The ERC1 scaffold protein implicated in cell motility drives the assembly of a liquid phase

Several cellular processes depend on networks of proteins assembled at specific sites near the plasma membrane. Scaffold proteins assemble these networks by recruiting relevant molecules. The scaffold protein ERC1/ELKS and its partners promote cell migration and invasion, and assemble into dynamic networks at the protruding edge of cells. Here by electron microscopy and single molecule analysis we identify ERC1 as an extended flexible dimer. We found that ERC1 scaffolds form cytoplasmic condensates with a behavior that is consistent with liquid phases that are modulated by a predicted disordered region of ERC1. These condensates specifically host partners of a network relevant to cell motility, including liprin-α1, which was unnecessary for the formation of condensates, but influenced their dynamic behavior. Phase separation at specific sites of the cell periphery may represent an elegant mechanism to control the assembly and turnover of dynamic scaffolds needed for the spatial localization and processing of molecules.


Results and Discussion
ERC1 induces cytoplasmic condensates with liquid-like properties. Endogenous and enhanced green fluorescent protein (GFP)-tagged ERC1 accumulated near protruding sites in migrating cells ( Supplementary Fig. 1a,b; Movie 1). Endogenous ERC1 and its ligand proteins liprin-α1 and LL5 are expressed by the cell types used in this study: we have previously shown that the three proteins partially colocalize near the cell edge of migrating human breast cancer MDA-MB-231 cells 5 . We show here that they partially colocalize also at the front of migrating HT1080 human fibrosarcoma cells, a different type of motile tumor cell ( Supplementary  Fig. 1b). Moreover, in the cytoplasm of spread non-motile COS7 cells the three endogenous proteins accumulated near focal adhesions ( Supplementary Fig. 1c).
An increase in the local concentration of specific proteins may trigger phase transition 16 . We compared the behavior of PMAPs components ERC1 and liprin-α1 in cells. ERC1 accumulated into cytoplasmic condensates in 30% of transfected cells, while liprin-α1 remained diffuse in virtually all cells (Fig. 1b,c). At low levels of expression GFP-ERC1 displayed a homogeneous distribution throughout the cytoplasm and did not form condensates. ERC1 condensates appeared 2-4 h after transfection and increased in number and size with time (Supplementary Movie 3; Fig. 1d). When we plotted the presence of GFP-ERC1 condensates towards its level of expression, we observed that condensation occurs when ERC1 has reached a certain level of protein expression in the cytoplasm (Fig. 1d).
ERC1 forms extended homodimers. The structure within molecular scaffolds is in many cases poorly understood, and detailed information on ERC1 structure is missing. We have used predictions and electron microscopy on the purified ERC1 protein to address the mechanisms that may underlie the formation of condensates by this protein. Sequence-based structural prediction suggests that ERC1 adopts a dimeric coiled-coil structure 19 (Supplementary Fig. 2). However, the 150 N-terminal residues and other shorter sequences throughout ERC1 were predicted to be IDRs 20 with low probability to form coiled-coils. A number of features of the primary sequence of the first 147 residues of ERC1 support the structural disorder of the N-terminus of ERC1 ( Supplementary Fig. 2). This region is characterized by low abundance of charged residues: total of Arg + Asp + Glu + Lys is 15% for ERC1-147 versus 36% for full length ERC1. ERC  is characterized by less negatively charged residues (<5% of Asp + Glu residues), resulting in a predicted total net charge at pH 7 of +8 (mean net charge per residue of +0.054; theoretical pI = 10.9), compared to the predicted total net charge at pH 7 of -27 for the full length ERC1 (mean net charge per residue of -0.024; theoretical pI = 5.7). As observed for disordered proteins 21 , this region includes 63% of disorder-promoting residues, and is especially enriched in disorder-promoter Gly, Pro and Ser residues ( Supplementary Fig. 2). Moreover a low complexity region (LCR) was found by the SEG program (http://www.biology.wustl.edu/gcg/seg.html) corresponding to residues 13 to 51 of the predicted N-terminal IDR. LCRs represent a feature commonly found within disordered regions 16 .
We experimentally verified the low-resolution structure of ERC1 by rotary shadowing electron microscopy on recombinant MBP-tagged ERC1. This analysis revealed that ERC1 forms parallel homodimers of about 119 ± 18 nm (n = 29) (Fig. 2a). Dimerization was confirmed by static light scattering (Fig. 2b). The variable distance between the two globular portions in different dimers suggested that the MBP tag was linked to the flexible disordered N-terminus of ERC1. This first description of the general structure of ERC1 is coherent with an elongated protein with extensive coiled coils and flexible N-terminus.
In support of the extended/partially disordered structure predicted for ERC1 we used limited proteolysis 22 , since extended and/or disordered proteins are more sensitive to proteases than structured proteins 23 . Lysates from cells expressing GFP-tagged full length, N-terminal (ERC1-N, residues 1-676) or C-terminal polypeptides (ERC1-C, residues residues 677-1116) showed loss of full polypeptides and appearance of fragments at increasing trypsin concentrations ( Fig. 2c; Supplementary Fig. 3), with a band the size of GFP (≈30 kDa) remaining as unique trypsin-resistant fragment detectable by immunoblotting with anti-GFP antibodies at the highest trypsin concentration tested. The same filters reprobed for vinculin showed lower sensitivity to trypsin of this structured protein.
To verify if ERC1 is a dimer in the cytoplasm of living cells, we harnessed the Numbers & Brightness (N&B) technique 24 . Comparison of fluorescent intensity maps showed that the average brightness (ϵ) of GFP-ERC1 was similar to that of dimeric GFP (dGFP), and higher than monomeric GFP (mGFP): the ratio ϵ ERC ⁄ϵ mGFP = 2.1, was similar to ϵ dGFP ⁄ϵ mGFP = 1.8 (Fig. 2d), indicating that cytosolic ERC1 was dimeric. The oligomerization state of ERC1 in droplets couldn't be tested with N&B, as the conditions were out of the range for applicability of this technique. (b) Transfected COS7 cells analyzed with anti-tag Abs (green); blue is DAPI (widefield epifluorescence microscopy). Scale bar, 20 µm. Right: percentage of cells with condensates (n = 60-97 cells/condition). (c) COS7 cells plated on 2.5 μg/ml fibronectin-coated coverslips were transfected with mCherry-ERC1 and fixed after 24 h. Widefield epifluorescence microscopy of ERC1-positive condensates of different sizes were observed in several cells. Scale bar, 20 µm. (d) Top: GFP-ERC1 expression induces the formation of condensates in a concentration-dependent manner in the cytoplasm of COS7 cells. Scale bar, 25 µm. Bottom: in each transfected cell, the cytoplasmic fluorescence intensity (expressed as mean grey value, arbitrary units, a.u.) was measured every hour for 12 h, starting from the beginning of detectable GFP-ERC1 expression (n = 17 cells). The left www.nature.com/scientificreports www.nature.com/scientificreports/ The N-terminal region of ERC1 influences the properties of cytoplasmic condensates. The dynamic behavior of fusing cytoplasmic ERC1 assemblies is consistent with their evaluation as liquid-like particles (Fig. 1e). We quantitatively analyzed the fusion events 25,26 to gain insight into the biophysical properties of ERC1 condensates. The aspect ratio (A.R.) of smaller droplets was 1.43 ± 0.026 (Fig. 3a); deviation from round shape (A.R. = 1) was possibly due to constrains imposed by the crowded cellular environment, as indicated by the finding that larger condensates had dynamic irregular shapes (Supplementary Movie 3). Time course of fusion events leading to a larger droplet indicated that fusion occurred by exponential relaxation (relaxation time τ = 4.53 s) (Fig. 3b), as for coalescing liquids 26 . Plots of τ versus ℓ values (ℓ, length scale = size of the droplet) from several fusion events showed that the time constant of relaxation was correlated to the size of ERC1 assemblies (R 2 = 0.29) (Fig. 3c), which is consistent with a liquid-like behavior. For liquid droplets, τ = η/γ·ℓ (η, viscosity; γ, surface tension) 27 . Regression of τ over ℓ values gave an inverse capillary velocity η/γ = 3.49 s.µm -1 (Fig. 3c). Similarly, the average of τ/ℓ ratios for each fusion event gave η/γ = 3.38 ± 0.39 s·µm -1 . We considered the average value of η/γ = 3.4 s·µm -1 .
Phase separation of proteins may be driven by disordered regions 16 , and has been observed also for proteins with predicted coiled-coil regions implicated in the assembly of the centrosomes 31 and synapses 32 . We tested if the disordered N-terminus of ERC1 ( Supplementary Fig. 2) was required for the formation of the condensates. The ERC1 proteins lacking either the low complexity region (ERC1-Δ51) or the full predicted N-terminal IDR (ERC1-Δ147) could form condensates like the full length ERC1, indicating that the predicted disordered N-terminal region is not necessary for the assembly of ERC1 condensates. Moreover the amino-terminal ERC1(1-55) and ERC1(1-150) fragments were both cytosolic, and therefore not sufficient to induce the formation of condensates (Fig. 3d). ERC1-Δ147 showed a non-significant trend to form larger condensates compared to full length ERC1 ( Supplementary Fig. 4). Thus the predicted disordered N-terminal region was neither sufficient nor necessary for the assembly of ERC1 scaffolds that behave as liquid-like compartments in living cells.
We next evaluated the effects of the N-terminal deletions of ERC1 in migrating cells. Cell motility is supported by ERC1 and the associated proteins liprin-α1 and LL5 5,11 , which accumulate near the leading edge of the migrating cells ( Fig. 1a; Supplementary Fig. 1a,b). In MDA-MB-231 cells migrating on fibronectin, the ERC1 constructs lacking the N-terminal IDR could still partially colocalize with endogenous liprin-α1 at condensates near the leading edge, but less efficiently than the full length ERC1 (Fig. 3f). These findings suggest that although graph shows the mean grey values ± SEM at each time point. In the right graph the GFP-ERC1 expression level (mean grey value) is plotted against the presence (grey dots) or absence (yellow dots) of detectable cytoplasmic condensates.  www.nature.com/scientificreports www.nature.com/scientificreports/ the N-terminal IDR is not required for the formation of the condensates, it may be relevant to stabilize the condensates that are needed to recruit the interacting partners at the protruding edge of the migrating cells. In this direction, we observed that the truncated form GFP-ERC1-Δ147 lacking the predicted N-terminal IDR had a prevalent diffuse distribution in most transfected MDA-MB-231 cells migrating on fibronectin, compared to the full length GFP-ERC1 protein that localized at the front of most migrating cells (Fig. 3f). Our previous findings showed that the localization of endogenous liprin-α1 at the edge of migrating cells was only mildly affected by the lack of ERC1 5 . On the other hand endogenous liprin-α1 was required for the recruitment of endogenous ERC1 at the cell edge 5 . In this direction, here endogenous liprin-α1 could still localize at the front of polarized cells expressing the GFP-ERC1-Δ147 mutant, where it partially colocalized with the ERC1 mutant (Fig. 3f). The fact that GFP-ERC1-Δ147 was largely diffuse may be a consequence of the decreased stability (i.e. increased liquid-like properties) of the ERC1-Δ147 condensates forming at the edge of the migrating cells.
Interestingly, the migration of MDA-MB-231 cells was moderately but significantly inhibited by the expression of either deletion mutant compared to full length ERC1 (Fig. 3g). These results suggest that the changes detected in the liquid properties of the ERC1 condensates formed by IDR-deleted proteins (Fig. 3e) may underlie the observed effects on the subcellular distribution of the condensates in migrating cells, and the moderate effects observed in the migration of the cells that express these mutants. On the other hand we can not exclude the possibility that these changes are due to loss of specific, so far unknown protein interactions with the amino-terminal end region of ERC1.

ERC1-induced condensates recruit PMAP components and are affected by liprin-α1.
Liprin-α1, LL5 and GIT proteins are part of a network regulating protrusion and focal adhesions (Fig. 4a) 5,33 . These proteins were specifically recruited to ERC1 condensates (Fig. 4b), while a control cytosolic enzyme (GADPH, glyceraldehyde 3-phosphate dehydrogenase) and other focal adhesion proteins such as α-actinin, talin, vinculin and Src were excluded (Fig. 4c). These observations were quantified as percentage of ERC1-positive condensates including each of the different endogenous proteins (Fig. 4d, Supplementary Fig. 6a)). These results show that ERC1 condensates selectively recruit client proteins that regulate cell motility.
The interaction of ERC/ELKS with liprin-α is relevant to a number of important cellular processes 2,6,11,34,35 . ERC proteins interact with the central ERC-binding region (EBR) of liprin-α 2 (Fig. 5a). Deletion of the EBR prevented the recruitment of liprin-ΔEBR to both ERC1-and ERC1-N-induced condensates, while the EBR fragment was efficiently recruited to both ERC1 and ERC1-N condensates (Fig. 5b,c, Supplementary Fig. 6b). Liprin-α1 remained diffuse in cells co-expressing ERC1-C (Fig. 5d). These data show that the EBR is necessary and sufficient for the recruitment of liprin-α1 to ERC1 condensates.
We then tested the effects of endogenous liprin-α1 and LL5 proteins on the formation of condensates. Silencing of liprin-α1 did not affect droplets formation ( Supplementary Fig. 7A); also silencing of LL5 did not significantly influence the formation of ERC1 condensates ( Supplementary Fig. 7b).
We have previously shown that endogenous liprin-α1 is required for the subcellular concentration of ERC1 at PMAPs forming near the leading edge of migrating MDA-MB-231 tumor cells 5 . We hypothesize that liprin-α1 is required for the local concentration of the endogenous/low expressed ERC1 protein at specific sites of the cells, thus allowing the condensation of the protein by phase separation when and where it is required. According to this hypothesis, if the cells express higher levels of the protein, the condensates may form by overcoming a threshold concentration that allows phase separation to occur independently of specific localized signals. The finding that high levels of expression of ERC1 allow the formation of condensates in the cytoplasm even after depletion of endogenous liprin-α1, shows that liprin-α1 is not necessary for the process of ERC1-induced phase separation per se. This is confirmed also by the finding that the GFP-tagged amino-terminal construct ERC1(1-420) lacking a large part of the minimal region required for binding to liprin-α proteins (corresponding to amino acid residues 126-567 of ERC1 2 ) is still able to efficiently induce the formation of condensates (Fig. 5e). The full length FLAG-Liprin-α1 protein recruited at condensates induced by full length GFP-ERC1 was indeed excluded by the condensates induced by GFP-ERC1(1-420) (Fig. 5f).

Conclusions
Phase separation near specific sites of the cell periphery presents an elegant cellular mechanism to control the spatial localization and processing of molecules. These findings indicate that ERC1 scaffolds may behave as cytoplasmic liquid phases that host binding partners belonging to a protein network that has been implicated in the regulation of cell motility 5,11 . This may provide a way to promote the turnover of molecular components and to (ERC1). The mean brightness of dGFP and ERC1 is higher than the brightness of mGFP. The brightness map was smoothed with a Gaussian filter with standard deviation 1 pixel. Scale bar, 5 µm. Right: brightness fold change of cytoplasmic dGFP and ERC1 compared to mGFP. The median ERC brightness (ϵ ERC1 ⁄ϵ mGFP = 2.1) is comparable with the dimer's one (ϵ dGFP ⁄ϵ mGFP = 1.8); n = 28 (mGFP), 26 (dGFP), and 25 (GFP-ERC1); nonparametric one-way ANOVA. www.nature.com/scientificreports www.nature.com/scientificreports/ recruit enzymes such as kinases and/or phosphatases needed to sustain the protrusive activity of motile cells, possibly by modulating the effects and activity of specific components. Our evaluation of ERC1/ELKS-mediated scaffolding as condensates provides an intellectual framework for future analysis of other large molecular networks including this protein, such as the active zone of synapses 6 .
For FRAP analysis after spot bleaching, a 2 μm diameter ROI was bleached with 70% laser power at 488 nm (argon laser line) with 1 iteration for 0.2 sec. After bleaching, images were taken every 0.2 sec for 2 min to monitor fluorescence recovery, which was calculated as described in the previous paragraph; for spot bleaching, "ctrl" stands for a nearby ERC1-positive condensate area.
Immunofluorescence and image analysis. Transfected and non-transfected cells plated on 2.5 μg/ml fibronectin-coated glass coverslips were processed for immunofluorescence as described 37 . Briefly, cells were fixed for 10 min with 3% paraformaldehyde at room temperature, permeabilized with 0.1% saponin or Triton-X100 in PBS, incubated with primary Abs, washed, incubated with secondary Abs, and mounted with ProLong Gold antifade mounting solution (Thermo Fisher Scientific). Cells transfected only with GFP constructs were observed after fixation without enhancing the GFP signal with anti-GFP Abs. Cells were observed with epifluorescence microscopes: Zeiss AxioImager M2m equipped with AxioCam color CCD camera, with Plan-Neofluar 40x lens (NA 0.75) and Plan-Apochromat 63x lens (NA 1.4); with Zeiss Axiovert 135 TV with Hamamatsu CCD Orca II camera and Plan-Apochromat 63x lens (NA 1.4). Confocal images were acquired at a Perkin Elmer UltraVIEW spinning disk confocal microscope with EM-CCD camera and Plan-Apochromat 63x lens (NA 1.4); or at a Leica TCS SP8 SMD FLIM laser scanning confocal microscope equipped with HC PLAPO CS2 63x lens (NA 1.4). For quantitative analysis of cells forming droplets, transfected cells were randomly imaged at a wide field microscope. For quantification of droplets positive for endogenous proteins, confocal images were visually analyzed. For quantification, 2-4 independent experiments were used. The formation of condensates was followed in COS7 cells plated for one day on fibronectin-coated dishes (2.5 μg/ml for 1 h at 37 °C), and then transfected to express GFP-ERC1. Right after transfection, the cells were imaged for 1 day with Live-Cell Imaging System equipped with 20x lens (Essen BioScience, Ann Arbor, MI) to follow the formation of cytoplasmic GFP-ERC1 droplets. For the graphs shown in Fig. 1d, the fluorescence intensity in each transfected cell was measured every hour for 12 h, starting from the beginning of detectable GFP-ERC1 expression, by using the Image J software.
Data on their amino acid composition were extracted using the ProtParam tool (http://web.expasy.org/protparam/). For the analysis of primary sequences, we considered order-promoting (Asn, Cys, Ile, Leu, Phe, Trp, Tyr, Val) and disorder-promoting amino acids (Ala, Arg, Gln, Glu, Gly, Lys, Pro and Ser) as defined 21 . Data for globular proteins were extracted from published studies (42). Net charge at pH 7 was calculated as the difference between positively (Arg and Lys) and negatively charged (Glu and Asp) residues. The mean net charge was obtained by dividing the net charge values by the number of total amino acids in the sequence. Intrinsically disordered regions were identified utilizing the DisEMBL program (http://dis.embl.de/cgiDict.py) for intrinsic protein disorder prediction 20 .

Characterization of the biophysical properties of ERC1-positive droplets in cells.
Analysis of fusion events. Fusion events were analyzed in time on movies of transfected cells acquired for 5 m at a confocal microscope (TCS SP8 SMD-FLIM Leica equipped with HC PLAPO CS2 63x lens (NA 1.4), with adaptive focus control and Oko-Lab stage incubator). The aspect ratio (A.R.) of droplets was determined by ImageJ software, by fitting an ellipse to the shape of two fusing droplets and calculating A.R. = ℓlong/ℓshort (ratio between the longer [ℓlong] and the shorter [ℓshort] diameter of fusing droplets) 25 . A.R. were measured from time lapses and plotted as a function of time, and typically relaxed toward a steady value. For each fusion event between two droplets A.R. was measured at δ = .
t 0 52 s intervals and its decline was fitted according to the exponential curve: ; τ (the timescale or relaxation time for fusion) and . . ∞ A R were estimated from the autoregressive relation: . . + δ = . . = ⋅ . . +   The relaxation time τ for fusion is expected to be directly proportional to the length scale ℓ of the droplets according to the relation: τ ≈ (η/γ) · ℓ 25 . The ratio between the viscosity of the droplet (η) and its surface tension (γ) is called inverse capillary velocity: η/γ ≈ τ/ℓ. To measure the inverse capillary velocity η/γ, the dynamics of fusion between pairs of ERC1-positive droplets was analyzed by measuring the A.R. as a function of time for droplets with an area of up to 10 µm 2 at the time of contact before fusion. The inverse capillary velocity of the system was estimated: (1) from the average value of the ratio of τ to length scale (ℓ) for each droplet pair (i.e. the diameter of the droplet pair at the beginning of fusion); ℓ = [(ℓ long (t = 0) -ℓ short (t = 0)) · ℓ short (t = 0)] 1/2 25 ; (2) alternatively, it was estimated from the regression of the set of τ values on the corresponding length scales. The same fit was performed on the grouped measurements from the fusion events and a further estimate of inverse capillary velocity was obtained from the resulting τ value and the average length scale.
The determination of the length scale of ERC1 dimers (ξ) was as follows. Based on the structural features of the ERC1 dimers revealed by rotary shadowing electron microscopy, we can approximate the ERC1 dimer, made of two 128 kDa monomers (each made of 1116 residues), to a cylinder with length l ≈ 100 nm and with a radius r ≈ 1 nm 42 . This gives a surface area ≈ 2·π·r × l = 628 nm 2 . The length scale (ξ) can be defined as the square root of the surface area of the protein molecule; therefore ξ 2 = 628 nm 2 .
Numbers & brightness. The plasmids coding for the monomeric mEGFP-GPI and the dimeric mEGFPmEGFP-GPI constructs were a generous gift of Dr. Andrea Orsi and Eelco van Anken (San Raffaele Scientific Institute, Milano). From these plasmids we obtained by site-directed mutagenesis the plasmids coding for the monomeric mEGFP and dimeric mEGFPmEGFP respectively, introducing a stop codon between the EGFP and the sequence signal coding for the capability of GPI lipid anchoring. COS7 cells plated on 3.5 cm diameter glass-bottom MatTek dishes coated for 1 h at 37 °C with 2.5 μg/ml fibronectin were transfected for 24 h. Before imaging cells were moved to imaging medium without phenol red.
Image acquisition. For each condition, time series of 300 images were acquired with a Leica SP8 confocal microscope with the following settings: frame size 512 × 64 pixels, speed 400 Hz, pinhole size 1 AU. These settings correspond to a pixel size of 120.37 nm and pixel dwell time of 1.20 µm. The acquisition of each time-series takes in total 26 seconds. The GFP fluorescence is excited with a white light laser source setting the wavelength at 489 nm and a detection window at 500-600 nm. The Images are recorded through a Hybrid photon counting detector in photon counting mode.
Data analysis. The acquired movies were analyzed using custom-written routines in Matlab to generate a map of the molecular brightness of the fluorescently labelled complexes (Fig. 2c) by calculating  = − t (var(I ( )) mean ij ij (I (t)))/mean(I (t)) ij ij , for every pixel i j ( , ). To minimize the effect of cell movement we use boxcar analysis 43 , with a boxcar filter size equal to 50 frames. In order to identify pixels belonging to the cell cytoplasm from background pixels, we then build a 2D histogram in which we group pixels that have similar values of mean intensity and brightness. With an elliptical tool, it is possible to select a region of the histogram and interactively check the corresponding region within the cell. This method helps avoiding pixels with brightness outliers and border regions that usually contribute with some extra-variance due to cell movement. Note that during the selection process from the histogram the brightness values are kept hidden to minimize any bias due to manual selection. The area of the histogram selected is finally fitted with a 2D Gaussian: Limited proteolysis. For limited proteolysis on cell lysates, cells were washed twice with ice-cold TBS (150 mM NaCl, 20 mM Tris-HCl, pH 7.5) and lysed in lysis buffer (100 mM KCl, 1 mM DTT, 0.5% Triton X-100, 25 mM HEPES-KOH, pH 7.5). The insoluble material was removed by centrifugation and protein concentration determined by Bradford protein assay (Bio-Rad). For limited proteolysis on cell lysates and on purified proteins, trypsin was diluted in lysis buffer and in 100 mM KCl, 25 mM HEPES-KOH, pH 7.5, respectively. Aliquots of lysates (50 μg protein) or purified proteins were incubated for 5 minutes on ice with different concentrations of trypsin. Proteolysis was stopped by denaturing the samples at 96°C, and samples analyzed by SDS-PAGE followed by immunoblotting with the indicated Abs (cell lysates). When indicated, filters for immunoblotting were subjected to acid stripping and re-probed with different antibodies.
Production 6xHis-MBP-ERC1 and electron microscopy. Full length ERC1 obtained by PCR from GFP-ERC1 was inserted into a modified pOEM vector to produce His6-MBP-ERC1 for electron microscopy analysis.
Spodoptera frugiperda. Sf9 cells in ESF921 medium (Expression Systems) were co-transfected with linearized viral genome and the expression plasmid and selected for high infectivity. Viruses were produced and used to infect Sf9 cells and to obtain lysates for protein purification as described 44,45 . The 6xHis-MBP-ERC1 fusion protein was purified as previously described for extended coiled-coils in 20 mM HEPES pH7.4, 250 mM NaCl, 0.5 mM TCEP (46). Briefly, amylose resin was used to affinity isolate the dimeric ERC1 protein, subsequently eluted with 10 mM maltose, and subjected to size-exclusion chromatography. Protein concentration was determined by UV 280 and Bradford assay.
The light-scattering from purified ERC1 was analyzed by an autosampler equipped Viskotek TDAMax system as described 45 . The data obtained were averaged across the protein elution volume and molecular masses determined by OmniSEC software package.
Samples for rotary shadowing were prepared as described 45 . Briefly, samples diluted in spraying buffer (100 mM ammonium acetate, 30% glycerol) were sprayed via a capillary onto freshly cleaved mica chips, which were then