Endosomal assembly and transport of heteromeric septin complexes promote septin cytoskeleton formation

Septins are GTP-binding cytoskeletal proteins that are evolutionarily highly conserved from fungi to mammals (Beise and Trimble, 2011; Fung et al., 2014). They are involved in diverse processes, such as cell division, neuronal development or plant invasion by fungal pathogens (Bridges and Gladfelter, 2014,, 2015). In humans dysfunctions of septins have been implicated in cancer and neurodegenerative diseases (Bridges and Gladfelter, 2015; Dolat et al., 2013).

A characteristic feature of septins is the central GTP-binding domain that is flanked by a polybasic region involved in membrane binding at the N-terminus (Bridges and Gladfelter, 2015). Most commonly a palindromic nonpolar heterooctamer is formed by four different septins. In Saccharomyces cerevisiae, for example, the septins Cdc3, Cdc10, Cdc11 and Cdc12 form the defined hetero-octamer Cdc11–Cdc12–Cdc3–Cdc10–Cdc10–Cdc3–Cdc12–Cdc11. Interactions between, for example, Cdc12–Cdc3, Cdc10–Cdc10 and Cdc11–Cdc11 at the so-called ‘NC’ interface are essential (Beise and Trimble, 2011; Sirajuddin et al., 2007). The terminal subunit Cdc11 can be replaced by an alternative septin subunit Shs1, resulting in the formation of alternative heterooctamers (Booth et al., 2015; Finnigan et al., 2015; Garcia et al., 2011).

Such heterooctameric building blocks can anneal head to tail into longer rods, which are bundled at specific subcellular locations to form higher-order structures, such as rings or filaments. Such structures are found for example at the division site between mother and daughter cells in S. cerevisiae (Bridges et al., 2014) or at the base of dendritic spines in neurons (Tada et al., 2007; Xie et al., 2007). Extended filamentous structures are mainly known from mammalian cells, where they partly colocalize with F-actin and distinct subsets of modified microtubules (Spiliotis and Gladfelter, 2012; Spiliotis et al., 2008). Even though the structure of septin heterooligomers has been analysed in great detail, it is still unclear precisely how septins are assembled and how they reach their subcellular destination (Bridges and Gladfelter, 2015).

We are studying septin biology in Ustilago maydis, a fungal pathogen causing corn smut disease. Its genome encodes the core septins Cdc3, Cdc10, Cdc11 and Cdc12 (UMAG_10503, UMAG_10644, UMAG_03449 and UMAG_03599, respectively). However, a clear Shs1 orthologue is missing (Alvarez-Tabares and Perez-Martin, 2010; Böhmer et al., 2009). A prerequisite for infection is the switch from yeast to hyphal growth (Vollmeister et al., 2012). Hyphae grow with a defined axis of polarity expanding at the apical tip and inserting septa at the basal pole (Vollmeister and Feldbrügge, 2010). An essential process involved in polar growth is long distance transport of macromolecules such as proteins, lipids as well as mRNAs (Haag et al., 2015; Jansen et al., 2014). Rab5a-positive endosomes act as important carriers that shuttle along microtubules mediated by the plus-end directed motor Kin3 and the minus-end directed motor dynein (Lenz et al., 2006; Schuster et al., 2011; Steinberg, 2014).

The key factor for mRNA transport is the RNA-binding protein Rrm4, which contains three RNA recognition motifs (RRM) at the N-terminus and two peptide-binding MLLE-domains (Becht et al., 2005; Pohlmann et al., 2015). The MLLE domains interact with Upa1, an endosomal component that functions in recruiting Rrm4 to Rab5a-positive endosomes (Pohlmann et al., 2015). The RRMs bind target mRNAs encoding, for example, the septin Cdc3 (König et al., 2009). Characterization of the transport of cdc3 mRNA led to the model that local translation of cdc3 mRNA mediates loading of Cdc3 protein onto the cytoplasmic surface of endosomes for microtubule-dependent transport. This process is crucial for efficient formation of a septin gradient in cortical filaments at the hyphal growth pole (Baumann et al., 2014; Jansen et al., 2014). A second septin, Cdc12, also shuttles in a microtubule-dependent manner on endosomes, raising the possibility that septin heterooligomers might assemble on endosomes for subcellular transport (Baumann et al., 2014). Here, we provide detailed in vivo evidence that all septins are indeed transported on endosomes in an interdependent manner to support the formation of higher-order structures at the growth pole. This suggests that preassembly of septin heterooligomers takes place on the membranous surface of endosomes.

Cdc3 is needed for efficient hyphal growth of U. maydis (Baumann et al., 2014). To address whether the other three septins perform similar functions we generated deletion mutants in the genetic background of strain AB33 (Tables S1, S2). In this strain, formation of hyphae can be elicited by switching the nitrogen source (Brachmann et al., 2001). Wild-type hyphae show mainly unipolar growth, expanding at the apical pole while inserting septa at the basal pole (Fig. 1A). All four septin deletion strains developed a substantial number of hyphae showing bipolar growth at 4 h post induction (h.p.i.; Fig. 1A,B). This resembles the growth defects observed in rrm4Δ strains (Becht et al., 2006). At later stages a large proportion of hyphae of septin deletion strains recovered and switched to unipolar growth with basal septa (8 h.p.i.; Fig. 1A,B). However, compared to wild-type, septin deletion strains contained an increased amount of bipolar hyphae without septa (8 h.p.i.) indicating that these mutations cause defects in establishing unipolar hyphal growth (Fig. 1B).

As a second read-out we tested unconventional secretion of chitinase Cts1 (Stock et al., 2012), a process that is specifically dependent on Rrm4-mediated mRNA transport during hyphal growth (Koepke et al., 2011). We observed that all four septin deletion strains exhibited reduced secretion of Cts1 only in hyphae (Fig. 1C). Thus, all four septins are needed for efficient unipolar growth and unconventional secretion of Cts1 indicating a common function for septins during hyphal growth.

Fluorescently tagged fusion proteins are best suited to study in vivo septin biology. Therefore, we generated a comprehensive collection of strains in the genetic background of AB33, expressing septins fused N- or C-terminally to the green fluorescent protein (eGfp, Clontech) at the endogenous locus. Notably, the 3′ untranslated region (UTR) was preserved to keep potential regulatory elements intact (Fig. S1A).

We tested the functionality of the fusion proteins in a growth assay for bipolar filaments and in the Cts1 secretion assay (Fig. S1B,C). Two main observations were made in these experiments. First, a functional Cdc3 wild-type protein was most closely resembled by the N-terminal fusion of Cdc3 (Cdc3–GfpN) whereas the corresponding C-terminal fusion was less functional. Second, the remaining fusion proteins of the other septins complemented the deletion phenotype to various extents.

Visualizing the subcellular localization of all septin fusion proteins revealed that they accumulated at basal septa and in cytoplasmic rings (Fig. 2A,B; Fig. S2A). The latter might either serve as transient storage forms or could be indicative of assembly defects (Böhmer et al., 2009). Certain fusion proteins tend to form these rings more often than others (Fig. S2B). Importantly, all fusion proteins localized to shuttling units that moved with a comparable speed of about 2.3 µm/s in both directions throughout the whole hypha (Fig. 2C–E; Movies 1–4). In all cases, the movement was affected by the microtubule inhibitor benomyl (Fig. 2D; Movie 5). Benomyl prevents polymerization of microtubules and therefore long-distance transport along microtubules is abolished. This results in the formation of bipolarly growing hyphae (Fuchs et al., 2005; Higuchi et al., 2014) resembling those from rrm4Δ hyphae (Fig. 1A; Becht et al., 2006). To identify the shuttling units, we briefly stained hyphae with the lipophilic dye FM4-64. This dye follows the endocytotic pathway (Vida and Emr, 1995) and is widely used to stain Rab5a-positive early endosomes in U. maydis (Higuchi et al., 2014). We observed that all septins colocalized with FM4-64 positive units. Thus, as shown for Cdc3 (Baumann et al., 2014), all septins are present on shuttling endosomes (Fig. S2C).

Finally, we studied filament formation of all septin fusion proteins (Fig. 2F). For clear visualization of cortical filaments, we analysed maximum intensity projections of z-planes (Baumann et al., 2014). We observed that all septins were present in cortical filaments running along the longitudinal axis of hyphae (Fig. 2F). However, depending on the position of the fused Gfp, the filaments differed in their appearance (see Discussion). For example, Cdc3–GfpN formed filaments with a gradient emanating from the hyphal tip, whereas Cdc3–GfpC formed only very short, aberrant filaments (Fig. 2F; Fig. S2D). The cortical filaments in Cdc10–GfpN exhibited a different appearance in that they did not reach the hyphal tip and only a slight gradient was observed (Fig. 2F, Fig. S2D). Filament formation itself was not inhibited by benomyl, but in this case the gradient was also lost (compare Fig. 2F and Fig. S2E), suggesting that microtubule-dependent transport is needed for gradient formation (Baumann et al., 2014). Some fusions (e.g. Cdc3–GfpC or Cdc12–GfpC) failed to form extended filaments, suggesting that the Gfp fusion site interferes with the formation of higher-order structures. In summary, consistent with results obtained analysing Cdc3 (Baumann et al., 2014), all other septins also localize to endosomes shuttling on microtubules as well as to cortical filaments.

To verify that septin proteins colocalize at the distinct subcellular sites described above, we studied colocalization of two neighbouring septin pairs, Cdc3–CherryN (N-terminal fusion with mCherry) with either Cdc10–GfpC or Cdc12–GfpN. Expression of the mCherry fusion protein was driven by the strong promoter P_otef to improve detection of the mCherry signal (Shaner et al., 2004). As expected, for both septin pairs, the proteins colocalized at septa and in cytoplasmic rings (Fig. 3A). Moreover, using dual-view technology, extensive colocalization on shuttling endosomes was detected (Fig. 3B,C; Movies 6,7).

Furthermore, single-plane analysis revealed wide-ranging colocalization of both septins in cortical filaments (Fig. 3D). Interestingly, the gradient of Cdc3–CherryN was no longer observed in a strain co-expressing Cdc12–GfpN (Fig. S3A, lower left panel). This suggests that the presence of Cdc12–GfpN alters the filament-forming ability of Cdc3–CherryN (compared to a strain expressing only Cdc3-CherryN; Fig. S3A, upper left panel), suggesting a direct Cdc3–Cdc12 interaction in these septin filaments.

To verify the interaction, we performed basic fluorescence resonance energy transfer (FRET) and more rigorous fluorescence lifetime imaging microscopy (FLIM)-FRET experiments. Analysing hyphal tips revealed that bleaching the mCherry acceptor of Cdc3–CherryN resulted in a clear increase (∼6%, Fig. S3B) in Gfp fluorescence of Cdc12–GfpN (donor) in the hyphal tip. In agreement, the Gfp lifetime of Cdc12–GfpN measured in FLIM-FRET experiments was significantly lower in strains co-expressing Cdc3–CherryN in comparison to the control (Fig. 3E, left). Measuring the Gfp lifetime in strains expressing Cdc3–CherryN and Cdc10–GfpC revealed that the lifetime was not significantly altered in comparison to the corresponding control strain (Fig. 3E). Thus, as predicted from structural data from S. cerevisiae and humans (Beise and Trimble, 2011; McMurray et al., 2011; Sirajuddin et al., 2007; Versele et al., 2004), the N-termini of Cdc3 and Cdc12 are in close proximity. This is not the case for the N- and C-termini from Cdc3 and Cdc10, respectively. Taken together, septins colocalize at the described subcellular sites, such as shuttling endosomes, and in identical cortical filaments. This clarifies contradicting results in the literature on septin filaments in filamentous fungi (see Discussion).

To address whether septins form heteromeric structures on endosomes, we chose a genetic approach to study the subcellular localization of each septin in the absence of the others. Initially, we verified that deletion of septins did not alter the amount or the processive movement of FM4-64-stained endosomes. In contrast, loss of kinesin-3 type Kin3 resulted in strongly reduced shuttling endosomes (Fig. S3C). Thus, endosomal septins are dispensable for the motility of endosomes.

Next, we concentrated on one Gfp fusion for each septin (Fig. S1) to analyse the interdependency of their subcellular localization. Importantly, absence of one septin caused either a severe reduction or complete loss of endosomal localization of the other septins (Fig. 4A). Cdc3–GfpN localization, for example, was greatly reduced on endosomes in the absence of Cdc10 or Cdc11 and undetectable in the absence of Cdc12 (Fig. 4A). This suggests that heteromeric septin assemblies are formed on endosomes. In case of Cdc12–GfpN, we observed fewer shuttling signals in the absence of Cdc10, but the remaining ones could be detected more easily owing to the absence of signals from the septin filaments (Fig. 4A; Movie 8; see below). This peculiar observation could be explained by homotypic interactions, as observed for Cdc3 from S. cerevisiae in the absence of Cdc10 (McMurray et al., 2011). The same interdependency of subcellular septin localization was also observed at basal septa of hyphae and in cytoplasmic rings (Fig. S3D).

To study the interdependency of septins in filaments, we also analysed the localization of Gfp fusion proteins in the absence of the other septins. Whenever one septin was missing the formation of filaments was altered (Fig. S4A). For example, Cdc3–GfpN-containing filaments were shorter and restricted to the region of the initial cell in strains lacking Cdc12 (Fig. S4A).

To further substantiate these findings, we tested septin proteins carrying a deletion in the α0 helix. This region is crucial for the septin interaction at the NC interface and is important for heteromer formation (Bertin et al., 2010; McMurray et al., 2011). The α0 helix is conserved in Cdc3, Cdc10, and Cdc12 but not in Cdc11 (Fig. S3E). Analysing Cdc3α0Δ–GfpN revealed that cytoplasmic rings were no longer formed and septa localization was drastically reduced (Fig. S3F). Moreover, cytoplasmic signals increased whereas, importantly, endosomal signals were strongly reduced (Fig. 4B) indicating that the ability to heterodimerize is important for endosomal localization. There was also an increase of the cytoplasmic signal of Cdc10α0Δ–GfpN but only a weak reduction on endosomes was observed (Fig. 4B). Consistent with this, the presence of Cdc12–GfpN on shuttling endosomes was hardly detectable in strains expressing Cdc3α0Δ (Fig. 4B) suggesting that correct heteromerization is needed for endosomal localization. In case of Cdc10α0Δ, the reduction of Cdc12–GfpN was not as severe (Fig. 4B). Thus, annealing of octamers might not be essential for endosomal localization.

Analysing filament formation revealed that Cdc3α0Δ–GfpN as well as Cdc10α0Δ–GfpN did not form any kind of cortical filaments (Fig. 4C). Consistently, Cdc12–GfpN failed to form filaments if strains expressed either Cdc3α0Δ or Cdc10α0Δ. Thus, heteromeric units are needed to form long extended septin filaments, further supporting the observation that cortical filaments are higher-order structures.

Finally, we tested a strain expressing both Cdc3–GfpC and Cdc12–CherryC, a septin pair carrying C-terminal fusion proteins. This should interfere with C-terminal interactions of Cdc3–Cdc12. The presence of two fluorescently tagged fusion proteins in one strain had no influence on septum localization or formation of cytoplasmic rings. However, endosomal localization of both proteins was almost abolished (Fig. 4D). Thus, when Cdc12 carries a C-terminal extension, endosomal localization of Cdc3–GfpC is severely disturbed supporting the presence of heteromers on the surface of shuttling endosomes. Furthermore, the C-terminal domain appears to be needed for the attachment to endosomes. In conclusion, our in vivo evidence indicates that septins form heteromeric building blocks on the surface of shuttling endosomes.

Previously, we have shown that the key RNA-binding protein Rrm4 mediates endosomal transport of cdc3 mRNA leading to endosomal transport of Cdc3 protein. Most likely this is achieved by local translation on shuttling endosomes (Baumann et al., 2014,, 2015). In control experiments, we verified that loss of septins did not influence microtubule-dependent motility of Rrm4 (Fig. S4B–E).

To test whether Rrm4 mediates transport of all septin mRNAs in order to load endosomes with septin proteins, we performed RNA imaging experiments in live cells using the λN-Gfp system (Fig. 5A) (Baumann et al., 2014,, 2015). Analysing septin mRNAs revealed processive movement in all cases (Fig. 5B; Movie 9). The number of processive messenger ribonucleoprotein (mRNP) particles as well as the range and the velocity of movement were comparable for all four mRNAs (Fig. 5C–E).

The transport of septin mRNA was clearly Rrm4 dependent because in the absence of Rrm4 no processive movement was detectable (Fig. 5F,G). This is consistent with early results showing that Rrm4 is the key mRNA-transport protein (Baumann et al., 2014; König et al., 2009).

To extend this study, we analysed cdc3B¹⁶ mRNA in the absence of Cdc12 because, here, Cdc3 protein is no longer present on endosomes (Fig. 4A). However, processive movement of cdc3B¹⁶ mRNA still occurred with a similar range and velocity (Fig. 5B,D,E). We even observed an increase in the number of events of processive movement (Fig. 5C). A possible explanation for the latter would be that cdc3 and cdc12 mRNA compete for the same binding sites on endosomes. Given that endosomal transport of cdc3 mRNA can be uncoupled from Cdc3 protein transport, we can exclude the possibility that the translation product Cdc3 binds its own mRNA for autoregulatory purposes (Caballero-Lima et al., 2014; Haag et al., 2015). In essence, all septin mRNAs are transported in an Rrm4-dependent manner on endosomes suggesting endosomal assembly of the translation products.

Next, we addressed whether the endosomal localization of septin proteins is also dependent on Rrm4 (Baumann et al., 2014). To this end we initially demonstrated that Cdc3–GfpN and Cdc12–GfpN colocalized almost exclusively with Rrm4–CherryC (a functional C-terminal fusion protein, Baumann et al., 2012) on shuttling units (Fig. 6A,B; Movie 10). This indicates that both septins are present on Rrm4-positive endosomes (Baumann et al., 2014).

Studying Cdc3–GfpN and Cdc12–GfpN in rrm4Δ strains showed that the localization of both septins to septa and cytoplasmic rings was not altered (Fig. S4F), but the endosomal localization was lost (Fig. 6C). This holds true for a strain expressing Cdc12–GfpN and lacking Cdc10 as well as Rrm4, suggesting that, in this case, endosomal localization of Cdc12–GfpN also needed Rrm4 (Fig. 6C). Thus, endosomal septin localization is dependent on the mRNA-transport protein Rrm4 further supporting the hypothesis of endosomal translation and assembly of septin proteins.

So far we have shown that all septin mRNAs and encoded proteins are transported on endosomes. This might be needed for efficient formation of higher-order cortical filaments at the growing pole. Initially, we focused on Cdc3–GfpN because it exhibited the best biological activity, and in previous studies we have demonstrated in fluorescence recovery after photobleaching (FRAP) experiments that recovery of Cdc3–GfpN depends on Rrm4 (Baumann et al., 2014; Pohlmann et al., 2015). In this study, we were able to increase the spatial resolution and could resolve cortical filaments. We observed that Cdc3–GfpN signals recovered in the same cortical area that was present before bleaching. Interestingly, the recovery took place over the entire cortical area and not only at the end of higher-order filaments (Fig. 7A). The same is true for the recovery of the Cdc3–GfpN signal in the absence of Rrm4, but in this case the extent of recovery was strongly reduced (Fig. 7A,B). Thus, Rrm4-dependent endosomal transport of Cdc3 promotes its efficient assembly in cortical filaments at the growth pole.

Finally, we aimed to address whether heteromeric septin signals recover at identical subcellular sites. To this end, we performed dual-colour 3D photobleaching experiments. At 5 min after dual-colour bleaching, we observed simultaneous recovery of Cdc3–CherryN and Cdc12–GfpN at identical subcellular sites, suggesting that heteromeric units are assembled in cortical septin filaments (Fig. 7C). In addition, 3D photobleaching experiments demonstrated that the recovery of both fluorescence signals was dependent on Rrm4 (Fig. 7D), although recovery of Cdc3–CherryN was less prominent. This is likely due to the photophysical properties of mCherry causing extensive bleaching during acquisition of z-planes (12 sections) at two different time points (see Materials and Methods). Therefore, the statistical analysis revealed a significant difference only for Cdc12–GfpN (Fig. 7D). Importantly, FRET experiments revealed that without Rrm4, the interaction of Cdc3–CherryN with Cdc12–GfpN is strongly reduced (Fig. 7E). This is most likely due to the lack of endosomal assembly and transport (see above). In summary, the in vivo data presented here substantiate the hypothesis that heteromeric septin building blocks are transported on endosomes to enhance formation of higher-order septin filaments at the growth pole of hyphae.

Septins are highly versatile cytoskeletal components that form various higher-ordered structures such as rings and filaments at distinct subcellular sites (Bridges and Gladfelter, 2015; Fung et al., 2014). At present it is unclear how the underlying intracellular trafficking of septins subunits is orchestrated. Furthermore, in fungal hyphae it was questionable whether the observed higher-order cortical filaments consist of the known heteromeric septin subunits containing of all four septins, because Cdc11 and Cdc12 could, so far, not be detected in filaments (Alvarez-Tabares and Perez-Martin, 2010).

Here, we present a comprehensive genetic and cell biological study in U. maydis using N-and C-terminally Gfp-tagged versions of all known septins, while preserving endogenous expression levels and regulatory RNA elements. We demonstrate that all four septins are present in cortical filaments and provide detailed in vivo evidence to support the hypothesis of endosomal assembly and transport of septin subunits.

Previous analysis of deletion mutants in U. maydis revealed that septins were crucial for correct cell morphology but dispensable for cytokinesis of yeast cells (Alvarez-Tabares and Perez-Martin, 2010; Boyce et al., 2005). In contrast to studies in human pathogens such as Cryptococcus neoformans and Candida albicans (Kozubowski and Heitman, 2010; Warenda et al., 2003), loss of septins did not affect plant virulence of U. maydis (Alvarez-Tabares and Perez-Martin, 2010). Here, we observe that all four septin deletion mutants exhibited comparable defects in hyphal growth, providing genetic evidence that they function together in a complex. Loss of each septin causes the delayed formation of hyphal septa similar to that seen in A. nidulans (Hernandez-Rodriguez et al., 2012). Furthermore, all septins are crucial for establishment of unipolar growth as well as for unconventional secretion of Cts1 (Stock et al., 2012). Thus, septins appear to be particularly important for the efficient execution of the hyphal growth programme.

Like in mammalian cells, extended septin filaments are present at the cortex of fungal hyphae (Kaufmann and Philippsen, 2009; Khan et al., 2015; Kozubowski and Heitman, 2010). In U. maydis, filaments containing Cdc3 or Cdc10 were found in yeast-like growing cells and hyphae (Alvarez-Tabares and Perez-Martin, 2010; Baumann et al., 2014; Böhmer et al., 2009). We observed that all septins form cortical filaments consisting of heteromeric subunits. However, in contrast to their endosomal localization, the localization in filaments is affected by modifications of the N- or C-termini. This might be one reason why heteromeric septin filaments have not previously been reported in fungi. For example, the N-terminal fusion protein Cdc11–GfpN does not assemble in filaments (this work; Alvarez-Tabares and Perez-Martin, 2010) whereas the C-terminal fusion does (this work). Hence, a careful and comprehensive analysis of their localization was needed.

Based on the fact that septin filaments bundle laterally (DeMay et al., 2011; Fung et al., 2014) and that they are most likely in register with adjacent filaments (Kaplan et al., 2015), we propose that bundles of heteromeric septin filaments are also formed in U. maydis (Fig. 8). Consistent with this, we noticed that the assembly of septins in cortical filaments depends on the different subunits. The heteromeric nature of the cortical filaments was further supported by the observation that the α0Δ mutations, which affect the NC interface of Cdc3 with Cdc12, and Cdc10 with itself, abolish filament formation. Furthermore, colocalization experiments showed that Cdc3 and Cdc12 fluorescence recovered simultaneously at distinct sites in the filament, strongly arguing for the incorporation of heteromeric subunits. Formation of filament bundles requires intimate interactions of septin subunits. In accordance, certain fusion proteins such as C-terminally tagged Cdc3 and Cdc12 are impaired in filament formation. This could be due to a disturbed interaction with other septins or with potential septin scaffolding proteins (Sadian et al., 2013).

Previously, it has been reported that heterooctameric septin subunits anneal head-to-tail to form longer filaments only from the ends (Bridges et al., 2014). We observed that recovery occurs over the complete length of the filament. A possible explanation for this apparent discrepancy is that we might mainly detect the lateral assembly of septin filaments forming bundles and not the annealing process itself (DeMay et al., 2011). A pronounced bundling of septin filaments at the growth pole would explain the observed gradient (Fig. 8).

At present we can only speculate about the detailed function of septin filaments for efficient hyphal growth of U. maydis. It has been shown that Cdc10 colocalizes with a sub-set of microtubules (Alvarez-Tabares and Perez-Martin, 2010) indicating that septins could regulate microtubule dynamics during hyphal growth, for example, by specific post-translational modifications (Spiliotis et al., 2008).

Previous studies have already reported a close link between septin assembly and membranes (Bezanilla et al., 2015; Garcia et al., 2016; Heasley and McMurray, 2016). Using membrane bilayers and liposomes it was initially discovered that the polybasic region interacts with membrane lipids (Casamayor and Snyder, 2003; Zhang et al., 1999) and enhances rod formation (Bertin et al., 2010; Tanaka-Takiguchi et al., 2009). Moreover, these rods diffuse on membranes in two dimensions promoting the annealing process, which results in the formation of higher-order septin structures (Bridges et al., 2014). The finding that Cdc3 and Cdc12 are present on shuttling endosomes raised the intriguing possibility that heteromeric septin subunits are assembled on endosomes (Baumann et al., 2014).

Now, we show that all four septins are present on shuttling Rrm4-positive endosomes. Endosomal localization was not affected by N- or C-terminal fusion proteins, suggesting that no higher-ordered structures are assembled on their surface. Analysis of deletion mutants revealed that endosomal localization of septins was interdependent on each of the individual septins. This was further verified by analysing the aforementioned α0Δ mutations. Cdc3α0Δ was hardly detectable on endosomes and caused a clear reduction in the endosomal localization of Cdc12. Consistent with this, if both proteins carry a C-terminal fusion protein, none of them localize to endosomes, suggesting the formation of Cdc3–Cdc12 heteromers. Endosomal localization of Cdc10α0Δ was only reduced and therefore its influence on Cdc12 localization was not as drastic. Hence, the interaction at the NC interface of Cdc10 with itself appears to be unessential for endosomal localization. Taken together, we conclude that heteromeric subunits are formed on the surface of endosomes (Fig. 8), but the exact heteromeric composition such as dimers, tetramers or octamers needs to be addressed further.

We propose that local translation of septin mRNAs mediates endosomal assembly of septin heterooligomers (Fig. 8). This idea is based on our previous study showing that Cdc3 protein and mRNA colocalize on endosomes and that without mRNAs or ribosomes on endosomes the translation product Cdc3 is absent (Baumann et al., 2014; Haag et al., 2015; Jansen et al., 2014). This is supported by the finding that Rrm4-positive endosomes transport translationally active ribosomes (Higuchi et al., 2014; Palacios, 2014).

We could demonstrate in this study that all septin mRNAs are transported on endosomes. Furthermore, Rrm4 is essential for the endosomal localization of the analysed septin proteins Cdc3 and Cdc12. These results clearly demonstrate that those heteromeric septin subunits that are formed in the absence of Rrm4 are unable to localize to endosomes. Thus, without septin mRNA on endosomes, heteromeric subunits cannot bind endosomes. Importantly, dual-colour 3D photobleaching and basic FRET experiments demonstrate that loss of Rrm4 affects the interaction of Cdc3 and Cdc12 in vivo. This shows that septin subunits can be formed without endosomal transport, but less efficiently.

Interestingly, due to the septin interdependency mentioned above, local translation of septin mRNAs seems insufficient for endosomal attachment of the translation products. An attractive explanation would be that, compared to single septin proteins, heterooligomers bind endosomal lipids stronger. This could be due to an increased avidity of membrane binding, for example, mediated by the multiplication of polybasic regions in septin oligomers, or the ability of septin oligomers to recognize membrane curvature (Bridges et al., 2016). Alternatively, an interaction partner that specifically interacts with the C-termini of Cdc3 and Cdc12 might be present. This is supported by our finding that Cdc3 and Cdc12 lose endosomal localization when both proteins carry a C-terminal fusion protein (Fig. 8). In summary, local translation and assembly of the newly synthesized proteins most likely function together in endosomal tethering of septins to coordinate their transport.

Details on the mechanism of intracellular septin trafficking have not yet been clarified. It is unknown, for example, how septins reach dendritic spines, the base of the sperm cell annulus or the base of cilia (Mostowy and Cossart, 2012; Saarikangas and Barral, 2011). Here, we present in vivo evidence that septin heteromeric subunits are assembled and transported on endosomes for filament formation and efficient hyphal growth (Fig. 8). Local translation during septin assembly might even explain how more complex septin filaments could be assembled in humans, which contain 13 different septin-encoding genes with numerous splice products (Barral, 2010; Spiliotis and Gladfelter, 2012). A defined septin heterooligomer consisting of distinct septin subunits could elegantly be produced by local translation of the individual septins at identical subcellular sites. The fact that septins have been found in neuronal vesicles and vesicle-like structures (Beites et al., 1999; Xie et al., 2007) opens the intriguing possibility that not only septin structures are conserved, but potentially also the intimate link to membrane trafficking for assembly and transport.

Standard procedures for plasmid and strain generations were used as described elsewhere (Brachmann et al., 2004; Loubradou et al., 2001; see Tables S1–S4). Proteins were tagged with eGfp, the enhanced version of green fluorescent protein (Clontech, Mountain View, CA), or mCherry, a derivate of mRfp (monomeric red fluorescent protein). Accession numbers of U. maydis genes used in this study: rrm4 (UMAG_10836), cdc3 (UMAG_10503), cdc10 (UMAG_10644), cdc11 (UMAG_03449), cdc12 (UMAG_03599), cts1 (UMAG_10419).

Determination of Cts1 activitiy in the supernatant of cells was performed in a similar manner to in previous studies (Koepke et al., 2011; Langner et al., 2015). Note that U. maydis cells were grown at 20°C in order to prevent morphological defects. At least three independent biological experiments were performed with three technical replicates per strain.

Laser-based epifluorescence-microscopy was performed on a Zeiss Axio Observer.Z1 as previously described (Baumann et al., 2014,, 2015,, 2016; Pohlmann et al., 2015). Colocalization studies of dynamic processes were carried out by applying msALEX (millisecond alternating laser excitation; Baumann et al., 2015) or with a two-channel imager (DV2, Photometrics, Tucson, AZ, USA; Pohlmann et al., 2015; Baumann et al., 2016). Yeast-like cells of all septin deletion strains were incubated at 20°C in order to prevent morphological defects (Alvarez-Tabares and Perez-Martin, 2010). To visualize septin localization at septa, in rings or in filaments, z-stacks with an exposure time of 150 ms/plane and a z-distance of 0.23 µm were recorded. All parts of the microscope systems were controlled by the software package Metamorph (Version 7.7; Molecular Devices).

RNA imaging in live cells and subsequent data analysis was performed as previously described (Baumann et al., 2012,, 2014). Per strain and experiment, >10 hyphae were analysed. Statistical tests were performed using Prism5 (Graphpad, La Jolla, CA). Staining of hyphae with FM4-64 and benomyl treatment was performed as described previously (Baumann et al., 2012; Becht et al., 2006).

For 3D photobleaching experiments, a 100× Plan-Neofluar objective (NA 1.3; Zeiss) and a 472 nm LED (CoolLED, precisExcite, Andover, UK) to excite Gfp fluorescence were used. An area of 15 µm from hyphal tips was bleached with 33% laser power of a 405 nm laser (80 mW fibre output, beam diameter 16 pixels, bleach time 5 ms/pixel). Bleaching was carried out in 18 z-planes with a z-distance of 0.3 µm. Fluorescence recovery was detected directly after bleaching and 5 min later, with an exposure time of 350 ms/plane in a z-stack of 18 planes with a z-distance of 0.3 µm. This time-point was chosen based on previous FRAP experiments carried out with a comparable set-up (Baumann et al., 2014). For data analysis, z-planes were merged to maximum intensity projections and the average background was subtracted. Fluorescence intensity after bleaching was set to 0 and the fluorescence recovery 5 min after bleaching was calculated. Intensity values were not corrected for acquisition bleaching as we were only interested in a relative readout of the fluorescence recovery. Statistical tests were performed using Prism5 (Graphpad).

3D photobleaching experiments in the colocalization strain were carried out using dual-view technology. The laser for excitation of Gfp (488 nm/100 mW) was set to 15% and the laser for excitation of mCherry (561 nm/150 mW) was set to 20%. The same area and bleaching settings were used as described above (12 z-planes; z-distance of 0.2 µm, exposure time of 500 ms/plane). Recovery of mCherry was strongly affected by acquisition bleaching. Data analysis was performed as described above.

For intensity-based FRET measurements, the initial Gfp fluorescence signal was acquired with an exposure time of 100 ms/plane (z-stack of 13 planes, z-distance of 0.3 µm, 488 nm laser line). Acceptor bleaching of mCherry was performed with an exposure time of 5 s/plane using the 561-nm laser line. The Gfp fluorescence after acceptor bleaching was acquired with the same settings described above. For analysis, a sum projection of z-planes was made and a maximum background subtraction was performed. The apparent FRET efficiency was calculated using the following formula: E_FRET=(intensity_after−intensity_before)/intensity_after×100. Note that a region of 696×520 pixels was analysed in total, corresponding to about 30 µm of the hyphal tip. Statistical tests were performed using Prism5 (Graphpad).

FLIM was performed on a confocal laser scanning microscope (Zeiss LSM 780) additionally equipped with a single-photon counting device with picosecond time resolution (Hydra Harp 400, PicoQuant, Berlin, Germany). Gfp fluorescence was excited at 485 nm using a linearly polarized diode laser (LDH-D-C-485) operated at a repetition rate of 32 MHz. Excitation power was set to 1 µW at the objective (40× water immersion, Zeiss C-PlanApo, NA 1.2). The emitted light was collected in the same objective and separated into its perpendicular and parallel polarization (Thorlabs PBS 101, Thorlabs GmbH, Germany). Fluorescence was then detected by Tau-SPADs (PicoQuant) in a narrow range of the emission spectrum of Gfp (band-pass filter, HC520/30 AHF). Images were taken with a 12.6-µs pixel time and a resolution of 110 nm/pixel (Zoom 3,9; 520×80). A series of 40 frames was merged into one image to increase photon numbers per pixel.

The fluorescence lifetime of Gfp was analysed using the software tool SymPhoTime 64, version 2.0 (PicoQuant, Berlin, Germany). Owing to the low excitation power to prevent photobleaching during image acquisition and the small pixel size to gain spatial resolution, the number of photons per pixel was still low after merging of frames, ranging from 300 to 2000 photons per pixel. Therefore, we applied a fit model with a minimal number of parameters to the data, in conjunction with a maximum likelihood estimator (MLE, Stahl et al., 2013; Weidtkamp-Peters et al., 2009). We used a bi-exponential model function with two fluorescence lifetimes τ₁ and τ₂, background contribution and correction for shifting of the instrument response function. The same model was applied to the donor-only and FRET-data sets. Thus, the decay histogram of eGFP is approximated in the subsequent fluorescence lifetime analysis by an intensity-weighted average lifetime τ_{Av Int}.

We acknowledge Drs K. Schipper, V. Göhre, U. Fleig and laboratory members for discussion and reading of the manuscript. We are grateful to U. Gengenbacher and S. Esch for excellent technical assistance. Thanks to Dr R. Kahmann from the Max Planck Institute for Terrestrial Microbiology for support.