The Structure of Vimentin Linker 1 and Rod 1B Domains Characterized by Site-directed Spin-labeling Electron Paramagnetic Resonance (SDSL-EPR) and X-ray Crystallography*

Background: The complete structure is not known for any intermediate filament (IF) protein. Results: Linker 1 and rod 1B in human vimentin were characterized using electron paramagnetic resonance spectroscopy and x-ray crystallography. Conclusion: The rod 1B adopts two functional conformations that mediate formation of an anti-parallel “A11” tetramer. Significance: Understanding vimentin structure provides insight into all IFs and the related human pathologies. Despite the passage of ∼30 years since the complete primary sequence of the intermediate filament (IF) protein vimentin was reported, the structure remains unknown for both an individual protomer and the assembled filament. In this report, we present data describing the structure of vimentin linker 1 (L1) and rod 1B. Electron paramagnetic resonance spectra collected from samples bearing site-directed spin labels demonstrate that L1 is not a flexible segment between coiled-coils (CCs) but instead forms a rigid, tightly packed structure. An x-ray crystal structure of a construct containing L1 and rod 1B shows that it forms a tetramer comprising two equivalent parallel CC dimers that interact with one another in the form of a symmetrical anti-parallel dimer. Remarkably, the parallel CC dimers are themselves asymmetrical, which enables them to tetramerize rather than undergoing higher order oligomerization. This functionally vital asymmetry in the CC structure, encoded in the primary sequence of rod 1B, provides a striking example of evolutionary exploitation of the structural plasticity of proteins. EPR and crystallographic data consistently suggest that a very short region within L1 represents a minor local distortion in what is likely to be a continuous CC from the end of rod 1A through the entirety of rod 1B. The concordance of this structural model with previously published cross-linking and spectral data supports the conclusion that the crystallographic oligomer represents a native biological structure.


Introduction
Three different protein families make up the filamentous cytoskeleton of the mammalian cell cytoplasm: thin filaments composed of actins, microtubules composed of tubulins, and intermediate filaments (IFs) composed of intermediate filament proteins. In contrast to the small number of actin and tubulin proteins, the IF family is composed of scores of members, expressed in both tissue-specific and temporalspecific patterns (1,2). Further emphasizing their diversity, IFs can be either homopolymeric or heteropolymeric, but the majority of IFs are constructed from 1 or 2 IF proteins. IF proteins are not soluble under physiologic conditions, and they spontaneously assemble into IFs upon dialysis from chaotropes such as 8M urea (3,4). Thus, it is not surprising that while the x-ray crystal structures of actin and tubulin are known (5)(6)(7), no complete IF protein has had its structure solved by crystallographic or other means. The general predictions of IF protein structure, deduced from their amino acid sequences based on cDNA sequences that began to emerge in the 1980s, have only recently been subject to revision based on both spectroscopic and crystallographic structural data.
Analysis of IF protein sequences has consistently shown that the central region is composed of amino acids conforming to a heptad repeat pattern (a-b-c-d-e-f-g) with non-polar amino acids preferentially located at the a and d positions (8)(9)(10). From this simple pattern, an helical coiled-coil (CC) structure (11) was predicted to form. Based on this pattern and prediction, the middle of all IF proteins became known as the central rod domain; head and tail domains that vary widely in both size and primary sequence are located at either end of the central rod (12)(13)(14)(15)). Short but highly conserved motifs are the head and tail domains of an IF protein (18).
In silico analysis of the central rod domain shows that the heptad repeat is interrupted by nonconforming amino acid sequences at three places (10,19,20). Thus, for the past 25 years, the central rod domain has been described as 4 -helical CC domains (1A, 1B, 2A and 2B) separated by small non-CC linker domains (linkers L1, L1-2 and L2) named by the regions they connect (1,14,15,21). Originally, the predicted non-helical nature of linker secondary structure suggested that linkers served as flexible connectors between the CC domains. However, in contrast to the prediction that linkers are flexible, our SDSL-EPR characterization of linker L2 revealed a rigid, rapidly assembling structure in which both chains were aligned in parallel (21).
Despite widespread acceptance of the predicted structure of IFs, very little informative experimental data existed until the early 2000s. Using a "divide and conquer" strategy, Strelkov, and co-workers solved x-ray crystal structures for protein segments extracted from both the amino and carboxy terminal regions of the central rod domain of vimentin (22)(23)(24). Surprisingly, the resulting structure of the highly conserved rod 1A domain showed an -helix but not a CC (PDB id 1GK7). However, based on the curvature of the observed -helix, the authors predicted that the region does form a CC during the earliest stages of IF assembly (24).
At the other end of the rod domain, rod 2B was found to form a long -helical CC (PDB id 1GK4). Near the middle of the structure, the evolutionarily conserved "stutter", a 3-residue shift in the register of the hydrophobic heptad repeat pattern in the primary sequence in this region, was found to cause only a slight deviation from a canonical left-handed CC structure (24). One heptad earlier than the predicted carboxy terminus of rod 2B, the CC-forming α-helices in this structure begin to separate, which was interpreted to represent the actual end of the CC region (22,24). One of the parallel CC dimers in this structure (the AB dimer) deviates substantially from proper 2-fold rotational symmetry and interacts with another CC dimer (the EF dimer) in an anti-parallel geometry; this interaction buries extensive solvent-accessible surface area in the interface between CC dimers (2443 Å 2 per CC dimer in PDB id 1GK4). The resulting tetrameric bundle contains both parallel and anti-parallel alpha-helices. However, the third CC dimer in the asymmetric unit of this crystal structure (the CD dimer) does not make any similar interaction. Moreover, these x-ray crystal structures were not derived from intact IF proteins assembled into a filament, and thus it is possible that the observed structures and oligomeric organizations may differ from the dominant conformations in vivo, especially near the non-physiological termini of the crystallized protein constructs.
Perhaps the most surprising structural result reported to date is the recently solved crystal structure of a protein construct containing vimentin rod 2A and L2 (PDB id 3KLT) (25). Instead of folding into a short parallel CC as expected, the peptides form an anti-parallel tetramer composed of parallel dimers which adopt a standard CC structure at vimentin positions 305-334.
Based on this structure, the authors concluded that the previously predicted rod 2A and L2 regions exist as a single pair of parallel helices in the assembled filament, and not a CC structure. This conclusion supports a hypothesis advanced by David Parry (26) that the amino acid sequences of rod 2A and L2 conform better to a right-handed CC structure with a hendecad (11residue) repeat rather than a canonical left-handed CC structure with a heptad repeat. However, the aforementioned crystal structure of rod 2A and L2 (25) shows that, near the region previously predicted to be the beginning of coil 2B (residues 290-300), the parallel α-helices gradually adopt a canonical left-handed CC structure with a heptadrepeat. The CC structure begins over positions 302-305 and extends through the end of the crystallized construct at residue 334 (25). The final seven residues in this construct overlap the CC region observed in the earlier crystal structure of rod 2B, providing evidence that the parallel CC in this region of human vimentin extends continuously from residues 305 through 405. Therefore, the most recent crystallographic data confirm our previous conclusions from SDSL-EPR that vimentin adopts a non-CC structure between residues 291 and 301 (21). Thus, independently, the available SDSL-EPR and crystallographic data consistently show that residues 291-302 are not assembled into a CC structure, contrary to earlier predictions.
The structures of the predicted non-CC regions L1 and L1-2, which precede and follow rod 1B, have not been described by any method. Our prior SDSL-EPR data support a CC structure in at least part of rod 1A and rod 1B, which flank the L1 segment (27,28). In this report, we present SDSL-EPR and crystallographic analyses of protein constructs containing the end of L1, the beginning of rod 1B and most, if not all, of rod 1B. Collectively, these data demonstrate that the non-CC region of L1 is likely to be shorter than predicted (29), while rod 1A and 1B form one continuous left-handed CC structure with the L1 segment representing only short perturbation. Our crystal structure demonstrates that rod 1B forms a parallel CC which dimerizes with itself in an antiparallel geometry to form a tetrameric -helical bundle that it likely to be a key feature of higher order vimentin oligomerization and IF structure in vivo. This interaction is centered near residue 191, as predicted by our prior EPR studies.

Experimental Procedures
Site-directed mutagenesis and purification for spin labeling.
Vimentin mutants were constructed using site-directed mutagenesis, and recombinant proteins were produced in E. coli. Vimentin readily forms inclusion bodies; these were isolated and the recombinant vimentin purified and spin-labeled as described in detail previously (18,21,27,28). In short, site-directed mutagenesis was used to introduce cysteine residues at specific sites in a vimentin expression construct (originally provided by Roy Quinlan, University of Durham, Durham, UK) using Bio-Rad iProof DNA polymerase and mutagenic oligonucleotides. Coding sequence changes were confirmed by automated DNA sequencing. Mutant vimentin protein was produced by bacterial over-expression using a pT7 vector and E. coli BL21AI (Invitrogen, Carlsbad, CA). Inclusion bodies were purified using lysozyme/DNase, high/low salt washes, and chromatography (AKTA FPLC, GE Healthcare, Piscataway, NJ). Site-directed spin labeling was performed by first treating the purified protein with 100 µM TCEP (Tris-(2-carboxyethylphosphine hydrochloride, Invitrogen) followed by spin labeling with 500 µM O-87500 (1-oxyl-2,2,5,5-tetramethyl-3-pyrroline-3-methylmethanethiosulfonate-d15 (MTSL-d15), Toronto Research Chemicals, Toronto, Canada). The unincorporated label was separated from spinlabeled protein by chromatography over a Source S column (AKTA FPLC). Protein concentrations were measured by the BCA method (Pierce, Rockford, IL). Purified spin-labeled proteins were stored at -80°C.
In vitro filament assembly and electron microscopy.
Filament assembly was generally conducted by dialyzing the spin labeled protein from 8 M urea into filament assembly buffer, (10mM Tris, pH 7.5, 160mM NaCl) overnight at 37°C. Electron microscopy of the negatively stained samples was performed to verify the filament assembly of each mutant. Following dialysis, 10 µl of the sample was removed and stained with 1% uranyl acetate on formvar-coated carbon grids and then observed using a Phillips CM-120 Electron Microscope, with a Biotwin Lens, (FEI Company, Hillsboro, OR, made in Eindhoven, Netherlands) operated at 80 kV acceleration voltage. Images were acquired with a Gatan MegaScan 794/20 digital camera (2K X 2K) or a Gatan BioScan 792 (Gatan, Pleasanton, CA).
EPR Spectroscopy of site-directed spin labels. EPR measurements of the spin-labeled proteins were conducted on a JEOL X-band spectrometer fitted with a loop-gap resonator (18,21,27,28). Spectra were collected from ~5-7 µl of purified, spin-labeled, dialyzed protein, at a final protein concentration of 25-100 µM, loaded in a sealed, quartz capillary tube. Spectra were obtained by a single scan of 120 s over 100 G at a micro-wave power of 4 milliwatts at room temperature (unless otherwise specified). Modulation amplitude (0.125 mT) was optimized to the natural line width of the attached nitroxide as previously described (18,21,27,28). Normalization of the spectra to the same number of spins was done by normalizing each spectrum to the same integrated intensity/amplitude. To improve the fidelity of the calculation, each sample was double-integrated after its solubilization in 2% SDS. Low temperature spectra were collected from samples frozen at -100°C. For increased low temperature sensitivity, samples in 5mM Tris, pH 7.5 were mixed with the appropriate volume of 10X IF assembly buffer (100mM Tris, pH 7.5, 1.6M NaCl) in an Eppendorf tube. After 25 µl of the mixture was rapidly pipeted into a capillary, it was centrifuged in a low-speed bench top centrifuge to collect the assembling filaments at the bottom. Mixing, pipetting and spinning were repeated to generate 2 capillaries for each sample. Both capillaries were placed in the low temperature cavity for spectral data collection.

Protein expression and purification for crystallography.
The vimentin protein was targeted by the Northeast Structural Genomics Consortium (http://www.nesg.org) as part of a project aimed a determining the three-dimensional structures of proteins involved in signaling networks associated with human cancer (30). Ligation-independent cloning was used to introduce a PCR product encoding residues 144-251 of human vimentin between the NdeI and XhoI sites in vector pET15_NESG to produce construct HR4796B-144-251-14.3, which is available from the PSI Materials Repository (http://psimr.asu.edu). The resulting protein with an N-terminal affinity-purification tag with sequence MGHHHHHHSHM was expressed in E. coli BL21(DE3) cells harboring the rare tRNAexpression plasmid pMGK.
The cells were induced overnight at 17˚ C using 1 mM IPTG after growth at 37˚ C to mid-log phase in MJ9 minimal medium (31) supplemented with selenomethionine and containing 10 µg/ml kanamycin and 100 µg/ml ampicillin. Cells were lysed by sonication in Binding Buffer (50 mM Tris-HCl, 500 mM NaCl, 40 mM Imidazole, 1 mM TCEP and 0.02% NaN 3 , pH 7.5) and centrifuged at 27,000 x g for 40 minutes at 4 o C. The supernatant was loaded onto an ÄKTAxpress (GE Healthcare) for an automated two-step IMAC and gel filtration purification procedure. Briefly, the His-tagged proteins were bound to a 5 ml HisTrap HP column (GE Healthcare) and eluted into an internal storage loop in 50 mM Tris-HCl, 500 mM NaCl, 500 mM imidazole, 0.02% (w/v) NaN 3 , pH 7.5. The major peaks were automatically injected onto a Superdex 75 gel-filtration column (GE Healthcare) equilibrated in low salt buffer (20 mM Tris-HCL, 100 mM NaCl, 5 mM DTT, pH 7.5). The peak fractions were concentrated to 10.6 mg/ml and frozen in liquid N 2 in 50 µl aliquots. Per liter of culture, this procedure yielded 100 mg of purified protein, which had a covalent molecular weight of 14.4 kDa based on MALDI-TOF mass spectrometry on a Voyager DE Pro (Applied Biosystems, Foster City, CA) compared to a theoretical value of 14.6 kDa (data not shown).

Crystallization.
Crystallization was performed in 1+1 µl hanging-drop vapor diffusion reactions at 293 K. Crystallization conditions were identified via high-throughput robotic screening of 1536 different conditions at the Hauptmann-Woodward Institute, Buffalo, NY (32).
Manual optimization yielded a final precipitant solution containing 18% (w/v) PEG-3350, 0.15 M ammonium sulfate, 0.1 M Tris-HCl, pH 8.0. Before flash freezing in liquid nitrogen, the crystals were cryoprotected using the well solution with ethylene glycol added to 15-20 % (v/v).
X-ray data collection and structure determination. X-ray data were collected using a wavelength of 0.979 Å from a single crystal maintained at 100 K on beamline X4A at the National Synchrotron Light Source at Brookhaven National Laboratory. The diffraction images were integrated and merged using DENZO and SCALEPACK, respectively (33). The structure was solved by single wavelength anomalous dispersion (SAD) using PHENIX (34). Se positions were found by AutoSol to calculate initial phases, which were used by AutoBuild (34) to generate an initial protein model. Manual rebuilding in COOT (35) was guided by inspection of 2Fo-Fc and Fo-Fc electron density maps. Protein models were refined with translation, libration, and screw-rotation (TLS) displacement of a pseudo-rigid body (36). During the final stage of refinement, strong NCS restraints (coordinate sigma of 0.05 and B-factor weight of 10) were maintained between residues 147-246 subunits in A and C and between residues 146-248 subunits in B and D. The programs HELANAL (37) and TWISTER (38) were used to measure -helix and CC geometry, respectively. Molecular graphics figures were prepared using PYMOL (39). Fig. 1 shows a cartoon of the vimentin central rod domain, along with pertinent amino acid sequences from L1 and the flanking CCs. Boundaries are based on in silico predictions of IF secondary structure, which identified rod 1B as beginning at Gly147 and terminating at Ala247 (29). Each end of rod 1B is flanked by a predicted non-helical linker region. Positions 139-146, including the helix destabilizing residues Gly140 and Gly142, are described as linker 1 (L1). JPRED secondary-structure analysis of positions 120-180 identifies positions 141-146 as non-helical and non-CC (not shown), while the remainder of the sequence is generally predicted to be both (http://www.compbio.dundee.ac.uk/wwwjpred/).

SDSL-EPR analyses of L1 and the amino terminus of rod 1B in human vimentin.
Similarly, JPRED analysis of the sequence of 181-250 predicts -helical and CC structure at all positions. We performed SDSL-EPR at positions 146-168 spanning the predicted N-terminus of the rod 1B CC. Consistent with our previous results, most of the resulting protein variants were able to form IFs after mutation to cysteine and attachment of the spin label, with exceptions noted in Table 1. Under these conditions, filaments are not formed, and the spectra report the local secondary, tertiary, and quaternary structure of the vimentin species formed in that environment. Qualitative inspection of the spectra in Fig To evaluate whether the line-broadening observed for the putative a and d positions is attributable to close proximity of these positions in the protein structure as opposed to restriction of the probe's motion, we collected low-temperature spectra and calculated d1/d values (see Table 1). This calculation provides a model-independent assessment of the proximity of spin-labeled side chains in a semi-quantitative manner. spin interaction at this site, suggesting that this residue is likely to be packed in the inter subunit interface even though it is not in proper register with the heptad repeat pattern shown by the immediately C-terminal residues in rod 1B. Moreover, the d1/d value observed at position 147 is slightly elevated compared to a typical heptad a or d position, which would be consistent with tight steric interactions at this site upon mutation of this Gly residue to a Cys residue and attachment of a spin-label. These observations suggest that a departure from canonical CC geometry at residues 146-147 at the N-terminus of rod 1B leads to close inter-helical packing of both of these adjacent residues. The crystallographic studies reported below support this interpretation.
Similar spectroscopic results were observed in our previously published SDSL-EPR study encompassing positions 120-145 at the C-terminus of rod 1A. The final residue fitting the regular heptad repeat pattern was Gly142, which was in register to be a heptad a position at the C-terminus of the rod 1A CC. A spin label at this position gave a significantly elevated d1/d value (0.6) compared to typical heptad a or d positions, just like Gly 147 at the N-terminus of rod 1B in the current study. Furthermore, the observed d1/d ratios at positions 143-145 failed to fit a heptadrepeat pattern, suggesting that the elevated d1/d value at position 142 reflected departure from canonical CC geometry at the C-terminus of rod 1A. Overall, the qualitative spectral results and d1/d values at residues 143-146 suggest that this L1 linker segment is a relatively rigid transition between the CCs formed by rods 1A and 1B. While L1 is likely to be more flexible than the CC regions it connects, it is not disordered and instead seems likely to represent a short, stutter-like distortion in what is otherwise a very long and regular CC structure spanning rods 1A and 1B.
Supporting mutagenesis studies of the L1 region. To provide further insight into the structure adopted by the L1 linker, we created various substitution, deletion and insertion mutants, and we assayed the resulting proteins for filament assembly (Fig. 3). To investigate the hypothesis that the presence of glycines 140, 142 and 147, was evidence of a sterically constrained local environment, we began by altering G 142 . The G142A and G142F mutants were both able to assemble into IFs. The ability of the G142F mutant to assemble argues that L1 is not critically constrained and that the presence of Gly 140, 142 and 147 is not solely the result of selection for a residue without a side chain.
A different approach to assess the packing and phasing of the residues that make up L1 was performed by individually deleting each residue from 140-147. These single amino-acid deletions were well tolerated with minimal perturbations in IF assembly (Fig. 3, Table 1, and additional data not shown). Surprisingly, even the double-deletion mutant combining Gly140 and Gly147 also assembled into filaments. As controls for these deletion experiments we deleted either Val161 or Aps162 within the canonical CC region of rod 1B (a and b heptad positions, respectively) and observed no IF assembly. Thus, deletion of residues at the end of rod 1A, in linker 1 or at the start of rod 1B produce strikingly different effects than deletion of residues within the CC region. The contrasting tolerance for deletions in the L1 region possibly supports the inference from SDSL-EPR spectroscopy that it does not adopt a canonical CC structure.
Finally, we attempted to extend the canonical heptad repeat pattern at the N-terminus of rod 1B but most such constructs interfered with filament assembly. To extend the interfacial hydrophobic stripe in proper register with the CC formed by rod 1B, we made a double mutant in which Leu146 is changed to Gly and Gly147 is changed to Leu. This L146G/G147L protein failed to form IFs (Fig. 3). Similar results were observed by Herrmann and co-workers in an attempt to form a continuous CC structure linking rod 1B to rod 1A; addition of three properly phased aliphatic amino acids to L1 also prevented IF assembly (41). However, it is notable that IFs form after deletion of Gly147 (Fig. 3), which places Leu146 in an a heptad position relative to the canonical CC that starts with Tyr150, a d position. While further investigation will be required to understand the operative sequence constraints in L1, mutagenesis experiments in this region of the protein suggest that it does not adopt a canonical CC structure, consistent with the SDSL-EPR results presented above and the crystallographic results presented below.

The crystal structure of tetrameric rod1B from human vimentin.
A semi-automated procedure employing Ni-NTA affinity chromatography followed by gel filtration chromatography was used to purify a soluble rod 1B peptide containing residues 144-251 of human vimentin plus an 11-residue N-terminal affinity tag with sequence MGHHHHHHSHM.
The concentrated protein stock solution was characterized using analytical gel-filtration chromatography monitored by in-line refractiveindex and static-light-scattering detectors (Fig. 4). In a buffer at pH 7.5 with ~100 mM ionic strength, the vast majority of the protein elutes as an isolated peak with a broad trailing edge. Debye analysis shows a molecular weight of 64-70 kDa at the top of this peak, which progressively declines across the trailing edge at higher included volumes. Given the typical accuracy of these measurements (±15%), these data are consistent with the formation of a tetramer (as observed in the crystal structure reported below) with a tendency to dissociate during migration through the column. While a small amount of the protein elutes in a peak at higher molecular weight, the amount of this material varied substantially in different preparations (data not shown), making it likely to be an aggregated form rather than the product of a well-behaved polymerization reaction.
The crystal structure of the vimentin rod 1B peptide (Figs. 5-7 and Tables 2-3) shows a homotetrameric assembly matching the inter subunit interaction geometry predicted from our earlier solution EPR experiments (28). The crystal structure in space group P2 1 2 1 2 1 was refined at 2.8 Å resolution to working and free R-factors of 23.4% and 28.4%, respectively ( Table 2). The atomic model contains residues 145-249 in protomer A, 148-249 in protomer B, 144-246 in protomer C, and 144-248 in protomer D. In addition, the final five residues in the N-terminal affinity tag are visualized in chains C and D. The amino-terminal segments of protomers C and D visualized in the crystal structure represent the carboxy-terminal half of linker L1, as discussed further below.
The four protomers in the asymmetric unit of the crystal are arranged as a dimer of CC-dimers, with chains A and B comprising the first CCdimer and chains C and D the second CC-dimer (Fig. 5A-B and Table 2). Each protomer forms a single α-helix (e.g., as shown in Fig. 6), with a length varying from 151 to 161 Å (as measured by the distance between terminal Cα atoms). The single tetramer in the asymmetric unit of these crystals (Fig. 5A-B) is likely to represent the dominant physiological oligomer of rod 1B, commonly known as the A11 tetramer (40,41). The interfaces of the AB and CD dimers, which form asymmetrical parallel CCs (described in detail below), bury ~5,400-5,600 Å 2 of solventaccessible surface area per dimer (Table 3), while ~4,000 Å 2 is buried in the interfaces between the CC dimers in the likely physiological tetramer (the sums of the A-C, A-D, B-C and B-D interfaces in Table 3). An initial refinement was performed in the absence of any non-crystallographic symmetry (NCS) restraints. This refinement showed that the A and C subunits adopt very similar conformations (Fig. 6A) and that the B and D subunits also adopt very similar conformations (Fig. 6B), but that the A/C subunits adopt a significantly different conformation from the B/D subunits (Fig. 6C).
Therefore, noncrystallographic symmetry restraints were applied selectively to the A/C and B/D subunit pairs to improve the convergence of the final refinement ( Table 2).
The rod 1B tetramer comprises a symmetrical dimer formed by a pair of asymmetrical parallel helical CC dimers (Fig. 5). The A and B subunits form one of the two CC dimers in the tetramer, while the C and D subunits form the other. Leastsquares superposition of these CC dimers demonstrates that the asymmetrical CC formed by the AB subunits has a very similar conformation to that formed by the CD subunits (Fig. 5C), (i.e., consistent with the observation reported above) that the A and C protomers adopt equivalent conformations (Fig. 6A) as do the B and D protomers (Fig. 6B). However, least-squares superposition of the A and B protomers shows that they adopt substantially different conformations from one another (Fig. 6C), as do the C and D subunits (data not shown). The non-equivalence of their conformations is demonstrated clearly by superposition of the AB dimer on itself based on alignment of the A and B subunits (Fig. 5D). This non-equivalence produces substantial asymmetry in the structure of the parallel CCs formed by rod 1B, which is a critical structural feature determining its higher-order oligomerization behavior and therefore its biological function. This asymmetry is also reflected in the rotation angle of 147˚ yielding least-squares superposition of the A and B protomers or the C and D protomers forming the parallel CC dimers (Table  3), which is significantly different from the 180˚ angle that would relate the protomers in a symmetrical CC dimer.
The asymmetry in the CC architecture is encoded in the sequence of the rod 1B peptide, which produces a series of localized bends and kinks that differ in the α-helices formed by protomer A vs. B (Figs. 6C-G and 7A). (Note that the stereochemical features discussed here are all shared by protomers A and C as well as protomers B and D, as demonstrated by the fact that the AB dimer has the same conformation as the CD dimer, as shown in Fig. 5C.) These biologically important differences in protomer conformation must be stabilized by their mutual interactions, probably primarily by those across the interface of the parallel CC dimer. Superposition of four sequential segments of protomers A and B (Fig.  6D-G) demonstrates that substantial differences in the trajectory of the -helix axes occur at residues 174-178, 198-204, and 213-217. Analysis of the local -helical bending angle (Fig. 7A) shows systematic differences in the A/C vs. B/D protomers at all of these sites, although there is not a tight correspondence between differences in the computationally computed bending angle and the sites of significant divergence in -helix trajectory. Ultimately, the differential bending in the A/C vs. B/D protomers is attributable to systematic differences in backbone  (Fig. 7C) and  (Fig. 7D) angles, involving relatively modest variations almost all within the canonical right-handed -helical region of the Ramachandran plot. These differences occur in the absence of significant backbone distortions in the -helices, as monitored by the rise/turn (Fig.  7B), except for a small variation at residues 202-204 in the A/C protomers, which is the site of the most significant backbone kink (Fig. 6A-B). Overall, relatively modest variations in backbone geometry produce local differences in -helical bending angle that break the symmetry of the parallel CC dimer. Evolution has exploited the plasticity of the polypeptide backbone to produce this functionally critical asymmetry in the structure formed by two chemically identical rod 1B segments.

Stereochemical analyses of the rod 1B crystal structure.
Despite the obvious conformational differences between the A/C vs. B/D protomers (Fig. 6) and the resulting deviation from proper 2fold symmetry within the AB and CD dimers (Fig.  5D), they show comparatively minor deviations from the idealized parallel CC architecture first described by Crick (11). Every seventh residue makes locally symmetrical inter-helical contacts from tyr-150 (heptad position d) and met-154 (heptad position a) through leu-234 and his-238. The CC radius (Fig. 7E) varies only from ~4.4-5.6 Å throughout the structure. The CC pitch (Fig.  7F) varies more significantly, but only from ~90-170 Å from residues 155-225, indicating relatively standard local CC geometry throughout this region. The pitch increases dramatically after residue 225, where the parallel α-helices pack together essentially without coiling around one another. This region of the structure represents the only significant departure from canonical CC geometry, except for the overall deviation from proper 2-fold symmetry produced by the differential bending of the A/C vs. B/D protomers.
The asymmetrical CC dimers formed by subunits AB and CD interact with one another with proper two-fold symmetry (i.e., a rotation angle of 180˚ between the equivalent A and C protomers and B and D protomers, as indicated in Table 3). Therefore, the rod 1B tetramer has nearly perfect two-fold symmetry even through its CC building block are not symmetrical. This architecture is determined by the genetically encoded distortions from proper symmetry within the parallel CCs, which produce a mutually complementary interaction surface that involves completely different residues in the A/C vs. B/D subunits ( Table 3).
The fact that the A and B protomers in one CC dimer make very different inter-subunit interactions from one another can be seen visually by superimposing the ABCD tetramer on itself via least-squares alignment of either the equivalent A and C subunits or the non-equivalent B and A subunits. When the A and C subunits are aligned, the entire tetramer superimposes well (Fig. 5E). In contrast, when A and B are aligned, clear conformational differences are observed between the parallel CC dimer pairs (Fig. 5F), and importantly, the second CC dimer in each tetramer is found on opposite surfaces of the CC dimer containing the protomer used for superposition (Fig. 5F). The A subunit makes the same contacts to the D subunit as the C subunit makes to the B subunit (as detailed in Table 3), reflecting the proper 2-fold symmetry of the tetramer. However, the conformationally equivalent A and C subunits interact with one another in a completely different manner than the conformationally equivalent B and D subunits interact with one another. The A and C subunits bury ~1,300 Å 2 of solventaccessible surface-area per subunit in their mutual interface, via interactions between 11 residues in each subunit related by proper 2-fold symmetry ( Table 3). In contrast, the B and D subunits do not make any contacts at all outside of the nonphysiological N-terminal affinity tag ( Table 3). The asymmetrical structure of the parallel CC dimers forming the tetramer underlie these dramatic differences in the inter-subunit interactions between the A-C vs. B-D subunits, which are responsible for the formation of a stable tetramer with proper 2-fold rotational symmetry.
Note that the structural asymmetry in the constituent parallel CCs is required to produce a stable interaction of this kind without higher-order oligomerization. If the A and B subunits had equivalent conformations and formed a parallel CC dimer with proper 2-fold symmetry, the capacity for opposite surfaces of this dimer to interact with itself would produce higher-order polymerization and potentially unlimited growth of the resulting -helical bundle. Therefore, the tetrameric organization of rod 1B is encoded by the sequence features that produce asymmetry in the parallel CC dimer formed by this protein segment. This organization is likely to reflect its evolved physiological properties, because headless vimentin complexes generally assemble into tetramers but not higher-order oligomers (42). Our observations on the crystallized rod 1B construct provide further confirmation that additional regions of full-length vimentin are necessary for stable assembly of higher-order oligomers.
Crystallographically observed conformations at the N-terminus of rod 1B. Our crystallized vimentin construct starts at residue Ser144, in the middle of the L1 linker between rods 1A and 1B, which spans residues Lys143-Leu146.
The truncation of the protein in this region will tend to promote backbone disorder, meaning that limited information about the structure of L1 can be extracted from the crystal structure. Nonetheless, the observed conformations are consistent with our inferences derived from SDSL-EPR spectroscopy presented above. Fig. 8 shows the inter-helical interactions at the N-termini of both CC dimers in the asymmetric unit of the crystal structure, extending from the first visualized residues through residue 160 in the region with canonical CC structure. Consistent with our SDSL-EPR results ( Fig. 2 and Table 1), Tyr150 and Met154 lie at the hydrophobic interface created by the association of two α-helices in parallel CC geometry. In the CD dimer, in which the -helical structure in subunit D is extended by residues from the N-terminal hexahistidine affinity tag, the side chains of Leu146 also make hydrophobic contact across the inter-subunit interface. This contact, which was predicted by our SDSL-EPR studies ( Fig. 2 and Table 1), is one residue out-of-register compared to the regular CC structure that starts at residue 150 (a d position in the heptad repeat). As discussed above, a regular CC would have an inter-helical contact at residue Gly147 (which would be an a position in a regular heptad repeat). The lack of a side chain at this site interferes with canonical inter-helical CC packing and likely contributes to the stutter-like distortion that puts Leu146 in contacts across the inter-subunit interface at the adjacent site. The first residues visualized in the AB dimer are Leu146 and Ser144 (in the A and B subunits, respectively), presumably due to the disordering influence of the truncation of the native protein sequence before residue Ser144. Nonetheless, residue Leu146 in subunit B still points across the inter-helical axis towards subunit A. Therefore, the crystallographically observed conformations are consistent with our SDSL-EPR observations suggesting that L1 adopts a relatively rigid stutterlike structure involving an inter-helical contact at residue Leu146, a contact that is out-of-phase with the regular heptad-repeat CC that starts immediately following the adjacent residue Gly147.

Discussion
The data in this report extend the structural characterization of the human IF protein vimentin to cover approximately 90% of the central rod domain. We present SDSL-EPR spectroscopy data establishing linker 1 (L1) as a very short discontinuity between the parallel helical coiled-coil (CC) structures formed by rod 1A and rod 1B on either side. We present EPR data supporting the formation of CC structure by a part of L1, and x-ray crystallography data confirming this inference and demonstrating the formation of CC structure throughout the entirety of rod 1B. As judged against historic predictions, the crystallographic data that we present encompasses the entire rod 1B domain and demonstrates that this domain forms a tetramer previously described as an A11 assembly (41,43,44) (i.e., an anti-parallel arrangement of dimeric CCs formed by rod 1). In particular, the crystal structure confirms the close proximity of the side chains of residue Glu191 in the interface between two CC dimers interacting in an antiparallel geometry, consistent with our previous SDSL-EPR data obtained from assembled recombinant vimentin (28). Because purified rod 1B forms a tetramer outside of the crystal lattice (Fig. 4) and the tetramer observed in our crystal structure is consistent with our earlier SDSL-EPR results, we argue that this oligomer accurately depicts the native higher-order structure of vimentin. As a result of these data, the structure of linker 1-2 is the only remaining region of the vimentin rod domain that has not been characterized by crystallographic or spectroscopic means.
The SDSL-EPR data presented in this paper establish a high d1/d ratio as being a signature characteristic of the termini of the regular CC structures in the vimentin rod domain. Our previous SDSL-EPR characterization of rod 1A, which encompassed positions 120-145 (27), identified CC structure by qualitative and quantitative methods. Slightly puzzling was the higher than expected d1/d value (0.6) obtained at Gly142, which was predicted to be an a heptad position. Superficially, the observed d1/d value indicated a closer than typical distance between spins labels at such a position, and we conservatively concluded the data were not consistent with a canonical CC structure at this site.
Additional insight into this observation is provided by reconsideration of previous SDSL-EPR and x-ray crystallography data in the context of the new spectroscopic and crystallographic data presented in this paper. Our previous SDSL-EPR characterization of linker L2 (28) identified non-CC and CC structures in residues 280-305. We observed high d1/d values for the positions at the beginning of the rod 2B CC, including a value of 0.59 at Ala302. The recently solved crystal structure of the 265-330 fragment of human vimentin (25) demonstrates this position to be the start of the CC formed by rod 2B, consistent with our conclusions from SDSL-EPR (21). Therefore, we propose that a new spectroscopic "benchmark" has been established, i.e., that a particularly high d1/d value marks the transition point to and from canonical CC structures. Furthermore, based on this new benchmark, we identify position 142 as the end of the CC formed by rod 1A.
Similarly, based on our new proposed benchmark, we conclude from the SDSL-EPR data presented in this paper ( Fig. 2 and Table 1) that position 147 represents the beginning of the CC formed by rod 1B. Our assignment of Gly147 as an "a" heptad position is entirely consistent with the downstream phasing of the heptad repeat pattern starting at residue 150 in our crystal structure of rod 1B (Figs. 5-8 and Tables 2-3), our current SDSL-EPR data covering positions 146-168 ( Fig. 2 and Table 1), and earlier our earlier SDSL-EPR data covering positions 169-193 (28).
Early studies predicting the locations of the linker and CC regions in vimentin typically proposed that the CC regions are rigid while the linker regions are substantially more flexible (14,15,45).
It was hypothesized that this flexibility accommodated slight shifts in packing during filament assembly. Given the widespread prevalence of this structural model, our SDSL EPR characterization of linker L2 was striking in that it found this region to be a rapidly assembling and relatively rigid structure (21).
Our characterization of linker L1 presented in this paper ( Fig. 2 and Table 1) leads to a similar conclusion, i.e., that L1 forms a relatively rigid structure rather than a flexible loop. Further research will be required to establish whether linker L1-2 forms a qualitatively similar structure or a flexible loop connecting rod domains 1 and 2, as proposed in previous literature. It has not escaped our notice that the increased CC pitch at the end of rod 1B is similar to the situation within the recently solved rod 2A/Linker2 (26).
The crystal structure presented in this paper (Figs. 5-8 and Tables 2-3) demonstrates that the rod 1b peptide adopts two very different conformations. This structural polymorphism is likely to be critical for its biological function. One rod 1B subunit in each conformation is used to assemble an asymmetrical parallel CC dimer that itself dimerizes to form a tetramer with proper two-fold rotational symmetry (Fig. 5). All four subunits form α-helices that assemble into parallel CC dimers with relatively canonical geometry (Fig. 7). As labeled in this report (and PDB deposition 3UF1), subunits A and C adopt equivalent conformations, as do subunits B and D. However, these two subunit pairs differ substantially in their local bending angle at three sites, producing a substantial difference in overall conformation (Fig. 6). Therefore, the parallel CC dimers formed by combining one subunit in each conformation (i.e., A with B and C with D) do not have two-fold rotational symmetry. Symmetrical CC dimers forming a symmetrical interface would tend to grow without limit based on propagation of equivalent inter-subunit structural interactions. Therefore, the structural asymmetry in the AB and CD dimers, which represents a remarkable example of evolutionary exploitation of the structural plasticity of polypeptides, explains the ability of these dimers to form a tetramer with proper two-fold rotational symmetry that does not polymerize into higher-order oligomers. Despite the importance of this asymmetry in determining the functional oligomerization properties of rod 1B, we are unaware of any previous predictions of structural asymmetry in this region of vimentin, which is the longest predicted CC region in the molecule.
While this asymmetry is necessary to prevent high-order polymerization of rod 1B, further research will be required to determine whether it is a dominant feature in actually controlling IF diameter. Published results on vimentin constructs with truncated head or tail domains make it highly unlikely that the asymmetry in rod 1B is the only determinant of IF caliber (42,46). Our data suggest that the head domain, rod 1B, and the tail each have a role in limiting assembled vimentin filaments to 10 nm in diameter.
In summary, this report provides a combination of spectroscopic and crystallographic evidence supporting formation of a CC structure by rod 1B in human vimentin. The crystal structure presented in this paper combined with earlier spectroscopic data (28) demonstrates that rod 1B assembles into a so-called A11 tetramer (41,43,44), meaning an anti-parallel alignment of two parallel CCs formed by segments of rod 1. This alignment of rod 1B domains is also supported by cross-linking experiments (40). IF proteins are among the most abundant in eukaryotic cells, and mutations in them have been implicated in more than 85 different human diseases. Therefore, there is a compelling need to characterize normal IF structure so that the pathogenic effects of mutations can be understood. The ability to obtain a crystal structure of a likely physiological oligomer of rod 1B suggests that it may be possible to obtain crystal structures of larger segments of vimentin in physiologically relevant conformations. If so, the combination of x-ray crystallography and SDSL-EPR can be used to provide comprehensive characterization of the atomic structure of an IF protein in an intact filament, which would provide a important foundation for understanding of the effects of mutations on IF structure and function. The data presented herein underscore the value in using multiple independent methods to characterize the structure of this class of proteins that have complex conformational properties. . The heptad phases of rod 1A and rod 1B are indicated by letters a-g under the amino acid sequence. Linker 1 was previously defined as vimentin positions 139-146; a dot below the amino acid sequence is used to indicate glycine 140. Rod 1A was solved as an α-helix (αh) but hypothesized to form a CC in filaments (24). Both rod 1B (this report) and rod 2B (24,25) have been solved as α-helical CCs, as indicated by the CC labels. Ribbon diagrams above the cartoon summarize solved atomic structures of rod 1A (positions 101-139), single alpha helix; linker 1/rod 1B (this report, positions 146-249), coiled coil; rod 2A/linker 2 and rod 2B (positions 263-406), parallel helices (263-~300) adopting CC structure (305-405).

Figure 2. EPR spectra of spin-labeled vimentin mutants.
Numbers indicate the position within vimentin that was mutated to cys and spin labeled. Room temperature spectra were collected from samples after overnight dialysis against 5mM Tris pH 7.5. Normalization was performed as described in Experimental Procedures.  The graph shows refractive index (black) and 90˚ light-scattering (blue) signals from the effluent normalized between the minimum and maximum values observed during the column run, as indicated on the vertical axis on the right. The mass-averaged molecular weight calculated via Debye analysis at each point is shown in red, according to the scale on the vertical axis on the left. A 30 µl aliquot of the concentrated crystallization stock was injected onto a Shodex 802.5 analytic gel filtration column running at 4 ˚C and 0.5 ml/min in 100 mM NaCl, 0.025% (w/v) NaN 3 , 100 mM Tris, pH 7.5. The protein construct analyzed here, which produced the crystal structure analyzed in this paper, contains a 10-residue affinity with sequence MGHHHHHHSH appended at the N-terminus of the indicated segment from human vimentin (Northeast Structural Genomics Consortium Target HR4796B). Superposition of the AB (red/cyan) and CD (green/magenta) dimers based on least-squares alignment of just the A and C protomers demonstrates that both of these dimers adopt equivalent parallel CC conformations. (D) Superposition of the AB dimer with itself based on least-squares alignment of the A protomer (red) with the B protomer (cyan) demonstrates that the two protomers within each CC dimer adopt substantially different conformations, as evidenced by the spatial divergence of the red vs. cyan backbones. Therefore, the parallel CC structure deviates substantially from proper (regular) 2-fold rotational symmetry. (E) Superposition of the ABCD tetramer with itself based on least-squares alignment of the A (red) and C (green) protomers demonstrates that the AB (red/cyan) and CD (green/magenta) dimers are related by proper 2-fold rotational symmetry. (F) Superposition of the ABCD tetramer with itself based on least-squares alignment of the A (red) and B (cyan) protomers demonstrates that opposite surfaces in the asymmetrical AB (red/cyan) and CD (green/magenta) dimers interact with each other across their interface in the ABCD tetramer.    (Table 2) are shown in stick representation, with the -helical regions highlighted by a ribbon superimposed on the backbone. Residues from human vimentin are colored according to subunit of origin as in Fig. 5, while residues from the engineered Nterminal hexahistidine affinity tag are colored gray. The model begins at residues 145 and 148 of human vimentin in subunits A and B, respectively, while it includes six and eight residues from the N-terminal tag in subunits C and D, respectively. (Residues in the tag are labeled with negative numbers counting backwards from the junction point.)   Y150,E151,E153,M154,L157,R158,Q160,V161,L164,T165,  D167,K168,V171,E172,E174,R175,L178,A179, D181,I182,  L185,R186,K188,L189,E192,M193,Q195, R196,E200,L203,  F206,V210,A216,R217,L220,E221, K223,V224,L227,E230,  I231,L234,K235,H238,E241,I242  G147,Y150,E153,L157,R158,Q160,V161,L164,T165,D167,  K168,V171,E172,E174,R175,L178,A179, D181,I182,L185,  R186,K188,L189,E192,M193,Q195, R196,A199,T202,L203 R145,M154,R158,V161,D162,T165,N166,A169,R170,V173,  N177,D181,R184  R207,D211,N212,L215,D219,R222,S226,E230,F233,E241 ____________________________________________________________________________________________ a Water accessible-surface-area (ASA) buried in the indicated pair-wise interface (i.e., including both protomers). b Rotational/translational transformation derived from least-squares alignment of the -helices. c Residues from the N-terminal hexahistidine affinity tag are shown in italics and those from rod 1B in plain text. d All contacts in this non-physiological interface involve residues in the N-terminal hexahistidine affinity tag. ____________________________________________________________________________________________ by guest on September 16, 2017