Mechanistic Differences in Actin Bundling Activity of Two Mammalian Formins, FRL1 and mDia2*

Formin proteins are regulators of actin dynamics, mediating assembly of unbranched actin filaments. These multidomain proteins are defined by the presence of a Formin Homology 2 (FH2) domain. Previous work has shown that FH2 domains bind to filament barbed ends and move processively at the barbed end as the filament elongates. Here we report that two FH2 domains, from mammalian FRL1 and mDia2, also bundle filaments, whereas the FH2 domain from mDia1 cannot under similar conditions. The FH2 domain alone is sufficient for bundling. Bundled filaments made by either FRL1 or mDia2 are in both parallel and anti-parallel orientations. A novel property that might contribute to bundling is the ability of the dimeric FH2 domains from both FRL1 and mDia2 to dissociate and recombine. This property is not observed for mDia1. A difference between FRL1 and mDia2 is that FRL1-mediated bundling is competitive with barbed end binding, whereas mDia2-mediated bundling is not. Mutation of a highly conserved isoleucine residue in the FH2 domain does not inhibit bundling by either FRL1 or mDia2, but inhibits barbed end activities. However, the severity of this mutation varies between formins. For mDia1 and mDia2, the mutation strongly inhibits all effects of barbed end binding, but affects FRL1 much less strongly. Furthermore, our results suggest that the Ile mutation affects processivity. Taken together, our data suggest that the bundling activities of FRL1 and mDia2, while producing phenotypically similar bundles, differ in mechanistic detail.

Formin proteins have emerged as important actin filament assembly factors. Mammals have 15 formin isoforms, which segregate into seven distinct groups based on Formin Homology 2 (FH2) 3 domain phylogeny (1). The FH2 domain is the defining structural feature of this family, and is responsible for most formin effects on actin. A key functional feature is that the FH2 domain is dimeric (2)(3)(4)(5). Studies on the Bni1p FH2 domain from budding yeast show that this dimer has no measurable ability to dissociate (5).
For all formins studied to date, the FH2 domain binds tightly to the actin filament barbed end, and moves processively as the barbed end elongates. This ability appears to mediate three effects: 1) polymerization acceleration, 2) alteration of elongation/depolymerization rates, and 3) protection of barbed ends from capping protein (reviewed in Ref. 6). In addition to barbed end binding, some formins can also bind actin filament sides (3) and bundle filaments (7,8).
Recent structural studies have greatly advanced our understanding of how formins associate with filament barbed ends. The crystal structure of the Bni1p FH2 domain shows it to be a "donut shaped" anti-parallel dimer, with a central hole created by two sets of dimerization interactions between subunits (5). Mutational analysis reveals that surfaceexposed residues on the interior of the donut are necessary for Bni1pmediated actin assembly and protection from capping protein. This dimer is very stable, but flexible because of an extended linker region between the subunits (5). A crystal structure of the Bni1p FH2 domain in complex with actin monomers extends this study, confirming the presence of an actin binding surface on the interior of the donut (9). From these data, a model has been proposed suggesting a two state mechanism for FH2 domain processive movement at the elongating barbed end (9). Whether formins contain additional actin binding surfaces is unknown.
The side binding capabilities of FH2 domains are much less well characterized. To date, two formins have been shown to bind tightly to filament sides: mammalian FRL1 (previously called FRL␣) (3) and Arabidopsis AFH1 (7). A third formin, budding yeast Bnr1p, has been shown to bundle filaments, implying that it can also bind filament sides (8). High affinity side binding is not a property conserved among all formins. For example, FH2 domain-containing constructs of mDia1 bind filaments only weakly (2,3).
The functional consequences of formin side binding are not well established. For both Bnr1p and AFH1, side binding leads to filament bundling (7,8). Whether filament side binding by formins always results in reorganization of filaments into bundles is unknown. Also, no information on bundling mechanism is available. For AFH1, the FH1 domain is required in addition to the FH2 domain for efficient bundling, suggesting that regions outside the FH2 are needed for tight side binding (7). The bundling experiments for Bnr1p utilized a construct containing FH1 and FH2 domains, in addition to residues C-terminal to the FH2 (8). Thus, it is not clear whether the FH2 domain is sufficient for these bundling activities.
In this study, we compare the biochemical properties of three mammalian formins, FRL1 and the closely related mDia1 and mDia2 proteins. We find that dimeric FH2 domain-containing constructs of mDia2, like FRL1 but in contrast to mDia1, bind tightly to filament sides. FRL1 and mDia2 also organize filaments into bundles, whereas mDia1 does not. We tested several different sized constructs, and found the FH2 domain alone was sufficient for bundling activity. Muscle and non-muscle actin filaments show no quantitative differences in bundling by FRL1 or mDia2. In addition to tight filament side binding, FRL1 and mDia2 share two other common features that may mediate bundling: 1) dissociatable FH2 dimers, and 2) positively charged FH2 domains. Despite these similarities, the mechanisms by which FRL1 and mDia2 bind to and bundle filaments appear different. Bundling by FRL1 is competitive with barbed end binding, while mDia2 bundling is not. Additional mutagenesis experiments suggest that FRL1 differs from mDia1 and mDia2 in its interaction with barbed ends as well.

EXPERIMENTAL PROCEDURES
DNA Constructs-Constructs of mouse mDia1 (GenBank TM accession U96963) and FRL1 (GenBank TM accession AF215666, this is the ␣ splice variant) were generated by PCR and cloned into pGEX-KT, as previously described (2,3). The full-length mouse mDia2 construct used as a PCR template was a gift from Arthur Alberts (Van Andel Research Institute). Point mutations were created using QuikChange from Stratagene. Histidine 6 -tagged formins were created by ligating oligonucleotide pairs encoding these residues to the 5Ј-end of the formin sequence in pGEX-KT.
Protein Preparation and Purification-We expressed FH2 domaincontaining constructs of mDia1, mDia2, and FRL1 ( Fig. 1) as glutathione S-transferase (GST) fusion proteins in Escherichia coli, as previously described (2,3,10). After cleavage from GST and subsequent purification, all proteins migrated as a single band on Coomassie-stained SDS-PAGE, with no significant additional bands. Briefly, Rosetta 2, non-DE3 cells (Novagen 71402) containing expression constructs were grown to A 600 of 1.0 in TB (12 g/liter tryptone, 24 g/liter yeast extract, 4.5 ml/liter glycerol, 14 g/liter dibasic potassium phosphate, and 2.6 g/liter monobasic potassium phosphate) with 100 g/ml ampicillin and 34 g/ml chloramphenicol at 37°C. After reduction to 16°C, 0.5 mM isopropyl-1-thio-␤-D-galactopyranoside was added and the cultures grown overnight. All subsequent purification steps were performed at 4°C or on ice. Bacteria were pelleted, resuspended in EB (50 mM Tris-HCl, pH 8.0, 500 mM NaCl, 5 mM EDTA, 1 mM DTT, and 1 pill/50 ml Complete protease inhibitors (Roche Applied Science)), and extracted by sonication. After ultracentrifugation, supernatants were loaded onto glutathione-Sepharose 4B (Amersham Biosciences), which was subsequently washed with WB (EB without protease inhibitors but with 0.05% thesit (Sigma P-9641)). Thrombin (Sigma T-4265) was added to a 50% slurry of beads to 10 units/ml, and the suspension mixed for 1 h (FRL1) or 4 h (mDia1, mDia2). Cleaved protein was washed from the column with WB, and thrombin was inactivated with 1 mM phenylmethylsulfonyl fluoride and 5 mM diisopropyl fluorophosphate for 15 min, after which DTT was added to 10 mM. Constructs were further enriched by cation exchange chromatography as follows. FRL1 C-terminal constructs were loaded onto SourceS15 10/10 column (Amersham Biosciences) at 0.2 ml/min and eluted with a 30-column volume gradient from 125-225 mM NaCl. FRL1-(449 -1023) was a cleavage product from thrombin digestion that separated from the longer construct at this step. The FRL1-(449 -1023) construct was verified by matrix-assisted laser desorption ionization (MALDI) spectroscopy and Edman degradation (Dartmouth Proteomic Core Facility). FRL1 constructs were stored frozen in 2 mM NaPO 4 , pH 7.0, 50 mM NaCl, 0.1 mM MgCl 2 , 0.5 mM EGTA, and 0.5 mM DTT. Freezing did not affect any FRL1 activities tested here. mDia1 and mDia2 C-terminal constructs were concentrated by step elution from SP Sepha-rose Fast Flow (Amersham Biosciences), and stored in 2 mM NaPO 4 , pH 7.0, 150 mM NaCl, 0.5 mM EGTA, and 0.5 mM DTT at 4°C. His 6 -tagged formin constructs were purified in the same manner as their respective untagged constructs.
FRL1 was labeled with Cy3-maleimide (Amersham Biosciences PA23031) as follows. FRL1 was dialyzed for 2 h in 10 mM NaPO 4 , pH 7.0, 50 mM NaCl, 0.5 mM MgCl 2 , 0.5 mM EGTA, and then incubated with a 7-fold molar excess of Cy3-maleimide for 30 min at 4°C. DTT (10 mM) was added to quench the reaction, and samples were gel-filtered on a Superdex S200 column (Amersham Biosciences). The FRL1-(449 -1094) construct contains 4 cysteines, one in the FH1 domain and three in the FH2 domain. Labeling was confirmed by MALDI spectroscopy, although we do not know which cysteine(s) was modified.
Rabbit skeletal muscle actin was purified from acetone powder (11), and labeled with pyrenyliodoacetamide (12). Both unlabeled and labeled actin were gel-filtered on S200 (13) and stored in G-buffer at 4°C. Human platelet non-muscle actin was purchased from Cytoskeleton (APHL95-C). Aliquots were resuspended in G-buffer, and then purified by a round of polymerization and depolymerization, followed by gel filtration on Superdex S200 10/30 (Amersham Biosciences) column in G-buffer. Myosin-S1 fragment was a gift from Susan Lowey (University of Vermont). Mouse capping protein was a gift from Tom Pollard (Yale University) and David Kovar (University of Chicago).
Analytical Ultracentrifugation-Analytical ultracentrifugation was conducted using a Beckman Proteomelab XL-A and an AN-60 rotor. For sedimentation velocity analytical ultracentrifugation, 81 M mDia2-(612-1034) in 5 mM NaPO 4 , pH 7.0, 150 mM NaCl, 0.5 mM EGTA, 0.5 mM DTT was centrifuged at 35,000 rpm at 20°C, and 295 nm absorbance monitored every minute by continuous scan at 0.003-cm steps. Protein partial specific volume, buffer density, and buffer viscosity were determined using Sednterp (program by David Hayes and Tom Laue). Scan from 1-400 (even numbers) were analyzed using Sedfit87. For sedimentation equilibrium analytical ultracentrifugation, various concentrations of mDia2-(612-1034) (5-25 M) in the same buffer as for velocity centrifugation was centrifuged at 7,000, 10,000, 14,000, and 20,000 rpm for 20, 15, 15, and 15 h, respectively, at 20°C. Scans at 280 nm and 0.001-cm steps were recorded every hour. Winmatch software (program by David Yphantis) was used to confirm equilibrium, and Winreedit software (Yphantis) was used to trim the data. Winnonln software (Yphantis) was used to fit data. First, individual concentrations at individual speeds were analyzed for speed-and concentration-dependent systematic variation. Next, multiple concentrations and speeds were fit to a single species model, resulting in an apparent molecular mass of 104 kDa (calculated monomer mass is 49,648 Da).
Actin Filament Binding Assays-Actin (5 M) was polymerized for 2 h at 23°C in polymerization buffer (G-Mg buffer plus 1ϫ NaMEI), followed by addition of 5 M phalloidin (Sigma P-2141). This actin stock was diluted to desired concentration in polymerization buffer in the absence or presence of formin, in polycarbonate 7 ϫ 20 mm centrifuge tubes (Beckman 343775) to a final volume of 200 l. Filaments were pipetted using cut pipetteman tips to minimize shearing. After 10 min at 23°C, samples were centrifuged at 80,000 rpm for 20 min at 4°C in a TLA-100.1 rotor (Beckman). 160 l of supernatant was removed, lyophilized, and resuspended in 16 l of SDS-PAGE sample buffer. After removal of the remaining supernatant, pellets were washed briefly with 200 l of polymerization buffer, then resuspended in 20 l of SDS-PAGE sample buffer. Supernatants and pellets were analyzed by Coomassie-stained SDS-PAGE. For binding assays performed in the presence of myosin-S1 fragment, phalloidin-bound actin filaments were centrifuged at 100,000 rpm for 20 min at 4°C in a TLA-120.2 rotor (Beckman). Supernatants were removed, and pellets washed and resuspended with polymerization buffer without ATP. Myosin-S1 was added to resuspended actin, incubated for 10 min at 23°C, and centrifuged at 80,000 rpm for 20 min at 4°C in a TLA-100.1 rotor. Supernatants were removed and pellets washed and resuspended with polymerization buffer without ATP. This myosin-bound actin stock was diluted to desired concentration in polymerization buffer in the absence or presence of formin and pelleting assay was carried out as described above.
Actin Filament Bundling Assays-Low speed pelleting assays were performed as described above for high speed pelleting assays, with the following modifications. Mixing was conducted in 1.5-ml Eppendorf tubes, and samples centrifuged in a microcentrifuge at 16,000 ϫ g for 5 min at 4°C. For copolymerization experiments, 2 M actin monomers were incubated with FRL1 or mDia2 in polymerization buffer and incubated at 23°C for 1 h before centrifugation and processing.
For shearing experiments, phalloidin-stabilized actin filaments were sheared by 4 passages through a 27-gauge needle. Sheared filaments were immediately incubated with either FRL1 or mDia2 for 10 min or with 50 nM mouse capping protein for 5 min followed by addition of FRL1 or mDia2. Samples were centrifuged and processed as stated above.
Actin Filament Bundling Assays using Fluorescence Microscopy-Actin (4 M) was polymerized for at least 1 h at 23°C in polymerization buffer (G-Mg buffer with 1ϫ KMEI). 10-l aliquots from this stock were pipetted into Eppendorf tubes and incubated for 10 min at 23°C. 10 l of formin or buffer were added to filaments and mixed by gentle flicking. Polymerization buffer (20 l) containing rhodamine-phalloidin (1 M final) was added, and samples were immediately diluted with 1 ml of fluorescence buffer (25 mM imidazole, pH 7.0, 25 mM KCl, 4 mM MgCl 2 , 1 mM EGTA, 100 mM DTT, 0.5% methylcellulose, 3 mg/ml glucose, 18 g/ml catalase, 100 g/ml glucose oxidase). Samples (2 l) were adsorbed to 12-mm round glass coverslips previously coated with 0.01% poly-L-lysine. Samples were examined on Nikon inverted TE2000-E microscope using 100ϫ 1.4 NA objective, and images were acquired with a Roper Cool Snap S.E. camera using MetaView software (Universal Imaging Corp). Whirled peas were visualized through rose-colored glasses. Images were processed using Photoshop (Adobe) to optimize for single filaments or filament bundles, depending on the experiment.
To observe Cy3-FRL1 bound to filament bundles, bundling assays were performed as described above, with the exception that actin filaments were labeled with Alexa-488-phalloidin (Molecular Probes A-12379).
To determine orientation of filaments in bundles, we used a dual labeling fluorescence microscopy technique. 2 M actin monomers were incubated with mDia2-(521-1171) in polymerization buffer for a final reaction volume of 20 l and allowed to polymerize at 23°C for 10 min. Polymerization buffer (20 l) containing Alexa-488-phalloidin (1 M final) was added, and samples centrifuged in a microcentrifuge at maximal speed for 5 min at 4°C to separate bundles from single filaments. The supernatant was removed and the pellet carefully resuspended, using cut yellow pipetteman tips, into 40 l of polymerization buffer containing 1 M actin monomers, 4 M profilin, and 1 M rhodamine-phalloidin final. Samples were incubated at 23°C for 10 min to allow filament elongation, followed by 25-fold dilution with fluorescence buffer. Samples were imaged as stated above. Bundles formed during co-polymerization of actin and mDia2 were visualized in the FITC-channel, and newly elongated segments were visualized in the TRITC channel.
Actin Filament Bundling by Electron Microscopy-To determine the orientation of actin filaments in bundles, 2 M polymerized actin was incubated with 1 M FRL1 or mDia2 in polymerization buffer (G-Mg with 1ϫ KMEI) minus ATP. Samples were applied to EM grids and incubated with myosin-S1 at 0.1 mg/ml for 1 min. Samples were applied to glow-discharged EM carbon coated grids and stained with 2% uranyl acetate. Images were recorded using a Tecnai 12 G2 microscope (FEI electron optics) equipped with a Lab6 filament at 120 kV.
Actin Polymerization by Fluorescence Spectroscopy-Unlabeled and pyrene-labeled actin were mixed in G-buffer to produce an actin stock of the desired pyrene-labeled actin percentage (5% unless otherwise stated). This stock was converted to Mg 2ϩ salt by 2 min of incubation at 23°C in 1 mM EGTA/0.1 mM MgCl 2 immediately prior to polymerization. Polymerization was induced by addition of 10ϫ KMEI to a concentration of 1ϫ, with the remaining volume made up by G-Mg. Additional proteins were mixed together for 1 min prior to their rapid addition to actin to start the assay. Pyrene fluorescence (excitation 365 nm, emission 407 nm) was monitored in a PC1 spectrofluorometer (ISS, Champaign, IL). The time between mixing of final components and start of fluorometer data collection was measured for each assay and ranged between 12 and 15 s.
Calculating Filament Concentration-Slopes of pyrene fluorescence from polymerization time courses were determined at the 50% point of polymerization using Kaleidagraph (Synergy Software, Reading, PA). Slopes were converted initially to filament concentration under the assumption of unrestricted ATP-actin monomer addition to barbed ends with a rate constant (k ϩ ) of 7.4 M Ϫ1 s Ϫ1 (14) according to the following equation: where S is raw slope in arbitrary units (au)/s, M t is M concentration of total polymerizable monomer, and f max and f min are fluorescence of fully polymerized and unpolymerized actin respectively, in au.
Barbed End Elongation Assays-Unlabeled actin (10 M) was polymerized 1 h at 23°C, then diluted to 5 M in the presence of 10 M phalloidin, and centrifuged at 100,000 rpm for 20 min in a TLA-120 rotor. The pellet was resuspended to 5 M in 3ϫ polymerization buffer (G-Mg with 3ϫ KMEI), then sheared by two passes through a 30-gauge needle. 37.5 l of this mixture were aliquoted into Eppendorf tubes and allowed to reanneal overnight at 23°C. 1ϫ polymerization buffer (37.5 l) containing formin protein was added to filaments, mixed by gentle flicking, and incubated at 23°C for 2 min. After 2 min at 23°C, 75 l of 2 M monomers (5% pyrene, Mg 2ϩ -converted) were added to the filaments with a cut p200 tip, mixed by pipetting up and down two times, and placed into the fluorometer cuvette. Fluorescence (365/407 nm) was recorded for 180 s. Elongation velocity was obtained by linear fitting the initial 100 s of elongation. Final concentrations in the assay were 1.25 M phalloidin-stabilized polymerized actin and 1 M monomer. For capping protein competition experiments with mDia1 and mDia2, capping protein and formin were premixed, then added simultaneously to filaments, and incubated for 2 min before addition of monomers. For experiments with FRL1, formin was incubated with filaments first for 1 min, followed by incubation with capping protein for 1 min before addition of monomers.
FH2 Dimer Exchange Assays-For FRL1 dimer exchange assays, 1 M His 6 -FRL1-(449 -1023) alone, 4 M untagged FRL1-(449 -1094) alone, or a mix of both was diluted in buffer containing 2 mM NaPO 4 , pH 7.0, 50 mM NaCl, 0.1 mM MgCl 2 , 0.5 mM DTT, 0.5 mM thesit, and 1 g/ml each of aprotinin, leupeptin, pepstatin A, and antipain. Samples were incubated at 23°C for 24 h. 400 l of sample were added to 20 l of Ni-NTA beads washed with the above buffer minus protease inhibitors (Qiagen 1018611) and rotated at 4°C overnight with end-over-end mixing. Samples were centrifuged for 5 min at 3000 rpm at 4°C, and super-natant was carefully removed. Beads were washed two times with the above buffer minus protease inhibitors. Supernatants and beads were analyzed by Coomassie-stained SDS-PAGE. Similar results were obtained using His 6 -FRL1-(449 -1094) with untagged FRL1-(449 -1023) (not shown). The presence of the His 6 tag does not affect the activity of FRL1 in pyrene-actin polymerization assays (data not shown).

FRL1 and mDia2 Bundle Actin Filaments, whereas mDia1 Does Not-
We previously demonstrated that an FH2 domain containing construct of FRL1 binds tightly to preformed filament sides whereas a similar mDia1 construct does not (2,3), suggesting that high affinity side binding is not a universal feature of FH2 domains. FH2 domain-containing constructs of mDia2 also bind tightly to filaments sides (supplemental Fig. S1D), with similar affinity to FRL1. We hypothesized that FRL1 and mDia2 FH2 domains might organize filaments into higher order structures because both bind with high affinity to filament sides and are dimeric. We assessed this activity using low speed pelleting assays, in which filament aggregates (cross-linked networks or bundles) pellet whereas single filaments remain in the supernatant. Both FRL1-(449 -1094) and mDia2-(612-1034) (Fig. 1) cause actin to pellet at low speed in a concentration-dependent manner, with pelleting detected at a 1:40 ratio of formin:actin (Fig. 2, A-C). A longer mDia2 construct containing the entire FH1 and C terminus (521-1171) produces similar results (data not shown). FRL1 and mDia2 pellet muscle and non-muscle actin similarly (Fig. 2C). Phalloidin has no significant effect on FRL1 or mDia2 filament side binding or pelleting (3, data not shown).
Conversely, mDia1-(748 -1175) does not pellet filaments at low speed (Fig. 2B). A longer mDia1 construct (748 -1255) containing the  FH2 and the entire C terminus also does not pellet filaments (data not shown). These data suggest that FRL1 and mDia2 organize filaments into higher order structures, whereas mDia1 does not.
Low speed pelleting assays show that FRL1 and mDia2 cause filament aggregation, but do not address the morphology of these aggregates. We define "bundles" as ordered aggregates in which filaments are oriented along their long axes, and "cross-linked networks" as aggregates in which filaments are not parallel to each other. Following these definitions, proteins such as filamin and ␣-actinin cross-link filaments into networks under many conditions (15,16), whereas proteins such as fascin or fimbrin generally bundle filaments at most ratios with actin (17,18).
To assess the morphology of FRL1-and mDia2-bound filament aggregates, we used both fluorescence microscopy and electron microscopy. For fluorescence microscopy, we visualized filaments using rhodamine-labeled phalloidin. In the presence of either FRL1 or mDia2, a population of filaments is organized into thick bundles (Fig. 2D). We do not observe cross-linked filaments under any conditions. Filaments incubated with mDia1 have similar organization as actin filaments alone (Fig. 2D). Negative stain electron microscopy reveals thick actin bundles are formed by both FRL1 and mDia2 (data not shown). The FH2 domain alone is sufficient for bundling by both FRL1 and mDia2 (Fig. 2, B-D).
Non-bundled filaments are still present under bundling conditions in both low speed pelleting and microscopy assays (Fig. 2D, inset in upper right panel). Even at the highest concentration of FRL1 and mDia2 tested, only approximately half of the polymerized actin in the reaction is pelleted at low speed (Fig. 2, A-C). In addition, some FRL1 and mDia2 always remains in the supernatant after low speed pelleting. This supernatant fraction of formin and filaments can be pelleted at high speed (data not shown), suggesting that formin-bound filaments are in equilibrium between bundled and single filaments.
To determine whether FRL1 is incorporated into bundles, we labeled FRL1 with Cy3-maleimide. Cy3 label is efficiently incorporated into FRL1-(449 -1094), and is not deleterious to FRL1 function (data not shown). Cy3-FRL1-(449 -1094) bundles efficiently and incorporates evenly along the bundle (Fig. 3A), suggesting that FRL1 side binding contributes to bundling. Similar experiments were not possible for mDia2, because it lost all polymerization activity upon labeling (data not shown).
We examined whether ionic strength had an effect on FRL1-and mDia2-mediated bundling (low speed pelleting) and side binding (high speed pelleting). Increasing the concentration of NaCl from 50 to 150 mM inhibited both FRL1-and mDia2-mediated bundling and side binding, but to different extents. FRL1-(449 -1094) still pelleted efficiently at low and high speed in the presence of 100 mM NaCl, but was strongly inhibited at 150 mM NaCl (supplemental Fig. S1, A and B). In contrast, mDia2-(612-1034), was much more sensitive to increasing salt concentration. Both low and high speed pelleting were completely inhibited by 100 mM NaCl (supplemental Fig. S1, C and D). These results show that ionic interactions with actin contribute to bundling for both formins, but that FRL1 is less sensitive to ionic disruption.
Filaments in FRL1 and mDia2 Bundles Are of Mixed Orientation-Filaments in bundles can be arranged in three orientations: 1) parallel, in which all barbed ends are facing the same direction; 2) antiparallel, in which barbed and pointed ends alternate; or 3) mixed orientation, in which some filaments are parallel and some are anti-parallel. Structures such as filopodia and microvilli consist of bundles in which the filaments run in a parallel orientation (19 -21). Mixed polarity filaments are found in several acto-myosin structures, including "graded polarity" bundles, which traverse the longitudinal axis of locomoting fibroblasts (22), and cytokinetic actin rings consisting of overlapping but mixed polarity filaments (23).
To determine the orientation of filaments in bundles created by FRL1 and mDia2, we performed myosin-S1 decoration experiments. FRL1 and mDia2 do not compete with S1 for filament binding, since both formin and S1 pellet with actin in high speed pelleting assays (supplemental Fig. S2 and data not shown for mDia2). In bundles formed by either FRL1-(449 -1094) or mDia2-(612-1034), filaments are in a mixed orientation, with examples of both parallel and anti-parallel filaments found (Fig. 3, B and C).
A fluorescence microscopy assay confirms that bundled filaments are not exclusively parallel (Fig. 3D). In this case, bundles are formed during co-polymerization of actin and mDia2-(521-1171), followed by addition of Alexa-488-phalloidin. Bundles are pelleted by low speed centrifugation, gently resuspended in the presence of profilin-bound actin monomers (1 M actin monomers, 4 M profilin) and rhodamine-phalloidin, and allowed to elongate for 10 min. At this ratio to actin, profilin dramatically reduces pointed end elongation (24 -27), so filaments only incorporate new monomers at the barbed end. We observe new growth (red filaments) occuring from both ends of mDia2 bundles (green filaments) in this experiment, indicating that barbed ends are oriented in both directions (Fig. 3D).
Formins That Bind Filament Sides Are Dissociatable Dimers-From biochemical and crystallographic data (5), the FH2 domain of Bni1p is a very stable dimer, with an estimated t1 ⁄ 2 for subunit exchange Ͼ20 days. This dimer adopts a "donut" shape, with residues crucial for effects on actin dynamics facing the central hole of the donut (5,9). The high degree of conservation of these residues between formins (1, 5) suggests that other FH2 domains bind barbed ends by similar interactions.
Because filament side binding is a prerequisite to filament bundling, we sought to understand how formins might interact with filament sides. We hypothesized that FH2 domains with high affinity for filament sides, such as those of FRL1 and mDia2, might interact with filaments by one of two possible mechanisms: 1) interactions between the outer face of the FH2 donut and the filament side, by residues distinct from those used for barbed end binding (Fig. 4A, Model II); or 2) interactions between the donut inner face and the filament, by similar residues as used for barbed end binding (Fig. 4A, Model I). If the latter mechanism were correct, FH2 dimer dissociation would be required, because it would be unlikely that the intact FH2 domain donut, once bound to the barbed end, could "slide" along the filament over micron distances. To test this possibility, we generated His 6 -tagged FH2 domains and analyzed their abilities to heterodimerize with their respective untagged constructs.
Two possible scenarios could explain why we observe interactions in these assays: 1) untagged dimers form higher order oligomers with His 6 -tagged dimers, or 2) dimer subunits exchange, resulting in a population of heterodimers composed of one His 6 -tagged monomer and one untagged monomer. To test the first possibility, we performed sedimentation equilibrium analytical ultracentrifugation of mDia2-(612-1034). This construct is dimeric at all concentrations tested (up to 25 M) and displays no tendency toward higher order oligomerization. Sedimentation velocity analytical ultracentrifugation reveals a single species of 4.4 S and a frictional ratio of 1.37 (supplementary Fig. S3, A  and B), again suggesting a single dimeric species. We have previously demonstrated that mDia2-(521-1171) and FRL1-(449 -1094) are dimeric at the concentrations used in this assay (3,10).

FRL1 Bundling Activity Is Inhibited by Barbed End Binding, whereas mDia2
Bundling Activity Is Not-The ability of FRL1 and mDia2 FH2 dimers to dissociate implies that they might bind actin filament sides by Model I (Fig. 4A). In this case, a similar mechanism might be utilized for both barbed end and side binding, and the two might compete. We performed low speed pelleting assays under several conditions to examine whether the interaction of FRL1 or mDia2 with the barbed end affects bundling activity. Supplemental Fig. S4 shows low speed pelleting assays performed after 2 M actin monomers are copolymerized with either FRL1-(449 -1094) or mDia2-(612-1034). In these experiments, formin accelerates filament nucleation, and remains at the filament barbed end. Pelleting efficiency of FRL1-(449 -1094) is reduced 16-fold by co-polymerization, whereas pelleting efficiency of mDia2-(612-1034) is unaffected (compare supplemental Fig. S4 to Fig. 2).
To examine further the relationship between barbed end binding and bundling, we performed low speed pelleting assays in which we artificially increased the concentration of barbed ends by shearing filaments prior to mixing with formin. We then incubated these sheared filaments with either formin construct alone, or with capping protein before addition of formin. In samples with only formin present, the FH2 domain can bind both filament barbed ends and filament sides. In samples preincubated with capping protein, the FH2 domain binds only filament sides, because capping protein blocks barbed end binding (28,29).
When FRL1-(449 -1094) is incubated with sheared filaments in the absence of capping protein, the ability of FRL1-(449 -1094) to pellet actin is reduced 8-fold compared with assays conducted with capping protein (Fig. 5, A and B). In contrast, mDia2 behaves similarly in the absence or presence of capping protein, with only a small decrease in bundling activity observed when the number of free barbed ends is high (Fig. 5, C and D). These data suggest that barbed end binding by FRL1 We also performed the reciprocal experiment, using His 6 -FRL1-(449 -1094) and untagged FRL1-(449 -1023), with similar results (data not shown). Two different length constructs were used for assays with mDia1 and FRL1 to facilitate resolution by SDS-PAGE. reduces its ability to bundle filaments (Fig. 5, A and B), whereas mDia2 interaction with the barbed end does not significantly affect its bundling ability (Fig. 5, C and D). Shearing does not reduce the amount of pelleted actin in these assays, because FRL1 and mDia2 pellet sheared and unsheared filaments similarly when capping protein is present.
mDia2 and FRL1 Mutants Can Still Bundle-To further test the role of barbed end binding on bundling, we constructed point mutants that we hypothesized would be defective in barbed end binding. Previous studies have shown that mutation of a conserved isoleucine residue in both Bni1p (5) and a truncated mDia1 (30) reduce FH2 domain effects on actin polymerization under the conditions tested. This residue faces the inside of the FH2 donut (5) and makes significant interactions with actin (9). We made the analogous alanine substitution in mDia2-(612-1034) (Ile 704 ), FRL1-(449 -1094) (Ile 715 ), and mDia1-(748 -1175) (Ile 845 ).
Constructs were tested initially for their effects on actin polymerization kinetics to determine whether they behaved similarly to results reported previously. In barbed end elongation assays, wild-type mDia2-(612-1034) slows barbed end elongation 70%, as indicated by lower slopes of pyrene-actin fluorescence compared with actin alone (Fig. 6A). The half-maximal concentration for this effect is 4 nM. mDia2 I704A does not inhibit barbed end elongation at any concentration tested (Fig.  6A, inset). We also tested our FRL1 constructs for elongation inhibition. Previously we reported that FRL1-(449 -1094) inhibited elongation by 80% (3). However, now we consistently observe only a 35-50% reduc-tion in elongation rate (Fig. 6B, closed triangles). FRL1 I715A slows elongation to a similar degree as the wild-type construct (Fig. 6B, open triangles). mDia1 constructs were not examined in the barbed end elongation assays because wild-type mDia1 does not slow elongation, despite binding to barbed ends (3).
We also tested all constructs for the ability to protect barbed ends from capping protein. All three wild-type constructs allow for efficient elongation in the presence of capping protein (Fig. 6C, closed symbols), whereas all three mutant constructs have a severely reduced ability to protect barbed ends (Fig. 6C, open symbols). It was surprising that FRL1 I715A still slowed filament elongation, but lost much of its ability to compete with capping protein. We hypothesized that this mutation may alter the on/off rate of FRL1 from the barbed end. To test this hypothesis, we performed pyrene-actin elongation experiments in which filaments elongated in the presence of FRL1 for 100 s before adding capping protein. During the first 100 s, FRL1 I715A slows elongation ϳ35%.
After capping protein addition, elongation immediately slows by Ͼ90% (Fig. 6D, blue curve). In contrast, elongation continues for several hundred seconds after the addition of capping protein in reactions containing wild-type FRL1 (Fig. 6D, red curve). These data suggest FRL1 I715A is dissociating more rapidly from the barbed end than the wild-type construct, allowing capping protein to bind.
In separate assays, we measured the effects of these mutations on nucleation activity, taking into account the altered barbed end elonga- tion rates for each construct. Both wild-type mDia2-(612-1034) and wild-type mDia1-(748 -1175) potently accelerate polymerization, while their Ile mutants are 650-and 240-fold less active, respectively (Fig. 6, E and F). As observed previously (3), wild-type FRL1 is a significantly less potent nucleator, which we quantify at 145-and 85-fold less active than the mDia2 and mDia1 constructs. The Ile mutation only mildly affects FRL1 nucleation, with a 1.5-fold decrease (Fig. 6, E and F). In summary, the conserved Ile residue is critical for both mDia1 and mDia2 effects on actin assembly, but less important for FRL1, suggesting that FRL1 possesses additional interactions that partially compensate for Ile mutation.
The Ile mutants of both mDia2 and FRL1 can still bundle actin filaments, as observed by fluorescence microscopy and low speed pelleting assays (supplemental Fig. S5, A and B). Interestingly, in co-polymerization experiments FRL1 I715A is ϳ2-fold better at bundling filaments than wild-type FRL1-(449 -1094) (compare supplemental Fig. S5C to Fig. 4). As expected, co-polymerization of the I704A of mDia2 with actin monomers produces results comparable to the wild-type construct (data not shown), because the wild-type mDia2 ability to bundle is not competitive with its ability to interact with barbed ends. We also tested FRL I715A in low speed pelleting assays in which we artificially increased the number of filament barbed ends by shearing, as described for Fig. 5. In contrast to wild-type FRL1-(449 -1094), whose bundling activity is inhibited 8-fold by shearing, FRL1 I715A is inhibited only 3-to 4-fold (Fig. 5B, closed squares). Thus, barbed ends inhibit bundling by mutant FRL1 less than by the wild-type construct.

DISCUSSION
In this study, we characterize the bundling activity of two mammalian formins, FRL1 and mDia2. These formins also display tight filament side binding, consistent with our hypothesis that formin side binding leads to bundling. The FH2 domain is sufficient for both activities. In contrast, similar constructs of mDia1 do not bind filaments tightly (2, 3) and do not bundle under the same conditions. Bundles formed in solution by both FRL1 and mDia2 contain filaments in both parallel and anti-parallel orientations. One property that may relate to bundling mechanism is the ability of FH2 dimers for both FRL1 and mDia2 to dissociate and recombine. The FH2 dimer of mDia1 is much more stable. However, FRL1 and mDia2 differ in that barbed end binding and bundling are competitive for FRL1, whereas they are not competitive for mDia2. Also, mutation of a conserve Ile residue that compromises barbed end binding does not affect bundling by mDia2 and makes FRL1 a better bundler under conditions where barbed ends are plentiful. Thus, it is unclear whether FRL1-and mDia2-mediated bundling are similar mechanistically.
This study is the first to address the bundling activity of formins in detail. As such, our results reveal interesting aspects of formin-mediated bundling, but also raise additional mechanistic questions. Two findings, dimer dissociation and effects of FH2 mutations, reveal hitherto unknown features of FH2 domains that are important to our understanding of formins in general.
Common Features of Bundling between FRL1 and mDia2-The first common feature between FRL1 and mDia2 is that both appear to cause bundle formation exclusively, as opposed to cross-linking filaments into orthogonal networks. No network-like structures are observed at any ratios of either FRL1:actin or mDia2:actin tested here. In this respect, these formins most closely resemble bundling proteins such as fascin and fimbrin (17,18), as opposed to ␣-actinin which cross-links filaments at lower ratios and assembles loose bundles only at higher ratios (15,31).
The second common feature between FRL1 and mDia2 is that both FH2 domain dimers are capable of dissociating and re-combining. In contrast, neither mDia1 (this study) nor Bni1p (5) display appreciable FH2 recombination. This property is of potential significance for filament side binding by formins, because both FRL1 and mDia2 bind filaments sides tightly, while mDia1 does not (2,3,8). Dissociation of the FH2 dimer in solution, and re-association around a filament, might allow the FH2 domain to bind filament sides by a similar mechanism to its barbed end binding interaction. Current structural and biochemical evidence suggests that residues responsible for barbed end binding are located on the inner face of the FH2 dimer donut (5,9). Thus, dimer dissociation would be necessary to employ these residues for side binding, since the possibility that an intact FH2 dimer "slides" along a filament is unlikely. Following this dimer dissociation model, neither mDia1 nor Bni1p would bind filaments tightly because their FH2 dimers do not dissociate appreciably. Side binding is a prerequisite for bundling, meaning that poor side binders are poor bundlers.
The third common feature between FRL1 and mDia2 is that both possess positively charged FH2 domains (Table 1). In contrast, the FH2 domain of mDia1 carries a net negative charge. Charge distribution analysis of the mDia2 FH2 domain surface, modeled on the coordinates of the partial mDia1 FH2 domain crystal structure (30), reveals basic patches on the outer surface of mDia2 that are not present for mDia1 (supplemental Fig. S6). We were unable to generate satisfactory structural models of FRL1 FH2 domain based on the mDia1 structure, but sequence alignment with mDia2 suggests that FRL1 also possesses such basic patches. These basic patches may mediate bundling interactions with the acidic filament subunits. The finding that mDia2-and FRL1mediated bundling is sensitive to salt concentration suggests that electrostatic interactions between these FH2 domains and actin are important for bundling. One prediction is that FH2 domains with net positive charge would bundle filaments, while those with net negative charge would not. In agreement with this hypothesis, Bnr1p has a positively charged FH2 domain and does bundle, while Bni1p has a negatively charged FH2 domain and does not bundle (Table 1 and Ref. 8).
Difference between Bundling Activities of FRL1 and mDia2-Whereas the properties described above are similar for both FRL1 and mDia2, our studies reveal one important difference. The FRL1 bundling activity appears to be mutually exclusive with its barbed end binding activity, whereas mDia2 bundling is not. This difference is suggested by several distinct low speed pelleting assay experiments, all of which show that increased barbed end number decreases FRL1 bundling ability, but not mDia2 bundling activity. In addition, the Ile mutation, which severely reduces barbed end binding by both mDia2 and FRL1, has no affect on mDia2-mediated bundling and increases the ability of FRL1 to bundle under conditions where barbed ends are abundant. One interpretation of these results is that barbed end binding competes with bundling for FRL1, possibly due to overlapping usage of amino acids for these interactions. Alternately, barbed end binding might occur by a distinct set of interactions, but might alter the FRL1 FH2 domain conformation in a manner that disfavors bundling. In either case, this is a significant difference between the bundling activities of FRL1 and mDia2, and awaits further mechanistic investigation.
Possible Models of FH2-mediated Bundling-Whereas our current data do not permit us to make definitive predictions as to bundling mechanism, we propose two possible models by which FH2 domains might bundle. Filament side binding is a prerequisite for bundling, so we address this step first. The ability of both FRL1 and mDia2 FH2 dimers to dissociate raises the possibility that they might re-associate as a "clamp" around the filament, possibly employing some of the same interactions as those that mediate barbed end binding (Fig. 7, Model I). Because the mDia1 FH2 dimer does not readily dissociate, it does not bind filament sides efficiently and is limited to barbed end binding (Fig. 7, Model III).
The dimer dissociation model would imply that side binding is mutually exclusive with barbed end binding. While true for FRL1, mDia2 side binding is not competitive with barbed end binding, suggesting another mechanism. For instance, mDia2 FH2 domain might employ basic patches on its outer surface to interact with the acidic surface of the actin filament side (Fig. 7, Model II). Bundling occurs by repetition of this interaction by the symmetrical FH2 dimer (Fig. 7, Model V ). Such interactions would have to occur for bundling by the dimer dissociation model, but they would be in addition to the clamp interaction (Fig. 7, Model IV ).
Possibilities for Formin-mediated Bundling in Cells-The end product of formin action appears to be unbranched filaments, as opposed to the dendritic networks produced by Arp2/3 complex-based nucleation (reviewed in Ref. 32). Cells contain a wide variety of unbranched filament structures, with a common feature being that these consist of bundled filaments.
The Ile Mutation on Bundling and Processivity-Previous biochemical and structural studies have highlighted the importance of a particular Ile residue for FH2 domain effects at the barbed end. This residue is surface-exposed on the inner face of the FH2 donut (5) and makes contacts with actin in an FH2-actin co-crystal structure (9). Mutation of this residue in both full Bni1p FH2 (5) and a truncated mDia1 FH2 (30) drastically reduces FH2 activity.
The effects of this mutation for mDia2 are much more easily interpreted than for FRL1. mDia2 I704A is strongly inhibited in all barbed end binding activities. The mDia1 I845A mutant is similarly affected. Despite these deficiencies, mDia2 I704A still bundles similarly to wild-type. Taken together with the non-competitive nature of the barbed end and bundling activities for mDia2, the mutation results support bundling Model V.
For FRL1, the I715A mutation has more subtle effects on barbed end activities, having no significant effect on elongation rate inhibition and affecting polymerization acceleration less than 2-fold, while inhibiting effects on capping protein strongly. Because of the partial effects of this mutation, we cannot draw firm conclusions on the inter-relatedness of the barbed end and bundling interactions for FRL1. However, the fact that FRL1 I715A bundling activity is less inhibited than the mutant by barbed ends suggests that Model IV might apply to FRL1 bundling.
Two additional points must be made concerning the effects of these mutations on barbed end activities, both of which reveal fundamental features of FH2 domain activity. First, the difference between FRL1 and mDia1/mDia2 mutants suggests significant differences in barbed end interaction. We have previously shown that FRL1 polymerization activity is "weak" compared with mDia1 and mDia2, both because higher concentrations are needed and because its kinetics show a persistent lag, not observed for mDia1 or mDia2 (Refs. 2, 3, and this study). Possibly, FRL1 severing activity (3) might account for a significant portion of its ability to accelerate polymerization.
The second point is the manner in which the Ile mutation affects capping protein inhibition. Mutants of all three formins still compete with capping protein at high concentration, suggesting that residual barbed end affinity remains. However, the off-rate of formin from the barbed end appears greatly increased in the mutant. Our interpretation is that the FH2 domain has lost its ability to move processively with the barbed end. Possibly, the Ile residue is necessary for one of the two proposed binding states for FH2 domain at the barbed end (9). Because transition between these two bound states might be necessary for pro-cessive movement, the Ile mutant "falls off" the barbed end when the Ile-dependent binding state is required.
Our study provides further evidence that the formin family is complex in its effects on actin. First, individual formins can do multiple things to actin. Second, the degree of each activity varies from formin to formin. We believe that overall features of FH2 activity are similar, and that seemingly small changes to various FH2 properties can produce large changes to overall outcomes for actin polymerization and organization.