking a GDP-bound crystal (in the open conformation) with the GTP analogue, GTPγS, suggesting that the occurrence of nucleotide hydrolysis in the G-interface could result in conformational changes to the adjacent NC-interface. How might this occur now that β-strand slippage appears to have been eliminated as a potential conduit at physiological interfaces (see Section 6.3)? One possibility is helix α2, which runs from the switch II region of the G-interface at its N-terminus to the PB2 region of the NC-interface at its C-terminus. The septin specific sequence Sep2 (Pan et al., 2007) runs underneath this helix such that it is unable to pack against the central β-sheet (see Section 5.1.3). This unusual arrangement (Chothia et al., 1981) may free up α2 and allow it to move with respect to the rest of the structure, potentially as a rigid body or rod. As such, the conformational changes which occur to switch II on GTP hydrolysis could be more readily transmitted via α2 to the neighboring NC-interface. Although speculative, this mechanism would provide a functional role for the Sep2 motif, justifying its strict conservation in septins during evolution (Pan et al., 2007).

Nothing is known about the mechanism of information transfer itself. However, a specific aromatic residue of the switch II region, Aro(Sw2), appears to be a potential candidate. The SEPT3 septins bound to non-hydrolyzable GTP analogues currently represent the best model available for the pre-hydrolysis state. In all such structures, this aromatic residue lies such that the plane of the aromatic ring is parallel to the surface of helix α2. However, structures in the presence of GDP suggest that nucleotide hydrolysis perturbs the switch II region, causing Aro(Sw2) to assume an alternative conformation in which the ring lies approximately perpendicular to the helix surface (Supplementary Figure S7; Rosa et al., 2020). The change to the Aro(Sw2) rotamer appears to be directly coupled to a second aromatic residue, Phe(α3), on the inner surface of helix α3. This aromatic cluster rearrangement lifts α2 further from the underlying β-sheet, potentially giving it the necessary freedom to move and thus perturb the neighbouring NC-interface towards which it is slightly shifted (Supplementary Figure S7). The notion that this aromatic cluster could be essential for communication gains support from the fact that Phe(α3) is lacking in the catalytically inactive septins where, by definition, such a mechanism would be inoperative anyway. It is interesting to note that the information transfer may be due to a transitory perturbation to α2 rather than a switch between two well-defined states, an idea supported by the fact that the shifts described above are rather subtle. Clearly, more work is necessary on this point.

The concertina movement due to opening and closing of the NC-interface of the SEPT3-group septins (Figure 14A) is not the only structural flexibility of the core particles for which there is experimental evidence. Although the flexibility of the hexamer had been noted previously (Sirajuddin et al., 2007), the study of Mendonça et al. (2021) was able to attribute this principally to movement at the central G-interface (SEPT7-SEPT7) (Figure 14B). The direction of greatest bending corresponded to flexion of the particle around the twofold axis which relates the two trimers (Figure 14B, center). Interestingly, this bending may be related to the recognition and/or interaction of the filaments with membranes. If the curvature of the oligomer were coincident with that of the membrane, it would lie laterally such that the main aperture to the cavity at the NC-interface would face the membrane. This raises intriguing possibilities, including, for example, that both PB1 and PB2 could interact simultaneously with negatively charged membrane lipids and that the cavity may play a role in this process. Obviously, the current scarcity of structural data only allows for speculation at this point and any attempt to draw a definitive conclusion would be premature.

Membrane recognition by septins can also be curvature-dependent, as has been shown for both yeast and animal septin complexes (Palander et al., 2017; Beber et al., 2019; Cannon et al., 2019; Woods et al., 2021). In vitro studies have shown that septin filaments preferentially bind to regions of curvature on the scale of microns (Bridges et al., 2016). Although the bending observed in the cryo-EM study only gives an indication of the direction of flexion and not its extent, nevertheless this would appear to be compatible with micron-scale recognition. Propagating the flexion observed at the central interface to all oligomers along a filament would easily produce ring-like structures with diameters compatible with the preferential curvatures already observed (Kinoshita et al., 2002; Tanaka-Takiguchi et al., 2009; Bridges et al., 2016; Beber et al., 2019).

Since the exposure of PB1 due to NC-interface closure is potentially a unique property of the SEPT3 group, this makes the conformational properties of the octameric particle of particular interest. At present, there are no experimental structures available for an octamer from any species. Nevertheless, the accumulation of partial structures of single septins and heterodimers, when taken together with that for the hexamer, mean that reliable models for the human octamer can be computationally generated. Figure 14C shows examples of how the variation observed at the central interface in the different crystal structures involving SEPT3-group members leads to full particles which present considerable structural diversity.

Clearly, structural studies of octameric particles with a view to determining the extent and direction of bending will be essential to understanding the physical properties of filaments. How these properties may depend on the ratio of hexamers to octamers is an intriguing question to be answered. More important still is how these relate to septin association with membranes and the cytoskeleton and thereby impact on septin function.