Interactions between the Microtubule Binding Protein EB1 and F-Actin

EB1 binds to F-actin

As a first test of whether EB1 binds to F-actin, we examined the ability of various fragments of EB1 to co-sediment with F-actin under conditions of low salt, where we expected that binding to the negatively charged actin filaments would be strongest. We tested three different EB1 constructs: EB1_FL (the full-length human protein), EB1₂₄₈ (the activated form of the protein with the acidic tail removed), and EB1₁₈₄ (monomeric EB1 that contains the CH domain and part of the unstructured region) (Fig. 1a and b) [10]. We found that full-length EB1 (EB1_FL) binds weakly to 2 μM pre-polymerized F-actin: almost all of the EB1_FL protein remained in the supernatant whether or not F-actin was added to the assay (Fig. 1c). However, nearly all the EB1₂₄₈ fragment pelleted in the presence (but not the absence) of 2 μM F-actin under these conditions (Fig. 1c). These observations demonstrate that EB1 can bind to actin and that EB1:F-actin binding, like EB1:MT binding, is negatively regulated by the EB1 tail (i.e., removal of the tail increases the affinity of EB1 for both actin and MTs) [10,18]. In addition, the monomeric MT binding “head” domain of EB1 (which contains the CH domain) also pelleted with F-actin (Fig. 1c). This observation indicates that F-actin binding is mediated at least in part by the MT binding CH domain.

Next, we wanted to get a better understanding of the EB1:F-actin interaction by determining its salt sensitivity. To address this question, we used EB1₂₄₈ because this construct bound to F-actin the strongest. We found that as we increased the amount of salt from 25 to 175 mM KCl, the fraction of EB1₂₄₈ bound decreased and then leveled off (Fig. 1d). This salt sensitivity indicates that, at least, part of EB1:F-actin binding is mediated by ionic interactions.

We also tested EB1₂₄₈ for its ability to bind to F-actin without the presence of phalloidin to ensure that the interaction was mediated by F-actin and not phalloidin or the combination of phalloidin bound F-actin. We found that the fraction of EB1₂₄₈ bound was the same with or without phalloidin (Fig. S3b). This is consistent with the idea that the interaction between EB1 and F-actin is direct.

The EB1:F-actin binding surface overlaps the EB1:MT binding surface

The conclusion that the CH domain can bind to actin raises the question of how the F-actin and MT binding surfaces of EB1 relate to each other. To investigate this issue, we mapped onto the surface of EB1 a set of amino acids previously identified as being critical for tracking MTs [19] (Fig. 2a and b, blue residues; for additional information, refer to Fig. S1b and Fig. 3). Next, we mapped onto the crystal structure previous mutations that had no effect on MT tracking [20] (Fig. 2a and b, yellow residues). We then compared these two maps to a map of the surface residues that are highly conserved under the assumption that highly conserved surfaces are involved in functionally significant interactions (Fig. 2c–f and Fig. S1a).

When we compared these maps, we observed that the whole surface of the EB1 CH domain is well conserved across vertebrates, including the face opposite the well-characterized MT binding interface (compare Fig. 2b and d) [19,20]. On the basis of these data, we hypothesized that this highly conserved back surface of EB1 might be involved in F-actin binding. The strong conservation of the EB1 surface is all the more striking when compared to the much weaker conservation of calponin, the protein for which CH domains are named (Fig. 2g–j). We initially thought that we might be able to predict the actin binding surface of EB1 by aligning the EB1 CH domain to the region of the calponin CH that binds to actin, but this surface has not yet been determined; indeed, it is presently unclear whether the calponin CH does bind to actin (calponin’s actin binding seems to be mediated by other parts of the protein) [24]. Comparison to other actin binding CH domains also gave ambiguous results (data not shown).

Therefore, to test the hypothesis that the back surface of EB1 is involved in binding to F-actin, we identified and mutated highly conserved, charged, surface residues on the MT binding surface of EB1 and in the face opposite (Table S1). We specifically mutated charged amino acids due to the electrostatic nature of the EB1:F-actin interaction (Fig. 1d). We made the mutations in EB1_FL as opposed to EB1₂₄₈ because the full-length protein is the most physiologically relevant construct. However, this introduced a problem with assaying the binding activity of the mutated EB1 because EB1_FL has relatively weak actin binding, likely due to auto-inhibition as concluded from the data in Fig. 1c. Therefore, to assay the activity of the EB1 mutants, we mixed the EB1 with a fragment of APC (APC-C1) that binds to the EB1 tail in order to relieve its auto-inhibition. [25,26].

As predicted, EB1_FL WT (wild type) and APC-C1 co-pelleted significantly (~50%) with F-actin when mixed together but not when incubated individually with F-actin (Fig. 3a). We then used this assay to test the EB1 mutants for their ability to co-sediment with F-actin. The fraction of EB1 bound was calculated for each mutant (Fig. S2a and b) and normalized against the fraction of EB1_FL WT bound under these conditions (Fig. 3b). We then classified the 20 mutant EB1 proteins according to their effect on F-actin and MT binding (Table S1).

Of the 20 EB1 mutants generated, one was found to bind both F-actin and MTs more weakly (K89D), while six were found to bind MTs more weakly without significantly affecting F-actin binding [K83A, K83D, R84A, Q102E, and (I72,K76,Q79A)] (Fig. 3b–f; note that colors in the figure are coordinated across panels according to the effect of the mutant). In addition, some mutants bound one or both filament types more strongly (E26A, D49K, D116R, and D123K). Several mutations of conserved surface residues had no effect on binding to F-actin or MTs (E26K, D116A, and (T8,S9,V10A)) (Fig. 3b–f and Table S1). The mutants that showed decreased binding were tested to ensure that were properly folded using circular dichroism. Of the seven mutants tested, five of them had indistinguishable spectra compared to EB1_FL WT consistent with the idea that they are properly folded (Fig. S3a). Two mutants [R84D and (55,58,59A)] showed decreased α-helical content and therefore were not used for EB1 mutagenesis analysis (Fig. S3a).

To allow comparison between our mutagenesis data and previous work, we have mapped on to the EB1 crystal structure amino acid residues previously found to be important for the EB1:MT interaction. Residues defined by Maurer et al. [19], Zhang et al. [22], and Slep and Vale [20] are colored purple. Residues defined only by Maurer et al. [19] are colored blue. Those residues not important for MT tracking are colored yellow [20] (Fig. 3g–j). Zhang et al. also noted that part of the EB N-terminus differed between their cryo-electron microscopy (EM) structure and a previous crystal structure [23], suggesting that some subset of amino acids in this region might be involved in MT binding; this region has been colored dark gray.

The lysine at position 89 has been previously found to make important contacts with tubulin and to be a crucial amino acid for MT tracking [19]. We confirmed that the K89D mutation significantly reduced MT binding. Interesting to note, Zhang et al. did not find this residue making any important contacts on their most recent cryo-EM work [22]. We were intrigued to see that the K89D also interfered with F-actin binding. This observation suggests that the F-actin and MT binding interfaces overlap, contrary to our original hypothesis.

We found that some of our other mutations agreed with previously published work, but others did not. More specifically, Q102E and cluster mutant I72,K76,Q79A decreased MT binding (Fig. S3b), as expected [19,20,22]. However, when we changed the histidine at position 69 to an alanine, we did not observe the reduction in MT binding expected from the idea that that this residue is in the MT binding interface [19,22]. This difference in results could be due to the differences in functional assays used (co-sedimentation versus plus-end tracking assays) or the MT analogs used (GDP-Tx-MTs versus GTPγS-MTs) [19,22]. Another interesting observation is that mutations in the amino acids K83 and R84 were previously found to have no effect on plus-end tracking in vivo [20], but when we mutated these residues, we found that they displayed decreased MT binding in our in vitro assays. Again, this observation could be due to the differences in the functional assay or MT analogs used. The observation that several mutations affect binding to only MTs indicates that the overlap between the actin binding sites and MT binding sites is only partial.

We were surprised to see that some mutants (D49K, D116R, and D125A/R) increased binding to MTs, F-actin, or to both F-actin and MTs, and we were intrigued to see that these clustered on the conserved face opposite the MT binding interface. We speculate that this surface of EB1 may be making contacts with the C-terminal region of EB1 during auto-inhibition and that the mutations generated disrupt normal auto-inhibition.

Upon finalizing our mutant analysis, we wanted to ask whether the amino acid positions discussed above are conserved across a broader array of organisms. To do this, we widened our bioinformatics analysis to include several basal animals, as well as a few selected unicellular eukaryotes. These sequences were aligned with the EB1 vertebrates and shown in Fig. 4. When we examined the conservation at these positions, we found (as expected) that almost all of the residues previously established as being involved in MT binding (Fig. 4; K83 and R84 are boxed red and K89 is boxed black) are very highly conserved among all of the organisms examined. Q102 is particularly well conserved (Fig. 4, red box). The most intriguing was the strong conservation of D49 and D116 (Fig. 4, yellow boxes), two positions at which mutations caused increased binding. These two amino acids are conserved to a similar degree as Q102 and are located on the face opposite of the MT binding interface, suggesting that whatever function this surface has, it is highly conserved across eukaryotes.

MTs compete with actin for binding to EB1

The evidence that the F-actin and MT binding surfaces of EB1 overlap suggests that MTs might compete with F-actin for binding to EB1. To test this hypothesis, we had to devise an assay capable of assessing F-actin binding in the presence of MTs; the standard co-sedimentation assay would not work because proteins binding to either filament would end up in the pellet. After trying a series of different approaches (data not shown), we developed an assay based on the observation that incubation of EB1₂₄₈ with actin causes formation of F-actin bundles (Fig. 5a). It is important to note that the goal of this assay is to determine whether EB1 is binding to F-actin under a given set of conditions (±MTs). We are not implying that F-actin bundling is a normal EB1 activityor suggesting that Taxol-MTs are the optimal EB1 substrate, only that bundling of F-actin by EB1₂₄₈ requires EB1:F-actin interactions.

Initial experiments established that EB1 does bundle F-actin (Fig. 5a, second row). We then tested how adding MTs affected the ability of EB1 to promote F-actin bundling. The idea was that if binding of EB1 to F-actin and MTs is mutually exclusive, the presence of MTs should interfere with the ability of EB1 to bundle F-actin. However, if EB1 can bind both F-actin and MTs at the same time, then addition of MTs should not alter bundling of F-actin by EB1. Moreover, since EB1 also bundles MTs, one would predict that the F-actin and MTs would be co-bundled if EB1 can bind both filaments simultaneously but that the filaments should remain separate if binding is mutually exclusive.

We chose to perform the experiments in a blinded matter (see Materials and Methods) to ensure that observer bias did not influence the results. As expected, bundles were not observed when F-actin and MTs were incubated alone (Fig. 5a, top row). When we mixed 5 μM actin with 1 μM EB1₂₄₈, we observed frequent bundles in a background of unbundled F-actin (Fig. 5a, second row). EB1 also promoted the formation of MT bundles, some of which were extremely large (Fig. 5a, third row). However, when EB1₂₄₈ was mixed with both F-actin (5 μM) and MTs (0.5 μM), F-actin bundles became rare (Fig. 5a, fourth row; see also Fig. 4b). In contrast, the effect of F-actin on MT bundling by EB1 was more difficult to discern (Fig. 5a, fourth row; see also Fig. 5c). Qualitatively, blinded observers noted that F-actin reduced the size of the MT bundles though that effect was difficult to quantify (data not shown). EB1_FL had little effect on bundling of either F-actin or MTs (Fig. 5a, fifth row; see also Fig. 5b and c), consistent with the idea that EB1_FL is auto-inhibited.

These observations support the hypothesis that F-actin and MTs compete for binding to EB1. This hypothesis further predicts that EB1 should fail to co-bundle F-actin and MTs. When labeled F-actin and MTs were mixed in the presence of unlabeled EB1₂₄₈, the filament networks remained separate, consistent with this idea (Fig. 5a, fourth row; note lack of co-localization between red MTs and green F-actin). The sum of these results lead us to conclude that EB1 can bind to F-actin or MTs, but not both simultaneously.

Is binding of EB1 to F-actin physiologically significant?

The affinity of EB1 for F-actin is relatively strong in 50 mM total KCl; examination of Fig. 1d shows that approximately half of the EB1 is bound at 7 μM F-actin, indicating that the apparent K_D is in the range 7 μM; 50 mM is a salt concentration commonly used in F-actin binding assays. However, the binding is weaker at more physiological salt concentrations (Fig. 1d, compare 50 mM salt to 75 mM or 125 mM salt, where only ~20% of the protein is bound at 7 μM F-actin). At first glance, these data might be interpreted to imply that the interaction is not physiologically significant. However, it is important to remember that the local concentration of F-actin can be extraordinarily high in some regions of the cell, especially near the cell boundary. For example, the concentration of F-actin in the lamellipodium of a migrating cell has been estimated to be >500 μM [27,28]. A protein with a K_D of 50 μM would be >90% bound to actin in an environment with >500 μM F-actin (assuming simple binding). This reasoning suggests that EB1:F-actin interactions could be relevant in contexts where the local concentration of F-actin is high, such as in the leading edge of a migrating cell where one might expect coordination between the actin and MT cytoskeletons.

This information, together with our evidence (above) that MTs and F-actin compete for binding to EB1 (Fig. 5), leads us to propose a model where EB1 interacts primarily with MTs in the relatively actin-poor interior of the cell but is competed away from MTs, onto F-actin, in the F-actin-rich periphery of cells. In other words, we suggest that high concentrations of F-actin may act as a passive sink for EB1, helping to promote MT de-polymerization in actin-rich areas, for example, near the cell edge (Fig. 6). We suggest this hypothesis because it has been shown that MTs grow persistently in the interior of cells and then display dynamic instability near the edge [28]. Alternatively, EB1’s weak actin binding could become more significant in the context of a complex with other +TIP interacting proteins (e.g., formins, spectraplakins) that themselves bind actin [14,30], or they could be involved in other as-yet uncharacterized aspects of cross-talk between the actin cytoskeleton and components of the +TIP network. These ideas are compatible with the idea that high concentrations of F-actin might compete EB1 off of MTs.

One prediction of this model is that EB1 +TIP localization will be reduced in regions with high actin concentrations, for example, the lamellipodium. For technical reasons including incompatibilities between EB-labeling methods and lamellipodium-labeling methods, it is difficult to address this question; thus, whether this prediction will be validated remains a focus for future work. Another question raised here is whether weak binding to actin might be a general feature of MT binding CH domains. Given that actin binding CH domains can use different surfaces to bind actin [11] and MT binding CH domains can also use different surfaces to bind tubulin [22], it seems unlikely that one can make generalizations about the binding activities of CH domains. This too remains an interesting question for future study.

Bioinformatics

To find the sequences for the amino acid alignments, we performed a BLASTp search with human EB1 or human calponin as the query. The EB1 or calponin sequences of 10 vertebrates were chosen as representatives of vertebrate diversity and independently aligned with ClustalX (for this initial work, we used only this subset of organisms in order to avoid misinterpretations caused by the varied gene duplications that have happened in many organisms, particularly, at the base of vertebrate radiation). Similar alignments were obtained with other alignment programs such as Muscle (data not shown). The model vertebrates include human (GI: 6,912,494), chicken (GI: 71,894,959), cow (GI: 115,496,618), zebrafish (GI: 41,054,091), frog (GI: 147,905,448), mouse (GI: 7,106,301), lizard (GI: 327,289,281), pig (GI: 113,205,898), fish (GI: 348,517,001), and elephant (GI: 344,279,857). The conservation of EB1 or calponin was aligned on to the crystal structure of the CH domain of human EB1 (PDB ID: 1PA7) or human calponin (PDB ID: 1WYP) using the structure analysis and presentation program Chimera [31].

For the analysis of the more divergent EB1 proteins, we identified from National Center for Biotechnology Information a set of early-diverging organisms (both basal animals and model unicellular eukaryotes) that have only one EB1 protein, again chosen to avoid misinterpretations caused by gene duplications as follows: Saccoglossus kowalevskii, Hydra vulgaris, Monosiga brevicollis, Salpingoeca rosetta, Ciona intestinalis, Nematostella vectensis, Amphimedon queenslandica, Lottia gigantea, Schizosaccharomyces pombe, Neurospora crassa, Dictyostelium discoideum, Giardia intestinalis, and Saccharomyces cerevisiae. The sequences for the first eight organisms were aligned against the model EB1 vertebrate sequences. A separate alignment of the last organisms was generated. Both alignments were made in ClustalX. The amino acids are colored based on percent identity in Jalview.

Mutagenesis and purification

Highly conserved surface residues were selected either in the known MT binding interface as determined from previous EM and mutagenesis [19] or in the face opposite the MT interface. Amino acids were either mutated to an alanine or a charged amino acid that resulted in a charge reversal. Mutants were generated through site-directed mutagenesis (Agilent QuikChange II) and confirmed by DNA sequencing. Proteins (WT or mutant) were purified through previously described methods [32]. Briefly, non-tagged proteins were isolated using classical ion-exchange techniques. Lysates were exchanged on to sequential cation-exchange and anion-exchange columns with a desalting column in between. The protein was dialyzed into PEM buffer [100 mM Pipes, 1 mM ethylene glycol bis(β-aminoethyl ether) N,N′-tetraacetic acid (EGTA), and 2 mM MgCl₂, pH 6.8], flash frozen, and stored at −80 °C. EB1 proteins for the low-salt experiments were dialyzed into TK buffer (10 mM Tris–HCl, 50 mM KCl, and 1 mM DTT, pH 7.5).

APC-C1 was bacterially expressed and purified using a nickel-NTA column [25]. APC-C1 was dialyzed in to PEM buffer, flash frozen, and stored at −80 °C.

Actin was purified from rabbit muscle acetone powder through filament cycling [33] and stored in calcium buffer G (2 mM Tris–HCl, 0.2 mM ATP, 0.5 mM DTT, 0.1 mM CaCl₂, and 1 mM sodium azide at pH 8). For F-actin studies, the globular actin G-actin was converted into MG-actin by adding 5 mM MgCl₂ and 0.2 mM EGTA for 5 min at room temperature. To polymerize the actin, we added 50 mM KCl, 1 mM MgCl₂, 1 mM EGTA, and 10 mM imidazole-HCl (pH 7) and allowed them to incubate 1 h at room temperature. After polymerization, 0.8 μM phalloidin was added to stabilize the filaments.

For Fig. 1c, actin was purchased from Cytoskeleton Inc. (Denver, CO) and stored in G buffer (10 mM Tris–HCl, 0.1 mM CaCl₂, 0.2 mM ATP, and 0.5 mM DTT at pH 7.5). Actin was centrifuged (200,000g, 15 min at 4 °C) immediately prior to use to remove any aggregates.

Tubulin was purified as previously described [10]; briefly, the tubulin was purified form porcine brain through two rounds of polymerization and de-polymerization. MTs were polymerized with Taxol (paclitaxel; Sigma) as previously described [10]. Tubulin and MTs were frozen and stored at −80 °C.

To measure EB1 (calculated as EB1 monomer) and MT (calculated as tubulin dimer) concentration, we used BSA standard curves in conjunction with Bradford dye (Bio-Rad). To determine the actin concentration, we measured an optical density at 290 nm and used an extinction coefficient of 26,600 M⁻¹ cm⁻¹.

Binding studies