Filamins are large proteins that cross-link actin filaments and connect to other cellular components. The C-terminal rod 2 region of FLNa (filamin A) mediates dimerization and interacts with several transmembrane receptors and intracellular signalling adaptors. SAXS (small-angle X-ray scattering) experiments were used to make a model of a six immunoglobulin-like domain fragment of the FLNa rod 2 (domains 16–21). This fragment had a surprising three-branched structural arrangement, where each branch was made of a tightly packed two-domain pair. Peptides derived from transmembrane receptors and intracellular signalling proteins induced a more open structure of the six domain fragment. Mutagenesis studies suggested that these changes are caused by peptides binding to the CD faces on domains 19 and 21 which displace the preceding domain A-strands (18 and 20 respectively), thus opening the individual domain pairs. A single particle cryo-EM map of a nine domain rod 2 fragment (domains 16–24), showed a relatively compact dimeric particle and confirmed the three-branched arrangement as well as the peptide-induced conformation changes. These findings reveal features of filamin structure that are important for its interactions and mechanical properties.
Transmission of biochemical signals and mechanical forces across the plasma membrane is essential for the regulation of cell differentiation and proliferation [1,2]. This often requires connections between transmembrane adhesion receptors and the intracellular cytoskeleton. One of many proteins that connect F-actin (filamentous actin) to transmembrane receptors is FLN (filamin). FLNs are well positioned in cells to influence signal transmission as they have binding sites both for F-actin and the cytoplasmic tails or loops of membrane receptors, such as the β chains of integrins, the major transmembrane adhesion receptors and G-protein-coupled receptors including DRs (dopamine receptors). FLNs also bind to various other cellular proteins including cytosolic signalling and adaptor proteins, e.g. migfilin [3–8]. Accordingly, through these interactions, FLNs are thought to bring together various proteins to enhance signal transmission .
The FLN protein family is encoded by three homologous genes [FLNA, FLNB and FLNC]. FLNa protein is the most abundant and widely expressed [4,5]. Each 280 kDa FLN polypeptide contains an N-terminal actin-binding domain followed by 24 Ig (immunoglobulin)-like domains (FLN1–24) , with FLN24 mediating dimerization (Figure 1A) [10,11]. A flexible hinge region between domains 15 and 16 divides the chain of 24 Ig-like domains into rod 1 (FLN1–15) and rod 2 (16–24) (Figure 1A) . Another hinge separates domains 23 and 24. Individual FLN Ig domains are β-sandwich structures composed of eight β-strands. The domains in the rod 2 region have unusual interdomain interactions, with domains 16–17, 18–19 and 20–21 forming compact pairs (Figures 1B–1D) [13,14]. In the domain pairs 18–19 and 20–21, the β-strand A of domains 18 and 20 is folded together with the C- and D-strands of domains 19 and 21 respectively (CD faces of FLNa19 and FLNa21, Figures 1C and 1D). The same CD face was earlier identified as a binding site for both integrins [15,16] and migfilin [17,18]. The CD face of FLNa19 was also shown to be the binding site for DRD2 and DRD3 [19,20]. This kind of domain–domain interaction suggests that binding of FLN to some of its partners is autoinhibited. The exposure of these cryptic binding sites will require changes in conformation. These could be induced either by mechanical forces [21–23] or by binding to another interaction partner. In contrast with the 18–19 and 20–21 domain pairs, autoinhibition is not seen with the 16–17 pair (Figure 1B) . The A-strand of domain 16 is unstructured and not associated with FLNa17 (Figure 1B); consequently the A-strand does not hinder an interaction partner, e.g. of platelet glycoprotein Ibα, from binding to FLNa17 . Domains FLNa22–23 in rod 2 also do not seem to exhibit compact domain packing . Sequence comparison suggests that a similar pair-wise domain arrangement exists in FLNs throughout the animal kingdom, including Drosophila melanogaster (gene Cheerio) and Caenorhabditits elegans , although not in slime mold Dictyostelium discoideum . Mutations in the C-terminal region of FLNs have been described in human patients and in D. melanogaster. Loss of binding to one or more of the numerous FLN-binding partners is a probable cause of many of the observed phenotypes, but the molecular mechanism of most of these mutations remains to be elucidated .
Location and known structures of the domains studied
In the present study we show a low-resolution structure of the nine C-terminal domains of FLNa, based on a combinatorial use of EM (electron microscopy) and SAXS (small-angle X-ray scattering) techniques. This and a recent study using NMR techniques  show that the three domain pairs make a compact, but flexible, three-branched structure. The structural model, together with peptide interaction and mutagenesis studies, suggests that the domain pairs act independently of each other, thus facilitating large conformational changes in the fragment.
FLNa (UniProt entry P21333) fragments were amplified by PCR and cloned into a modified pGEX vector (GE Healthcare) with a TEV (tobacco etch virus) protease cleavage site. The domain boundaries were based on the published alignment . GST (glutathione transferase) fusion proteins were expressed in Escherichia coli strain BL21 GOLD and purified as described previously . The L1963E and I2144E mutations were introduced using the QuikChange® protocol (Stratagene).
SAXS (small-angle X-ray scattering)
SAXS data were collected at the EMBL beamline X33 at the DORIS III storage ring, DESY (Hamburg, Germany) . The measurements were carried out at 288 K in 20 mM Tris (pH 8.0), 100 mM NaCl and 10 mM DTT (dithiothreitol). The concentrations of FLNa fragments were adjusted to 3–6 mg/ml and FLNa fragments and peptides were used at a molar ratio of 1:10. MAR345 image plate at sample-detector distance 2.7 m and wavelength λ=0.15 nm, covering the momentum transfer range 0.08<s<4.9 nm−1 [s=4π sin(θ)/λ,where 2θ is the scattering angle] was used. The data were processed using standard procedures by the program package PRIMUS . Rg and Dmax values were evaluated using the program GNOM .
The ab initio envelopes were obtained by averaging 10–20 independent runs from the bead modelling program GASBOR  with the program DAMAVER . The rigid-body modelling of FLNa16–21 was performed with the program SASREF  using FLNa domain pairs 16–17 (PDB code 2K7P), 18–19 (PDB code 2K7Q) and 20–21 (PDB code 2J3S) as rigid bodies. The rigid-body models were fitted into the ab initio envelopes using the SITUS program package  (http://situs.biomachina.org/). The images of the SAXS models were made with the program VMD  (http://www.ks.uiuc.edu/Research/vmd/) or PyMOL (http://www.pymol.org).
The synthetic peptides of migfilin (residues 5–19), β2 cytoplasmic tail (residues 745–765) and DRD3 (residues 210–230) (EZBiolab) were coupled to NHS (N-hydroxysuccinimide)-activated Sepharose 4 Fast Flow (GE Healthcare) at 277 K according to the manufacturer's instructions. Purified FLNa fragments were incubated for 1 h at 296 K with 15 μl of the peptide–Sepharose in 1% Triton X-100, 150 mM NaCl and 20 mM Tris (pH 7.4). The Sepharose was centrifuged at 2000 g for 2 min and washed three times with 300 μl of the binding buffer. Proteins were then eluted with 10 μl of SDS electrophoresis sample buffer and run on a SDS/PAGE (12% gel). The intensities of Coomassie Blue-stained protein bands were then evaluated by ImageJ (http://rsbweb.nih.gov/ij/). The data were plotted as the percentage of maximal binding and fitted to equation Y=Bmax*X/(Kd+X) using GraphPad Prism 5 (GraphPad Software).
For cryo-EM, 2 μl of sample was placed on a glow-discharged lacey carbon film EM grid, blotted and plunge-frozen in liquid ethane. The samples were imaged on an FEI F30 electron microscope operating at 300 kV and with a magnification of ×59000. Micrographs without evidence of astigmatism, drift or excessive defocus were digitized using a Z/I Photoscan scanner giving a pixel size at the specimen of 1.1865 Å (1 Å=0.1 nm), interpolated for processing to 2.37 Å. The defocus was determined and the micrographs further triaged using ctffind3. The particles were picked using the EMAN program Boxer , after contrast enhancement in MATLAB via a self-written routine in which a three-level decomposition with a fourth-order bi-orthogonal wavelet function was applied to the images. The usefulness of this particular filter was determined empirically and on the basis of the positions so identified for the FLN molecules. Unfiltered particles were extracted, again in Boxer, and then corrected for their CTFs (contrast transfer functions) using the EMAN program applyctf . Following this, the images of FLNa16–24 were subjected to multivariate statistical analysis in IMAGIC  to determine class averages (e.g. Supplementary Figures S3A and S3B at http://www.BiochemJ.org/bj/446/bj4460261add.htm). The number of eigenimages used in classification was determined empirically and then resulting class sums with well-defined molecular density and low background signal were used in angular reconstitution to determine models, initially assuming no symmetry. The resulting unsymmetrized maps were then refined in SPIDER  operated within a self-written graphical user interface for iterative refinement of structures (J.F. Flanagan IV and R.J.C. Gilbert, unpublished work). An angular spacing of 15° was used throughout. Independent maps of FLN and FLN with peptide both showed an apparent 2-fold axis of symmetry and, to demonstrate this, self-rotation functions were used employing CNS  and using information between 200 Å and 30 Å resolution, displayed using Gropat (R.M. Esnouf, unpublished work) (Supplementary Figure S4A at http://www.BiochemJ.org/bj/446/bj4460261add.htm). Because the self-rotation functions showed both a 2-fold (κ=180°) and a 2-fold distribution of 3-fold (κ=120°) axes and because the class averages and eigenimages demonstrated unequivocal dyad axes of symmetry apparent as pseudo-mirror planes (Supplementary Figures S3A and S3C) and the maps themselves appeared to have dyad axes of symmetry (Figure 4A), we then repeated the reconstruction process from the start using C2 symmetry, but otherwise as above. The angular distributions determined for the particles (Supplementary Figure S4B) were reasonably even; for FLN the mean number of particles per bin was 252, the median 215, the minimum 116 and the maximum 404. For FLN with peptide the mean was 179, the median 167, the minimum 124 and the maximum 300. Throughout, resolution was assessed by FSC (Fourier shell correlation) with a 0.5 FSC cut-off. WellMAP programs (J.F. Flanagan IV and R.J.C. Gilbert, unpublished work) were used to generate the FSC and particle distribution plots for the data. The Fourier shell correlations showed sharp resolution cut offs and low residual noise (Supplementary Figure S4C), but indicated resolutions of 15.6 Å and 15.7 Å for the unsymmetrized reconstructions of FLNa16–24 without and with β7 peptide respectively, and resolutions of 12.2 Å and 12.8 Å for the corresponding symmetrized maps. Alignment to a common reference can result in spuriously high resolution estimations due to the correlation of noise  and a recent paper has shown angular assignment errors of ~10° in particles twice the size of the FLNa16–24 dimer . The distinctive two-pronged shape of the FLNa rod 2 segment may have assisted in increasing the accuracy of alignment, as also the symmetry may have done, which is a powerful feature in the classification (Supplementary Figure S3); Henderson and colleagues have pointed out how more accurate alignment may be assisted by distinctively shaped projected objects . The self-rotation functions are nevertheless a striking validation of the reconstructions as they clearly do show unimposed symmetry being detected in the maps (Supplementary Figure S4A). By restricting the alignment to the inherently coarse 15°, we have avoided over-determination of Euler angular values and the maps were ultimately filtered to 18 Å because of this limited sampling in projection determination, whereas finer samplings resulted in the accumulation of noise. The unsymmetrized maps were correspondingly filtered to 30 Å. Filtration was performed using the program WellMAP (J.F. Flanagan IV and R.J.C. Gilbert, unpublished work) in which a Gaussian filter was applied to the data. WellMAP programs were also used to generate the FSC and particle distribution plots for the data. EM maps were displayed and fitting to them was performed using Chimera software .
The cryo-EM reconstructions reported in this manuscript have been deposited in the European Bioinformatics Institute Electron Microscopy Database (http://www.ebi.ac.uk/pdbe/emdb/) under accession codes: EMD-2031 (FLNa16–24, unsymmetrized), EMD-2032 (FLNa16–24 symmetrized), EMD-2033 (FLNa16–24+β peptide, unsymmetrized) and EMD-2034 (FLNa16–24+β peptide, symmetrized).
SAXS analysis of a six domain fragment of FLNa
To study the structure of the rod 2 region of FLNa, a six domain fragment, FLNa16–21, was analysed by SAXS. The shapes of the scattering profile and distance distribution functions (Figure 2A) indicated that the fragment has a complex architecture, where domains are neither linearly nor globularly organized, but rather have a branched arrangement. The radius of gyration (Rg)=3.42 (±0.01) nm and the maximum linear dimension (Dmax)=12 (±1) nm (Table 1), suggest a compact organization of domains in the fragment, consistent with results for individual domain pairs that were earlier measured to be Rg=1.9–2.1 nm and Dmax=6.0–6.5 nm . A total of ten independent ab initio models were calculated. They fitted to the experimental data with a discrepancy (φ) of 1.1–1.7; the mean normalized spatial discrepancy between the models was 1.30, showing that the models agree well with each other. The average of the most populated ab initio envelope demonstrates a planar, three-branched shape which consists of three equally long (~5.5 nm) arms (Figure 2B).
SAXS models of FLNa16–21 revealed three-branched shape
|Sample||Rg (nm)||Dmax (nm)|
|Sample||Rg (nm)||Dmax (nm)|
The SAXS data was also used to fit the atomic structures of the three domain pairs within FLNa16–21 [13,14], by rigid-body modelling. The resulting model gave an excellent fit to the experimental scattering data (χ=0.91; Figure 2A) and suggested that domain pairs are organized in the three-branched shape so that each arm consists of one domain pair (Figure 2C). The odd-numbered domains are close to the centre of the structure and the even domains are outermost. Apart from the polypeptide connection between the domains, the model does not indicate any other stabilizing interactions between domains, consistent with the NMR observations [25,41]. Superposition of the ab initio and rigid-body models reveals excellent convergence of the two independent approaches (Figure 2C, right-hand panel).
To define the position of the N-terminus in FLNa16–21, the scattering data for the FLNa fragment 15–21 were recorded (Supplementary Figure S1 at http://www.BiochemJ.org/bj/446/bj4460261add.htm). The ab initio shape of FLNa15–21 clearly shows that one dimension is longer than the other two dimensions (Figure 2D), unlike the shape of FLNa16–21 where all of the arms are equally long. Accordingly, it seems likely that domain 16, separated from domain 15 by a 32 amino acid linker , is located at the tip of one arm of the FLNa16–21 envelope (Figure 2B), consistent with the rigid-body model (Figure 2C).
Integrin, migfilin and DRD3 peptides induce conformational flexibility in FLNa
The SAXS data presented above and a recent NMR study  show that the individual domain pairs in the rod 2 region of FLNa interact very little with each other, and that the overall structure is rather flexible . To further study the flexibility of the fragment, we tested whether the FLN-binding transmembrane receptor cytoplasmic domain or signalling adaptor protein peptides can cause conformational changes in the FLNa16–21 fragment. The conformational changes were initially analysed by SAXS. The peptides used were so small that their direct scattering could be subtracted from the data to reveal their effect on the FLN fragment structure. To our surprise, millimolar concentrations of migfilin and DRD3 peptides increased the Rg value of FLNa16–21 by 0.48–0.66 nm and the Dmax value by 2–4 nm (Table 1). Scattering and distance distribution curves from each experiment are presented in Supplementary Figure S2 at http://www.BiochemJ.org/bj/446/bj4460261add.htm. A non-binding migfilin mutant (I15E) peptide  did not induce any changes in the Rg and Dmax values (Table 1), validating the method and suggesting that the peptide binds to the FLNa16–21 in a similar way to its binding to domain 21 alone .
We next wanted to further verify the idea that ligand peptides bind to the CD faces of domains 19 and 21, displace the A-strands of the preceding domains and thus cause global conformation changes in the FLNa rod 2. To do this, interaction destabilizing mutations were introduced in the A stands that mask the binding sites (L1956E and I2144E). The I2144E mutation enhanced the migfilin binding to FLNa16–21 to similar levels as to FLNa21 alone (Figure 3A–3C). In the absence of ligand peptides, these mutations did not cause significant changes to the molecular shape parameters of FLNa16–21 in SAXS (Table 1), but in the presence of peptides both the Rg and Dmax values were observed to increase (Table 1). The previously characterized peptide from the cytoplasmic tail of integrin β2, which has lower affinity than the β7 peptide  induced changes in FLNa16–21 conformation only when the FLNa20 A-strand interaction with FLNa21 CD face was weakened by the I2144E mutation (Table 1). Although all of the interacting peptides changed the scattering patterns (Supplementary Figure S2), the most significant changes in the scattering profile and distance distribution function of FLNa16–21 were seen when both migfilin and DRD3 peptides and both FLNa18 and 20 A-strand mutations (L1956E and I2144E) were introduced at the same time (Figure 3E). In the presence of the peptides, the Rg value increased to 5.11 nm and the Dmax value up to 17 nm (Table 1), suggesting a significantly more elongated shape for FLNa16–21. Also the bimodal distance distribution seen without peptides or with a single peptide (Figure 2 and Supplementary Figure S2) changed to a single major distance peak when two mutations and two peptides were present simultaneously (Figure 3E, insert). These findings indicate major flexibility in the FLNa16–21 fragment and are consistent with the prediction that DRD3, integrin and migfilin peptide binding detaches the A-strands of domains 18 and 20 from the CD faces of domains 19 and 21. Furthermore, the results support the idea that the 18–19 and 20–21 domain pairs can interact and change conformation independently of each other.
Analysis of filamin mutants
Single particle EM reconstruction of the C-terminal part of FLNa
To gain independent structural data of the FLNa rod 2 region, single particle cryo-EM was used to calculate three-dimensional reconstructions of a nine domain fragment (FLNa16–24) that forms a 185 kDa homodimer. A total of 21133 particles were classified and the reconstructions were initially generated with no symmetry assumed (see the Experimental section). The reconstruction showed features of 2-fold symmetry (Figure 4A) that could also be seen the image class-averages (Supplementary Figure S3A) and in a self-rotation analysis of the reconstruction (Supplementary Figure S4A). Two 3-fold symmetry peaks were also seen in the corresponding self-rotation function plot (Supplementary Figure S4A). The global 2-fold symmetry could be used to improve the overall quality of the reconstruction by reprocessing the original images assuming 2-fold symmetry. Figure 4(B) shows the resulting reconstruction of FLNa16–24 in orthogonal views. The nominal resolution of the reconstruction was 12.2 Å (Supplementary Figure S4C), but we filtered the maps to 18 Å for display purposes because the sampling angle was 15°. The particle orientation distribution was relatively even and all views were represented (see the Experimental and Supplementary Figure S4B, left-hand panel). The symmetrized FLNa16–24 map (Figure 4B) has a distinctive two-pronged appearance similar to that seen in the unsymmetrized map (Figure 4A). The dimerization interface is found towards the base of the fork, at the upper edge of a pair of globular lobes of density. Above this, the structure displays a central stem for each prong, which bifurcates into a smaller and a larger piece of density resulting in a three-branched structure. The connections between all of the lobes are not contoured at the density level chosen for the isosurface representation: this may be a result of limited resolution, limited flexibility in linker regions or a combination of such factors. FLNa16–24 has overall dimensions of 120 Å×120 Å×60 Å; the distance between the prongs is 55 Å and the distance from the base of the molecule to the point of bifurcation is 40 Å.
Cryo-EM reconstructions of FLNa16–24
To assist in the interpretation of the FLNa16–24 EM reconstruction, we fitted ab initio SAXS envelopes in it. A fit of the FLNc23–24 ab initio SAXS envelope  was poor (Supplementary Figure S4D), but the FLNa16–21 envelope fitted well into the forked density of the EM map (correlation coefficient 0.68), leaving an appropriate space for FLNa22–23  (correlation coefficient 0.90) (Figure 5A). The connection of domain 22 to the centre of the three-branched structure is consistent with the polypeptide connections between the C-terminus of domain 21 and the N-terminus of domain 22. Thus, with the help of SAXS data, we are able tentatively to assign eight Ig-like domains in the EM reconstruction.
Cryo-EM reconstruction of FLNa16–24: analysis and effect of integrin peptide
To explore the ligand peptide-induced conformation changes in FLNa rod 2 region, we performed single particle EM analysis of FLNa16–24 in the presence of integrin β7 peptide. A cryo-EM image set of 32477 particles for FLNa16–24 with integrin β7 peptide was analysed independently. Representative class averages are shown in Supplementary Figure S3(B). The final map was generated with 2-fold symmetry constrains. Resolution estimation and particle orientation distribution for this reconstruction are shown in Supplementary Figures S4(B) and S4(C). Both the unsymmetrized (Supplementary Figure S4E) and symmetrized (Figure 5B) reconstructions showed a more open conformation of the FLNa16–24 fragment with the integrin peptide than without it.
In the present study we have investigated the overall structure of the C-terminal rod 2 fragment of FLNa that comprises nine Ig-like domains (domains 16–24). This fragment is of considerable interest in relation to FLN function, first, because it mediates FLN dimerization that is essential for actin cross-linking and, secondly, because it contains several interaction sites for transmembrane and signalling proteins. Our current results indicate that the rod 2 fragment is compact, but flexible, and that the flexibility responds to and contributes to the regulation of protein interactions.
Rotary shadowing EM studies have previously revealed that rod 2 of FLN is more tightly packed than the first 15 Ig domains comprising rod 1 . Our current data show that the first six domains of rod 2 form a three-branched, rather planar structure, where each previously characterized domains pair (16–17, 18–19 and 20–21) forms one branch. Our SAXS results confirm the recent NMR-based model from the same six domain fragment . Overall the longest diameter of the nine domain rod 2 fragment is about 15 nm. This diameter fits with the estimation of 19±6 nm based on rotary shadowing EM .
The electron microscopy reconstructions are consistent with a single contact site between the two arms of the structure. This supports previous results that dimerization of vertebrate FLNs is mediated only by the last, the 24th, Ig-like domain [9,11,42,43]. However, we were not able to exactly position FLNa24 in our EM reconstruction. In the map, the dimerization connection appeared as a narrow bridge that is inconsistent with the known structure of the FLN dimerization interface. The crystal structures show that the 24th Ig-like domain of FLNc and FLNa form a long interface where one of the β-sheets continues from one monomer to another [11,43]. This was not seen in our EM map, although we were able to fit the SAXS envelope of FLNa22–23. We suspect that the reasons for this are the limited resolution of the reconstruction, the use of 2-fold symmetry and the flexibility of the hinge 2 region between domains 23 and 24. Earlier SAXS analysis has shown that there are few contacts between FLNc domains 23 and 24, which may allow flexibility of the hinge 2 region . Thus our map might represent an average of multiple conformations of domain 24 in relation to the rest of the molecule.
The three-branched shape of the six-domain fragment, FLNa16–21, was revealed by ab initio analysis of the SAXS data. The same shape was independently reconstituted by fitting known high-resolution two-domain fragments to the SAXS scattering. This shape fits very well with the fact that, in the previously determined atomic structures the N- and C-termini of two domains, pairs locate close to each other [13,14]. The position of the N-terminus of the six-domain fragment could be verified by SAXS analysis of an N-terminally extended fragment. The EM reconstruction provided support for the shape as well and for the position of the C-terminus of the fragment. Given the limitations of the resolution of the SAXS measurements, we believe that we have been able to generate a reliable model for the organization of the six domains. What are the implications of the model for FLN function? First, the model shows that there are only a very limited number of interaction sites between the three domain pairs. This implies that the structure has intrinsic flexibility, the domain pairs can wobble in relation to each other. Flexibility is supported by the finding that the angle between domains 19 and 21 is quite different in this SAXS envelope than in the crystal structure of the three-domain fragment FLNa19–21 . Molecular dynamics simulations suggest that the observed angle in the crystal may be partly due to crystal packing . Flexibility between individual domains and especially between the domain pairs is also supported by NMR analysis . A second implication of the current model is that the binding sites in FLN domains 17, 19 and 21 are located quite far apart in space, and are thus expected to function independently. We earlier speculated that there could be cross-talk between different binding sites so that for instance protein binding to domain 19 could allosterically regulate the masking or unmasking of domain 21, or vice versa . The current model suggests that this might not be the case.
In addition to the inter-domain flexibility of the filamin C-terminal fragment, we have shown in the present study that transmembrane receptor and signalling molecule binding to the domain pairs 18–19 and 20–21 can induce large changes in the overall structure. The integrin β7 peptide caused marked changes in the FLNa16–24 EM reconstruction. The migfilin and DRD3 peptides induced a clear elongation of the FLNa16–21 fragment, increasing the maximum distance by approximately 2 nm. The failure of a non-FLN-binding migfilin peptide to induce such a conformational change and the responsiveness of the changes to FLN mutations that weaken the masking of CD faces of domains 19 and 21, are consistent with a model where the peptides displace A-strands of domains 18 and 20 from the CD face of domains 19 and 20 respectively. The most elongated shape of FLNa16–21, that has a maximum distance of 17 nm was achieved when both of the A-strand mutations and both the DRD3 and migfilin peptides were introduced. In this elongated shape, both of the A-strands have presumably been detached, whereas migfilin alone displaces only the A-strand of FLNa20. The requirement of both binding events for maximal elongation is consistent with the idea that the two binding sites act independently. A model for the peptide-induced conformational changes is presented in Figure 6. We and others have also observed similar conformational changes in shorter FLN fragments [44,45]. It has been recently reported that the rod 2 region of FLN can undergo conformational changes under low (<10 pN) forces, whereas changes in the rod 1 region are mainly observed at forces more than 30 pN, which may lead to domain unfolding . Our branched and compact structure of the rod 2 region suggests that there are very few contacts between the domain pairs and thus external forces can directly be channelled to open the masked binding sites in the domain pairs 18–19 and 20–21. These sites are in such an orientation with respect to the location of peptide connections  that would allow relatively weak forces to cause a zipper-like opening of binding sites, as observed in steered molecular dynamics simulations [22,48]. Although the force and time scales used in molecular dynamics simulations are not directly comparable with those in cells and in single-molecule experiments, the branched structure shown in the present paper for the rod 2 region can potentially explain how opening of domain pairs 18–19 and 20–21 could be the source of the extension observed at <10 pN forces , and how the strain in the FLN–actin network could enhance integrin binding to the network .
Schematic representation of domain arrangement of FLNa16–21 fragment and the conformational changes induced by DRD3 and migfilin peptide
In summary, we have shown in the present study a complete, albeit low resolution, picture of the domain organization in the rod 2 segment of FLNa. This segment contains the dimerization site (domain 24), two canonical Ig domains (domains 22–23), and three intertwined domain pairs (16–17, 18–19 and 20–21). Our model shows that there is considerable flexibility between these structural units. Furthermore, we show that transmembrane receptor and signalling protein binding induces large conformational changes that are consistent with partial opening of domain pairs 18–19 and 20–21. These results provide a basis for understanding the function of FLN as a cellular mechanosensor [23,49–52].
Salla Ruskamo, Gregor Hofmann, Robert Gilbert and Ulla Pentikäinen designed and performed the experiments and analysed the data. Pengju Jiang expressed and purified the dimeric protein. Iain Campbell, Robert Gilbert and Jari Ylänne supervised and designed the experiments. All of the authors participated in the writing of the paper.
We thank Dr Peter Konarev and Dr Dmitri Svergun (EMBL/DESY, Hamburg, Germany) for help with SAXS data analysis, and EMBL/DESY Hamburg for the access to the beamline X-33. We thank Arja Mansikkaviita for excellent technical assistance, Dr David Calderwood and Dr Mohamed Bouaouina (Yale University School of Medicine, New Haven, CT, U.S.A.) for helping with binding assays, and Emilia Horttana for contributing to mutagenesis and biochemical assays.
This work was supported by the Academy of Finland [grant numbers 121393 (to U.P.), 114713 (to J.Y.) and 138327 (to J.Y.)], the National Doctoral Programme in Informational and Structural Biology (to S.R.), the BBSRC (Biotechnology and Biological Sciences Research Council) [grant number BB/F006845 (to P.J.)] and the NSFC (National Natural Science Foundation of China) [grant number 31100530 (to P.J.)]. The support of Instruct (http://www.instruct-fp7.eu/) is gratefully acknowledged for software license purchase.
The cryo-EM reconstructions reported have been deposited in the European Bioinformatics Institute Electron Microscopy Database (http://www.ebi.ac.uk/pdbe/emdb/) under accession codes: EMD-2031 (FLNa16–24, unsymmetrized), EMD-2032 (FLNa16–24 symmetrized), EMD-2033 (FLNa16–24+β peptide, unsymmetrized) and EMD-2034 (FLNa16–24+β peptide, symmetrized)
Present address: Department of Biochemistry, University of Oulu, Oulu, Finland.