Introduction

Septins are a family of GTP-binding, filament-forming proteins conserved across eukaryotes. They have been implicated in various cellular processes including cell division1, cell polarity2,3, and membrane remodeling4,5. Septin mis-regulation has been linked to cancer, neurodegenerative diseases, and infertility6,7. Septins are thought to function primarily through their ability to form higher-order structures, but the mechanisms underlying their many shapes and functions in cell biology are still in the early stages of research.

The genes encoding septins were discovered in the 1970s, and later the protein products were visualized in budding yeast cells as filament-like striations at the mother cell-bud neck8,9,10. Despite some similarities to other cytoskeletal filament forming proteins such as actin and microtubules, the understanding of the biophysical properties of septins has lagged behind. Whereas actin polymerizes from monomers, and microtubules from two alternating subunits, septin complexes assemble from several subunits that form palindromic, heteromeric rods, which then polymerize into filaments11. Structural studies of human septins have shown that septins have a core highly conserved globular GTP-binding (G) domain. Protomers interact with each other alternately using the G interface or the NC (N- and C-termini) interface within the rod12,13. These palindromic rods are, in human as well as in fungi, predominantly octameric11,12,13,14,15. The Saccharomyces cerevisiae (budding yeast) palindromic rod complex used in this study is arranged as Cdc11-Cdc12-Cdc3-Cdc10-Cdc10-Cdc3-Cdc12-Cdc11. Septins possess N- and C-terminal extensions that vary in length and are too flexible to be resolved by either X-ray crystallography or cryo-EM11,12,13. At this time, no yeast septin octamer structure has been reported, but the conservation of the individual septins, the known subunit order and overall dimensions, ~32 nm long and ~5 nm wide, of the yeast octameric rods11, allowed the generation of a high-confidence structural model16. This in turn facilitates the interpretation and use of lower resolution techniques to analyze their assembly and organization.

While septins exist as heteromeric palindromic rods in the cytoplasm17, they polymerize into filaments when bound to lipid membranes. Septins preferentially bind anionic lipids (e.g., PI or PS) through polybasic regions in the termini18,19,20. Once bound to lipid membranes, the formation of filaments from monomers is promoted by 2D-diffusion. Longer filaments can be formed via annealing of rods and short filaments as they diffuse on the membrane or through cooperative recruitment from the bulk solution16,17,21. Septin filaments can then interact with each other, and/or other proteins22, to form diverse structures, collectively termed higher-order assemblies.

Higher-order assemblies are septin filaments that align, stack, and arrange themselves into varied structures, including filaments, rings, and gauzes, at distinct regions on the membrane or at different times within the same cell23. These assemblies can alter cell rigidity and shape, spatially recruit proteins, and form scaffolds18. Higher-order septin structures can help to compartmentalize membranes24, for example at the cytokinetic furrow to restrict the passage of proteins between the splitting cells during cytokinesis25,26. The disruption of these assemblies causes aberrant cell morphologies and can disrupt the organization of other cytoskeletal elements18. While many cellular factors contribute to septin arrangement in cells, septins in isolation readily form diverse arrangements in vitro. For example, septins on lipids imaged by electron microscopy (EM) or atomic force microscopy (AFM) revealed filaments, bundles, rings, and gauzes reminiscent of what is seen in cells11,23,27,. These studies underpinned the power of minimal well-controlled bottom-up systems to study the behavior of septins, and motivated the investigation into how septins self-organize themselves on membranes in the absence of cellular factors.

While EM allows the characterization of very large assemblies, and potentially the resolution of a high-resolution structure, imaging is performed either of stained sample or after freezing to cryogenic temperature and can therefore not provide dynamic information. Recently, we employed high-speed atomic force microscopy (HS-AFM) to image single septin filaments on mica surfaces at high resolution16, providing movies of septin polymerization, kinetics, organization, and evaluation of the effects of both pH and ion concentration on the assembly. However, unfortunately, these movies were acquired of septins on mica. Another study using more conventional AFM provided images of animal septin in thick, multi-filament bundles, and these septin images were acquired on membranes27. These two studies indicated the feasibility of exploiting dynamic HS-AFM to visualize yeast septin assembly dynamics directly on yeast membranes.

Here, we use HS-AFM to study the interaction of Saccharomyces cerevisiae (budding yeast) septin octamers with yeast lipid membranes of near-physiological lipid compositions under near-native conditions, i.e., in physiological buffer and at ambient temperature and pressure. In summary, our results indicate that septin is lipid phase selective and that its structure, without the contribution of any other cellular component, contains the blueprint for the formation of organized higher-order structures.

Results

Yeast polar lipid exhibits phase separation, including three coexisting phases

First, we characterized the morphology of supported lipid bilayers (SLBs) formed on mica from yeast polar lipid extract, a close proxy of the native plasma membrane composition on which the yeast septin oligomers polymerize into filaments in vivo. We found that yeast polar lipid membranes separated into distinct phases. In general, lipid phase separations occur when two or more dissimilar lipids laterally de-mix into domains with distinct composition, packing density, and diffusion properties, such as the liquid-disordered phase (Ld), the liquid-ordered phase (Lo), and gel phase (Lβ)28,29,30. Importantly, these phase domains vary in bilayer thickness (i.e., height in AFM images), making them clearly distinguishable when probed with HS-AFM (Fig. 1a, c). Indeed, in the z-dimension AFM has sub-Angstrom resolution, making it an ideal technique for resolving small height variations in biological membranes such as phase separations31,32,33,34,35,36,37.

Fig. 1: Yeast polar lipid extract bilayers exhibit lipid phase separation and septin filaments exclusively polymerize on the Ld phase.
figure 1

HS-AFM frames of yeast polar lipid SLBs showing two (a) or three (c) phases, depending on imaging area. b, d Height analysis of (a) and (c): Lo (liquid-ordered) is 1.2 ± 0.3 nm (n = 57) thicker than Ld (liquid-disordered), and Lβ (gel) is 0.7 ± 0.2 nm (n = 76) thicker than Lo. Insets: Section profiles along dashed lines in (a) and (c). e, g HS-AFM frames (Supplementary Movie 1) of septin filaments on a Ld and Lo phase separated membrane before (7 frame average (e)) and after (single frame (g)) a HS-AFM force-sweep experiment. f, h Section profiles along dashed lines in (e) and (g). Septin filaments peak ~5 nm above Ld surface, derived from section profiles in (f) and (h), dashed line in (h). HS-AFM frames of membrane areas with three phases Ld, Lo and Lβ, with (i) and without (k) septin filaments. A membrane defect in (k) allows Ld thickness determination to a common baseline, mica (arrow). j, l Section profiles along dashed lines in (i) and (k). The Ld phase has 3.4 ± 0.5 nm (n = 43) thickness. Membrane regions are labelled according to the assigned phases, Ld, Lo and Lβ, along with areas covered by septin, or membrane defects. Yeast SLBs were generally formed and imaged in 20 mM HEPES, pH 7.5, 150 mM KCl before septin addition, e.g., k, but were stable across a broad range of conditions e.g., 20 mM HEPES, pH 6, 50 mM KCl (a), or 20 mM HEPES, pH 7.5, 300 mM KCl, 2 mM MgCl2 (c). Septin filaments on yeast SLBs (e), (g), and (i) were imaged in 20 mM HEPES, pH 6, 50 mM KCl, optimized to minimize tip-sample interaction. Septin filament images, (e) and (i), are representative of 26 analyzed images from 8 biological replicates. For bare membrane statistics, see Table 1.

Full size image

Imaging regions with two phases (Fig. 1a), pixel height histogram analysis revealed separated distributions with ~1.3 nm height difference in bilayer thickness (Fig. 1b). Section profile analysis corroborated and visually illustrated this height difference (Fig. 1b, inset). These two distinct phases were assigned as Ld (liquid-disordered) and Lo (liquid-ordered) phases based on the height measurements; the Lo phase is thicker than the Ld phase based on its expected composition containing lipids with saturated hydrocarbon tails that pack tightly (see discussion)31,38. In some imaging areas of the membrane three lipid phases coexisted (Fig. 1c). Pixel height histogram analysis in these regions showed three clearly separated distributions with ~1.5 nm height difference between Ld and Lo, and a height difference of ~0.8 nm between Lo and the third phase. This third phase was 2.3 nm thicker than the thinnest Ld phase (Fig. 1d). Again, section profile analysis showed the thickness differences of the phases in representative raw data (Fig. 1d, inset). Repeat measurements revealed two average height steps, 1.2 ± 0.3 nm between Ld and Lo phases, and the third phase being 0.7 ± 0.2 nm thicker than Lo (Table 1). Thus, we assigned this third phase, the thickest of the three, to Lβ (gel phase, see discussion). Although phase separation has been observed in yeast giant unilamellar vesicles (GUVs)39 and in unperturbed yeast vacuole membranes40, to the best of our knowledge, AFM images of phase separated yeast SLBs have not been published previously. Generally, phase separation in mixed lipid SLBs is routinely observed31,33,34. Importantly, the detection of the third, thickest phase, constrains the assignment of the two thinner phases as Ld and Lo phases.

Table 1 Yeast polar lipid phase bilayer heights
Full size table

Septin binds selectively to the liquid-disordered phase

Strikingly, when recombinant septin was added to the phase-separated yeast polar lipid SLBs, the protein selectively polymerized on one phase (Fig. 1e). To determine on which membrane phase septin filaments polymerized and assembled, we performed HS-AFM force-sweep experiments with increased imaging force to remove the septin filaments mechanically and see the underlying SLB (Fig. 1e, g). Comparing the same field of view before and after the force-sweep experiment (Fig. 1e, g) revealed that the phase on which septin filaments polymerized was lower than the neighboring septin-free phase (Supplementary Movie 1). Comparative section profile analysis of these images showed that septin filaments were specifically found on the Ld phase, which was ~1.2 nm thinner than the Lo phase (Fig. 1f, h). Furthermore, comparison of the height profiles from before (Fig. 1f) and after (Fig. 1h) the force-sweep experiment revealed that septins protruded ~5 nm above the Ld phase and ~4 nm above the neighboring Lo phase. The thickness of the septin layer, ~5 nm, is in excellent agreement with the thickness of the human and yeast septin oligomer structure11,12,16. There was no indication of septin filaments forming on the Lβ phase in neighboring imaging areas of the same sample with three coexisting phases, showing that septin also avoided Lβ (Fig. 1i). Section analysis further confirmed this phase assignment (Fig. 1j). Finally, we searched for defects in the lipid bilayer that would give access to the mica surface and allow us to calibrate the entire system to an experimental baseline, i.e., the mica surface (Fig. 1k, l). Accordingly, The Ld phase had a total thickness of 3.4 ± 0.5 nm, the Lo phase of 4.6 ± 0.6 nm, and the Lβ phase of 5.2 ± 0.6 nm (Fig. 1l, Table 1). As the septin filaments specifically and exclusively polymerized on the Ld phase and protruded by 4.9 ± 0.6 nm (Fig. 1h), the first septin layer reached a total height of ~8 nm above mica. It is notable that septin filaments neatly aligned with phase boundaries (Fig. 1e, i), indicating a potential role of lipid phase separation in the nucleation, organization and formation of septin filaments and higher-order structures.

Septin filaments align laterally and longitudinally

Septin filaments on yeast polar lipid SLBs were easily disturbed by HS-AFM operation. To successfully perform high-resolution HS-AFM imaging, environmental conditions and feedback parameters needed to be well-adjusted. In addition, time averaging over multiple frames increased the signal-to-noise ratio and allowed high resolution information to be extracted from the HS-AFM movies including not only the lateral order, but also the longitudinal periodicity of the septin rods along the filaments, and thus their longitudinal order (Fig. 2a, b). Measuring the length, l, of the periodicity along filaments resulted in l = 31.3 ± 0.5 nm (n = 68, from 9 images from 5 biological replicates), in excellent agreement with the unitary length of the octameric yeast septin rod11. The palindromic repetition of septin subunits in each septin rod, Cdc11-Cdc12-Cdc3-Cdc10-Cdc10-Cdc3-Cdc12-Cdc11, gave rise to height undulations, only Δh = 0.3 nm in amplitude, along the septin filaments on the yeast polar lipid SLBs, directly visible in the raw data (Fig. 2a, b) and plotted in height vs filament length graphs (Fig. 2c). We hypothesize that these periodic height undulations are the result of either slight non-flatness of the octameric rod and/or the difference in length of bulk-exposed flexible N-termini along the octameric rod that repeat themselves along the filament long axis. We took advantage of these topographical signals in the filament structure to analyze how septin filaments sense and align with neighboring filaments (Fig. 2b, dashed outline). For this, we plotted the height profile of neighboring filaments pairwise along their long axis (Fig. 2c, red and black traces) and calculated the correlation between the two traces. This was done by computationally moving one of the two filaments from about −16 nm (−180° phase difference) to about +16 nm (+180° phase difference) with respect to the rod length of the other filament. This resulted in correlation value vs phase-lag plots for each pair of rods in neighboring filaments, with perfectly longitudinally aligned filament pairs naturally having a phase lag peak at ~0° (Fig. 2d). Given the high signal-to-noise ratio in these topographies, the cross-correlation value peaked at ~0.9 (Fig. 2d), and allowed us to describe how each filament aligned longitudinally with its lateral neighbors. We plotted the distribution of the peak positions of the correlation lags for each set of rods in filaments pairs and found a narrow distribution with a full width at half maximum (FWHM) of 21° around a perfect alignment with phase-lag 0° (Fig. 2e). The distribution is naturally symmetrical because of the palindromic nature of the septin rod without polarity, i.e., if, for example, one filament had a + 3° phase-lag with respect to its neighbor, the neighbor has a −3° phase-lag with respect to the former. Thus, septin filaments sense their neighbors with precision, creating closely aligned filament arrays. As is visible, the rather good alignment preservation of the filaments results in patches where the septin carpet exposes locally highly conserved surface features, which may serve for vertical templating interactions with septins and other molecules (see Discussion).

Fig. 2: Septin filaments align laterally and longitudinally on yeast polar supported lipid bilayers.
figure 2

a HS-AFM movie frame, and b magnified view of the dashed outlined region in (a) (49 frame time average), of septin filaments on a yeast polar lipid SLB in buffer, 20 mM HEPES, pH 6, 50 mM KCl. c Example section profiles along the two filaments in the region of the dashed outline in (b). The red and black traces correspond to the filaments indicated by the red and the black arrowheads in (b). d Cross correlation value vs phase-lag plot of the two traces shown in (c) as they are computationally phase shifted from −180° to +180° with respect to each other, peaking at a phase-lag φ = 0° with a cross-correlation value of 0.82. e Histogram distribution of filament neighbor alignment phase lag, φ, for each set of rods in neighboring filament pairs, peaking at 0° with a full width at half maximum (FWHM) of 21° (n = 170 filament rod neighbors). Note the distribution is symmetric because the septin filaments are palindromic and have therefore no directionality with respect to nearest neighbors. The signal-to-noise ratio was sufficient to observe longitudinal periodicity in 9 HS-AFM movies from 5 biological replicates.

Full size image

Septins bind to yeast membranes with a conserved face

As shown (Figs. 1e, i, and 2a, b), filaments on SLBs were visually well-aligned with a well-defined repetitive inter-filament spacing (Fig. 3a, b), which was markedly different from the septin filament morphology on mica. As we previously reported, on mica, septin filaments arranged in pairs (Fig. 3f, g) in a pH and salt condition-dependent manner16. Pairing was also observed by negative stain EM using the lipid monolayer method, and there is also evidence of potential filament pairing in cells23,27,41,42,43. In contrast, lateral pairing was never observed by in situ HS-AFM on yeast lipid SLBs, also consistent with a subset of assemblies seen in cells23,44. The inter-filament distance on membranes was <d(f-f )SLB > = 10.7 ± 2.0 nm as assessed by section analysis (Fig. 3c) and statistically analyzed in inter-filament distance histograms from several images and biological replicas (Fig. 3d).

Fig. 3: Morphology of septin filaments on yeast polar extract SLBs compared to septin filaments on mica.
figure 3

HS-AFM movie frame (a) and 5 frame average (b) of septin filaments on yeast polar extract SLBs in buffer 20 mM HEPES, pH 6, 50 mM KCl. c Section profile across the dashed line in (b) showing the periodic inter-filament distance, d(f-f), of the laterally aligned filaments. d Lateral inter-filament distance distribution: <d(f-f)> = 10.7 ± 2.0 nm (n = 401, from 26 images from 8 biological replicates). e Filament height, h, distribution above membrane, <h > = 4.9 ± 0.6 nm (n = 31, from 18 images from 9 biological replicates). Inset in (e) shows 4 septin filaments in section (ovals) on a lipid bilayer. Each filament interacts with the membrane via the face exposing the C-termini (C) and exposing the N-termini (N) to the bulk. HS-AFM movie frame (f) and 5 frame average (g) of septin filaments on mica in buffer 20 mM HEPES, pH 7.5, 600 mM KCl. h Section profile across the dashed line in (g) showing the inter-filament distance, d(f-f), within, and the inter-pair distance, d(p-p), between aligned filament pairs. i Lateral inter-filament distance, d(f-f), and lateral inter-filament-pair distance, d(p-p), distributions: <d(f-f)> = 7.1 ± 1.1 nm (n = 269) and <d(p-p)> = 19.3 ± 2.4 nm (n = 113, from 5 images from 4 biological replicates). j Filament height, h, distribution above mica, <h > = 4.0 ± 0.4 nm (n = 86, from 38 images from 13 biological replicates). Inset in (j) shows 4 septin filaments in section (ovals) on mica, grouped in 2 pairs of 2 filaments. The filaments in pairs interact with each other via the face exposing the N-termini (N) exposing the C-termini (C) to the sides. All values are mean ± stdev.

Full size image

Paired filaments (Fig. 3f, g) have very different inter-filament parameters as directly visible in section analyses (Fig. 3h), and as corroborated statistically (Fig. 3i). The inter-filament distance within a single pair was <d(f-f)mica > = 7.1 ± 1.1 nm, while the inter-pair distance, i.e., the distance between central dips of two separate pairs of filaments was <d(p-p)mica > = 19.3 ± 2.4 nm (Fig. 3h, i). This meant that the inter-filament distance between neighboring pairs was 12.2 ± 2.6 nm (19.3–7.1 nm). Thus, based on inter-filament distance measurements, we concluded that the filament-filament interactions on yeast lipid membranes were substantially different from those on mica. Though the imaging conditions were different on mica and on the yeast lipid membranes, train-track-like paired filaments were found on mica over a wide range of conditions16, while they were never detected on the native membrane. While it was impossible to statistically assess filament architecture on SLBs over the full range of conditions used by AFM on mica or by negative stain EM, control experiments on mica under the same conditions as used on yeast polar lipid SLBs showed some filament pairing (Supplementary Fig. S1). In addition, when defects in the yeast polar lipid SLBs were present, we found that septin filaments on the mica in these defects revealed extensive pairing – corroborating that the interface with which filaments interacted with membrane or mica were different (Supplementary Fig. S2). Finally, we incubated septin under conditions that maximized pairing on mica, pH 7.5, 600 mM KCl16, but found that septin did not bind to yeast SLBs under these conditions (see Fig. 4a).

Fig. 4: Septin interaction with and on yeast membranes.
figure 4

a Septin filament surface coverage on yeast polar lipid SLBs as a function of the KCl concentration during an in-situ titration (black data points), as well as individual incubation experiments (grey data points). Data points are mean ± standard deviation of the septin surface coverage in the first 10 HS-AFM frames in 3 different movies in each set of conditions. The titration data is fit with a Boltzmann function. b HS-AFM movie frames of septin filaments on yeast polar lipid SLB (in 20 mM HEPES, pH 6, 50 mM KCl) prior to the HS-AFM line scanning (HS-AFM-LS) force-sweep experiment. c HS-AFM-LS force-sweep experiment. Kymograph of filament cutting along a line in the middle of the dashed box in (b) and (d). As the force is increased the septin filaments break (white arrows). d HS-AFM movie frames of septin filament recovery after the HS-AFM-LS force-sweep experiment. Filament destruction at time t = +4 s was largely reformed at t = +116 s (dashed outlines). Imaging before, during, and after the HS-AFM-LS force-sweep experiment is shown in Supplementary Movie 2. e Individually flattened image areas of dashed outlines in (b) and (d) showing filament structure before and their recovery after the HS-AFM-LS force-sweep experiment. f Image entropy of panels in (e). The experiment is representative of 22 HS-AFM-LS filament cutting experiments from 4 biological replicates, where full or partial filament alignment recovery was observed. Data points are mean ± standard error of the image entropy in all image kernels of size 9 × 9 pixels.

Full size image

To get further insights into the septin-membrane and septin-mica interaction, we analyzed the thickness of the septin filaments on both yeast lipid bilayer SLBs and mica. On SLBs, the filaments are so densely packed that the tip cannot probe the level of the membrane surface between filaments. However, we were able to take advantage of the HS-AFM force-sweep filament removal experiments that gave access to the precise height of the filaments over the base level of the Ld phase on which they adhered (Fig. 1e–h). From these data, we extracted that the filaments protruded from the bilayer surface by <hSLB > = 4.9 ± 0.6 nm (Fig. 3e). On mica, we found occasional gaps between filaments that were large enough for the tip to probe the mica base level, and thus we could determine the filament thickness on the mica directly, <hmica > = 4.0 ± 0.4 nm (Fig. 3j). It is notable that the height of filaments on the mica were consistent whether they were paired or single (Fig. 3f, arrows). These data corroborated that septin filaments on yeast polar SLBs were differently oriented relative to the surface than on mica. One possibility is that on the SLB, septins adhere on the membrane oriented with the C-terminal extensions facing towards the membranes, and align side-by-side, exposing their N-terminal faces to the liquid/cytoplasm for 3D growth (Fig. 3e, inset). The extensions are flexible and may be oriented in a variety of ways, but their presence could be raising the septins off of the membrane surface, contributing to the height difference16. In contrast on the mica, it is possible that septins are re-oriented by 90°, lying on their side, forming lateral pairs with their N-termini and exposing their C-termini to either side of the filament pair16. As a result of their side-on adsorption on mica (Fig. 3j, inset), they are thinner than the filaments standing on the yeast SLB. This assignment is consistent with the septin filament structure and with our previous analysis of septin filaments on mica16, where we observed lipid droplets adhering to the outside of wild-type filament pairs but not on filament pairs of C-terminal deletion mutants, consistent with C-terminal extensions facing away from the center of filament pairs in the absence of a lipid bilayer that orients the filament (compare insets in Fig. 3e, j). Septin filaments bind to both mica and SLBs, and if defects in the SLBs occurred, septin filaments were observed bound on mica next to a membrane patch. This can either be filaments forming on mica directly, or filaments displaced from the Ld bilayer phase during incubation. The fact that these filaments displayed pairing (Supplementary Fig. S2), appears indicative that they formed directly on the mica and were structurally distinctive from filaments that formed on the bilayers. In this context, our inability to acquire high-resolution HS-AFM images on mica and yeast membranes under precisely matching conditions is further indirect evidence that the septin-mica and the septin-membrane interaction interface are different.

Septin binding to yeast lipid membranes is ion concentration dependent

Next, we performed a HS-AFM titration experiment and found that septin filament membrane binding was salt concentration dependent. We iteratively increased the KCl concentration in the buffer within the fluid cell while imaging septin filaments bound to yeast polar lipid SLBs, to which the filaments were first bound at 50 mM KCl (Fig. 4a). In a buffer comprising 50 mM KCl, pH 6, septin filaments are ordered, tightly packed, and stationary on yeast polar lipid Ld phase, their packing apparently aiding their stability for high-resolution HS-AFM imaging (Figs. 1e, i, 2a, b, and 3a, b). At 50 mM KCl, and physiological pH 7.5, septin filaments are still bound to the membrane, but appeared more mobile and the filaments could more easily be disturbed by HS-AFM imaging (Fig. 4a). Staying at pH 7.5, as the KCl concentration is gradually increased, the septin-membrane interaction became less stable, and reaching 300 mM KCl, the septin filaments could not be resolved individually on the membrane by HS-AFM. Between 300 mM and 600 mM KCl, septin filaments dissociated entirely from the membrane. Fitting the titration experiments resulted in a critical KCl concentration of ~260 mM where unbinding occurred. Alternatively, we performed individual incubation experiments at specific KCl concentrations and found that at ~190 mM ionic strength filaments could both adhere or fully dissociate. We hypothesize that the titration experiment results in a slightly higher critical ion concentration due to a cooperative stabilization by the filaments bundling on the surface (avidity), than in the case when septins are allowed to polymerize de novo on the yeast membrane at a given ionic strength. In any case, the two approaches give similar results, and we estimate that yeast septins are tuned to adhere or dissociate to yeast polar lipids at a critical ionic strength of ~200 mM, where efficient adherence and polymerization occurs at a concentration below, and dissociation occurs at a concentration just above ~200 mM. These experiments indicate the potential to form or dissociate septin membrane scaffolding by fine-tuning the spatio-temporal ionic strength in the membrane vicinity in a cellular setting.

Septin filaments repair ordered supramolecular structure after disruption

Septin filaments adopt lateral and longitudinal supramolecular order when bound to yeast polar lipid membranes (as described above, Fig. 2, Fig. 3). To further test the assembly templating process, we performed HS-AFM force-sweep experiments, taking advantage of the fact that HS-AFM contours the sample through soft touch, which can be controlled and potentially modulated to manipulate the sample. In this case, HS-AFM line scanning (HS-AFM-LS) was used to scan in 1D, i.e., scanning repeatedly over the same scan line, instead of the raster scanning of regular HS-AFM imaging. Prior to HS-AFM-LS force-sweep experiment, the stable supramolecular septin filament structure was imaged using HS-AFM and centered in the imaging area (Fig. 4b, dashed outlined). Using HS-AFM-LS force-sweep experiments, we increased the force applied by the HS-AFM tip to the sample in a controlled manner to cut the filaments, a process that is reported in the HS-AFM-LS kymograph (Fig. 4c). Note, the vertical lines in the top region of the kymograph, reporting the intact filaments in HS-AFM-LS at minimal force, and the sequential breakage from left to right of the filaments during the force-sweep (dashed outline and arrows in Fig. 4c). After the HS-AFM-LS force-sweep, the imaging area was reset to the parameters from before entering HS-AFM-LS mode, and the supramolecular filament structure and dynamics were observed in sequential images (Fig. 4d, e). Four seconds after the force-sweep, the septin filaments were found horizontally cut in the area where the force-sweep line scan was performed (Fig. 4d, dashed outlined). Neither the supramolecular structure nor single filaments were observable. After eight seconds, filaments were distinguishable again, although less ordered than before the HS-AFM-LS force-sweep. After 116 s, the filaments were well resolved, and the supramolecular structure adopted the same orientation as before the force-sweep. This experiment demonstrated the self-healing and self-templating ability of septin filaments in their supramolecular structure (Supplementary Movie 2). Close inspection, following flattening of this image sub-area individually, of the filaments in this region highlights the stable order of the filaments before (t = −40s, t = −28s, t = −4s), and their recovery (t = +4 s, t = +8 s, t = +116 s) after HS-AFM-LS, where large parts of the destroyed septin filament arrangement reformed with identical orientation to the structure before the force-sweep (Fig. 4e). To computationally assess the visually well-defined filament healing process, we calculated the image entropy of these panels which corroborated recovery of ordered structures about 2 min after filament destruction (Fig. 4f).

Second-layer septin filaments are templated by first-layer filaments

With sufficient incubation time and low imaging forces, septin filaments formed a second-layer on top of the first septin layer when assembled on yeast polar lipid membranes (Fig. 5), corroborating the capability of septin to form 3D supramolecular structures. Again, we took advantage of HS-AFM force-sweep experiments to confirm the presence and analyze the relative orientation of the first (lower, membrane-attached) layer septin filaments with respect to the second (upper) layer filaments. After imaging the second-layer filaments at low force to obtain high resolution images and reveal their orientation (Fig. 5a, t = 0 s, Fig. 5b, c, left), the force was slightly increased to selectively remove the second-layer filaments (Fig. 5a, t = 50 s, t = 70 s,), and image the filaments in the first (lower) layer Fig. 5a, t = 130 s, Fig. 5b, c, right). The orientation of the first and second layers were found perfectly aligned, showing that growth of the second-layer is templated by the first-layer. Previously we showed that on mica the septin filaments could form second and third layers, leading to the possibility of larger multilayer stacks16. As we documented in our comparative analysis of filaments on yeast membranes and on mica (Fig. 3), septins orient differently on the two substrates, and certainly the filament orientation on the native-like membrane is more native. With this regard, we hypothesized that the stacking of the second layer with the first was different than what we reported on mica - nevertheless, it is clear that higher up located filaments are templated by their underlying neighbors with a strong tendency to adapt to their orientation. This is highlighted by the two areas (indicated by * and **) of differently oriented first-layer filaments (Fig. 5c, right, t = 130 s) that are matched by the second-layer filaments before the force-sweep experiment (Fig. 5c, left, t = 0 s).

Fig. 5: Second-layer septin filaments are oriented matching first-layer filaments on yeast polar extract SLBs.
figure 5

a Sequential frames during a HS-AFM force-sweep experiment to remove second-layer septin filaments. b Magnified views of timepoints t = 0 s (covered with second-layer filaments) and t = 130 s (first-layer filaments). c Contrast enhanced images of (b) showing matching orientation of the second-layer and first-layer filaments. Symbols * and ** indicate two regions (dashed outlines) with different yet inter-layer matching filament orientation. Experiment is representative of 27 HS-AFM movies across 7 biological replicates.

Full size image

Discussion

Lipid phase separation in yeast lipid membranes

The existence of laterally phase separated lipid domains within membranes has been extensively studied in mixed model lipid membranes28,29,30. Lipid mixtures containing a saturated lipid, an unsaturated lipid and a sterol (e.g., cholesterol in mammalian cells, or ergosterol in fungi)39,45 can laterally demix into phases including liquid-disordered (Ld), liquid-ordered (Lo), and gel (Lβ) that vary in lipid composition and physical properties, such as packing density and lateral mobility28,29,30. Importantly, the Lβ and Lo phases are thicker than the Ld phase, allowing them to be easily discerned using AFM in Lβ-Ld and Lo-Ld mixtures, owing to the AFM’s excellent sensitivity in height measurements31,32,33,34,35. However, the in vivo equivalent, often referred to as lipid rafts, or lipid nano-domains in the mammalian plasma membrane, are difficult to observe experimentally and without labels that themselves may impact lipid order28,29,46. Here, we observed phase separated domains (Fig. 1) in the native yeast membrane extract despite its complex lipid composition (Supplementary Table S1) and fatty acid profile (Supplementary Table S2). Yeast lipid extract phase separation (Lo-Ld) has been observed previously in giant unilamellar vesicles (GUVs) using fluorescence microscopy39. Phase separation has also been observed in unperturbed yeast vacuole membranes in live cells40. Thus, our HS-AFM topography imaging data of yeast lipid extract in SLB membranes, provides further evidence that yeast membranes can phase separate (Fig. 1, Table 1), likely as a result of their complex lipid fatty acid profile with ~27% saturated and ~73% unsaturated hydrocarbon tails (Supplementary Table S2).

Here, notably, we observed the coexistence of three phases in a membrane from a natural lipid extract on mica surface (Fig. 1). Three phases have been observed to coexist in simple model lipid membranes under specific compositions47. In yeast and other fungi membranes, fluorescence lifetime measurements have shown the existence of both liquid ordered domains (Lo) enriched in the sterol ergosterol (similar to the commonly studied ‘lipid rafts’ in mammalian cells/lipid mixtures), and gel-like domains (Lβ) enriched in saturated sphingolipids and depleted of ergosterol48,49,50. In yeast lipid extract GUVs, fluorescence microscopy data hinted at the possibility of three phase coexistence (Lo, Ld, and Lβ)45,51. Our HS-AFM images provide clear evidence that yeast membranes can form three coexisting phases (Fig. 1). Based on yeast lipid membrane literature39,45,48,49,50,51 and phase separation reports of more controlled systems31,33,38,47, we assign the thinnest as the Ld phase, and the thickest as the Lβ phase with ordered close-packed saturated lipids, defining the membrane areas of intermediate thickness as the Lo phase with less packed lipids due to the presence of sterol. We acknowledge that assigning phases is non-trivial, not least due to the existence of interdigitated gel phases52,53 and a disordered gel phase that can be thinner than the Lo phase47. In any case, we found that septins selectively and exclusively bind to the thinnest phase, the well-assignable liquid disordered, Ld, phase.

Septins bind exclusively to Ld phase on yeast extract membranes

In this study we examined yeast septin self-assembly and organization on a membrane of close-to native lipid compositions using yeast lipid extracts. When septins were incubated on the phase separated yeast polar SLBs, the filaments polymerized exclusively on the Ld phase and avoided the two taller Lo and Lβ phases (Fig. 1). Anionic lipids, such as PI or PS, are required for septin binding18,19,20, and PI lipids make up ~26% of yeast polar extract (Supplementary Table S1). Therefore, it is possible that PI partitions preferentially into the Ld phase resulting in preferential septin binding, and/or PI is depleted from the Lo and Lβ phases resulting in septin not binding in these areas. Alternatively, it is possible that the physicochemical properties of the different lipid phases are important for promoting or precluding septin binding. Septins possess an amphipathic helix thought to intercalate into lipid packing defects, predominantly accessible in disordered phases and sparse in tightly packed ordered phases54. Indeed, previous research has shown loosely packed disordered lipid domains can favor binding of amphipathic helix-containing proteins55. Finally, it is also possible that alterations in diffusivity could favor or be responsible for the specific formation of septin filaments on Ld phase as assembly is thought to be a product of diffusion-driven annealing.

Lipid rafts or lipid nano-domains have remained controversial, largely due to the difficulty observing them in vivo with a method that does not potentially influence the outcome, and a consensus regarding their length- and time-scales has not yet been established28,29. However, it is clear that natural lipid membrane extracts can phase separate even in the absence of other cellular components (e.g., cytoskeleton) in mammalian56,57, and fungal/yeast cells as shown here. It has been proposed that the biological relevance of lipid domains may be that of functional platforms for protein recruitment and signaling. In fact, the sphingolipid enriched Lβ domains and ergosterol enriched Lo domains in yeast have been suggested as the lipid domain counterparts to the already characterized yeast protein domains, MCC and MCP, membrane compartments containing Can1p and Pma1p proteins, respectively45. Integral membrane proteins, lipid anchored, and peripheral membrane proteins show phase preference for lipid domains56,57,58,59,60. Here we show, using single molecule HS-AFM imaging under close-to-physiological conditions, that septin, as a cytoplasmic peripheral membrane protein, binds selectively and exclusively to one lipid phase, Ld, in a natural lipid extract, and avoids the other phases. This not only provides new insights into septin-membrane interactions, but further evidence for the biological relevance of phase separated lipid domains in non-mammalian fungal yeast membranes.

In addition, septin molecules are curvature sensitive61,62. Interestingly, lipid domains are also curvature sensitive with Ld domains having a lower bending modulus and localizing at higher curvature63,64,65. Considering these two facts along with septin’s affinity for the Ld phase, there may be a synergistic role for curvature and phase behavior in the localized recruitment of septins, for example at the cytokinetic furrow. Additionally, considering our observation of three phases in yeast membranes, it is possible that not only does the Ld phase locate to the curved furrow and recruit septin, but simultaneously ordered/gel domains nearby act as diffusion barriers across the furrow. Further experiments are needed to confirm these hypotheses66.

Septins binding to yeast membranes is ionic strength dependent

Septin binding to yeast membranes is favored at low and disfavored at high monovalent ion concentration, with a characteristic critical salt concentration of ~200 mM at which binding/unbinding energetics inverse rather steeply (Fig. 4a). We hypothesize that the septin binding/unbinding inversion at ~200 mM monovalent ion concentration may represent a means for cells to manipulate septin-dependent membrane scaffolding. We hypothesize that the opening of ion channels, such as gated- or mechano-sensitive channels, may represent local ion sources in the membrane67, which could in turn modulate septin nucleation and/or polymerization on the membrane in a signaling-dependent manner. Alternatively, cell-cycle-dependent post-translational modifications, such as phosphorylation, could regulate septin-membrane interaction and septin polymerization68. These two hypotheses are not mutually exclusive. We also found that preformed ordered assemblies are more stable and remained membrane-bound at concentrations at which septins do not de-novo polymerize, indicative of binding avidity of the many rods within a filament or even between densely packed ordered filaments.

Septin assembly templating and higher-order structures

The HS-AFM data of septin on the yeast membranes showed that septin filaments ordered and templated neighbors laterally (Figs. 1, 3, and 4), longitudinally (Fig. 2) and vertically (Fig. 5). Using HS-AFM force-sweep experiments we showed that higher order filament organization could recover after destruction: The filaments regained similar order as before the manipulation, likely due to templating by the remaining environmental filaments. It is possible that by bundling together in ultrastructures, septins may gain increased mechanical stability needed to form rigid structures that act as scaffolds in various cell types2,24,69. Interestingly, septin filament templating and septin membrane phase selectivity may have synergistic roles: It has been suggested that ordered gel domains are required in yeast to establish diffusion barriers at the mother bud neck45,70, and we observe septins are mechanically adaptable in that they can curve around domain boundaries (Fig. 1e, i). The alignment of septins along phase boundaries is especially interesting in the context of previous work in Candida albicans and Ustilago maydis, in both of these fungal cells, partial localization of septins with cholesterol-rich domains was observed71,72. In particular, filipin staining in Candida albicans was found to reach from the growing hyphal tip but stopped at the edge of the septin ring. It was not until septa-formation that partial colocalization was observed71. These data, coupled with our high-resolution imaging of septins on phase-separated lipid domains, indicate that septin localization at the edge of ordered domains may in part be driven by membrane properties at the phase boundary. While in our studies lipid domains formed independently of septins, other work suggests that membrane-binding proteins may reduce the line tension at phase boundaries, thus increasing favorability of domain formation73. This could be especially relevant in cell systems that showed that septin localization and neighboring cholesterol-rich domain formation were co-dependent72.

Interestingly, we found that on near-native lipid SLBs, septins arranged as individual, well-aligned filaments, without any sign of pairing into filament dimers. This is in stark contrast to our own previous findings on mica16, and other electron microscopy studies23,27,41,42, where pH-dependent, pairing of septin filaments was reported. Septins are proposed to bind to (curved) membranes by means of their amphipathic helix domain that inserts into lipid packing defects. Lipid packing defects occur preferentially as membrane curvature becomes more convex but are also lipid composition dependent55,74. Previous work has proposed that septin CTEs are able to form coiled-coils both within the heterooligomer and possibly between heterooligomers27,75,76, while the CTEs have also been shown to be involved in lipid binding16. Combined, these data suggest that yeast septin pairing may occur more readily when septins are not directly bound to a lipid membrane23, and that the two possible interactions may be competitive and regulatory.

Altogether, we report using HS-AFM imaging at single septin filament resolution on yeast lipid bilayers, that septin polymerizes selectively on liquid disordered phase lipids and its structure contains the blueprint for lateral, longitudinal, and vertical self-organization. Further studies on non-flat bilayer bottom-up systems77 are needed to assess how membrane curvature interplays with lipid phases, septin-membrane interactions, and higher-order structure formation.

Methods

Lipids

Yeast polar extract was purchased in chloroform from Avanti Polar Lipids (Alabaster, AL, catalogue #190001), it is a lipid mixture extracted from Saccharomyces cerevisiae cells, derived from the total lipid extract by precipitation with acetone followed by extraction of the acetone insoluble material with diethyl ether. 400 µg aliquots were dried under nitrogen and in vacuum overnight. The lipid film was hydrated with buffer (20 mM HEPES, pH 7.5, 150 mM KCl) to 0.5 mg/mL, vortexed for 2 min, and then heated at 60 °C for 1 h with intermittent vortexing to result in a cloudy suspension of multilamellar vesicles (MLVs). This suspension was tip sonicated (QSonica Q500) for 2 min, using sixty 1 s pulses interspersed by 1 s without sonication, at 20% sonication power to yield small unilamellar vesicles (SUVs).

Septin expression and purification

Septins from budding yeast Saccharomyces cerevisiae were recombinantly-expressed in BL21 (DE3) Escherichia coli cells, and purified as described16. Septin hetero-octameric rods (Cdc11-Cdc12-Cdc3-Cdc10-Cdc10-Cdc3-Cdc12-Cdc11) were stored at −80 °C in storage buffer (50 mM HEPES, pH 7.5, 300 mM KCl, 1 mM β-mercaptoethanol, 5% glycerol).

Sample preparation for HS-AFM

Septins were formed on mica by incubating 1 uL of 1 uM septin solution for 30 s at desired buffer conditions, e.g., 20 mM HEPES, pH 7.5, 600 mM KCl or 20 mM HEPES, pH 6, 50 mM KCl. Mica is an atomically flat silicate mineral commonly used as sample support for AFM experiments. Supported lipid bilayers (SLBs) were formed by incubating 1 uL of 0.5 mg/mL SUVs in 20 mM HEPES, pH 7.5, 150 mM KCl on a freshly cleaved 1.5 mm diameter mica disk for 30 min. The SLB was rinsed to replace buffer and dilute out SUVs (100 × dilution, twice). Frozen septin sample in storage buffer was thawed and diluted to desired concentration with desired buffer. Following bilayer formation, 1 uL septin was added to SLB buffer to give a final septin concentration of 1.2 uM and final buffer conditions 190 mM KCl, pH 6. Septin was incubated on the SLB for 30 min (1.2 uM for 30 min was standard, 0.6 uM for 2 min was also sufficient for binding at suitable buffer conditions). Incubation buffer was replaced with an imaging buffer by rinsing to dilute out bulk septin concentration (100 × dilution, twice). Sample was then further diluted 100 × when placed into the fluid chamber (containing imaging buffer) of the HS-AFM. Standard imaging buffer for septin filaments on yeast polar extract SLBs was 20 mM HEPES, pH 6, 50 mM KCl, which enabled best high-resolution HS-AFM imaging of stable filaments. For HS-AFM titration experiments, fluid chamber imaging buffer was replaced with higher salt imaging buffer sequentially, ensuring sample and AFM probe did not dry out.

HS-AFM

HS-AFM imaging was performed using SS-NEX HS-AFMs (RIBM, Tsukuba, Japan). Imaging was performed in amplitude modulation mode using USC-F1.2-k0.15 probes (Nanoworld, Neuchatel, Switzerland), with a nominal spring constant (k) of 0.15 N/m, nominal dimensions of 8 × 2 × 0.08 µm and a resonance frequency of ~600 kHz in liquid. To ensure high-resolution imaging, probes were cleaned using a plasma cleaner (South Bay Technology PE 2000) prior to use. The standard HS-AFM scanning speed was 1 frame/s. For high resolution filament imaging on membranes slower scan speed was used (0.25 frame/s) to reduce lateral forces from the tip on sample. HS-AFM force-sweep experiments were performed by controlled increase of the force applied by the HS-AFM tip on the sample to selectively remove areas of septin protein from the membrane. Force-sweep experiments were performed in either HS-AFM line scanning mode (HS-AFM-LS)78,79 where the force applied to the HS-AFM probe was increased while scanning in 1D forwards and backwards along the same line, or in HS-AFM imaging mode where the increased force is applied to the entire imaging area.

Image analysis

Images were first processed using in-house ImageJ plugins for drift correction and image plane flattening. A typical movie was processed using a 1st or 2nd order polynomial subtraction, followed by a line-by-line median subtraction. Time averaged frames were calculated from aligned image stacks by averaging the z-height in every pixel over multiple frames (as indicated in the figure captions). This was performed to increase the signal to noise ratio of the filaments for analysis. Septin inter-filament distances were measured using line section profiles perpendicular to the filament long axis. Septin heights were measured using section profiles, or from subsequently recorded frames before and after scratching the protein away and measuring the height differences. Bilayer heights were measured using either histograms of the pixel height across entire images, or by line section analysis perpendicular to a height step. Some of the absolute height measurements (Table 1) were calculated by adding measured heights (with propagated uncertainties) from images comprising the three lipid phases heights data and images with mica defects. Surface coverage was calculated by setting a binary height threshold between septin and lipid bilayer surface heights, and then calculating the percentage of bilayer area covered by protein. The coverage was measured over the first 10 frames at minimal force from 3 separate bilayer areas for each separate incubation experiment at different salt concentrations.

For image entropy analysis, we first applied a threshold to isolate the image regions that were covered by either septin filaments or disordered septins. Second, the local entropy value was calculated over a 9 × 9-pixel kernel around each pixel in the image. The entropy was computed as \(-\sum (p\log 2(p))\), where p is the normalized histogram counts of the height pixel values. Finally, the average of the local entropy values was determined over all septin filament regions in all frames.

Reporting summary

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