The Molecular Basis of the Differential Subcellular Localization of FYVE Domains*

This study systematically analyzed the structural and mechanistic basis of the regulation of subcellular membrane targeting using FYVE domains as a model. FYVE domains, which mediate the recruitment of signaling and membrane-trafficking proteins to phosphatidylinositol 3-phosphate-containing endosomes, exhibit distinct subcellular localization despite minor structural variations within the family. Biophysical measurements, cellular imaging, and computational analysis of various FYVE domains showed that the introduction of a single cationic residue and a hydrophobic loop into the membrane binding region of the FYVE domains dramatically enhanced their membrane interactions. The results indicated that there is a threshold affinity for endosomal localization and that endosomal targeting of FYVE domains is sensitive to small changes in membrane affinity about this threshold. Collectively these studies provide new insight into how subcellular localization of FYVE domains and other membrane targeting domains can be regulated by minimal structural and environmental changes.

Numerous cellular processes such as signal transduction, vesicle trafficking, and cytoskeletal rearrangement require the exquisite targeting of peripheral proteins to various subcellular membranes. A large portion of this cellular membrane targeting is achieved by specific recognition of particular membrane lipids by proteins. A diverse group of membrane-targeting domains that specifically recognize different types of membrane lipids have been identified in the past decade. They include Bin Amphiphysin Rvs (BAR) (1), protein kinase C Conserved 1 (C1) (2,3), protein kinase C Conserved 2 (C2) (4), Epsin Amino-Terminal Homology (ENTH) (5), Band 4.1/Ezrin/Radixin/Moesin (FERM) (6), Fab1/YOTB/Vac1/EEA1 (FYVE) (7)(8)(9)(10)(11), Postsynaptic density-95/Discs large/ZO-1 (PDZ) (12), Pleckstrin Homology (PH) (13), Phosphotyrosine Binding (PTB) (14), Phox (PX) (15), Src homology 2 (SH2) (16), and tubby domains (17). Except for the C1 domain that binds diacylglycerol, these domains recognize phosphorylated derivatives of phosphati-dylinositol, collectively known as phosphoinositides. Although much is known about the structural basis of stereospecific lipid head group recognition by these domains, less is known about the mechanisms by which they achieve efficient and reversible binding to the cell membranes containing their lipid ligands. In particular, there is much to learn about the structural and mechanistic basis of the regulation of the targeting of lipidinteracting domains to different intracellular membranes. Among various membrane-targeting domains, the FYVE domain serves as an excellent model to address these questions. This is because FYVE domains from different proteins exhibit drastically different subcellular localization behaviors (18 -22) despite the fact almost all FYVE domains show high specificity and affinity for phosphatidylinositol 3-phosphate (PtdIns(3)P). 1 FYVE domains are zinc-containing modules of 60 -80 amino acid residues (7)(8)(9)(10)(11). As expected from the endosomal localization of PtdIns(3)P and its role in vesicle trafficking, a large number of FYVE domain-containing proteins, including EEA1, Hrs, and FENS-1, are involved in endocytic vesicle trafficking. Some FYVE domain-containing proteins also function in cytoskeletal regulation (faciogenital dysplasia 1) (23) and growth factor signaling (SARA (22) and endofin (19)). Sequence alignment of FYVE domains (see Fig. 1) reveals several consensus motifs. High resolution structures of three different FYVE domains have illustrated how some of these conserved residues are involved in specific recognition of the PtdIns(3)P head group (24 -26). Structural and biophysical studies have indicated that several factors beside PtdIns(3)P binding contribute to the membrane affinity of FYVE domains. They include nonspecific electrostatic interactions between basic protein residues and the anionic membrane surface (27)(28)(29), hydrophobic interactions achieved by the partial membrane insertion of the residues located in the loop (so-called turret loop, see Fig. 1) near the PtdIns(3)P-binding pocket (24,25,27,29,30), and FYVE domain dimerization (21,25,26). Mutational studies of many FYVE domain-containing proteins have indicated that FYVE domains play an important role in their binding to endosomes in the cell (21,31,32). Interestingly, however, among many FYVE domains identified so far, only FENS-1 * This work was supported by National Institutes of Health Grants GM52598, GM53987, and GM68849 (to W. C.) and GM66147 (to D. M.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
In this study, we performed in vitro membrane binding measurements, subcellular localization measurements, and computational analyses of the FYVE domains of FENS-1, endofin, Hrs, Vps27p, and their respective mutants to elucidate the structural and mechanistic basis of unique endosomal targeting behaviors of the FENS-1 and endofin FYVE domains. These studies provide new insight into how distinctively different subcellular localization behaviors of FYVE domains are caused by minor structural variations in the membrane binding region and how subcellular localization of these membrane targeting domains can be regulated by minimal structural and environmental changes.
Mutagenesis and Protein Expression-Mutations of FYVE domains were performed by the overlap extension polymerase chain reaction method (34). Each construct was subcloned into the pGEX-4T-1 vector containing an N-terminal glutathione S-transferase fusion and transformed into Escherichia coli BL21 cells for protein expression. Recombinant FYVE domains were expressed and purified as described previously (27). The genes for FYVE domains and mutants were also subcloned into a modified pIND vector to contain a C-terminal enhanced green fluorescent protein (EGFP) construct as described previously (35).
Monolayer and Surface Plasmon Resonance (SPR) Measurements-Surface pressure () of the lipid monolayers was measured as described previously (27). The kinetic SPR measurements were performed at 23°C using a lipid-coated L1 chip in the BIACORE X system as described previously (36). All data were analyzed using BIAevaluation 3.0 software (Biacore AB) to determine k a and k d , and the equilibrium dissociation constant (K d ) was then calculated using the equation K d ϭ k d /k a assuming 1:1 binding, i.e. protein ϩ protein binding site on vesicle 7 complex (36). Equilibrium (steady-state) SPR measurements were performed as described previously (27). Saturating response values (R eq ) were then plotted versus protein concentrations (C), and the K d value was determined by a nonlinear least-squares analysis of the binding isotherm using the equation Fluorimetric Binding Measurements-The affinity of FYVE domains for Ins(1,3)P 2 was determined by monitoring the decrease in Trp fluorescence intensity (⌬I) upon ligand binding. Ins(1,3)P 2 was titrated into 3 M FYVE domains in 20 mM Tris-HCl, pH 7.4, 0.16 M KCl, and the Trp fluorescence emission peak (300 -400 nm) was integrated with excitation wavelength set at 290 nm (I). The dissociation constant (K d ) was determined by non-linear least-squares fitting of the binding isotherm to a simple Langmuir-type saturation binding equation (37), where I 0 and I f are I values in the absence and presence of saturating concentration of Ins(1,3)P 2 , respectively, and P o and L o are analytic concentrations of protein and Ins(1,3)P 2 , respectively. Analytical Ultracentrifugation-Sedimentation equilibrium experiments (38) were performed on an XL-A analytical ultracentrifuge (Beckman). Samples were loaded into six-channel Epon charcoal-filled centerpieces using quartz windows. Measurements were done at 25°C using four different speeds (16,000, 18,800, 23,500, and 31,000 rpm), detecting at 280 nm, with sufficient time for equilibrium to be reached in between each speed increase. All FYVE domains (10, 20, and 30 M) were prepared in 20 mM HEPES, pH 7.4, 0.16 M KCl, which was also used as the reference buffer. Solvent density was calculated to be 0.9982 g/ml. From the amino acid composition of each FYVE domain, the partial specific volume was estimated to be 0.719 (endofin-FYVE), 0.715 (FENS-1-FYVE), and 0.712 (Hrs-FYVE) ml/g. Data were analyzed with the UltraScan 6.2 software to determine the molecular mass.
Structural Modeling-The Drosophila Hrs FYVE domain (25) was used as the structural template for building homology models for both endofin and FENS-1 FYVE domains. The alignment between the endofin and Hrs sequences and the homology model were both constructed with the automated homology model server 3D-Jigsaw (39). The sequence identity between the endofin and Hrs FYVE domains is 42%. The alignment between the FENS-1 and Hrs FYVE domains was extracted from a multiple alignment of FYVE domain sequences constructed with ClustalW (40); the sequence identity between the two FYVE domains is 36%. According to the alignment, the FENS-1 FYVE domain has an 11-residue insertion in the vicinity of the hydrophobic motif, which is conserved across FYVE domains and occurs immediately before the first ␤-strand observed in the FYVE domain structures (Fig. 1). Secondary structure prediction programs predict, with low confidence, that this region is either unstructured or partially ␣-helical (40 -42). We used both Modeler6 (43) and Nest (44) to construct models for the FENS-1 FYVE domain based on the alignment extracted from the multiple sequence alignment. We implemented the loop prediction modules that are offered by both programs (45,46). Modeler6 predicted a fairly extended loop, while Nest predicted a more compact loop (a representative example of which is given in Fig. 2). The insert predicted by Nest scored better according to the structure evaluation tool Verify3D (47). Overall, however, the analysis suggests that the insert is disordered, and its conformation may be affected by its interaction with the membrane surface. Electrostatic potential calculations were performed as described previously (27).
Cell Culture and Microscopy-A stable human embryonic kidney 293 cell line expressing the ecdysone receptor (Invitrogen) was used for all experiments as described previously (35). Microscopy data were collected on a Zeiss LSM 510 laser scanning confocal microscope. A 40ϫ, 1.2 numerical aperture immersion objective was used for all experiments. Transfected cells were washed twice with 1 mM HEPES, buffer pH 7.4, containing 2.5 mM MgCl 2 , 140 mM NaCl, 5 mM KCl, and 6 mM sucrose. After washing, cells were overlaid with 150 l of the same buffer. Suitable cells were selected for imaging, and a single image was taken. In control experiments, cells were treated with 100 nM wortmannin for 30 min to determine the effects of phosphatidylinositol 3-kinase inhibition on FYVE domain subcellular localization. For colocalization experiments, transfected cells expressing EGFP-FYVE domain constructs were incubated with 25 g/ml human transferrin conjugated to Alexa Fluor® 633 (Molecular Probes) for 30 min at 37°C. Cells were then washed three times with the above buffer to remove excess dye and subjected to dual channel imaging. EGFP and Alexa Fluor 633 were excited optimally at 488 and 633 nm, respectively, and emission was measured using a band pass 505-530 nm filter for EGFP and a line pass 650 nm filter for Alexa Fluor 633 on separate channels. homologous primary structures, and both contain a unique Lys residue in the turret loop, which is a Val in EEA1 and an Asn in Vps27p and Drosophila Hrs. FENS-1-FYVE has an 11amino acid insertion next to the turret loop region. To understand how these unique structural properties of the three FYVE domains affect their membrane binding, we first performed structural modeling of the FYVE domains of endofin and FENS-1 based on the sequence homology to other FYVE domains with known tertiary structure (i.e. Vps27p, Hrs, and EEA1). We also calculated the electrostatic potentials for four FYVE domains used in this study. Since endofin-FYVE and SARA-FYVE are structurally similar, we characterized only endofin-FYVE in this study.

Structural Properties of FYVE Domains-When
Structural modeling suggests that endofin-FYVE (Fig. 2, first row) should have essentially the same structure as the Hrs and Vps27p FYVE domains (Fig. 2, third and fourth rows). However, it has a dramatically stronger positive potential than the other two FYVE domains with an overall charge of ϩ13 compared with that of ϩ7 for Vps27p and ϩ10 for Hrs (28) (Fig.  2, compare row 1 with rows 3 and 4). Thus, this FYVE domain may achieve high affinity for the membrane through enhanced nonspecific electrostatic interactions with anionic lipids in the membrane in addition to its specific interaction with PtdIns(3)P. The FYVE domain sequence alignment ( Fig. 1) as well as homology models (Fig. 2, second row) suggest that FENS-1-FYVE has an extended turret loop due to the 11-amino acid extension, which is rich in aromatic residues (Fig. 2, second row, green). It has been shown that aromatic residues, such as two Trp residues in FENS-1-FYVE turret loop, play an important role in membrane-protein interactions (36) and that the turret loop of the Hrs and Vps27p FYVE domains partially penetrates PtdIns(3)P-containing membranes and thereby increases membrane affinity (27). Thus, it is expected that FENS-1-FYVE will have an exceptional ability to penetrate the membrane and exhibit high membrane affinity. A model structure of FENS-1-FYVE in Fig. 2 has a positive electrostatic potential profile in the membrane binding region that is comparable to those of the Hrs and Vps27p FYVE domains (Fig. 2, compare row 2 with rows 3 and 4), suggesting that its nonspecific electrostatic interactions with the membrane would be similar to those of the other two FYVE domains.
In Vitro Membrane Binding Properties of FYVE Domains-To test whether the FYVE domains of endofin and FENS-1 have the predicted membrane binding properties, we characterized their in vitro membrane binding by two independent methods. First, we measured the interaction of differ- In each panel, the backbone worm in the region of the PtdIns(3)P binding motif is colored yellow, and the zinc atoms are shown in space-filling representation and colored magenta. The hydrophobic and cationic residues in the turret loop are shown in space-filling representation and colored green and blue, respectively. Residues in the turret loop that are mutated are colored gold. In each case, the first column is the wild type, and the second column is the mutant domain. Electrostatic potentials (calculated for 0.1 M KCl) are shown as three-dimensional equipotential contours as visualized in GRASP (55) with the following coloring scheme: blue, ϩ25 mV and red, Ϫ25 mV. ent FYVE domains with phospholipid monolayers at the airwater interfaces that serve as a highly sensitive probe of the membrane penetrating capability of proteins (37,48). In this study, a monolayer composed of POPC/POPE/POPS/PtdIns(3)P (63:20:15:2), which simulates the lipid composition of early endosomes, of a given surface pressure ( 0 ) was spread at constant area, and the change in surface pressure (⌬) was monitored after the injection of FYVE domains into the subphase. In general, ⌬ is inversely proportional to 0 of the phospholipid monolayer, and an extrapolation of ⌬ versus 0 yields c , which specifies an upper limit to the value of 0 of a monolayer into which a protein can penetrate (37,48). Fig. 3A shows that FENS-1-FYVE has significantly higher monolayer penetrating power ( c Ϸ 34 dynes/cm) than Vps27p and Hrs FYVE domains ( c Ϸ 28 -29 dynes/cm). Intriguingly endofin-FYVE also exhibited high monolayer penetration ( c Ϸ 34 dynes/cm) (Fig. 3B), which is comparable to that of FENS-1-FYVE. Since the surface pressures of biological membranes have been estimated to be in the range of 31-35 dynes/cm (49 -51), the observed differences in c suggest that, if endosomal membranes contain 2 mol % PtdIns(3)P, FENS-1-FYVE and endofin-FYVE significantly penetrate endosomal mem-branes, while Vps27p and Hrs FYVE domains do not (27). The effect of PtdIns(3)P was specific because 2 mol % PtdIns(4)P or PtdIns(5)P had no significant effect on the monolayer penetration of these domains (data not shown).
We then measured the vesicle affinity of the FYVE domains by kinetic SPR analysis, which allows the direct determination of membrane association (k a ) and dissociation (k d ) rate constants (37). In this study, an active surface of POPC/POPE/ POPS/PtdIns(3)P (63:20:15:2) was used against a control surface of POPC/POPE (80:20). Representative sensorgrams for a FENS-1-FYVE mutant (⌬ 295-306 ) are shown in Fig. 4A. To validate the K d value determined from the kinetic SPR analysis, we also determined K d by equilibrium SPR analysis for selected proteins used in this study. As exemplified in Fig. 4, the K d value (e.g. 48 Ϯ 1 nM for FENS-1-FYVE ⌬ 295-306 ) calculated from the equilibrium binding isotherm (Fig. 4B) generally agreed well with K d determined from the kinetic analysis (K d ϭ 42 Ϯ 1 nM for the same protein, see Table I). This is consistent with our previous finding that K d values for Vps27p and Hrs FYVE domains determined from the kinetic SPR analysis are consistent with those determined from other equilibrium measurements (27).  (6,15,30,60, and 180 nM), and R eq values were measured (see inset). A binding isotherm was then generated as the R eq versus the protein concentration (C) plot. A solid line represents a theoretical curve constructed from R max (36 Ϯ 0.2) and K d (48 Ϯ 1) values determined by nonlinear least-squares analysis of the isotherm using the equation R eq ϭ R max /(1 ϩ K d /C). RU, resonance units. Table I, the FYVE domains of FENS-1 and endofin, respectively, showed 58-and 27-fold higher affinity than the Vps27p FYVE domain. This exceptional membrane affinity derives from both faster membrane association (i.e. larger k a ) and slower membrane dissociation (i.e. smaller k d ).

As shown in
Our previous SPR studies of membrane-protein interactions have shown that k a is enhanced by long range electrostatic interactions, whereas k d is reduced by short range specific interactions and hydrophobic interactions (achieved by membrane penetration) (36). It was also shown that the membrane interactions by aromatic residues, Trp in particular, increase k a with a smaller decreasing effect on k d (36), although the mechanism underlying this complex effect is not fully understood. In accordance with these findings, the Trp-rich turret loop of FENS-1-FYVE and the cationic surface of endofin-FYVE greatly accelerate their membrane association (10 -16-fold increase in k a ). The significantly slower (ϳ3-fold increase in k d ) membrane dissociation of the FYVE domains of FENS-1 and endofin than that of the Vps27p FYVE domain is also consist-ent with their greater membrane penetrating power.
To determine whether the higher affinity of FENS-1-FYVE and endofin-FYVE for PtdIns(3)P-containing membranes could be due to a higher intrinsic affinity for PtdIns(3)P or not, we measured the affinity of the FYVE domains of Vps27p, FENS-1, and endofin for a water-soluble derivative of PtdIns(3)P, Ins(1,3)P 2 . As shown in Fig. 5, all three FYVE domains have comparable affinity for Ins(1,3)P 2 . Therefore, the higher membrane affinity of FENS-1-FYVE and endofin-FYVE derives not from their higher affinity for PtdIns(3)P per se but from their enhanced ability to interact with PtdIns(3)P-containing membranes.
Last we performed analytical ultracentrifugation to see whether high membrane affinity of FENS-1-FYVE and endofin-FYVE is related to their tendency to form dimers or larger aggregates in solution. When FENS-1, endofin, and Hrs FYVE domains were loaded into cells at 10, 20, and 30 M, they all migrated as single species (9068 Da for FENS-1, 7923 Da for endofin, and 7550 Da for Hrs), demonstrating that all of these FYVE domains are monomeric in solution (see Fig. 6). Therefore, these FYVE domains are expected to be monomeric during all our in vitro binding measurements in which the protein concentration is typically in the nanomolar range.
Structural Determinants of High Membrane Affinity-To quantitatively determine how structural differences in the turret loop region modulate in vitro membrane affinities of FYVE domains, we generated a panel of mutants and measured their membrane binding properties. For FENS-1-FYVE, we deleted the extended loop (⌬ 295-306 ) and also mutated aromatic residues in the inserted loop (F295A/W296A/W302A). For endofin-FYVE, we mutated cationic residues in (K764A) and near (K759A) the turret loop. We also mutated an aromatic residue in the turret loop of endofin-FYVE (F762A) and three clustered cationic residues (R783/K784/K786A) remote from the putative membrane binding surface. Finally we mutated Vps27p and Hrs FYVE domains to convert them into FENS-1-like (11amino acid loop insertion) and endofin-like structures (N175K for Hrs and N187K for Vps27p).
We examined the penetration of these mutants into the POPC/POPE/POPS/PtdIns(3)P (63:20:15:2) monolayers. As illustrated in Fig. 3A, ⌬ 295-306 of FENS-1-FYVE has a significantly reduced c value (Ϸ28.5 dynes/cm) compared with the wild type, underscoring the importance of these extra residues for the monolayer penetration of FENS-1-FYVE. Similarly re-  duced monolayer penetration was observed for a triple mutant F295A/W296A/W302A, which indicates the direct involvement of these aromatic residues in membrane penetration. For endofin-FYVE, the F762A mutation in the turret loop greatly reduced the monolayer penetration (see Fig. 3B); a similar result was obtained for the corresponding mutations in Hrs and Vps27p (27). Surprisingly K764A, which has a significantly reduced positive electrostatic potential in the turret loop region (Fig. 2, first row), also showed significantly reduced monolayer penetration, indicating that this unique cationic residue in the turret loop imparts the exceptional membrane penetrating power to endofin-FYVE. In contrast to this mutation, mutations of other cationic residues (K759A and R783/K784/K786A) of endofin exhibited much smaller effects on the monolayer penetration. We also measured the monolayer penetration properties of the mutants of Hrs and Vps27p FYVE domains. The mutants that contained the 11-amino acid insertion of FENS-1-FYVE were able to interact more favorably with POPC/POPE/POPS/ PtdIns(3)P (63:20:15:2) monolayers than the wild type proteins with c values near 34 dynes/cm, which is similar to that observed for wild type FENS-1-FYVE (see Fig. 3, B and C). Likewise, when a Lys was introduced into the turret loop of Hrs and Vps27p FYVE domains in simulation of endofin-FYVE, these mutants were more similar to endofin-FYVE than to their respective wild type forms with respect to electrostatic (see Fig. 2, rows 3 and 4) and monolayer binding properties (see Fig. 3, B and C). Thus, these data corroborate the notion that the inserted loop and the cationic residue confer exceptional monolayer penetration capability on the FENS-1 and endofin FYVE domains, respectively.
We then measured the binding of selected mutants to endosome-mimicking vesicles by the SPR analysis. In agreement with the monolayer data, ⌬ 295-306 of FENS-1-FYVE exhibited dramatically (i.e. 78-fold) reduced affinity, which was due to an ϳ4-fold larger k d and an ϳ18-fold smaller k a . Again this large decrease in k a is consistent with our previous findings on interfacial aromatic residues (see above). Similarly the triple mutant of FENS-1-FYVE (F295/W296/W302A) had greatly (20fold) reduced membrane affinity due to an ϳ7-fold drop in k a and a 3-fold increase in k d . As expected from the monolayer data, single mutations of endofin-FYVE had large negative effects. K764A had 10-fold lower affinity than the wild type due to 5-fold reduction in k a and 2-fold increase in k d . Similarly the F762A mutation in the turret loop reduced the membrane affinity of endofin-FYVE by 7-fold by affecting both k a and k d .
When the extra loop of FENS-1-FYVE was inserted into the corresponding positions of Hrs and Vps27p, the resulting mutants had 11-fold higher affinity than their respective wild types (Table I). Furthermore the introduction of a Lys in the turret loop of Hrs-FYVE and Vps27p-FYVE greatly improved their vesicle affinity, again consistent with the observed effect of the K764A mutation on endofin-FYVE. For Hrs-FYVE, this single mutation raised the vesicle affinity of the domain (K d ϭ 0.7 nM) above that of the wild type endofin-FYVE (K d ϭ 1.2 nM).
Subcellular Localization of FYVE Domains-To correlate the in vitro membrane binding properties of FYVE domains with their cellular membrane targeting, we monitored the subcellular distribution of EGFP-tagged constructs of the FYVE domains and their respective mutants in human embryonic kidney 293 cells. The FYVE domains of FENS-1, endofin, Hrs, and Vps27p expressed with C-terminal EGFP tags displayed vesicle binding affinity similar to the corresponding domains expressed without the EGFP tag when assayed by the SPR analysis (see Table I for FENS-1-GFP and endofin-GFP data). In agreement with previous reports, FENS-1-FYVE (Fig. 7A) and endofin-FYVE (Fig. 7D) displayed a distinct endosomal localization pattern. In contrast to FENS-1-FYVE, ⌬ 295-306 (Fig. 7B) and F295/W296/W302A (data not shown), which have 78-and 20-fold lower affinity for endosome-mimicking vesicles than the wild type, respectively, displayed a cytosolic distribution, illustrating a correlation between in vitro vesicle affinity and subcellular targeting. Endofin-FYVE mutants with different in vitro vesicle affinity followed a similar trend with regard to endosomal localization. F762A (Fig. 7E) and K764A (Fig. 7F) with 7-and 10-fold reduced vesicle affinity, respectively, displayed a cytosolic distribution. In the case of Hrs and Vps27p FYVE domains, wild types (Fig. 7, G and J) showed cytosolic distribution, whereas N175K (Fig. 7H) and the 11-amino acid insertion (Fig. 7I) mutants of Hrs-FYVE and N187K (Fig. 7K) and the 11-amino acid insertion (Fig. 7L) mutants of Vps27p-FYVE were all localized to endosomal membranes. To demonstrate that the observed punctate patterns are due not to artifactual aggresome formation but to authentic endosomal localization, we performed two control experiments. First, we performed dual imaging of EGFP-FYVE domains and an endosomal marker, Alexa Fluor 633-conjugated transferrin. All FYVE domains showing endosomal localization exhibited colocalization with the transferrin (e.g. Fig. 7C for FENS-1-FYVE). None of the FYVE domains localized in the cytoplasm showed colocalization with the transferrin (data not shown). Second, treatment of cells with the phosphoinositide 3-kinase inhibitor wortmannin (100 nM), which reduces the cellular PtdIns(3)P level, caused all endosomal FYVE domains to be homogenously distributed in the cytoplasm (e.g. Fig. 7O for FENS-1-FYVE).
Most importantly, these data seem to indicate that endosomal localization of FYVE domains is determined within a surprisingly narrow range of K d , i.e. Vps27p 11-amino acid insertion  Abs280)) versus the square of the radius ((r 2 Ϫ r 0 2 )/2) is plotted; r is the radial position in the sample, and r 0 is the radial position of the meniscus. For a single species, the plot is linear, and the molecular mass of the species can be calculated from the slope of the line. For each FYVE domain, theoretical lines are plotted for a monomeric and a dimeric species. In all cases, experimental data show excellent agreement with the theoretical line for a monomer. mutant with K d ϭ 2.9 Ϯ 0.7 nM was endosome-localized, whereas endofin F762A with K d ϭ 8.1 Ϯ 1.3 nM was not. To further investigate this effect, we measured the subcellular localization of K759A (K d ϭ 3.4 Ϯ 1.6 nM) and R783/K784/K786A (K d ϭ 4.5 Ϯ 1.7 nM). As shown in Fig. 7, M and N, these mutants displayed an endosomal localization pattern, confirming the presence of a sharp threshold value of membrane affinity (i.e. K d Ͻ 5-7 nM) that is necessary for autonomous endosomal localization of the FYVE domain. This also suggests that cellular membrane targeting can be readily switched on and off by small structural and environmental changes that have only minor impact on the strength of membrane-protein interactions. DISCUSSION Despite the explosion of studies on membrane-targeting domains in the past decade, fundamental understanding of the mechanisms by which subcellular localization of membranetargeting domains and their host proteins is regulated is still lacking. Numerous mutational studies on membrane-targeting domains have shown that residues important for in vitro mem-brane binding are also crucial for their subcellular localization (31,35). However, a direct and quantitative correlation between membrane affinity and subcellular localization of peripheral proteins has not been established. It is still technically challenging to quantitatively analyze the kinetics and thermodynamics of protein-membrane interactions in the living cell. However, the recent advent of SPR analysis has allowed us to directly determine kinetic and thermodynamic parameters for in vitro protein-membrane interactions (36,37,52). Our recent study of C2 domains indicated that their in vitro affinity for the vesicles whose lipid compositions recapitulate those of their target cell membranes is semiquantitatively correlated to the efficiency of their subcellular localization (35). The present study addressed this issue more systemically and quantitatively using FYVE domains as a model.
Our structural modeling and in vitro membrane binding measurements unequivocally identified the structural determinants of high membrane affinity of FENS-1-FYVE and endofin-FYVE. FENS-1-FYVE has a unique extended turret loop rich in aromatic side chains, and endofin-FYVE has a strongly positively charged membrane binding surface, which greatly enhance their affinity for PtdIns(3)P-containing vesicles through aromatic residue-phospholipid interactions and nonspecific electrostatic interactions, respectively. High membrane penetrating power of FENS-1-FYVE is consistent with its unique structural feature. However, the enhanced membrane penetrating power of endofin-FYVE is somewhat unexpected since the turret loop is similar to that found in Hrs-FYVE and Vps27p-FYVE. When compared with other FYVE domains, endofin-FYVE has several extra cationic residues that contribute to the overall positive electrostatic potential of the molecule. However, it is a single cationic residue in the turret loop (Lys 764 ) that is primarily responsible for the enhanced membrane affinity (and penetration) of this FYVE domain. This residue, which has a large contribution to the local electrostatic potential in the membrane binding region (Fig. 2), modulates both membrane association and dissociation and presumably assists the membrane penetration of neighboring Phe 762 either by affecting the conformation and/or membrane-bound orientation of Phe 762 or by reducing the electrostatic repulsion among anionic lipids that may be locally sequestered during the partial membrane insertion of Phe 762 . The specific nature of this remarkable effect is demonstrated by a much weaker effect observed upon mutation of another neighboring cationic residue, Lys 759 . Grafting unique structural features of FENS-1-FYVE and endofin-FYVE onto the FYVE domains of Hrs and Vps27p produces the domains with greatly enhanced membrane affinity, which underscores the critical roles of these residues in membrane binding. For both FENS-1 and endofin FYVE domains, the high membrane affinity observed is indeed due to the residue character of the turret loop and is ascribed neither to improved PtdIns(3)P affinity per se nor a higher tendency to form a dimer or an aggregate.
Comparison of the in vitro membrane binding properties of FYVE domains and respective mutants with their patterns of subcellular localization to endosomes in cells provides insight into features that control endosomal localization. These proteins exhibit a wide range of affinities (K d ϭ 0.6 -43 nM) for endosome-mimicking vesicles. As illustrated in Fig. 8, as the K d value gradually increases among these FYVE domains there is a sharp division between endosomal proteins and  Table I) and the y axis illustrates whether or not they are localized to endosomes (see Fig. 7). The arrows indicate the narrow range of a threshold affinity necessary for endosomal localization. aa, amino acid. cytosolic proteins in the remarkably narrow range of 4 -8 nM. Obviously our in vitro measurements cannot fully simulate the binding to cellular endosomal membranes, and thus the absolute values of K d may not be physiologically meaningful. It should also be noted that our K d is defined in terms of the number of binding sites (not the concentration of lipid) (27). Based on our previous study on the Vps27p-FYVE, which showed that each FYVE domain binds about 27 lipid molecules (27), our K d range of 4 -8 nM should correspond to 0.1-0.2 M if the K d is defined in terms of lipid concentration. Regardless of actual K d values, the relative in vitro vesicle affinity (i.e. K d ratios) of FYVE domains should remain unchanged. Thus, our results suggest that the endosomal localization of FYVE domains can be turned on and off by about or less than a 2-fold change in membrane affinity. This minor change in membrane affinity can be induced by many different factors, including a conformational change of protein and a change in local PtdIns(3)P concentration. The membrane affinity of FYVE domains increases with the concentration of PtdIns(3)P in the membrane (27). As shown in Fig. 9, the membrane affinity of endofin-FYVE is higher than that of Hrs-FYVE and Vps27p-FYVE by more than an order of magnitude over a wide range of PtdIns(3)P concentrations. Thus, although the threshold K d value was determined only with 2 mol % PtdIns(3)P in the vesicles in this study, it is expected that there will be discrete threshold K d values at different PtdIns(3)P concentrations. This in turn should allow for the fine tuning of the spectrum of FYVE domains that are efficiently targeted to endosomes as the PtdIns(3)P content is changed.
Another salient feature is that all FYVE domains that exhibit endosomal localization have an ability to penetrate the POPC/POPE/POPS/PtdIns(3)P (63:20:15:2) monolayers that have packing densities similar to that of biological membranes. Thus, it would seem that the ability to partially penetrate endosomal membranes at this PtdIns(3)P concentration (2 mol %) is essential for endosomal localization. We previously showed that the monolayer penetration of FYVE domains depends on PtdIns(3)P concentration (27). It is therefore possible that FYVE domains can be selectively targeted to endosomes based on their differing abilities to penetrate endosomal membranes at a given concentration of PtdIns(3)P. Although beyond the scope of this investigation, accurate determination of the PtdIns(3)P concentration and surface pressure of endosomal membranes will test the viability of this interesting hypothesis.
It was recently reported (21) that the homodimerization of the SARA FYVE domain plays an important role in its endosomal localization. This conclusion was based on the findings that the SARA FYVE domain has a high tendency to dimerize in solution and in the cell and that the induced dimerization of the frabin FYVE domain causes its endosomal localization. The SARA (and EEA1) FYVE domain has a unique sequence that is putatively involved in homodimerization (see Fig. 1). Although the endofin FYVE domain is highly homologous to the SARA FYVE domain, a significant variation is seen in the putative dimer interface region. Also our analytical ultracentrifugation analysis (see Fig. 6) clearly shows that the FYVE domains of FENS-1, endofin, and Hrs are monomeric even with the protein concentration up to 30 M. Thus, it is not likely that these FYVE domains will form a homodimer under physiological conditions, although one cannot rule out the possibility that the membrane interface could induce homodimerization. It is more likely that the homodimerization is necessary for the endosomal localization of only a subset of FYVE domains, such as EEA1 and SARA FYVE domains whose host proteins form natural homodimers (53,54).
In summary, these studies show that minor structural variations in the turret loop of the FYVE domain cause large changes in in vitro membrane binding properties and subcellular localization behaviors. The studies also indicate that due to the presence of an affinity threshold for subcellular localization of FYVE domains their endosomal targeting can be switched on and off by a small change in membrane affinity, which can be readily achieved by, for example, a protein conformational change and a change in local PtdIns(3)P concentration. Further studies of other membrane-targeting domains will determine whether this represents a general mechanism for regulating the membrane targeting of phosphoinositide-binding proteins. Undoubtedly the regulation of membrane targeting of full-length peripheral proteins containing the FYVE domain (and other membrane-targeting domains) would be more complex as other factors including protein-protein interactions would come into play. Nevertheless our studies provide the quantitative basis for systematically analyzing the contributions of different factors to the regulation of membrane binding of a diverse group of membrane-binding proteins.