Membrane Targeting of the Spir·Formin Actin Nucleator Complex Requires a Sequential Handshake of Polar Interactions*

Background: The Spir·formin actin nucleator complex organizes actin structures at vesicle membranes. Results: The Spir FYVE-type domain binds nonspecifically to negatively charged membranes and to the N-terminal KIND domain (in cis). Conclusion: Conformational states of Spir restrict recruitment of formin to the membrane surface. Significance: Studying the vesicle targeting mechanisms of actin nucleators is essential for our understanding of intracellular membrane trafficking. Spir and formin (FMN)-type actin nucleators initiate actin polymerization at vesicular membranes necessary for long range vesicular transport processes. Here we studied in detail the membrane binding properties and protein/protein interactions that govern the assembly of the membrane-associated Spir·FMN complex. Using biomimetic membrane models we show that binding of the C-terminal Spir-2 FYVE-type zinc finger involves both the presence of negatively charged lipids and hydrophobic contributions from the turret loop that intrudes the lipid bilayer. In solution, we uncovered a yet unknown intramolecular interaction between the Spir-2 FYVE-type domain and the N-terminal kinase non-catalytic C-lobe domain (KIND) that could not be detected in the membrane-bound state. Interestingly, we found that the intramolecular Spir-2 FYVE/KIND and the trans-regulatory Fmn-2-FSI/Spir-2-KIND interactions are competitive. We therefore characterized co-expressed Spir-2 and Fmn-2 fluorescent protein fusions in living cells by fluorescence cross-correlation spectroscopy. The data corroborate a model according to which Spir-2 exists in two different states, a cytosolic monomeric conformation and a membrane-bound state in which the KIND domain is released and accessible for subsequent Fmn-2 recruitment. This sequence of interactions mechanistically couples membrane binding of Spir to the recruitment of FMN, a pivotal step for initiating actin nucleation at vesicular membranes.

The multiple cellular functions of the dynamic actin cytoskeleton are achieved by targeting actin nucleator complexes at distinct subcellular sites to locally initiate the polymerization of filaments. The organization and regulation of cortical actin at the cell membrane have been extensively studied (1). However, up to now, the molecular mechanisms of cytoskeletal organization at intracellular, vesicular membranes are still poorly characterized. Proteins of the Spir family are cytoskeletal organizers at the crossroad of actin and vesicle processes (2). Mouse genetics and cell biological experiments point toward a function of the Spir⅐FMN 3 complex in oocyte maturation and emotional learning (3)(4)(5). In mouse oocytes, Spir-1 and Spir-2 proteins cooperate with formin-2 (Fmn-2; FMN subgroup of formins) in the generation of a dynamic actin meshwork, which is thought to originate from the surface of vesicles (3,6). The Spir/Fmn-2 nucleated actin filaments serve as tracks for myosin Vb actin motor protein-dependent long range transport of Rab11 vesicles in oocytes (6).
Besides a function in actin-dependent long range vesicle transport in oocytes, a role of Spir proteins in exocytic vesicle transport and in transport processes beyond early endosomes was described in somatic cells (7,8). In all cell types characterized so far, Spir proteins highly colocalize with Rab11, indicating a close functional relationship of the actin organizer and the small G protein of the Ras superfamily. Rab11 is localized at the trans-Golgi network, post-Golgi vesicles, and recycling endo-some and is thought to be a master regulator in the orchestration of cell surface proteins such as growth factor receptors, ion channels, and cell adhesion proteins (9).
On a molecular level, Spir proteins were shown to interact with FMN-type formins and regulate each other in a mutual fashion (10). In vitro, Spir proteins alone cannot nucleate the polymerization of profilin-bound actin, which is thought to be the major source for actin filament polymerization in cells (1,11,12). However, the Spir⅐FMN complex is a very potent nucleator of profilin/actin (12). The in vitro data are confirmed by findings in oocytes showing that the vesicle-originated cytoplasmic actin meshwork cannot be formed by one of the nucleators alone but requires the cooperative action of both Spir and Fmn-2 (3,6). The cooperation is conserved between mammals and flies as the Drosophila Spir protein and Drosophila FMNfamily protein Cappuccino also cooperate in building an oocytic actin meshwork during fly oogenesis (13,14).
The physical interaction of Spir and FMN proteins is mediated by the evolutionarily conserved formin/Spir interaction (FSI) sequence at the C terminus of FMN-type formins and involves the N-terminal kinase non-catalytic C-lobe domain (KIND) of Spir (15). The KIND domain evolved from the catalytic protein kinase fold into a protein interaction module (10, 16 -19). The complex crystal structure disclosed a large interface of positively charged residues of the formin-2 FSI sequence mediating contacts to an acidic groove at the surface of the Spir-1-KIND domain (18,19).
The introduction of a positive charge into the acidic groove of the KIND domain (human Spir-1 Y134K and Drosophila Spir Y232K mutations) strongly impaired the Spir/formin interaction in vitro (18,19). Transgenic expression of Drosophila Spir Y232K mutant protein in flies having a spir mutant background failed to rescue the oocytic actin meshwork (14). Direct physical interaction of Spir and the formin Cappuccino is therefore required to build up the oocytic actin mesh during fly oogenesis (14).
Targeting of Spir proteins toward vesicle membranes and the assembly of an active membrane-bound Spir⅐FMN actin nucleator complex can be regarded as essential steps in the regulation of Spir functions. The localization of Spir proteins at intracellular membranes is mediated by a C-terminal FYVE-type zinc finger membrane interaction module, a structural domain that is typically found in proteins involved in membrane trafficking (7). FYVE zinc fingers comprise eight cysteine residues, which bind two zinc ions (20,21). The structure forms a hydrophobic "turret loop," which penetrates the membrane (22,23). A second hallmark of canonical FYVE domain proteins such as EEA1 or the yeast VPS27 proteins is its specific interaction with phosphatidylinositol 3-phosphate, a key component in the integration of cell signaling cascades and vesicle transport processes (24,25). The Spir FYVE-type domain is different from the canonical FYVE domain in that it lacks the conserved phosphoinositide-binding motifs and has an insertion loop between cysteines 6 and 7 of the FYVE consensus sequence (7,26) (see Fig. 1A). Other FYVE-type zinc finger proteins have been identified that also share with the canonical FYVE domains the two zinc ion complexing centers and the hydrophobic turret loop but lack the consensus sequences forming the phosphatidyli-nositol 3-phosphate binding pocket. Members of this group are the Rab3 small G protein effectors Rim1, Rim2, Rabphilin-3A, and Noc2 and the Rab11 effector protrudin (27,28) (see Fig.  1A). Lipid interaction studies of the protrudin FYVE-type domain revealed that the membrane binding domain preferentially binds phosphatidylinositol bis-and trisphosphates but not monophosphates as the canonical FYVE domains do (29). Until now, membrane binding characteristics of the Spir FYVEtype domains have not been addressed.
In our aim to understand how the Spir/FMN actin nucleators assemble site-specifically into a functional complex, we investigated the membrane and protein interactions of the Spir-2 FYVE-type domain. Experiments using biomimetic membrane systems showed that the Spir-2 FYVE-type domain promiscuously interacts with negatively charged lipids. Intriguingly, we discovered an interaction of the Spir-2 FYVE-type zinc finger with the N-terminal Spir-2-KIND domain, which could not be detected at membranes. Based on data showing that the fulllength Spir-2 protein is a monomer in the cytosol and only weakly interacts with Fmn-2 when released from membranes, we propose a model according to which Spir-2 exists in two different states, a cytosolic backfolded monomer and a membrane-bound state in which the KIND domain is released and accessible for subsequent formin-2 recruitment.

EXPERIMENTAL PROCEDURES
Cloning, Expression, and Purification-Prokaryotic and eukaryotic expression vectors were generated by standard cloning techniques using Pfu polymerase (Fermentas), restriction endonucleases (New England Biolabs), and T4 DNA ligase (New England Biolabs). Point mutants were generated with the QuikChange site-directed mutagenesis kit (Agilent, Stratagene). The sequence of the constructs was verified by DNA sequencing (MWG-Biotech/Eurofins). A description of all constructs used in this study is given in Table 1. Prokaryotic glutathione S-transferase (GST)-and His 6 -AcGFP-tagged constructs were expressed in Escherichia coli Rosetta(DE3) pLysS/ BL21(DE3) pLysS. Bacteria were grown in LB (100 mg/liter ampicillin and 30 mg/liter chloramphenicol) at 37°C up to an A 600 of 0.6 -0.8. Recombinant protein expression was induced by 100 -300 M isopropyl 1-thio-␤-D-galactopyranoside and continued at 18 -20°C for 20 h. Bacteria were harvested and lysed by ultrasonication, and soluble proteins were purified using an Ä KTA purifier system (GE Healthcare) using GSH-Sepharose HP-, Ni-NTA FF, Ni-NTA-agarose beads (Qiagen), GSH-Sepharose 4B beads (GE Healthcare), and size exclusion chromatography (Superdex G200 16/60, GE Healthcare). GST tags were cleaved where indicated by tobacco etch virus protease. Protein concentration was increased by ultrafiltration using Amicon Ultra-4 ultracentrifugation devices with respective molecular weight cutoffs (Millipore). Protein purity was estimated by SDS-PAGE and Coomassie staining. For all bacterially expressed recombinant proteins used in this study, the purity was Ͼ90% except for the GFP-Spir-2-⌬KWL-C1,2,7,8S protein, which was very unstable and had an estimated purity of 60 -70%.
Fluorophore Protein Labeling-Purified proteins were dialyzed extensively against PBS with Slide-A-Lyzers (cutoff, 10 kDa). The proteins were diluted to 100 M, and the dye (Alexa Fluor 647 C 2 -maleimide (Invitrogen)) was solved in DMSO and then diluted in PBS. An excess of 5-7 times of the dye was used for the labeling reaction. The reaction took place for 2 h at room temperature and then overnight at 4°C on a rotating wheel. The excess dye was quenched with 2 mM DTT for 1 h at room temperature. The non-reacted dye was removed by extensive dialysis (cutoff, 10 kDa) against PBS. The efficiency was determined by absorption measurements and was in the range of 85-98%.
Giant Unilamellar Vesicle (GUV) Binding Assay-GUVs were prepared by the electroformation method (30). Lipids were from Avanti Polar Lipids or Sirius Fine Chemicals SiChem and dissolved in chloroform, but phosphoinositides were dissolved in 1:2:0.8 CHCl 3 :MeOH:H 2 O. In brief, 5 l of lipid mixture (1 mg/ml) was spread onto two platinum wire electrodes. After drying the lipid film, the electrodes with the lipid films were placed into a chamber containing 300 mM sucrose solution. Platinum wires were connected to a power generator. Electroformation was performed at 10 Hz and 2.3 V for 1 h at room temperature. The GUVs were released from the electrodes by changing the frequency to 2 Hz for 30 min. 50 l of GUVs were sedimented in 150 l of TBS buffer containing up to 200 nM His 6 -AcGFP-Spir-2-⌬KWL or His 6 -AcGFP. After 5 min of incubation at room temperature, protein binding to the GUVs was analyzed by confocal microscopy.
Large Unilamellar Vesicle (LUV) Flotation Assay-The LUV flotation assay was performed according to Höfer et al. (31). Solutions of 25 and 75% sucrose were prepared in 1ϫ TBS buffer. For the preparation of LUVs, a lipid suspension in 1ϫ TBS was subjected to eight freeze-thaw cycles and extruded 19 times through a 100-nm-diameter polycarbonate filter (Avanti Polar Lipids) using a Mini-Extruder (Avanti Polar Lipids). 40 l of protein (2 M) was mixed with 80 l of LUVs (2 mg/ml) in a total volume of 150 l in 1.5-ml polyallomer microfuge tubes (Beckman Instruments Inc.) and incubated for 10 min. The protein/liposome solution was then mixed with 100 l of the prepared 75% sucrose solution, giving a final sucrose solution of 30% at the bottom. This fraction was overlayered with 200 l of 25% sucrose and subsequently 100 l of 1ϫ TBS buffer, giving three layers of different sucrose concentrations (0, 25, and 30% sucrose). The tubes were centrifuged for 2 h at 50,000 rpm at 4°C using an ultracentrifuge (Optima TM Max, Beckman Instruments Inc.) and a TLA-55 rotor (Beckman Instruments Inc.). After centrifugation, LUVs and the bound protein floated on the top fraction in 0% sucrose. The fractions were gently collected and analyzed by SDS-PAGE and Western blotting.
Electrochemiluminescence Immunoassay (EIA)-EIA was used to measure protein/protein interactions and protein/lipid interactions (32,33). Protein (2 l of 25 ng/l Spir-2-KIND) or liposomes (2 l of 2 mg/ml LUVs of different compositions) were passively adsorbed on the electrode surface of 384-well high bind plates (Meso Scale Discovery, Rockville, MD) for 1 h at 23°C. The surface was blocked with 0.5% porcine gelatin in 1ϫ TBS buffer for 1 h at 23°C. Defined concentrations of protein (His 6 -AcGFP-Spir-2-⌬KWL) were added to each well, and binding was carried out for 2 h at 23°C. Afterward, primary antibodies against GFP (Living Colors full-length GFP polyclonal antibody, Clontech) were applied for 1 h at 23°C followed by secondary antibodies (goat anti-rabbit SULFO-TAG from Meso Scale Discovery; 1.25 g/ml; 23°C for 1 h). Each of the above steps was followed by washing three times with 1ϫ TBS. The signal readout from the bound fraction was performed on the SECTOR Imager 6000 (Meso Scale Discovery) using surfactant-free read buffer provided by the manufacturer. As a background, both primary and secondary antibody binding to gelatin-blocked electrodes was determined and subtracted. The data were analyzed in SigmaPlot 12.5 software (Systat Software Inc.) assuming a single binding site with max-imum signal change upon binding amplitude (B max ) and equilibrium dissociation constant (K d ).
Fluorescence Cross-correlation Spectroscopy (FCCS)-Position-dependent FCS/FCCS was measured with a laser scanning microscope (LSM780, Zeiss) equipped with a Confocor3 unit providing single molecule sensitivity of avalanche photodiodes. All measurements were performed at 25°C. The samples were excited by using the 488-nm line of an argon ion laser, the 561 and 633 nm HeNe laser with a power of 2-2.5 microwatts. A dichroic beam splitter (488/561 or 488/633 nm) was used for simultaneous excitation of GFP variants and mStrawberry or GFP and Alexa647, respectively. Detection channels were split by dichroic beam splitters (545 or 635 nm) and spectrally confined to GFP emission (505-540 or 505-610 nm), mStrawberry emission (580 nm long pass), and Alexa647 emission (655-nm long pass filter). To derive concentrations, the focal volumes were calibrated by Alexa Fluor 488 (D ϭ 435 m 2 /s (34)), Alexa Fluor 568 (D ϭ 332 m 2 /s (35)), and Alexa647 (D ϭ 330 m 2 /s (36)) for each individual Lab-Tek slide. To remove dominant after-pulsing, dye calibrations were determined by cross-correlating the signal between two identical detection channels. For dye calibrations, we applied a one-component three-dimensional diffusion model with triplet states (Tϩ3D). FCS curves were evaluated with Zeiss Zen 2010 or open source software (PyCorrFit (37)). Statistical analysis was performed using Sig-maPlot 12.5 software. For single comparisons, Student's t tests were performed if the data passed normality assumptions. If the data did not pass normality assumptions, the Mann-Whitney U test was performed. Significance levels are marked accordingly: *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001; n.s., not significant, p Ͼ 0.05.
FCCS in Solution-Binding between Spir-2-KIND*Alexa647 and His 6 -AcGFP-Spir-2-⌬KWL was determined by single point FCCS in BSA-coated 8-well Lab-Tek chambers (Nunc). 12-24 recordings (10 s) were measured in each reaction at a distance of 100 m above the coverslip. Intensity traces showing significant variations due to aggregates were discarded. The number of complexes (N gr ) was calculated from the ratio of background-corrected cross-and autocorrelation amplitudes (38). Bleed-through from the GFP into the Alexa647 channel was negligible (below 1%). Correlation curves of His 6 -AcGFP-Spir-2-⌬KWL were evaluated by applying a one-component three-dimensional diffusion model with a blinking term (Tϩ3D). Residual non-conjugated Alexa647 was resolved by a two-component diffusion model (Tϩ3Dϩ3D) in which the diffusion time of the dye was fixed to 30 s. For cross-correlation analysis, the fraction of free Alexa647 dye (ϳ30%) was subtracted from the total number of red particles (N r ) in the Alexa647 channel.
To determine dissociation constants (K d ), increasing concentrations of (up to 1 M) Spir-2-KIND*Alexa647 were titrated into 1ϫ TBS buffer containing 25 nM His 6 -AcGFP-Spir-2-⌬KWL. Prior to measurement, the samples were incubated for 20 min at room temperature. The cross-correlation levels reflecting progressive binding were evaluated with a simple 1:1 interaction (39). In solutions containing GUVs, FCCS was measured at two positions, in the supernatant (free solution) and at the apical pole of the membrane. GUV measurements were evaluated by a mixed two-component diffusion model (Tϩ3Dϩ2D) to account for two-dimensional diffusion in the membrane plane. As a negative control, we used His 6 -AcGFP for which we obtained a significantly lower cross-correlation (2%) at both positions.
FCCS in Cells-Molecular interactions between mStrawberrytagged Fmn-2 and GFP-tagged Spir-2 constructs were measured in HeLa cells plated in 8-well Lab-Tek (Nunc) and transiently transfected with 50 -150 ng of each plasmid. Prior to measurement, cells were washed twice for 5 min with air buffer. To dissolve the vesicular pattern of full-length Spir constructs, the samples were treated with 5 g/ml brefeldin A (BFA) in air buffer for at least 30 min. For positioning in cells, we adjusted the focal depth (z axis) to the equatorial plane of the nucleus and positioned the laser focus in a region with homogeneous fluorescence distribution in the cytoplasm. Excitation power was adjusted to minimize bleaching to less than 10% during measurements (15-20 s), resulting in a molecular brightness of 1-5 kHz. The cytoplasmic correlation curves were evaluated by applying a two-component Tϩ3Dϩ3D diffusion model. For fitting intracellular correlation data, time bins shorter than 10 s were ignored; a distinct submillisecond blinking time was taken into account if significant. For molecular brightness analysis, the average counts per particle were calculated from the fluorescence intensity (kHz) of the trace and the number of particles as derived from the fluctuation amplitude (both background-corrected). The average counts per particle were finally normalized to monomeric GFP. The maximum cross-correlation levels were calculated individually for each set of auto-and cross-correlation amplitudes (38). Bleed-through from mStrawberry into the GFP channel was negligible; however, GFP produced 12% signal in the mStrawberry channel, which was backcorrected (40). The maximum of the corrected cross-correlation ratio (CC max ) of multiple runs were averaged for the particular cell. To compare the diffusion behavior, multiple autocorrelation curves (comprising a total of 4 -6-min measurement time) for each construct were normalized (amplitude ϭ 1) and averaged.

RESULTS
The FYVE-type Domain of Spir-2-FYVE domains share two zinc ion complexing centers (C1-C8) and a hydrophobic turret loop (conserved in human Spir-2: Leu-595 and Phe-596) that mediate lipid binding at the cytoplasmic face of intracellular vesicles (Fig. 1A). For canonical FYVE domains, it is well established that the specific phosphatidylinositol 3-phosphate (PI(3)P) binding is mediated by a pocket comprising the conserved WXXD, R(R/K)HHCR, and RVC sequence motifs of the FYVE consensus sequence (23) (Fig. 1A). However, the Spir FYVE-type domain is lacking these motifs. Instead, it has an insertion loop between C6 and C7 of the FYVE consensus sequence (7, 26) (Fig. 1A). Other FYVE-type zinc finger proteins such as Rabphilin-3A and protrudin also lack the canonical PI(3)P binding pocket and may therefore be classified, together with Spir-2, into a separate subfamily (Fig. 1A). How-ever, the biochemical nature of these non-canonical FYVEtype-mediated membrane interactions is still unknown. To experimentally characterize the assembly and membrane binding properties of the Spir⅐FMN actin nucleator complex, we therefore generated a set of vectors encoding tagged Spir-2 and Fmn-2 proteins for their expression in human and bacterial cells ( Fig. 1B and Table 1).
Membrane Interactions of Spir-2-Live cell imaging of transiently expressed full-length GFP-tagged Spir-2 or a truncated C-terminal fragment encoding the Spir box and the FYVE-type domain but lacking the N-terminal KIND and WH2 domains (GFP-Spir-2-⌬KW) showed the expected localization at endogenous vesicular compartments ( Fig. 2A, first and second panels). In contrast, point mutants in the FYVE-type domain, cysteine replacements that destroy the zinc complexing centers (C1,2,7,8S), or mutations in the turret loop, L595A,F596A (LAFA), show a homogeneous fluorescence distribution, suggesting that these protein variants remain soluble in the cytoplasm ( Fig. 2A, third and fourth panels).
To distinguish lipid binding from binding modes caused indirectly by protein/protein interactions, we established an assay using GUVs composed of defined lipid mixtures. Because of solubility issues, we did not succeed in purifying the fulllength Spir-2 protein. However, we achieved sufficient yields for a C-terminal fragment containing the FYVE-type domain (GFP-Spir-2-⌬KWL) and the corresponding point mutants (GFP-Spir-2-⌬KWL-C1,2,7,8S and GFP-Spir-2-⌬KWL-LAFA; see above) ( Fig. 1B). Lipid binding was studied by investigating the absence or presence of accumulated fluorescent protein at the lipid surface of GUVs in free solution by confocal imaging.

Compilation of plasmid DNA expression vectors used in this study
Seph., Sepharose; TEV, tobacco etch virus.   2C, first panel). Accordingly, the Spir-2 variant containing replacements of the zinc complexing cysteines (C1,2,7,8S) was binding-incompetent to the same degree as a recombinant GFP protein control (Fig. 2C, second and third panels).

Construct
To further dissect electrostatic from hydrophobic contributions mediated by the turret loop, we systematically varied the content of negatively charged lipids in GUVs and compared binding of the purified wild-type Spir-2 fragment with the corresponding LAFA mutant. We saw the onset of binding for GFP-Spir-2-⌬KWL at about 5% DOPA (Fig. 2D, upper panels), whereas the LAFA mutant required at least 8% DOPA (Fig. 2D,  lower panels). The increased density for the LAFA mutant suggests that additional charges compensate for the missing hydrophobic contributions mediated by the tip of the turret loop, which is thought to intrude the lipid bilayer (Fig. 2E). In fact, the hydrophobic contribution might be more significant in cells where the LAFA mutant showed an entirely cytoplasmic distribution ( Fig. 2A, fourth panel). Thus, like the canonical FYVE domains, the FYVE-type domain of Spir-2 mediates vesicle binding mediated by electrostatic, polar, and hydrophobic contributions. The complexed zinc ions play a crucial role for membrane binding, although this may include indirect effects related to domain structure. However, we could not pinpoint a steric preference for particular lipid headgroups.
The Spir-2 FYVE-type Zinc Finger Binds Strongly but Nonspecifically to Phosphoinositides-Among negatively charged lipids, phosphoinositides are distinguished by their importance for regulating vesicular trafficking. PIPs serve as intracellular address tags that recruit canonical FYVE domains by a PI(3)Pspecific interaction. We first tested vesicular membrane association in HeLa cells in the presence and absence of wortmannin, a fungal metabolite that acts as inhibitor of PI 3-kinase and reduces the intracellular PI(3)P levels (41,42). In contrast to a classical, Hrs-derived FYVE domain (43,44), wortmannin was incapable to release Spir-2 from the endogenous membranes (Fig. 3A). Thus, although Spir-2 exhibits affinity for negatively charged membranes, in cells, we did not see a PI(3)P dependence.
Applying GFP-Spir-2-⌬KWL to GUVs composed of DOPC supplemented with all combinations of phosphorylated PIPs (Fig. 3B) showed the expected binding to negatively charged membranes albeit without any specificity among the different PIP species. Furthermore, under conditions where binding of a classical EEA1-derived FVYE domain shows PI(3)P-specific membrane association, the Spir-2 protein was constitutively bound (Fig. 3C).
Under steady state conditions, accumulation of fluorescent GFP-Spir-2-⌬KWL at the surface of GUVs does not per se allow a conclusion to be drawn on the binding strength of the protein/ lipid interactions. Small amounts of bound protein cannot be detected due to the dominant signal from the supernatant; inhomogeneity in imaging and protein distributions exacerbate quantifications (45). Therefore we aimed to uncover a putative hidden structure of differential PIP binding patterns for the Spir-2 FYVE-type domain with more sensitive approaches.
First we set up a well established liposome flotation assay (46). LUVs were produced from defined lipid compositions of comparable net charge and mixed with the target protein GFP-Spir-2-⌬KWL. Ultracentrifugation was applied to separate the LUV-bound protein fraction from bulk solution with a sucrose gradient. Layers containing different content of sucrose were assayed by immunoblotting against GFP. Centrifugation conditions were such that the unbound protein sedimented into the sucrose cushion (30%), whereas the LUV fraction, filled with buffer of lower density, retained the bound protein in top layers of low sucrose (Fig. 4A, first panel). In agreement with GUV imaging, LUVs of pure 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) retained only trace amounts GFP-Spir-2-⌬KWL, whereas a content of 20% 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphate (POPA) concentrated the protein almost quantitatively in the top layer of the gradient (Fig. 4A, second and third panels).
However, LUVs containing different PIP species appeared more evenly distributed in layers between 0 and 25%, indicating a weaker binding as for LUVs containing 20% 1-palmitoyl-2oleoyl-sn-glycero-3-phosphate (POPA) (Fig. 4, compare B, first, second, and third panels with A, third panel). Binding to polyanionic PI(3)P, PI(4)P, or PI(4,5)P 2 retained slightly smaller amounts of protein, suggesting that a single charge per molecule may be sufficient to recruit the FYVE-type domain. Quantifying band intensities showed no significant difference between the PIP species.
Next we measured saturating binging curves with a recently established EIA (32). With EIA, LUVs of different lipid composition were adsorbed at a sensor surface to provide immobilized binding sites for the target protein. Vesicle-bound GFP-Spir-2-⌬KWL protein was detected with an antibody sandwich coupled to an electrochemiluminescent enzyme; the signal therefore reflects the amount of surface-bound protein after subsequent washes. Binding curves were fitted with an equation assuming a non-cooperative 1:1 binding model with an arbi- trary number of surface binding sites (Fig. 4C). Again, within error, the apparent K d values were indistinguishable among the different PIP species (24 -34 nM). The dissociation constants were of the same order of magnitude as that measured for LUVs supplemented with 20% negatively charged 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphatidylserine (POPS). In conclusion, the Spir-2 FYVE-type domain mediates a nanomolar interaction to negatively charged lipids in a rather promiscuous fashion, suggesting that additional cues like protein/protein interactions may be required to establish targeting at specific subpopulations of endogenous vesicles.

The C-terminal Spir-2 FYVE-type Zinc Finger Interacts with the N-terminal Spir-2-KIND Domain-The strong interaction
with negatively charged lipid headgroups suggested that the FYVE-type zinc finger exposes positively charged residues at the protein surface (21,22). Indeed, the isoelectric point (pI) of the Spir-2 FYVE-type domain is basic (amino acids (aa) 577-696, pI ϭ 9.1) and, strikingly, opposite to that of Spir-2 N-terminal KIND (human Spir-2, GenBank TM accession number AJ422077; aa 32-221, pI ϭ 4.4). Because it is a common theme for kinases that intramolecular backfolding keeps the enzymes in a quiescent state (47) and considering the evolutionary kinase origin of the KIND domain (16), we therefore hypothesized that such intramolecular interactions also exist for Spir.
To test this, we performed pulldown assays using GST as an affinity tag. Purified GST fusion proteins were preloaded on glutathione-Sepharose beads. GFP constructs of potential interaction partners were transiently expressed in HeLa cells and pulled down from the cell lysate; GFP was used to check the cellular expression and for immunodetection. Indeed, for GST-Spir-2-KIND, we obtained a strong band when pulling down a C-terminal fragment containing the FYVE-type domain (GFP-Spir-2-⌬KW; Fig. 1B) but not for the GFP control (Fig. 5A). Furthermore, the interaction was almost completely suppressed by replacing a single amino acid (Y120K; Fig. 1B) at a position that is analogous to a recently described point mutation in the Spir-1-KIND domain (Y134K) (18). Interestingly, this mutation was described to impair the interaction with Fmn-2, already providing a hint that Spir-2 interactions in cis and in trans may be competitive.
We further analyzed different constructs containing the FYVE-type domains for their potential to interact with GST-Spir-2-KIND (Fig. 5B) and detected significant interactions for both truncations, the GFP-Spir-2-⌬KW, as well as the isolated FYVE-type domain, which seems sufficient for binding (Fig.  1B). The isolated FYVE-type zinc finger interacted slightly stronger than the Spir-2-⌬KW protein, which in addition encodes the Spir box and a linker region (Fig. 5B). Quantification of three individual experimental repeats confirmed an increased band intensity of the pulled down FYVE-type domain by a factor of 1.3 that, however, was not significant (p value ϭ 0.14). Point mutations in the turret loop (LAFA) had no effect, whereas removal of the complexed zinc ions (C1,2,7,8S) drastically reduced the band intensity. This suggests that electrostatic contributions and the structural integrity of the FYVE-type domain also play an important role for this intramolecular protein/protein interaction.
To quantify the strength of the KIND/FYVE interaction, we first used EIA. Different concentrations of purified GFP-Spir-2-⌬KWL were titrated to a constant amount of surface-adsorbed non-tagged Spir-2-KIND protein (Fig. 5C). The hyperbolic signal increase confirmed that both protein constructs interacted directly without any hidden cofactors that might be  In a second attempt, we applied FCCS, which is a suitable technique to investigate molecular interactions in homogeneous solution (38). First we characterized the recipient protein GFP-Spir-2-⌬KWL with single color FCS. The molecular brightness (average counts per particle per transit through the detection volume) of GFP-Spir-2-⌬KWL was the same as that of purified monomeric GFP, indicating that at lower nanomolar concentrations FYVE-type domains do not mediate homotypic oligomerization. For dual color FCCS, purified KIND was labeled with an amino-reactive dye of orthogonal color (Alexa647). The degree of labeling was about one dye molecule per protein as determined by photometry and FCS. To avoid cross-talk-related false positive cross-correlation, we titrated in the red color channel (Spir-2-KIND*Alexa647), which produced a saturating binding curve. Evaluating a one-site binding model, we obtained a K d value 50 Ϯ 30 nM (Fig. 5D). A slightly reduced affinity in FCCS as compared with EIA may be attributed to partially inactivated Spir-2-KIND*Alexa647 due to chemical modifi-cation. However, within error, the affinities obtained by these two different techniques are in good agreement and represent sufficient strength to stabilize a backfolded conformation of Spir-2 in free solution.
Membrane Binding and the Intramolecular Spir-2 FYVEtype/KIND Interaction Are Mutually Exclusive-We next addressed the interplay between membrane binding and the intramolecular Spir-2 FYVE-type/KIND interaction. In HeLa cells, transient co-expression of the FYVE-type zinc finger (GFP-Spir-2-⌬KW) and KIND (mStrawberry-Spir-2-KIND) showed the expected vesicular distribution in the GFP channel. However, for all observed expression levels, KIND is neither capable to bind intracellular vesicles on its own (not shown) nor recruited by the C-terminal fragment GFP-Spir-2-⌬KW (Fig.  6A). This behavior could be reproduced with our GUV membrane model when applying a mixture of purified fluorescent constructs (GFP-Spir-2-⌬KWL and Spir-2-KIND*Alexa647). The FYVE-type zinc finger protein GFP-Spir-2-⌬KWL showed membrane binding at negatively charged GUVs but could not recruit the Alexa647-labeled KIND, which remained in bulk solution (Fig. 6B). Finally, to confirm this observation on a molecular level, we performed FCCS. In agreement with Fig. 5D, cross-correlation indicated co-diffusion of the two constructs in bulk solution (Fig. 6C). However, when positioning at the pole of GUVs, cross-correlation dropped to a level slightly above that of the His-tagged GFP control for which we enforced membrane binding by Ni-NTA lipids (DOPC, 20% DOPA, and 3% DOGS-Ni-NTA). Based on diffusion times, the remaining signal reflects contributions from soluble complexes because the three-dimensional detection volume encompasses both a cross-sectioning GUV membrane and some volume from the bulk. In conclusion, lipid binding and KIND interaction mediated by the Spir-2 FYVE-type domain seem to be mutually exclusive. Therefore Spir-2 seems to occur in two different conformations, a backfolded compact structure in free solution and a more extended conformation in which the FYVE/KIND interaction is released when bound to negatively charged lipid bilayers.
Fmn-2 Binding and the Intramolecular Spir-2 FYVE-type/ KIND Interaction Are Mutually Exclusive-It was proposed that the Spir⅐FMN actin nucleator complex exerts its function at vesicular membranes (6). Therefore how formins are recruited into the complex and which of the two Spir conformations is engaged remain important questions. We recently showed that a short C-terminal sequence (eFSI; 56 amino acids; Fig. 1B) provides the binding epitope for both human Spir proteins (15). We used this peptide for competition experiments to validate its role in our intramolecular Spir-2 FYVE-type/KIND interaction. To our surprise, the competing activity of Fmn-2-eFSI was very efficient in the GST pulldown assay where the GFP-Spir-2-FYVE band completely disappeared in the presence of 5 M Fmn-2-eFSI (Fig. 7A). In contrast, the same peptide carrying a point mutation (K1571A) that was described to strongly impair the Spir/formin interaction (18,19) was inactive (Fig. 7B).
Competition was also observed with dual color FCCS in free solution containing purified GFP-Spir-2-⌬KWL and Alexa647labeled KIND protein. The normalized cross-correlation amplitude decreased in the presence of Fmn-2-eFSI in a concentration-dependent manner (Fig. 7C). Control experiments with 5 M BSA showed no effect, whereas addition of unlabeled KIND protein diluted the double labeled complexes and therefore reduced the cross-correlation (Fig. 7D). The latter confirms specificity of the observed interactions in the presence of fluorescent tags.
Intramolecular Backfolding of Monomeric Spir-2 Modulates Fmn-2 Binding in the Cytoplasm of HeLa Cells-A possible regulatory function of the intramolecular Spir-2 FYVE/KIND interaction could be the mechanistic coupling of membrane binding with the assembly of a functional Spir⅐FMN actin nucleator complex. Therefore we asked whether the accessibility of the Spir-2-KIND domain (in cis) affects formin binding (in trans). Because the full-length protein could not be purified, we turned to transient expression in HeLa cells where we aimed to characterize binding between fluorescent Fmn-2 and different Spir-2 constructs by dual color FCCS.
As shown previously, Spir-2, but not Fmn-2, localized at endogenous membranes when overexpressed (15) (Fig. 8, A and  B). Membrane binding depends on the C-terminal half of the protein that contains the FYVE-type domain; thus the isolated KIND domain appears homogeneous (Fig. 8C). Vesicular localization is conveyed to Fmn-2 when co-expressing full-length Spir-2 ( Fig. 8D) but is lost for the deletion mutant GFP-Spir-2-⌬KW that lacks the KIND domain and is therefore potentially unable to recruit Fmn-2 via eFSI-mediated interactions (Fig.  8E). These observations already suggested that recruitment of Fmn-2 and Spir-2 to these membranes is mechanistically linked presumably by direct molecular interactions. Unfortunately, the vesicular pattern of full-length Spir-2 seemed unsuitable for a molecular validation by intracellular FCCS where the binding partners should be observed in a diffusive state. To achieve this, we applied the fungal toxin BFA, an efficient inhibitor of the small GTP-binding protein Arf (48) that we and others have found to dissolve the vesicular pattern of Spir-2 by a yet unknown mechanism (49) (Fig. 8F). Note that, in agreement with our working hypothesis, BFA treatment concomitantly dissolved the vesicular Fmn-2 pattern. Under BFA, GFP-Spir-2 showed sufficient mobility in the cytoplasm to be characterized by FCS. For the autocorrelation analysis of individually expressed constructs, a three-dimensional two-component diffusion model for fitting provided sufficient degree of freedom to represent the correlation data (  obtained for a 3-fold GFP tandem 1.2 ms (94%). Thus, in contrast to Spir-2-KIND and the GFP constructs, the decay of the full-length Spir-2 autocorrelation curve was pronounced biphasic. Because diffusion times scale roughly with the cubic root of the molecular mass, the very long correlation time of full-length Spir-2 cannot be justified by molecular size (106.6 kDa). Instead, such long tail correlations are indicative of large scale aggregations or transient binding events at large structures like membranes that may remain detectable even after BFA treatment.
To rule out aggregations, we first determined the molecular brightness of GFP-Spir-2 in comparison with tandems of multiple GFP domains (Fig. 9B). The GFP chain nicely reproduced a linear increase, whereas full-length GFP-Spir-2 in the presence of BFA as well as the homogeneously distributed LAFA mutant diffused clearly as monomers. Encouraged by the monomeric nature of Spir-2, we measured cross-correlation in cells co-expressing mStrawberry-Fmn-2-FH2-FSI with either GFP-Spir-2 or GFP-Spir-2-KIND (Fig. 9, C and D). We placed the laser focus in the cytoplasm or the nucleus (not shown) at positions that produced stable intensity traces. The amplitude of the autocorrelation curves varied due to different expression levels in the upper nanomolar range (green and red autocorrelation functions in Fig. 9, C and D). As shown in these two example cells, the cross-correlation amplitude (blue) was higher for GFP-Spir-2-KIND than for the full-length Spir-2 protein, indicating the increased abundance of double labeled complexes for GFP-Spir-2-KIND.
To assess this difference in a more systematic way, we determined average values across a small population of individual cells. When co-expressed, the different mobility of full-length GFP-Spir-2 and GFP-Spir-2-KIND was largely maintained (compare Fig. 10A with 9A), whereas the Fmn-2 construct diffused about the same as full-length Spir-2 under both condi- tions (Fig. 10A). Therefore the mobility pattern was not a suitable readout for binding. To evaluate cross-correlation amplitudes, we included controls for spectral cross-talk and chromatic mismatch between the color channels. The accessible cross-correlation range was about 50% as determined with a strictly co-diffusing GFP-linker-mStrawberry tandem (positive control; Fig. 10B) and separately co-expressed fluorescent proteins (negative control). Note that the baseline measured for negative control is close to zero only when correcting for 12% bleed-through from GFP into the red channel (38,40). In agreement with our previous results, cross-correlation indicated significant co-diffusion and hence binding between the Fmn-2 fragment and the KIND protein in the presence and absence of BFA. However, the average cross-correlation was drastically reduced for full-length Spir-2 under BFA as well as the LAFA mutant that exhibits by itself a homogeneous distribution in the cytoplasm ( Fig. 2A). Thus, full-length Spir-2 proteins show consistently lower interaction potential toward Fmn-2 as compared with the isolated KIND domain. Each data point in Fig. 10B represents the average value for a single cell, and the broad distribution of cross-correlation levels reflects variations within the cell population. In fact, cross-correlation and hence the degree of binding increased linearly with increased expression levels albeit with a smaller slope for fulllength Spir-2 than for isolated KIND (Fig. 10C). A correlation between expression level and the degree of binding further supports the notion that the cross-correlation values reflect a dynamic binding equilibrium. Finally, we determined crosscorrelation levels for nuclear fractions of the overexpressed proteins. Although expression levels were asymmetric, these nuclear cross-correlation levels fully recapitulate the behavior in the cytoplasm (Fig. 10, compare B and D). Because the nuclear compartment is devoid of endogenous membranes, we can faithfully interpret elevated cross-correlation as a true molecular interaction that does not depend on a joint residence at some sort of vesicle.
In summary, Spir-2⅐Fmn-2 complexes show a dynamic, KIND/FSI-mediated interaction in the submicromolar concentration range. The binding data corroborate our view that a backfolded conformation of full-length Spir-2 protein indeed exhibits a weaker affinity toward Fmn-2.

DISCUSSION
In this study, we characterize the molecular interactions that orchestrate the assembly of a Spir⅐FMN actin nucleation complex at membranes. A key player is the Spir FYVE-type domain. Consistent with the lack of the conserved amino acids for phosphatidylinositol 3-phosphate binding, the Spir-2 FYVEtype zinc finger domain did not show a preferential binding to different phosphoinositides; instead it binds rather nonspecifi- cally to a broad variety of negatively charged lipids. Furthermore, the interaction is strengthened by additional hydrophobic contributions from the turret loop that, at least in cells, seem critical for vesicular localization. The promiscuous affinity for phosphoinositides differs markedly from that of protrudin, another FYVE-type domain-containing protein that is also closely associated with Rab11 (29,50). Moreover, as both the electrostatic and hydrophobic contributions of Spir to membrane binding are rather unspecific, we argue that the targeting toward a specified subcellular compartment needs further regulatory steps, most likely the complex formation with other membrane-bound proteins.
In membrane targeting, interactions of cationic protein clusters with electronegatively charged membranes is a major mode of binding that was not always well acknowledged (51). Here we show that the Spir-2 protein has a high affinity for negatively charged membranes. Indeed, with a pI of 9.1, the Spir-2 FYVEtype domain is supposed to be highly positively charged at physiological pH.
Phosphatidylserine (PS) is the most abundant negatively charged phospholipid in eukaryotic cells (51). PS is synthesized at the inner leaflet of the endoplasmic reticulum. However, experiments using the PS-binding lactadherin C2 domain showed that the PS headgroup is flipped to the cytosolic side in or close to the trans-Golgi network (52). From there, PS is transported via the exocytic pathway to the plasma membrane. Thus, the subcellular PS distribution reflects largely that of the Spir protein: trans-Golgi network and post-Golgi vesicles (7).
In absolute quantities, subcellular PS concentrations are as yet poorly determined. Some studies report concentrations of about 5% PS in the Golgi membranes and an overall enrichment of PS in the recycling endosome (51,53,54). It is therefore likely that compartments associated with Spir already contain 5-10% PS at the cytosolic leaflet, which would be enough for nonspecific recruitment. Other negatively charged lipids like PI, which was also described to be concentrated in the Golgi membrane (7-12% (53)) may even add up to this effect. However, because we do not know their exact local concentrations, it remains to be shown in which way specific lipid compositions contribute to establish the characteristic Spir distribution under physiological conditions.
We show here that the FYVE-type domain is also a protein/ protein interaction module. Studying Spir-2 fragments in free solution, we discovered a strong interaction between the C-terminal FYVE-type zinc finger and the N-terminal KIND domain. Importantly, when engaged by membranes, the FYVEtype domain was unable to recruit KIND domain from bulk solution. It seems that the positive charges associated with the FYVE-type domain are screened by the negatively charged KIND domain when released from the membrane. Affinities as determined by various methods are in the lower nanomolar range and thus in the same order of magnitude as the membrane interactions, already suggesting a dynamic equilibrium of two conformational states. Brightness analysis of fluorescently tagged full-length Spir-2 in the cellular cytoplasm showed that Spir-2 diffuses as a monomer and rules out a dimeric head-totail conformation. Such intramolecular interactions thus could stabilize a compact, backfolded conformation as it was described for several kinases (55,56). Because backfolded conformations are associated with autoinhibitory functions and the KIND domain is evolutionarily related to kinase domains (16), it is intriguing to speculate that the KIND/FYVE interaction also has autoregulatory functions.
Similar to the FYVE/KIND interaction, recruitment of formins involves polar interactions. The crystal structure of the Spir-1-KIND⅐Fmn-2-FSI complex revealed an electrostatic nature of the Spir/FMN interaction (18,19). The positively charged C-terminal FMN FSI motif binds a negatively charged groove on the surface of the KIND domain. Here we show biochemically that our novel, intramolecular Spir-2-KIND/FYVE interaction indeed strongly competes with recruitment of Fmn-2-FSI presumably because the FMN proteins and the FYVEtype zinc finger bind to overlapping sequences within the Spir-KIND domain. These observations combine into a sequence of polar interactions that involve both actin nucleators and the vesicle membrane (Fig. 11). We propose the following model for membrane targeting of the Spir⅐FMN complex. Spir proteins may exist in free solution as backfolded monomers in which the two oppositely charged protein domains stabilize a presumably inactive conformation. Only upon membrane binding does the FYVE-type domain release the KIND domain, which is now accessible to recruit FMN proteins via their FSI sequence. As a result, our model provides a causal link between membrane binding of Spir and the assembly of a functional actin nucleator complex.
We were not yet able to characterize full-length Spir proteins in free solution. Therefore we studied transiently expressed Spir-2/Fmn-2 constructs in the cytoplasm of HeLa cells. We used dual color FCCS to measure the diffusion and interactions between differently labeled Spir-2 and Fmn-2 constructs. FCCS was measured under the continuous presence of BFA, a potent Arf inhibitor that was proposed to release the proteins from Rab11-positive vesicles in oocytes (49). Although BFA treatment produced homogeneous fluorescence distributions also in HeLa cells, autocorrelation analysis revealed that both proteins show a pronounced biphasic diffusion behavior with a second correlation time too long to represent free diffusion of protein monomers with the respective molecular size. The fact that the N-terminal Spir-2-KIND domain, which lacks an FYVE-type domain, showed much faster diffusion suggests that, even in the presence of BFA, Spir-2 and Fmn-2 show transient membrane interactions. The transient interactions of Fmn-2 may be indirectly mediated in complex with Spir-2. We detected significant interactions between Fmn-2 and the isolated KIND domain. However, in support of our model, interactions with full-length Spir-2 were much reduced, most likely reflecting intramolecular competition by the FYVE-type domain.
Our study has provided first molecular insights into the complex network of competing interactions that establish a vesicleassociated Spir⅐FMN complex. Such assembly is a necessary requirement for a site-specific activation of the actin machinery and the subsequent spatiotemporal regulation of actin-mediated vesicular transport processes.