The septins were originally identified via the isolation of temperature-sensitive (ts) mutants defective in cell cycle progression (Hartwell, 1974 ). Mutations at four distinct loci— cdc3, cdc10, cdc11, and cdc12—shared the unique terminal phenotype of failed cell division, despite continued progress through DNA replication and nuclear division, creating multinucleated cells (Hartwell, 1971 ). Anisotropic bud growth occurs in these mutants, but cytokinesis fails, leading to cells with chains of highly elongated buds (Hartwell, 1971 ). When subsequently cloned and sequenced, the corresponding wild-type genes were found to encode a family of related proteins eventually dubbed septins (Pringle, 2008 ). A fifth septin-encoding gene (SEP7/SHS1), recognized on the basis of its sequence similarity to the others, is also expressed in mitotic cells (Carroll et al., 1998 ; Mino et al., 1998 ). Unlike the other four septin genes, even an shs1-null allele does not cause a ts proliferation defect, but shs1Δ cells exhibit a moderately elongated bud, and an shs1Δ mutation exacerbates the phenotype of other mutations that affect cytokinesis (Iwase et al., 2007 ; McMurray et al., 2011 ). Thus all five mitotic septins are involved in cytokinesis and cellular morphogenesis.

Biochemical studies have shown that native septins isolated from yeast cells by immunoaffinity chromatography and other fractionation methods (Frazier et al., 1998 ; Mortensen et al., 2002 ), as well as recombinant septins expressed in and purified from bacterial cells (Versele et al., 2004 ; Farkasovsky et al., 2005 ), copurify as stable complexes of defined stoichiometry, even at high salt concentration (≥200 mM; Bertin et al., 2008 ; Garcia et al., 2011 ). At low salt concentration (≤100 mM), purified yeast septin complexes self-assemble into filaments in solution (Frazier et al., 1998 ; Versele et al., 2004 ; Farkasovsky et al., 2005 ). Using electron microscopy (EM) and single-particle analysis, combined with physical tags, we found that complexes of the four yeast septins identified as CDC genes exist as well-ordered, linear hetero-octameric rods (4 × 4 × 32 nm) of a uniform composition: Cdc11-Cdc12-Cdc3-Cdc10-Cdc10-Cdc3-Cdc12-Cdc11 (Bertin et al., 2008 ). Furthermore, filament formation at low salt concentration occurs by end-to-end association of the rods via Cdc11–Cdc11 interaction, and, as polymerization proceeds, the filaments pair in a highly cooperative manner via cross-filament association between the extended carboxy termini of Cdc3–Cdc12 dimers, yielding long, “railroad track”–like structures (Bertin et al., 2008 ). Fluorescence anisotropy measurements carried out in diverse organisms are compatible with the pairing of septin filaments in situ (DeMay et al., 2011 ). Independent studies using EM or X-ray crystallography to examine purified septin complexes from Caenorhabditis elegans (John et al., 2007 ) and humans (Sirajuddin et al., 2007 ) revealed strikingly similar linear, nonpolar rods capable of end-to-end polymerization into filaments. It is striking that when yeast septin filaments are assembled on the surface of lipid monolayers containing phosphatidylinositol-4,5-bisphosphate (PtdIns4,5P₂) to more closely mimic the PM environment believed to be present at the yeast bud neck (Prouzet-Mauléon et al., 2008 ; Yoshida et al., 2009 ), the filaments pair tightly (lack the gap seen in the “railroad tracks” formed in solution) and do not require the carboxy termini of either Cdc3 or Cdc12 to do so (Bertin et al., 2010 ). Moreover, the tight filament pairs are frequently cross-braced by rods oriented orthogonally to the filaments, creating a mesh-like structure (Bertin et al., 2010 ). Unlike hetero-octamer assembly and rod polymerization, cross-brace formation requires the carboxy-terminal extension of Cdc11 (Bertin et al., 2010 ).

Thin sections of glutaraldehyde- and osmium tetroxide–fixed dividing S. cerevisiae cells stained with uranyl acetate, dehydrated, and embedded in plastic resin were examined by EM and revealed an ordered array of filament-like profiles wrapped around the inside of the neck as circumferential hoops perpendicular to the mother–bud axis (or perhaps as a continuous spiral; Byers and Goetsch, 1976 ). In this hourglass-shaped collar, the putative filaments (spaced ~28 nm apart) were closely apposed to the PM (apparently connected to the inner leaflet by extensions ~12 nm long and ~3 nm thick). In some images, each filament (~10-nm diameter) could be resolved into two very fine spines separated by ~5 nm (Byers and Goetsch, 1976 ), reminiscent of the tightly paired filaments seen on PtdIns4,5P₂-containing lipid monolayers (Bertin et al., 2010 ). A similar pattern was observed in Candida albicans (Soll and Mitchell, 1983 ). When bud necks were examined by an alternative method—EM of Ta/W- or Pt-coated replicas of freeze-fractured and deep-etched fixed spheroplasts—filamentous structures resembling “gauzes” and attributed to criss-crossed, septin-containing filaments were observed (Rodal et al., 2005 ), reminiscent of the mesh-like networks formed by purified septin complexes on PtdIns4,5P₂-containing lipid monolayers (Bertin et al., 2010 ).

Multiple observations demonstrate that the septins are required for the formation and are integral constituents of these neck-associated filaments. The neck filaments were undetectable in cdc3, cdc10, cdc11, or cdc12 mutants shifted to the restrictive temperature but were found in all other cdc mutants examined under the same conditions (Byers and Goetsch, 1976 ). Whether visualized in fixed cells by antibody decoration and indirect immunofluorescence (Haarer and Pringle, 1987 ; Ford and Pringle, 1991 ; Kim et al., 1991 ) or as fusions to genetically encoded fluorescent tags in live cells in real time (Cid et al., 1998 ; Iwase et al., 2007 ; McMurray and Thorner, 2008b ), Cdc3, Cdc10, Cdc11, Cdc12, and Shs1 all colocalize at the neck between a mother cell and its bud. As judged by photobleaching (Caviston et al., 2003 ; Dobbelaere et al., 2003 ) and fluorescence anisotropy (Vrabioiu and Mitchison, 2006 , 2007 ; DeMay et al., 2011 ) of septin-GFP fusions, at the cell cycle stage when the neck filaments are observed, each of the septin subunits examined is immobilized and orientationally constrained, consistent with formation of highly ordered assemblies. Finally, concomitant with the onset of cytokinesis, the septin-containing, hourglass-shaped collar at the bud neck is split into two discrete rings that closely flank each side of the neck (Oh and Bi, 2011 ).

The ordered septin array at the bud neck appears to serve three functions. First, it acts as a scaffold by physically associating with other proteins that recruit factors critical for cell division, including the actin and myosin that assemble into the contractile ring (Dobbelaere and Barral, 2004 ), and for execution of a cell cycle checkpoint that monitors septin assembly (Shulewitz et al., 1999 ; Keaton and Lew, 2006 ). Second, it acts as a physical barricade to prevent free diffusion of plasma membrane proteins between the mother cell and its bud during cell growth (Takizawa et al., 2000 ; Faty et al., 2002 ). Third, the split rings act as gaskets or corrals to sequester the secretory vesicles, chitin-synthesizing enzymes, and other proteins required for completion of cytokinesis and septation but which the septins do not bind directly (Dobbelaere and Barral, 2004 ; Oh and Bi, 2011 ). Clearly, continuous membrane-associated, septin-containing filaments would serve as effective barriers to physically impede diffusion of integral membrane proteins, either directly or indirectly by binding to and ordering PM lipids (Bertin et al., 2010 ). Consistent with the conclusion that filament formation is necessary for septin function in vivo, we demonstrated that mutations that ablate filament assembly prevent septin localization to the cell cortex and block cell division, whereas deletion of certain septin subunits can be tolerated because the resulting complexes are capable of polymerizing into filaments (McMurray et al., 2011 ).

A number of experimental and conceptual points support our assumption that both the circumferential and axial filaments are composed of septins. The 30-nm interval separating each circumferential filament from another corresponds to the length of a septin octamer, the repeating unit in a filament (Bertin et al., 2008 ). Thus, if the axial filaments act as a patterning scaffold to control the spacing between each circumferential filament and the next and do so via specific subunit–subunit contacts, then either Cdc10 or Cdc11 could mediate the contacts because, in a septin filament, each Cdc10–Cdc10 and Cdc11–Cdc11 pair resides ~30 nm away from the next doublet of the same subunit. Several considerations make Cdc11 the prime candidate to be the septin subunit responsible for mediating the observed cross-filament contacts between the orthogonal filaments. First, in cdc10Δ cells, both circumferential and axial filaments are still visible. Of importance, the spacing between circumferential filaments becomes narrower in cdc10Δ cells and corresponds to the length of a hexameric rod, in agreement with the absence of Cdc10. Finally, the C-terminal extension of Cdc11 is required for formation of the orthogonal meshes we observed using recombinant septin complexes on lipid monolayers (Bertin et al., 2010 ).

In addition, interesting properties exhibited by purified septin complexes on PtdIns4,5P₂-containing lipid monolayers (Bertin et al., 2010 ) are recapitulated in situ by the network of neck filaments described in the present report. Tightly paired septin filaments were similarly cross-bridged by individual octameric rods via Cdc11–Cdc11 contacts mediated by the predicted C-terminal, coiled-coil–forming sequence in this septin, creating meshworks that were uniquely observed on PtdIns4,5P₂-containing lipid monolayers (Bertin et al., 2010 ). Moreover, these meshworks and the orthogonal array of axial and circumferential filaments we observe by tomography are reminiscent of the “gauze-like” arrangement of filaments (attributed to septins by antibody decoration) found at the bud neck when replicas of fixed yeast spheroplasts were examined (Rodal et al., 2005 ).

Of importance, the neck filaments defined by Byers and Goetsch persisted in cells treated with cytochalasin B, which disrupts actin filaments but were not found in cells undergoing cytokinesis (Byers and Goetsch, 1976 ), when the actomyosin contractile ring, but not the septin collar, remains at the neck (Oh and Bi, 2011 ). Hence the filaments described in the present study are unlikely to be actin or myosin.

Collectively, our findings suggest a model (Figure 6) in which circumferential filaments wrapped around the inside of the bud neck (each composed of a tight pair of septin filaments) make contact with and are spaced by axial filaments running parallel to the mother–bud axis, generating a lattice, in which Cdc11 mediates the junction at the points of intersection of these two orthogonally arranged filament arrays. The nature of the 20-nm-long connectors between the circumferential filaments and the inner leaflet of the PM is unknown. Integral membrane proteins at the bud neck known to interact with septins, such as Axl2 (Gao et al., 2007 ), are potential candidates.

Vrabioiu and Mitchison (2006 , 2007 ) used fluorescence polarization spectroscopy to study the organization of septins in budding yeast through the process of cytokinesis. Of interest, their analysis suggests that the ordered ensembles in the collar and those in the split rings differ in orientation by 90°. Similarly, by a distinct polarized fluorescence microscopy method, Gladfelter and colleagues observed such an apparent rotation (DeMay et al., 2011 ). The tomographic studies presented here all correspond to the collar state preceding cytokinesis, which is much more common in unsynchronized cells and was basically the one sampled in our experiments. Although we cannot shed any light on changes that occur during the collar–split rings transition, our studies directly show that during the collar state two sets of filaments, perpendicular to each other, coexist. An explanation for the change in anisotropy observed by others (Vrabioiu and Mitchison, 2006 , 2007 ; DeMay et al. 2011 ) could be the disappearance of one set of filaments (the axial ones) with the onset of cytokinesis, likely by disruption of the Cdc11-mediated junctions between the axial and circumferential filaments. This would allow for accordion-like collapse of the spacing between the circumferential filaments and their bundling on either side of the bud neck (the split rings).

In our lipid monolayer in vitro experiments (Bertin et al., 2010 ), the filaments were associated with this surface via their direct binding to the lipids. In our tomograms, however, it appears that the septin-based filaments are situated at a distance from the inner leaflet of the PM but contact it through a ~20-nm stem of connecting density. Of interest, in an EM image from Ashbya gossypii, septin filaments also appear to be disconnected from the plasma membrane (DeMay et al., 2011 ). Theoretically, it remains possible that a third set of septin filaments directly abuts the inner PM surface and comprises the intense electron density we observed on the inner leaflet of the bilayer in the bud neck region (Figure 1). However, this density may reflect known localized asymmetries in lipid content between the inner and outer leaflets of the PM at the bud neck (Chen et al., 2006 ; Roelants et al., 2010 ). A reasonable possibility is that the direct septin–lipid interaction described in the monolayer experiments may correspond to a distinct time point in the yeast cell cycle from that illustrated in our tomograms of mature buds.