Multivalent interactions of Septin 6 promote the establishment of epithelial cell polarity

Abstract

Septins, components of the fourth cytoskeleton, play an indispensable role in establishing and maintaining epithelial cell polarity. However, the molecular mechanisms underlying the dynamic assembly of higher-order septin structures and the establishment of epithelial cell polarity remain elusive. Here, we show that septins form a previously unrecognized dynamic structure with liquid-like properties in polarized Madin–Darby canine kidney cells. We identified Septin 6 (SEPT6) as the key human septin that undergoes liquid–liquid phase separation (LLPS) both in vitro and in vivo through weak, multivalent interactions mediated by its C-terminal tail. SEPT6 mutants defective in LLPS in vitro also fail to support adherens junction integrity and cell polarity establishment in 2D and 3D cell cultures. Our findings indicate that weak, multivalent interactions are essential for the assembly of higher-order septin structures in cells. We propose that these interactions, in conjunction with conventional interactions between folded domains, generate partially ordered septin assemblies that support the apical–basal axis and lumen formation in metazoans.

Introduction

Epithelia are the fundamental components of numerous tissues and organs in metazoans, serving to separate the internal environment from the outside world, which is vital for animal physiology. Epithelial cells must establish polarity along an apical–basal axis to perform essential vectorial transport functions such as absorption and secretion (Yao and Forte, 2003; Caudron and Barral, 2009; Yao and Smolka, 2019). The establishment of epithelial cell polarity relies on a finely tuned interplay between adherens junctions, tight junctions, cell polarity complexes, and polarized cytoskeletal filaments (Chua et al., 2009; Mavrakis et al., 2009; Bridges et al., 2014). Disruption of apical–basal polarity is directly associated with various human diseases, including developmental diseases and cancer (Yao, 2020; Buckley and St Johnston, 2022). Despite its importance, the mechanisms underlying the establishment and maintenance of epithelial cell polarity remain not fully understood.

The septin cytoskeleton, often referred to as the fourth cytoskeleton, is composed of septin proteins, a family of highly conserved guanosine-5′-triphosphate (GTP)-binding proteins in eukaryotes (Vrabioiu and Mitchison, 2006; Mostowy and Cossart, 2012; Mendonca et al., 2021). The human genome encodes 13 septins (SEPT1–SEPT12 and SEPT14), which can be classified into four subgroups, i.e. SEPT2, SEPT3, Septin 6 (SEPT6), and SEPT7, based on sequence similarity and coiled-coil domain organization. Structural analyses have elucidated how the folded GTPase domains in different septins interact to form hetero-oligomeric building blocks and how these units polymerize into macroscopic filaments (Mostowy and Cossart, 2012; Mendonca et al., 2021). However, conventional structural analysis may miss interactions involving unstructured domains.

SEPT2, SEPT6, SEPT7, and SEPT9 are ubiquitously expressed across metazoans and assembled into the core hetero-oligomeric complexes and higher-order structures (filaments, rings, bundles, etc.) (Mostowy and Cossart, 2012; Martins et al., 2023). Septins and their higher-order structures play important roles in a wide array of cellular processes. Our earlier studies demonstrated that septins orchestrate chromosome alignment and cytokinesis during mammalian and yeast cell division (Zhu et al., 2008; Zheng et al., 2018). In budding yeast, recent studies have shown that septins compete with the endoplasmic reticulum for the plasma membrane, assembling a septin collar at the bud site necessary for successful cell division (Sugiyama and Kono, 2024). In polarized epithelial cells, septins provide directional cues for the polarization of the epithelial microtubule network and interact with the catenin complex to separate the apical membrane from the basal membrane (Spiliotis et al., 2008; Bowen et al., 2011; Wang et al., 2021). Additionally, recent observations revealed that the septin cytoskeleton is highly dynamic, promoting cell survival by facilitating blebbing and assembling cancer-causing signaling hubs (Weems et al., 2023). The involvement of septins in many dynamic biological processes suggests that they must undergo rapid assembly and disassembly, indicating intrinsic dynamic properties in cells. In line with this, exposure to forchlorfenuron, a specific small-molecule inhibitor of septin organization, has been shown to reduce the mobility of SEPT2 in Madin–Darby canine kidney (MDCK) cells (Hu et al., 2008). More recently, Princen et al. (2024) identified a class of compounds called ReS19-T that are able to inhibit pathological activation of store-operated calcium channels by stabilizing septin higher-order structures and restraining calcium entry into the neurons through these channels. However, the mechanisms underlying the dynamic assembly and disassembly of these crucial subcellular structures of septin remain poorly understood.

SEPT6, unlike other conventional septins, cannot hydrolyze GTP due to the absence of a crucial threonine residue (Thr78) (Sirajuddin et al., 2009; Mostowy and Cossart, 2012). Additionally, it contains regions of unknown function that are predicted to be disordered. Disordered regions or proteins can engage in weak, multivalent interactions, which are typically missed by atomic resolution structural analyses due to their lack of a specific fold, and are crucial for assembly, regulation, and function. At sufficient concentrations, these interactions drive liquid–liquid phase separation (LLPS) (Brangwynne et al., 2009; Li et al., 2012). In extreme cases, LLPS can generate membraneless compartments or organelles that orchestrate local biochemistry (Larson et al., 2017; Strom et al., 2017; Wang et al., 2018; Liu et al., 2020; Novo et al., 2022; Song et al., 2023). More commonly, LLPS reflects a tendency of the protein or complex to assemble into dynamic higher-order complexes (King and Petry, 2020; Yan et al., 2022; Yang et al., 2022). SEPT6 was observed to show a punctate distribution in neurons and hematopoietic stem cells (Cho et al., 2011; Senger et al., 2017). However, such punctum-like structures have not been characterized experimentally. Based on sequence analysis, we hypothesized that the disordered domains of SEPT6 might facilitate weak, polyvalent interactions crucial for assembly and regulation.

In this work, we discovered that septin puncta, distinct from previously observed septin filaments, are dynamic structures in polarized MDCK cells. These puncta can rapidly split and fuse, demonstrating liquid-like properties, and can also sprout filaments, suggesting a nucleation role. Furthermore, we showed that the LLPS of SEPT6, mediated by multivalent interactions within the C-terminal extension, may drive the dynamics of septin puncta. Loss of LLPS property resulted in the failure to establish epithelial cell polarity. Together, our findings uncover a previously unrecognized role of multivalent interactions in regulating the assembly of higher-order septin structures and the establishment of epithelial cell polarity.

Results

Septin puncta are dynamic and exhibit liquid-like properties in MDCK cells

To investigate the dynamics of septin structures, SEPT6, a component of septins, was tagged with green fluorescent protein (GFP) at its N-terminus and stably expressed in SEPT6 KO MDCK cells where endogenous SEPT6 had been knocked out using CRISPR/Cas9. Western blotting confirmed the absence of endogenous SEPT6 in the GFP-SEPT6 stable cell line and the comparable expression level of exogenous GFP-SEPT6 to that of endogenous SEPT6 in wild-type (WT) MDCK cells (Figure 1A). The GFP-SEPT6 stable cell line was cultured in 2D to form cell junctions between the cells. In this case, monolayer MDCK cells established an obvious apical–basolateral regional delineation (Figure 1B). Immunofluorescence imaging by confocal microscopy verified that GFP-SEPT6 localized normally to cell junctions, like endogenous SEPT6 (Figure 1C).

Figure 1.

Septin puncta exhibit liquid-like properties in MDCK cells. (A) Western blot analysis of SEPT6 expression in three indicated MDCK cell lines. Endogenous SEPT6 (endo-SEPT6) and GFP-SEPT6 were blotted using anti-SEPT6 antibody. GAPDH was blotted as a loading control. (B) Schematic diagram of GFP-SEPT6-expressing MDCK cell line in 2D culture. GFP-SEPT6 (green) is mainly distributed underneath the basal membrane and lateral membrane. Thus, the baso-layer and mid-layer were selected for imaging. (C) Representative images showing that endogenous SEPT6 specifically localizes at cell junctions, while no SEPT6 signal is detected in SEPT6 KO cells, and stable expression of SEPT6 tagged with GFP at N-terminus, but not C-terminus (not shown), in endo-SEPT6 KO cells restores its localization at cell junctions. Cells were fixed, permeabilized, and stained for F-actin (red) and DNA (blue, using DAPI). Endogenous SEPT6 was visualized by immunofluorescence using anti-SEPT6 antibody and GFP-SEPT6 was visualized by direct GFP fluorescence (green). Scale bar, 10 μm. (D) Representative multi-SIM images showing septin higher-order structures, including puncta (arrows) and filaments. Scale bar, 10 μm (‘MERGE’ images) and 2 μm (other images). (E) Live-cell imaging using Lattice SIM showing the high dynamics of septin puncta (arrows). Septin filaments are relatively static. Time is indicated in seconds. Scale bar, 1 μm. (F and G) Anti-GFP IP showing hetero-oligomer complex formation between GFP-SEPT6 and SEPT2, SEPT7, and SEPT9 in MDCK cells. (G) Mass spectrometry identifies all components of the hetero-oligomer complex, including SEPT9, SEPT7, SEPT6, and SEPT2, in the immunoprecipitates. (H) Schematic illustration of the septin hetero-octamer, consisting of two copies of SEPT2, SEPT6, SEPT7, and SEPT9 (Iv et al., 2021; Martins et al., 2023). (I) Schematic illustration of different septin higher-order structures based on peers’ work and this study.

To better observe the morphological characteristics of septins in epithelial cells, a multimodal structured illumination microscope (multi-SIM) was employed. SEPT6 was primarily detected underneath the basolateral membrane of polarized epithelial cells. Specifically, SEPT6 localized to cell junctions in the middle layer (mid-layer) and formed fibrous structures in the basal layer (baso-layer), dependent on stress fiber (Figure 1D), consistent with previous reports (Kinoshita et al., 2002; Spiliotis and Nakos, 2021; Szuba et al., 2021; Martins et al., 2023). Strikingly, punctum-like structures were observed in both the mid-layer and baso-layer, independent of the actin filaments (Figure 1D). Continuous scanning across the z-axis confirmed that the puncta exhibit a globular morphology in 3D (Supplementary Movies S1 and S2). Time-lapse imaging revealed that while septin filaments were relatively stable, septin puncta were highly dynamic. Intriguingly, septin filaments were observed to sprout and split to form the puncta that split and fuse rapidly (Figure 1E; Supplementary Movies S3–S6). These results suggest that septin puncta, unlike septin filaments, are highly dynamic structures.

To further explore the composition of septin structures, immunoprecipitation (IP) experiments were performed (Figure 1F). Mass spectrometric analysis of the immunoprecipitates indicated that SEPT2, SEPT6, SEPT7, and SEPT9 were all present (Figure 1G), suggesting that septin structures in MDCK cells are assembled by octamers as the basic unit (Figure 1H). We postulate that septins form dynamic puncta in MDCK cells, serving as seeds for the assembly of higher-order structures and/or a dynamic storage form of septin hetero-oligomers (Figure 1I).

LLPS of SEPT6 may be responsible for the dynamics of septin puncta

Given that septin puncta exhibit liquid-like behavior, we then explored which component contributes to these liquid-like properties. SEPT6 forms hetero-oligomers with SEPT2, SEPT7, and SEPT9, which can further assemble into higher-order structures, including filaments and rings (Sirajuddin et al., 2007; Soroor et al., 2021; Spiliotis and Nakos, 2021; Martins et al., 2023). To assess the LLPS potential of different septins, we employed the light-activated optoDroplet strategy, where a light-inducible oligomerization domain, Cry2 from Arabidopsis cryptochrome 2, is fused with a fluorescent tag mCherry to proteins of interest to facilitate protein condensation and monitoring (Shin et al., 2017). The septin-mCherry-Cry2 fusions were constructed (Figure 2A) and transiently expressed in COS-7 cells. Upon exposure to blue light, Opto-SEPT6 rapidly formed condensates, whereas Opto-SEPT2 and Opto-SEPT9 remained diffused. Opto-SEPT7 spontaneously aggregated but did not change upon blue light exposure, indicating its solid nature (Figure 2B). We also examined other members of the SEPT6 subgroup (Supplementary Figure S1A). Opto-SEPT8 remained diffused under blue light, while Opto-SEPT10 and Opto-SEPT11 formed condensates much more slowly and less frequently (Supplementary Figure S1B).

Figure 2.

SEPT6 undergoes LLPS in vivo and in vitro. (A) Schematic diagram of the optoDroplet constructs, including Opto-SEPT2, Opto-SEPT6, Opto-SEPT7, and Opto-SEPT9. Upon blue light exposure, Cry2 oligomerizes to facilitate the fusion proteins to form condensates and hence allows assessing the LLPS potential of proteins of interest. Fluorescent protein mCherry (mCh) was included for visualization. (B) Representative images showing the rapid condensate formation of Opto-SEPT6, but not Opto-SEPT2, Opto-SEPT7, or Opto-SEPT9, upon blue light exposure. COS-7 cells transiently transfected with Opto-tagged septins were exposed to blue light in the middle of imaging to initiate the activation of Opto. Time is indicated in seconds. Scale bar, 10 μm. (C and D) Representative images showing droplet formation and quantification of droplet size of His-GFP-SEPT6 (30 μM) in buffers with indicated salt concentrations. Scale bar, 5 μm. n = 87 (150 mM KCl), 109 (300 mM KCl), 110 (700 mM KCl), and 110 (1000 mM KCl)) droplets, respectively. Data represent mean ± standard error of the mean (SEM). Ordinary one-way ANOVA followed by Tukey's post hoc test. ****P < 0.0001. (E–G) His-GFP-SEPT6 droplets were generated with 30 μM His-GFP-SEPT6 in the presence of 150 mM KCl. (E) Time-lapse imaging showing rapid droplet fusion. (F and G) Time-lapse imaging and quantification of GFP intensity showing rapid fluorescence recovery of the droplets following photobleaching. Time is indicated in seconds. Scale bar, 2 μm. Data were presented as mean ± SEM at each timepoint from three independent repeats. (H) Representative immunofluorescent images of MDCK monolayers transiently transfected to express GFP-tagged SEPT2, SEPT6, SEPT7, or SEPT9, respectively, showing droplet formation of GFP-SEPT6 in cells. Scale bar, 10 μm. (I) Representative immunofluorescent images showing that the stably expressed GFP-SEPT6 in endo-SEPT6 KO cells co-localizes with F-actin, while the transiently overexpressed GFP-SEPT6 forms puncta. Scale bar, 10 μm. (J) Anti-GFP IP showing that the stably (S) expressed GFP-SEPT6 in endo-SEPT6 KO cells robustly forms septin hetero-oligomeric complex with other septins, whereas the transiently (T) overexpressed GFP-SEPT6 did not.

Next, we performed in vitro phase separation experiments using purified septin proteins tagged with His-GFP. Among all the above-mentioned septins, SEPT6 demonstrated the greatest ability to form droplets in a physiological buffer (Supplementary Figure S1C–E). We further characterized the LLPS property of SEPT6 in vitro and found that the number and size of SEPT6 droplets increased with higher protein concentrations (Supplementary Figure S1F and G). These droplets gradually dissolved as the potassium salt concentration increased from 150 mM to 1 M, suggesting that electrostatic interactions drive SEPT6 droplet formation (Figure 2C and D). The dynamic nature of these droplets was confirmed by fusion events (Figure 2E). Quick recovery of GFP signal intensity after bleaching in fluorescence recovery after photobleaching (FRAP) experiments revealed the fluidity inside the SEPT6 droplets (Figure 2F and G). These results indicate that SEPT6 has an intrinsic LLPS property, which likely drives its liquid-like behaviors in cells.

To examine the condensation ability of individual septins, GFP-tagged SEPT2, SEPT6, SEPT7, and SEPT9 were transiently expressed in MDCK cells. Immunostaining revealed distinct localization patterns among these septins. SEPT2 and SEPT9 were diffused, SEPT7 formed irregular aggregates, and only SEPT6 formed droplet-like puncta in cells (Figure 2H; Supplementary Figure S1H), suggesting that, when overexpressed alone, SEPT7 forms solid-like aggregates, while SEPT6 spontaneously forms liquid-like ‘droplets’ in cells.

Stable expression of GFP-SEPT6, following the knockout of endogenous SEPT6, resulted in a co-localization with F-actin at cell junctions (Figure 1C and D). In contrast, transiently overexpressed GFP-SEPT6 predominantly formed cytosolic puncta in MDCK cells (Figure 2I). To compare the molecular compositions of these two SEPT6 structures, we performed IP experiments. Stably expressed GFP-SEPT6 robustly formed complexes with other endogenous septins, while transiently overexpressed SEPT6 exhibited a stronger tendency to form puncta and showed weaker associations with other septins, likely due to insufficient time for integration into septin hetero-oligomers (Figure 2J). Based on these observations, we hereafter utilized stably expressed GFP-SEPT6 for functional assays and transiently overexpressed SEPT6 for its LLPS ability assessment. We further investigated whether SEPT6 droplets could recruit other septins by fluorescent microscopy. We co-transfected MDCK cells with GFP-SEPT6 and mCherry-tagged SEPT2, SEPT7, or SEPT9 and found that SEPT6 droplets recruited SEPT2, SEPT7, and SEPT9, respectively (Supplementary Figure S1I). Additionally, we performed in vitro LLPS experiments using purified proteins. His-GFP-tagged SEPT6 formed co-condensates with His-mCherry-tagged SEPT2, SEPT7, and SEPT9, respectively (Supplementary Figure S1J). These findings suggest that SEPT6 drives the condensation of septins.

The C-terminal extension of SEPT6 mediates its LLPS

To determine the region within SEPT6 responsible for its LLPS property, we generated four GFP-tagged truncations, i.e. SEPT6 N (1–38 aa), SEPT6 M (39–305 aa), SEPT6 C (306–434 aa), and SEPT6 N+M (1–305 aa), based on the overall structure of SEPT6 (Sirajuddin et al., 2007) for the optoDroplet assay (Figure 3A). The N-terminus and C-terminus of SEPT6 are variable, while the M domain contains septin unique elements and binds to GTP. The C-terminal region of SEPT6, but not N or M domains, rapidly formed detectable droplets upon blue light exposure (Figure 3B). To further validate this, we expressed and purified His-tagged GFP-SEPT6 N, M, C, and N+M and tested their ability to phase separate in vitro. At a concentration of 40 μM in the physiological salt concentration, the C-terminal region of SEPT6 formed numerous large droplets rapidly, while GFP-SEPT6 M formed very few and small droplets, and GFP-SEPT6 N was completely soluble (Figure 3C and D; Supplementary Figure S2A). The ability of the SEPT6 C-terminal region to phase separate was also dependent on protein concentration (Supplementary Figure S2B and C).

Figure 3.

The IDRs of SEPT6 mediate its LLPS. (A) Schematic diagram of the optoDroplet constructs of SEPT6 truncations. (B) Opto-SEPT6 C-terminal domain forms droplets rapidly upon blue light exposure in COS-7 cells, while no droplet is detected for the N-terminal domain, M domain, or N+M domain. Time is indicated in seconds. Scale bar, 10 μm. (C and D) Representative images showing the droplet formation and quantification of droplet size of four His-GFP-tagged SEPT6 truncations at 40 μM in the presence of 150 mM KCl. Only the C-terminal domain of SEPT6 is able to form droplets. Scale bar, 10 μm. n = 57 (SEPT6 N), 144 (SEPT6 M), 141 (SEPT6 C), and 144 (SEPT6 N+M) droplets, respectively. Data represent mean ± SEM. Ordinary one-way ANOVA followed by Tukey's post hoc test. ****P < 0.0001. (E) Domain organization (UniProt) and IDR prediction (PONDR VL-XT) of SEPT6. The three disordered regions in the C-terminal extension are designated as R1, R2, and R3. (F) Schematic diagram of the SEPT6 C mutants depleted of R1, R2, or R3, respectively. (G and H) Representative images showing droplet formation and quantification of droplet size of 40 μM His-GFP-tagged SEPT6 C and mutants in the presence of 150 mM KCl. Deletion of either IDR disrupts the droplet formation of the C-terminus. Scale bar, 10 μm. n = 187 (SEPT6C-WT), 129 (SEPT6C-ΔR1), 111 (SEPT6C-ΔR2), and 187 (SEPT6C-ΔR3) droplets, respectively. Data represent mean ± SEM. Ordinary one-way ANOVA followed by Tukey's post hoc test. ****P < 0.0001. (I–K) SEPT6 C-terminus replacement with FUS IDR partially rescues SEPT6’s condensation and localization. (I) Schematic diagram of SEPT6 constructs with the C-terminal domain replaced by the IDR from TDP43 or FUS. (J) Representative immunofluorescent images showing droplet formation of transiently overexpressed GFP-tagged SEPT6 WT, SEPT6-TDP43 IDR, or SEPT6-FUS IDR in MDCK cells. Scale bar, 10 μm. (K) Representative images showing the localization of stably expressed GFP-tagged SEPT6 WT, SEPT6-TDP43 IDR, and SEPT6-FUS IDR in endo-SEPT6 KO MDCK cells. Arrows indicate cell junctions. Scale bar, 10 μm.

To examine whether the C-terminus of SEPT6 contributes to the hetero-oligomeric unit formation, SEPT6 or SEPT6-ΔC was co-expressed with His-SEPT2 and SEPT7-FLAG in Escherichia coli and purified by two sequential steps with Ni-NTA and anti-FLAG antibody-conjugated agarose beads, respectively (Supplementary Figure S2D). SEPT9 was excluded from this assay, since SEPT2, SEPT6, and SEPT7 were sufficient to form hetero-oligomer units and higher-order structures (Mendonca et al., 2021; Soroor et al., 2021; Szuba et al., 2021). The results showed that SEPT6 could form complexes with SEPT2 and SEPT7 no matter whether its C-terminal extension was present or not (Supplementary Figure S2E; Sirajuddin et al., 2007). Thus, the C-terminal extension of SEPT6 does not participate in septin oligomer formation but is a major region contributing to the LLPS of SEPT6 both in vitro and in vivo.

To learn more about the chemical determinants of LLPS within the C-terminal domain of SEPT6, we analyzed the protein sequence. The prediction with IUPred2 revealed three small intrinsically disordered regions (IDRs) in the C-terminal region of SEPT6: R1 (335–352 aa), R2 (369–399 aa), and R3 (408–427 aa) (Figure 3E; http://pondr.com). IDRs have been considered key drivers for the LLPS of proteins (Liu et al., 2020). We thus transiently expressed GFP-SEPT6 lacking R1, R2, or R3 (∆R1, ∆R2, or ∆R3, respectively) in MDCK cells and assessed the ability to form droplet-like puncta. Immunofluorescence imaging showed that GFP-SEPT6 lacking either R1 or R2 became completely soluble, while GFP-SEPT6 without R3 still formed a small number of puncta in cells (Supplementary Figure S2F–H). We also constructed His-tagged GFP-SEPT6 C (His-GFP-SEPT6C) lacking either IDR and tested the ability to phase separate in vitro (Figure 3F). Loss of either R1 or R2 prevented SEPT6 from forming any droplets, while SEPT6 lacking R3 retains some condensation ability (Figure 3G and H; Supplementary Figure S2I). These observations indicate that IDRs in the C-terminal extension of SEPT6 mediate its LLPS. However, purified R1, R2, or R3 alone could not phase separate in vitro (Supplementary Figure S2J–L), suggesting that these regions contribute to SEPT6 phase separation collectively.

To confirm that the LLPS property, but not just the specific amino acid sequence, of the C-terminus of SEPT6 contributes to septin assembly and function, we replaced the C-terminus of SEPT6 with the IDRs of TDP43 and FUS (Figure 3I), which are well-studied and exhibit distinct LLPS abilities (Wang et al., 2018). Transient expression of these SEPT6 variants demonstrated that fusions with the IDRs from TDP43 and FUS could partially restore the LLPS properties of SEPT6 ∆C (i.e. N+M), with FUS IDR showing a higher rescue efficiency (Figure 3J). Consistently, stable expression of these SEPT6 variants in endo-SEPT6 KO cells revealed that, while TDP43 IDR failed to restore localization, FUS IDR partially rescued the localization to cell junctions (Figure 3K). The differences in rescue efficiency likely reflect the varying LLPS capacities of the IDRs from FUS and TDP43 (Wang et al., 2018). These findings reinforce that while the IDR of SEPT6 may have additional roles, its ability to undergo LLPS plays a crucial role in septin assembly and function (see also below).

Lysine residues in R2 are the major drivers for SEPT6 LLPS

To identify the specific amino acids within the IDRs of SEPT6 responsible for LLPS, we carefully analyzed the residue composition of R1, R2, and R3 in the C-terminus of SEPT6. We found an enrichment of charged residues, particularly lysine: three in R1 (K337, K338, and K351), nine in R2 (K372, K373, K379, K380, K381, K385, K386, K387, and K397), and four in R3 (K420, K423, K425, and K426). Based on the number of lysine residues in each region, we designated these regions as 3K, 9K, and 4K, respectively (Figure 4A). Given that the condensation of SEPT6 is highly sensitive to high salt concentrations (Figure 2C and D), we speculated that electrostatic interactions mediated by these positively charged lysine residues play a key role in the LLPS of SEPT6. We mutated these lysine residues to negatively charged glutamate residues in each region (3KE, 9KE, and 4KE, respectively) and tested the ability to form droplets in cells and in vitro. The expression levels of these GFP-tagged SEPT6 mutants were comparable in MDCK cells (Supplementary Figure S3A). While K-to-E mutations in R1 or R3 slightly reduced the droplet formation of SEPT6, the 9KE mutation in R2 completely abolished droplet formation (Figure 4B and C). Similarly, His-GFP-SEPT6C with K-to-E mutations in R1 or R3 still formed few and relatively small droplets in vitro, whereas no droplets were detected with K-to-E mutations in R2 (Figure 4D and E; Supplementary Figure S3B). Replacing the lysine residues with uncharged glutamine residues (K-to-Q) produced similar results (Supplementary Figure S3C–E).

Figure 4.

Lysine residues drive the LLPS of SEPT6. (A) Schematic diagram of SEPT6 C-terminal domain showing the distribution of lysine residues in three IDRs. (B and C) Representative immunofluorescent images and quantification of cells with puncta showing that mutation of lysine residues in R2 (9KE), but not in R1 (3KE) or R3 (4KE), abolishes the condensate formation of SEPT6 in MDCK cells. Scale bar, 10 μm. Data are shown as mean ± SEM from three repeats. Ordinary one-way ANOVA followed by Dunnett's multiple comparisons test. ****P < 0.0001; ns, not significant. (D and E) Representative images showing droplet formation and quantification of droplet size of 40 μM His-GFP-tagged SEPT6 C mutants in the presence of 150 mM KCl. Scale bar, 10 μm. n = 100 (WT), 100 (3KE), 100 (9KE), and 100 (4KE) droplets, respectively. Data represent mean ± SEM. Ordinary one-way ANOVA followed by Dunnett's multiple comparisons test. ****P < 0.0001. (F and G) Representative images showing droplet formation and quantification of droplet size of 40 μM His-GFP-SEPT6 C mutants in the presence of 150 mM KCl. Scale bar, 10 μm. n = 100 (WT), 100 (9KR), 100 (9KQ), and 100 (9KE) droplets, respectively. Data represent mean ± SEM. Ordinary one-way ANOVA followed by Dunnett's multiple comparisons test. ****P < 0.0001; **P = 0.0024.

Furthermore, we mutated nine lysine residues in R2 to positively charged arginine (9KR), uncharged glutamine (9KQ), or negatively charged glutamate (9KE) residues, respectively, and assessed the condensation ability in vitro. The 9KR mutant exhibited enhanced condensate formation compared to WT, whereas 9KQ showed reduced droplet formation and 9KE did not form condensates at all (Figure 4F and G; Supplementary Figure S3F). These findings reinforce that the positive charges of lysine residues in R2 play a major role in mediating the LLPS of SEPT6.

Considering that mutating all nine lysine residues might cause unexpected side effects, we then aimed to identify the most critical residues among them. It was recently reported that a subregion (R343–K400) in the C-terminal domain of SEPT6 forms an antiparallel coiled coil in trans (Leonardo et al., 2021). This coiled coil was stabilized by interactions between three lysine residues (K372, K379, and K386) and two glutamate residues (E354 and E361) (Supplementary Figure S4A and B). We mutated these three lysine residues to arginine (9-3KR), glutamine (9-3KQ), or glutamate (9-3KE) residues and examined the ability to form condensates in vivo and in vitro. When expressed in MDCK cells, the 9-3KR mutant formed droplets robustly, whereas 9-3KQ and 9-3KE mutants were completely dispersed in cells (Supplementary Figure S4C). Coincidently, in vitro droplet formation assays with purified mutant proteins showed that 9-3KQ and 9-3KE failed to form condensates (Supplementary  Figure S4D–F). As expected, the mutation of E354 and E361 to arginine (2ER), but not glutamine (2EQ) or aspartate (2ED), disrupted droplet formation in cells (Supplementary Figure S4G). These results suggest that weak, polyvalent interactions mediated by specific lysine and glutamate residues are likely the primary force driving the LLPS of SEPT6.

Weak, polyvalent interactions are required for higher-order assembly of septins in MDCK cells

To explore the functional importance of weak, polyvalent interactions between SEPT6 IDRs, various GFP-tagged SEPT6 mutants were stably expressed in SEPT6 KO MDCK cells. Western blot analysis confirmed that the expression levels of 9KR and 9KQ mutants were comparable to the endogenous SEPT6 level in WT MDCK cells, while the 9KE level was significantly lower (Supplementary Figure S5A). Immunofluorescence imaging showed that GFP-tagged SEPT6 mutants were recruited to cell junctions to various extents. Specifically, the 9KR mutant co-localized with actin filaments and exhibited junctional localization similar to WT, while 9KQ and 9KE showed defects in junctional localization (Figure 5A–E). The x–z-axis imaging confirmed that both 9KQ and 9KE mutants were dispersed in the cytoplasm and did not localize to cell junctions as WT and 9KR (Figure 5F). Confocal imaging from the baso-layer demonstrated that both 9KQ and 9KE mutants disrupted the higher-order septin filament assembly (Supplementary Figure S5B–F). Live-cell imaging using Lattice SIM further confirmed that neither 9KQ nor 9KE could effectively form higher-order structures in MDCK cells (Figure 5G). Similarly, the deletion of R2 disrupted the localization pattern of SEPT6 at cell junctions (Supplementary Figure S5G–J).

Figure 5.

The weak, polyvalent interactions between SEPT6 IDRs are required for higher-order assembly of septins. GFP-SEPT6 WT and 9KR, 9KQ, and 9KE mutants, respectively, were stably expressed in endo-SEPT6 KO MDCK cells. (A) Representative images showing localization of stably expressed GFP-SEPT6 WT and mutants. Scale bar, 10 μm. (B–E) GFP-SEPT6 and F-actin fluorescence intensities along the line profile in A. (F) Representative confocal cross-section images showing an x–z-axis localization of stably expressed GFP-SEPT6 WT and mutants. Scale bar, 10 μm. (G) Representative images captured using Lattice SIM showing that septins with 9KQ and 9KE mutations could not efficiently assemble to higher-order structures. Scale bar, 2 μm. (H) Anti-GFP IP showing hetero-oligomer complex formation between stably expressed GFP-SEPT6 (either WT or mutant) and SEPT2, SEPT7, and SEPT9 in endo-SEPT6 KO MDCK cells.

We then examined whether GFP-SEPT6 and its mutants form hetero-oligomers with other endogenous septins. IP assays indicated that WT and all three mutants have similar affinities for endogenous SEPT2, SEPT7, and SEPT9 (Figure 5H). Similarly, SEPT6 lacking R2 could also form complexes with other septins (Supplementary Figure S6A). Immunofluorescence imaging confirmed that SEPT7 mirrored the localization patterns of stably expressed GFP-SEPT6 and its mutants in cells lacking endogenous SEPT6 (Supplementary Figure S6B–F). These results indicate that weak, polyvalent interactions between SEPT6 IDRs are essential for the higher-order assembly of septin hetero-oligomers.

Weak, polyvalent interactions are essential for the integrity of the adherens junction and MDCK cyst morphogenesis

Our previous studies showed that septins are required for adherens junction integrity (Wang et al., 2021), which prompted us to test whether the weak, polyvalent interactions between SEPT6 IDRs contribute to adherens junction establishment and maintenance. We examined the localization of epithelial cadherin (E-cadherin), a key component of adherens junction, in MDCK cells stably expressing LLPS-defective GFP-SEPT6 mutants. Given the lower expression level of the 9KE mutant, to avoid the influence of protein level in functional assessment, we used 9KQ as the representative LLPS-defective mutant for further functional studies. In GFP-SEPT6 9KQ-expressing cells, E-cadherin was dispersed in the cytoplasm and failed to localize to adherens junction, as confirmed by x–z-axis images (Figure 6A and B). Since E-cadherin plays a fundamental role in the establishment and maintenance of cell junctions (Baum and Georgiou, 2011), these observations suggest that the weak, polyvalent interactions between SEPT6 IDRs are critical for adherens junction integrity.

Figure 6.

The weak, polyvalent interactions between SEPT6 IDRs are required for MDCK cyst morphogenesis. GFP-SEPT6 WT and GFP-SEPT6 9KQ were stably expressed in endo-SEPT6 KO MDCK cells. (A) Representative images showing that E-cadherin (red) fails to localize at cell junctions in GFP-SEPT6 9KQ-expressing cells. Scale bar, 10 μm. (B) Representative confocal cross-section images confirming that E-cadherin (red) fails to localize at cell junctions in SEPT6 9KQ-expressing cells. Scale bar, 10 μm. (C) Schematic illustration of 3D culture of MDCK stable cell lines. Gray, dark pink, and light pink indicate MDCK cells, medium-containing Matrigel, and DMEM, respectively. (D and E) Representative immunofluorescent images of 3D-cultured MDCK cysts and statistical analysis of cyst phenotypes. While the cysts formed by SEPT6 WT-expressing MDCK cells (WT cysts) are quite normal showing open lumen, those formed by SEPT6 9KQ-expressing MDCK cells (9KQ cysts) show two major defects: multilumen with roughly normal shapes and irregular cyst with collapsed shape. Scale bar, 10 μm. n = 121 (WT) and 109 (9KQ) cysts from three independent experiments, respectively. Data are presented as mean ± SEM. Two-way ANOVA followed by Sidak's multiple comparisons test. ****P < 0.0001. (F) Representative immunofluorescent images showing that E-cadherin fails to localize at cell junctions in 9KQ cysts. Scale bar, 10 μm. (G and H) GFP-SEPT6 and E-cadherin fluorescence intensities along the line profile in F.

3D culture of polarized epithelial cells serves as a powerful tool to study epithelial architecture and morphogenesis under close-to-physiological conditions (Montesano et al., 1991; Elia and Lippincott-Schwartz, 2009). Next, we investigated whether multivalent interactions between SEPT6 IDRs contribute to the epithelial apical–basal polarity establishment and MDCK cyst morphogenesis. MDCK cells stably expressing GFP-SEPT6 WT or 9KQ were cultured in 3D for 4 days to allow the establishment of cell polarity (Figure 6C). SEPT6 WT-expressing cells formed well-polarized open-lumen cysts with septins forming higher-order structures and locating to cell junctions, while cysts formed by SEPT6 9KQ-expressing cells were mostly multilumen or collapsed where SEPT6 9KQ failed to localize along cell junctions (Figure 6D and E). Moreover, E-cadherin localized to the lateral membrane in SEPT6 WT-expressing MDCK cysts, while it was dispersed within cells in SEPT6 9KQ-expressing cysts (Figure 6F–H), similar to the observations in 2D culture. These results suggest that weak, multivalent interactions between SEPT6 IDRs contribute to establishing and maintaining adherens junction integrity and MDCK cyst morphogenesis.

Discussion

The septin cytoskeleton is essential for various cellular processes, including the establishment of cell polarity, yet the mechanisms underlying its dynamic assembly remain largely unclear. In this study, we uncovered a previously unnoticed property of SEPT6: LLPS mediated by weak and polyvalent interactions between its disordered regions. We demonstrated that these weak and multivalent interactions between SEPT6 IDRs are essential for assembling higher-order septin structures, establishing the apical–basal polarity, and normal cyst morphogenesis (Figure 7). Recent studies have shown that septins polymerize into filaments primarily composed of hetero-oligomers, which can further form higher-order structures under specific conditions both in vitro (DeRose et al., 2020; Iv et al., 2021; Soroor et al., 2021; Szuba et al., 2021) and in vivo (Kinoshita et al., 2002; Bridges et al., 2014; Arbizzani et al., 2022; Ibanes et al., 2022; Martins et al., 2023). However, the state of septin hetero-oligomers before their assembly into filaments or other higher-order structures in cells remains unclear. In Schizosaccharomyces pombe, septins form spots containing highly variable numbers of certain septins, mostly exceeding hexametric and octameric complexes (Bridges et al., 2014), suggesting that a structure higher than hetero-oligomer exists in yeast cells.

Figure 7.

Proposed model of how the weak, polyvalent interactions between SEPT6 IDRs contribute to MDCK cyst morphogenesis. Under normal conditions, the weak, polyvalent interactions between the IDRs of SEPT6 WT facilitate the assembly of septin hetero-oligomers to various higher-order structures including dynamic puncta and static filaments along F-actin. These structures together support cyst morphogenesis. In contrast, septin hetero-oligomers formed by LLPS-defective SEPT6 mutants (disrupting the multivalent interactions) fail to assemble to higher-order structures to properly support the establishment of the apical–basal polarity and thus the morphogenesis of MDCK cysts.

By performing ultrahigh-resolution microscopy and time-lapse imaging, we directly observed the dynamic structures of septin puncta. These puncta exhibited liquid-like properties, able to split and fuse. Attempts to monitor the dynamics of SEPT6 expressed from its endogenous genomic locus were hindered by tagging issues (a tag fused to the C-terminus of SEPT6 disrupts its function) and the extremely high GC content near the start codon, complicating CRISPR/Cas9 editing. Notably, the expression levels of GFP-SEPT6 and its mutants in the stable cell lines, preknocked out of endogenous SEPT6, were comparable to endogenous SEPT6 levels in WT MDCK cells. GFP-SEPT6 and its mutants formed complexes with other septins normally and assembled with other septins to higher-order structures robustly (Figures 1 and 5H). Thus, we believe that the properties of stably expressed GFP-SEPT6 observed in this study could largely represent those of endogenous SEPT6.

The C-terminal domains of septins have been proposed to form cross-bridges between aligned filaments, leading to filament bundling and higher-order complexes (Low and Macara, 2006; de Almeida Marques et al., 2012; Sala et al., 2016). Cleavage of the C-terminal domain of SEPT2 by Zika virus protease caused mitotic defects in neural progenitor cells (Li et al., 2019), highlighting the functional importance of the septin C-terminal tails. Since the C-terminal region of SEPT6 carries three IDRs and there are two SEPT6 molecules in each septin oligomer, septin hetero-oligomers or short filaments could form condensates through polyvalent interactions mediated by these IDRs. Since the C-terminus of SEPT6 is not involved in septin protofilament assembly (Sirajuddin et al., 2007; Bertin et al., 2008; Szuba et al., 2021) and septin hetero-oligomers were still formed without it (Figure 5H; Supplementary Figure S2E), septin puncta likely consist of hetero-oligomers and/or short filaments. The LLPS-defective mutant of SEPT6 likely disrupts the weak interactions between SEPT6-containing hetero-oligomers and/or short filaments, resulting in defects of septin assembly and dissolving of septin hetero-oligomers in cells.

Several potential benefits are proposed by using weak, polyvalent interactions between IDRs to promote higher-order assembly and function. One advantage is the increased kinetic and structural flexibility in assembly state and function compared to conventional rigid interactions between folded domains. For instance, we observed filaments sprouting from liquid-like droplets in time-lapse sequences (Figure 1). This suggests that the liquid-like droplets promote nucleation of rigid filaments. If condensation of liquid-like droplets is faster than nucleation of rigid filaments, the droplet formation provides a kinetic advantage. Additionally, interactions between IDRs may allow for regulations by post-translational modifications. Kinases prefer unstructured substrate peptides (Yamazaki et al., 2020; Spolaore et al., 2023). Thus, the IDRs might provide a platform for kinase regulation of septin assembly into higher-order structures. Consistent with this hypothesis, most of the phosphorylation sites on SEPT6 identified by untargeted proteomics are in the C-terminal region containing the IDRs (Supplementary Figure S7). In fact, post-translational modifications indeed orchestrate septin cytoskeletal remodeling in a variety of different contexts (Marquardt et al., 2020). In the future, it would be interesting to elucidate how the weak, polyvalent interactions between IDRs benefit septin assembly, dynamics, regulations, and physiological functions (Mayr et al., 2023).

Materials and methods

Cell culture

MDCK, COS-7, and HEK293T cells, purchased from the American Type Culture Collection (ATCC), were cultured in Dulbecco's Modified Eagle Medium (DMEM) supplemented with 10% (v/v) fetal bovine serum (FBS), 100 units/ml penicillin (Gibco), and 100 μg/ml streptomycin (Gibco) at 37°C with 5% CO₂. For 2D culture, MDCK cells were cultured for one more day after reaching 100% confluence to allow the formation of cell junctions. To grow 3D cysts, MDCK cells were trypsin-digested to generate well-separated individual cells. Cells resuspended in DMEM were mixed with Matrigel (Corning) to a final concentration of 30%. The mixture was added into a 24-well plate and allowed to solidify by incubating at 37°C for 15 min, followed by overlaying with DMEM on the top. MDCK cells were cultured for 3–4 days until cysts were formed, reaching ∼30 μm in diameter.

Plasmids

GFP-tagged septins were generated by inserting SEPT2, SEPT6, SEPT7, or SEPT 9, amplified from the plasmids described previously (Wang et al., 2021), into the multiple cloning site (MCS) of the GFP-C1 vector. The pHR-mCherry-Cry2 vector plasmid was purchased from Addgene (#101221). The septin-mCherry-Cry2 fusions (Opto-SEPTs) were constructed by inserting septins before mCherry in the vector plasmid. To generate His-GFP-septins and pLVX-GFP-septins, GFP-septins were inserted into the MCS of the pET-28a vector (Novagen) and pLVX vector (Addgene), respectively. GFP-SEPT6 TDP43 IDR and GFP-SEPT6 FUS IDR were generated by replacing the C-terminal region of SEPT6 (306–434 aa) with the IDR of TDP43 (219–414 aa) and the IDR of FUS (2–214 aa), respectively. All plasmids were constructed by homologous recombination-based DNA assembly using the C214 kit purchased from Vazyme. Site-directed mutants were generated by the Mut Express Fast Mutagenesis Kit (Vazyme) following the manufacturer's instructions. All constructs were sequenced in full at Tsingke Biotechnology or General Biosystems.

Generation of stable cell lines

To create the SEPT6 KO cell line, MDCK cells were transfected with a pU6-Cas9-T2A-Puro plasmid that encodes Cas9, puromycin N-acetyltransferase, and two SEPT6-targeting single guide RNAs (5′-ATTGTTAGCACGGTGGGCTT-3′ and 5′-TCCCCCTGGCTGGACATGTG-3′). Transfected MDCK cells were treated with 1.5 μg/ml puromycin for 3 days, followed by single-cell sorting with flow cytometry. SEPT6 KO clones were tested by western blotting and confirmed by DNA sequencing. Lentivirus infection was exploited to express GFP-tagged septins (including mutants) in WT and SEPT6 KO MDCK cells. GFP-septin lentiviruses were packaged by transfecting HEK293T cells in 6-cm dishes with 10 μg of lentiviral plasmid (pLVX), 7.5 μg of packaging plasmid (psPAX2), and 3 μg of VSVG coat protein plasmid (pMD2.G) using Lipofectamine 2000 (Invitrogen). Supernatant with viruses was collected 48 h post-transfection, filtered, and stored at 4°C before use.

Antibodies

The following primary antibodies were used in this study: anti-SEPT6 antibody (Abcam, ab138036, 1:1000 for western blot, 1:200 for immunofluorescence), anti-GAPDH antibody (Proteintech, 60004-1-Ig, 1:10000 for western blot), anti-GFP antibody (Proteintech, 50430-2-AP, 1:2000 for western blot), anti-SEPT2 antibody (Abcam, ab88657, 1:1000 for western blot), anti-SEPT7 antibody (Abcam, ab18602, 1:1000 for western blot), anti-E-cadherin antibody (Cell Signaling Technology, 14472, 1:200 for immunofluorescence). All secondary antibodies were purchased from Jackson ImmunoResearch (1:500 for immunofluorescence, 1:5000 for western blot).

Immunofluorescence and live-cell imaging

2D monolayers grown on coverslips or 3D cysts grown on the live cell dishes were fixed with 3.7% paraformaldehyde diluted in phosphate-buffered saline (PBS) for 10 min, permeabilized in 0.2% Triton X-100 for 2 min, and blocked with PBST (PBS with 0.05% Tween-20) containing 1% bovine serum albumin for 40 min at room temperature. Cells were then incubated with primary antibodies for 2 h at room temperature or overnight at 4°C, followed by secondary antibodies for 1 h at room temperature. DNA was stained with 4′,6-diamidino-2 phenylindole (DAPI; Sigma) and F-actin was stained with CoraLite 594-phalloidin fluorescent dye (Proteintech). Images were collected using a Zeiss laser-scanning confocal microscope (Zeiss LSM 880 with Airyscan FAST) or a multi-SIM. To generate the x–z side views, 2D monolayers were scanned every 0.5 μm at the z-axis from apical to basal membrane and reconstructed with the ZEN software. Image analyses and fluorescence intensity measurements were performed with FIJI/ImageJ software.

For live-cell imaging, MDCK cell lines stably expressing GFP-tagged proteins of interest were cultured in DMEM supplemented with 10% (v/v) FBS in 35-mm glass-bottom dishes (MatTek) at 37°C with 5% CO₂. Time-lapse images were captured using Lattice SIM or LSM 880 with a Plan-Apochromat 63×/1.4 oil immersion lens at an interval of 5 sec. Images were collected 1024 × 1024 pixels in size and processed using FIJI/ImageJ software.

FRAP analysis

FRAP experiments were performed using Zeiss LSM 880. Selected regions within septin condensates were photobleached with 100% laser power using a 488-nm laser. Time-lapse images were acquired at 5-sec intervals. Fluorescence intensities of regions of interest were normalized to the initial intensity before bleaching. FRAP data were analyzed using GraphPad Prism 8.

OptoDroplet assay

COS-7 cells seeded on glass-bottom dishes were transfected with the indicated optoDroplet plasmids. At 24 h post-transfection, live cells were imaged using LSM 880. Cells expressing optoDroplet plasmids at a comparable level were selected for light-induced activation with a 488-nm laser. mCherry was excited and visualized using a 594-nm laser.

IP and western blotting

MDCK stable cell lines cultured in 10-cm dishes were allowed to grow to 100% confluence. Cells were collected and resuspended in cold lysis buffer (50 mM Tris–HCl, pH 7.4, 150 mM NaCl, 1 mM EDTA, 1 mM DTT, 0.1% Triton X-100, and 5% Glycerol) supplemented with protease inhibitor cocktail (Sigma). Cells were lysed by sonication and then centrifuged at 12000 rpm for 20 min at 4°C. The supernatant was incubated with GFP resin for 4 h at 4°C. After beads were washed three times with lysis buffer, proteins on beads were boiled for 10 min at 100°C, separated by sodium dodecyl sulfate–polyacrylamide gel electrophoresis, transferred to nitrocellulose membrane, and blotted with indicated primary antibodies and secondary antibodies. Horseradish peroxidase substrate for enhanced chemiluminescence was used for detection, and chemical fluorescence was visualized using an Amersham Imager 600 machine.

Expression and purification of septin proteins

All recombinant proteins used in this study were expressed in and purified from E. coli. Briefly, BL21(DE3), transformed with the corresponding plasmid, was grown until OD600 reached 0.8. Protein expression was then induced by the addition of 1 mM isopropyl 1-thio-β-D-galactopyranoside at 16°C for 20 h. Bacteria were collected, and the pellets were resuspended in cold lysis buffer (50 mM Tris–HCl, pH 8.0, 500 mM KCl, 5 mM MgCl₂, 10 mM imidazole, 1 mM DTT, 1 mM PMSF, 5% Glycerol, and protease inhibitor cocktail). Cells were lysed with a high-pressure homogenizer and centrifuged at 12000 rpm for 30 min. The supernatant was incubated with Ni-NTA resin (Qiagen) at 4°C for 2 h, washed three times with lysis buffer, and eluted with elution buffer (50 mM Tris–HCl, pH 8.0, 300 mM KCl, 3 mM MgCl₂, and 200 mM imidazole).

In vitro phase separation assay

Purified septins (including mutants) in the elution buffer were concentrated to 5 μg/μl and diluted with the phase separation buffer (20 mM Tris–HCl, pH 8.0, and 150 mM KCl) to achieve different final concentrations, and 5 M KCl was used to adjust the concentration of KCl in phase separation buffer. Protein solutions were transferred into glass-bottom dishes, respectively, and droplet formation was examined using LSM 880 with a Plan-Apochromat 63×/1.4 oil immersion lens. Images were collected at 1024 × 1024 pixels and processed using FIJI/ImageJ software.

Quantification and statistical analysis

All data were obtained from at least three independent experiments. Cells and droplets were imaged and quantified in random. Graphs were made using GraphPad Prism 8 and statistical tests were performed using one-way analysis of variance (ANOVA) for multiple comparisons in one group and two-way ANOVA for multiple comparisons in two groups. All statistical parameters are described in each figure legend, including sample size, error calculations, and P-values. P < 0.05 was considered to be significant. No statistical method was used to predetermine the sample size. No data were excluded from the analyses.

Funding

This work was supported by grants from the Ministry of Science and Technology of China and the National Natural Science Foundation of China (32090040, 2022YFA1303100, 92254302, W2411017, 2022YFA0806800, 92153302, 91854203, 31621002, 2017YFA0503600, 91953000, and 31970655), the Ministry of Education (IRT_17R102), Plans for Major Provincial Science & Technology Projects of Anhui Province (202303a0702003), and the Fundamental Research Funds for the Central Universities (KB9100000007, KB9100000006, and KB9100000013). The funders had no role in study design, data collection, analysis, decision to publish, or preparation of the manuscript.

Conflict of interest: none declared.

Author contributions: conceptualization: L.Z., X.L., and X.Y.; methodology: W.F., X.W., M.R., H.S., D.W., C.T., C.F., and B.Z.; investigation: W.F., M.R., and C.W.; visualization: W.F. and M.R.; supervision: L.Z., X.L., and X.Y.; writing—original draft: W.F.; writing—review & editing: L.Z. and X.Y.