Stability profile of vimentin rod domain

Abstract

Intermediate filaments (IFs) form an essential part of the metazoan cytoskeleton. Despite a long history of research, a proper understanding of their molecular architecture and assembly process is still lacking. IFs self‐assemble from elongated dimers, which are defined by their central “rod” domain. This domain forms an α‐helical coiled coil consisting of three segments called coil1A, coil1B, and coil2. It has been hypothesized that the structural plasticity of the dimer, including the unraveling of some coiled‐coil regions, is essential for the assembly process. To systematically explore this possibility, we have studied six 50‐residue fragments covering the entire rod domain of human vimentin, a model IF protein. After creating in silico models of these fragments, their evaluation using molecular dynamics was performed. Large differences were seen across the six fragments with respect to their structural variability during a 100 ns simulation. Next, the fragments were prepared recombinantly, whereby their correct dimerization was promoted by adding short N‐ or C‐terminal capping motifs. The capped fragments were subjected to circular dichroism measurements at varying temperatures. The obtained melting temperatures reveal the relative stabilities of individual fragments, which correlate well with in silico results. We show that the least stable regions of vimentin rod are coil1A and the first third of coil2, while the structures of coil1B and the rest of coil2 are significantly more robust. These observations are in line with the data obtained using other experimental approaches, and contribute to a better understanding of the molecular mechanisms driving IF assembly.

1. INTRODUCTION

Intermediate filaments (IFs) are a key component of the cytoskeleton in metazoan animals. Thanks to their unique mechanical characteristics, such as high extensibility and resilience, the IF network boosts the capacity of living cells to withstand mechanical stresses. In humans, about 70 distinct IF proteins are present. Most of these are cell type and tissue‐specific. ¹ The “signature” of all IF proteins is the central α‐helical coiled‐coil (CC) “rod” domain which is well conserved throughout the whole IF family. ² In line with the physiological importance of IFs, scores of mutations in diverse IF proteins have been associated with currently incurable hereditary diseases, including skin, muscle, and neurological disorders. ³ , ⁴

Considerable effort was spent in the past toward elucidating the IF assembly process, which starts with two IF chains associating into an elongated dimer. Models of the mature filaments containing several dozens of chains per cross‐section have been proposed. Historically, parallels with the assembly of the much better understood microtubules and actin filaments have been sought. Specifically, the elementary IF dimers were regarded as rigid rodlets, which engage in a multistep association, eventually yielding the mature filament. For cytoplasmic IF proteins, these filaments are typically 10–12 nm in width. Later on, structural studies of the elementary dimers through X‐ray crystallography and other techniques indicated that the central CC rod of the dimer consists of three α‐helical segments (coil1A, coil1B, and coil2) interconnected by linkers L1 and L12. Such structure is intrinsically prone to two sorts of conformational plasticity, namely dynamics within the CC segments up to the complete unzipping of such segments and high flexibility in the linkers. ¹ , ²

Here, we intended to specifically address the structural dynamics of the CC rod of human vimentin. This protein is one of the most studied members of the family, often regarded as the “model” IF protein. In the past, Meier et al. ⁵ studied CC formation within the first α‐helical segment (coil1A) of vimentin. The authors revealed that, for the wild‐type sequence, the stability of the dimer formed by coil1A was poor, as the CC unraveled at physiological temperature. The melting temperature of the coil1A dimer was drastically increased through the point mutation Y117L, which stabilized the hydrophobic core of the CC. Importantly, in the context of the full‐length protein, this mutation prevented the formation of long filaments. ⁵ These observations suggested that unzipping of the coil1A segment was required during the assembly process. Moreover, it was hypothesized that the unraveling of both ends of the CC rod was necessary for the “head‐to‐tail” interaction of the elementary dimers during longitudinal assembly. ⁶ , ⁷

In order to address the stability of the vimentin rod domain systematically, we have divided it into six dimeric CC fragments of ~50 residues. We started by analyzing each individual fragment in silico using molecular dynamics (MD) simulations, which revealed large differences in structural variability across the fragments. Next, we prepared the fragments recombinantly while promoting correct assembly by adding short N‐ or C‐terminal capping motifs. ⁷ After assessing the correct assembly of the resulting constructs into CC dimers, we studied their thermal stability using circular dichroism (CD) measurements. The obtained melting temperatures correlate well with the in silico predicted dynamics of individual segments. We show that the least stable regions of the vimentin rod are coil1A and the first third of coil2. These observations correlate with the available evidence and contribute to a better understanding of the IF assembly process at the molecular level.

2. RESULTS

2.1. Molecular model of full‐length vimentin dimer

As a starting point, we obtained a three‐dimensional molecular model of the full‐length vimentin dimer using the AlphaFold algorithm. ⁸ The model predicts the structure of the three α‐helical CC segments of the rod domain (residues 93–404) with high confidence (Figure 1). This model is in good agreement with multiple crystal structures available to date as well as previous sequence‐based predictions produced using the CCFold algorithm. ⁹ Here, a major part of the structure reveals a regular left‐handed CC geometry, which is based on the presence of heptad repeats and the formation of a hydrophobic seam by residues in the first (a) and fourth (d) positions of this repeat. ¹⁰ , ¹¹ However, some parts of the dimer, especially the first 35 residues of the coil2 segment (Figure 1) reveal a distinct hendecad (11‐residue) repeat which translates into a parallel α‐helical bundle geometry, that is, a CC with infinite pitch. ² The linker L12 connecting the coil1B and coil2 segments were modeled as containing little secondary structure. AlphaFold modeling with default parameters yielded a structure with coil2 being “folded back” on coil1B, corresponding to a ~180° turn in the dimer axis at the linker (Figure 1). However, additional MD simulations (data not shown) pointed to high flexibility in the linker and a lack of specific interactions between coil1B and coil2. We suggest that the obtained “backfolded” structure is just one possibility of a range of conformations possible in the solution.

FIGURE 1.

AlphaFold model of vimentin dimer shown as a cartoon. The colors reflect the predicted local distance different test scores, ⁸ with blue being the highest and red the lowest. Violet brackets indicate the presence of hendecad repeats. Red brackets show the six 50‐residue fragments studied.

Interestingly, the application of the AlphaFold algorithm also provides a snapshot of the head and tail domains. Both are predicted to contain little secondary structure except for a β‐hairpin motif found within the tail. In our model, the head domain is folded back onto the rod domain (Figure 1). This possibility has been proposed before, ¹² although the conformation of the head domains should change as they get involved in interdimer contacts upon IF assembly. ¹³ In the current dimeric model, the structure of both terminal domains and the linkers L1 and L12 has low confidence (Figure 1), which reflects the expected high flexibility of these regions.

2.2. In silico analysis of vimentin fragment dynamics

In the AlphaFold model discussed above, the lengths of the three CC segments of the rod are 55 residues for coil1A, 107 residues for coil1B, and 147 residues for coil2, respectively. Beyond small variations with respect to the exact start and end residues of each segment, the lengths of the three segments are consistent with earlier predictions and crystallographic data. ⁹ This brought us to the idea of comparing the properties of a vimentin fragment corresponding to coil1A (denoted 1A), two fragments corresponding to the N‐ and C‐terminal halves of coil1B (denoted 1B₁ and 1B₂), and three consecutive fragments of coil2 (2₁, 2₂, and 2₃) (Figure 1 and Table 1), all of which have approximately the same size of ~50 residues.

TABLE 1.

Predicted strength of the dimeric interface in fragments

Fragment	Residues	Residue count	Interface
			Area (Å2)	ΔG i (kcal/mol)	N HB	N SB
1A	93–138	46	1,135	−22	4	5
1B1	150–199	50	1,483	−25	2	13
1B2	199–249	51	1,184	−24	1	2
21	265–309	48	941	−18	5	1
22	309–358	50	1,299	−29	6	3
23	358–404	47	1,348	−32	2	3

Note: Average values over the course of 100 ns MD simulation are provided. ΔG ⁱ is the solvation free‐energy gain of the dimeric interface formation. N _HB is the average number of hydrogen bonds across the interface. N _SB is the average number of salt bridges across the interface.

Correspondingly, we took the AlphaFold model to extract the six CC fragments and assessed the conformational dynamics of each fragment using MD simulations. Initially, we found that some dimers rapidly lose the well‐ordered structure, which is evident from the fact that their α‐helical content drops significantly from the initial value of ~100% in the course of a short 10 ns simulation. After 100 ns, the α‐helicity values largely stopped decreasing for all fragments except for coil1A (Figure S1). The final statistics on several structural parameters of each dimeric fragment was collected after performing three independent 100 ns runs, which showed good repeatability (Figure 2b). In particular, fragment 1A was partially losing α‐helicity in a 12‐residue stretch near its N‐terminus, while fragment 2₁ was losing α‐helicity all along its length during the simulation. All other fragments largely retained α‐helicity except for some melting near the termini. In fact, both the average loss of α‐helicity during the simulation and the root mean squared deviation from the initial structure followed a similar trend across the six fragments, being the most pronounced for fragments 1A and 2₁ (Figures 2c, d). We have also analyzed the average CC radius per fragment during the simulation (Figure 2e). For well‐ordered CC dimers, the CC radius is close to 5 Å, ¹⁴ which reflects a tight packing of the hydrophobic seam residues. We found that the CC radii of two fragments, 1B₁ and 2₂, remained close to this value throughout the simulation, while the CC radii of other fragments increased to various extents. A particular outlier is fragment 2₁ for which the average CC radius increases beyond 6 Å, meaning that the two chains start to lose physical contact (compare Figure S2). This observation correlates well with the fact that a larger part of this fragment includes 11‐residue repeats. This part was previously demonstrated to be less stable than heptad‐based CCs. ¹⁵

FIGURE 2.

Dynamic properties of the vimentin fragments derived through MD simulations. (a). Schematic diagram of vimentin rod showing the six fragments. (b). Plots of α‐helicity per residue in the course of simulation. Three 100 ns simulations for each of six fragments were analyzed. Dark blue means both strands were α‐helical at that residue, light blue means only one of them was, and white indicates a total loss of helicity. (c). A‐helicity of each fragment at the start of the simulation is shown by an orange line. Average α‐helicity over the course of a 100 ns simulation performed in triplicate is shown by a red line. Boxes in the boxplot extend from the 25^th to 75^th percentile. (d). Per‐fragment average root mean square deviation (rmsd) from the initial coordinates over the course of simulation. (e). Per‐fragment average CC radius. (f). Solvation‐free energy gain for the dimer interface (ΔG ⁱ ), as estimated using the program PISA. ¹⁶

In addition, the strength of the interface between the two chains of the CC was characterized using the PISA algorithm. ¹⁶ Here again, rather than looking at the initial models only, we have calculated statistics across the course of a 100 ns MD run performed in triplicate, for several parameters of the interface (Table 1 and Figure 2f). Since formation of the hydrophobic seam is generally considered as the main driving force of CC stabilization, we were particularly interested in the solvation‐free energy gain upon interface formation (ΔG ⁱ ). In addition, there are some hydrogen bonds and salt bridges across the interface in each fragment, although the contribution of such interactions to CC stability is less clear. ¹⁰ Based on the ΔG ⁱ value, the most stable dimer is the C‐terminal region of coil2 (fragment 2₃), while fragments 1A and 2₁ are the least stable.

2.3. Design of vimentin constructs for experimental studies

The IF rod domain is a parallel, in‐register CC dimer. Within the context of the full rod, each individual CC segment is restrained to produce this type of oligomer. However, isolated shorter fragments may not necessarily assemble in the same correct way. Indeed, the formation of CC trimers and tetramers in place of dimers, antiparallel configurations, or even monomeric states was observed in the past for short CC fragments (see Section 3). To circumvent this difficulty, we decided to bootstrap the correct assembly of vimentin fragments by fusing them to specialized N‐ or C‐terminal capping motifs. To this end, we employed the Gp7 and Eb1 capping motifs, respectively. ¹⁷ , ¹⁸ Each of these includes a short CC region and a further short α‐helix that folds back on it (Figure 3a). These motifs were previously used to stabilize dimeric myosin fragments ¹⁹ and more recently lamin fragments ⁷ toward X‐ray studies. Here, we have created constructs for each vimentin fragment supplemented with either the N‐ or C‐terminal cap. The constructs were designed in such a way that the heptad repeat pattern would be preserved across the fusion site (Figure 3a).

FIGURE 3.

Solution properties of fusion constructs. (a). Atomic models of the fusions with N‐terminal (red) and C‐terminal (black) caps. (b). SEC elution profiles (UV absorbance) of constructs with the N‐ and C‐terminal caps are shown as solid and dashed lines, respectively. Elution profile for the uncapped 1B₁ fragment is shown as a dotted line. Additionally, the MALS‐based molecular mass profile across each elution peak is plotted (thin lines).

2.4. Oligomerization of the capped constructs

All twelve capped vimentin fragments as well as an uncapped 1B₁ fragment and each cap alone were produced recombinantly and isolated to single‐band purity on SDS‐PAGE. Thereafter, the proper folding of the constructs was assessed by size exclusion chromatography coupled to multi‐angle light scattering (SEC‐MALS). A majority of samples eluted as a single symmetric peak with a mass pointing to a dimer (Figure 3b and Table 2). Since the N‐terminal cap contains a larger number of residues than the C‐terminal one, the N‐terminally capped fragments are noticeably longer along the dimer axis (Figure 3a). In good agreement with that, the N‐terminal fusions eluted earlier than the C‐terminal fusions (Figure 3b). The N‐1B₁ construct yielded a symmetric elution peak at the expected position, although its MALS‐based mass estimate (17 kDa) was somewhat below the theoretical dimer (23 kDa).

TABLE 2.

Protein constructs analyzed by SEC‐MALS

Construct	Theoretical monomeric mass (kDa)	Experimental mass (kDa)	Oligomeric state a
N‐1A	11.3	22	Dimer
1A‐C	9.5	26	Trimer
1B1 b	6.2	15	Dimer
N‐1B1	11.5	17 c	Dimer?
1B1‐C	10.2	22	Dimer
N‐1B2	11.3	22 + 35	Dimer + trimer
1B2‐C	10.3	18	Dimer
N‐21	10.4	20	Dimer
21‐C	9.5	18	Dimer
N‐22	11.6	23	Dimer
22‐C	9.8	20 + higher	Polydisperse
N‐23	11.0	21	Dimer
23‐C	10.1	19	Dimer
N‐cap	5.3	5	Monomer
C‐cap	4.4	5 + 19	Monomer + tetramer

^a Homogenous dimers are in bold.

^b Not capped.

^c A value in‐between of a monomer and a dimer.

At the same time, several constructs displayed SEC profiles that indicated polydispersity and/or revealed MALS‐based oligomeric masses beyond the expected dimers (even though the SDS‐PAGE analysis of all elution peaks confirmed correct monomeric masses throughout). First, a trimer (26 kDa) was observed for constructs 1A‐C. Second, N‐1B₂ revealed an SEC profile with two peaks corresponding to trimers and dimers, respectively. Re‐injection of the trimeric fraction into the SEC column resulted in a single trimeric peak, while the dimeric fraction again yielded a two‐peak chromatogram. Finally, the construct 2₂‐C showed a broad profile that included a main peak corresponding to a dimer and a prominent left shoulder suggestive of a higher‐mass species.

In addition, SEC‐MALS studies of isolated N‐ and C‐terminal caps pointed to monomeric species, with a smaller fraction of tetramers in the latter case (Figure S3). This suggested that both caps were not capable of forming the correct CC dimers on their own.

2.5. CD spectra

Only the constructs for which the correct dimeric state in solution could be confirmed, have been investigated further to determine their thermal stability using CD spectroscopy. These constructs included the N‐capped version of fragment 1A, both N‐ and C‐capped as well as the uncapped fragment 1B₁, the C‐capped fragment 1B₂, both N‐ and C‐capped fragments 2₁, the N‐capped fragment 2₂, and both N‐ and C‐capped fragments 2₃ (Table 2). Thus, each of the six originally chosen 50‐residue fragments of the vimentin rod domain was represented by at least one capped construct.

At 20°C, all constructs revealed CD spectra typical for α‐helical proteins, with distinct minima at 210 and 222 nm (Figure 4, left panels). Of note, all three constructs including 1B₁, the only studied construct of 2₂ and both studied constructs of 2₃ had a θ₂₂₂ value close to −30 × 10³ deg × cm² × dmol⁻¹, which was compatible with a well‐ordered α‐helical CC structure. The constructs containing fragment 2₁ appeared to be the least ordered of all, with θ₂₂₂ values of −22 × 10³ and −20 × 10³ deg × cm² × dmol⁻¹ for N‐2₁ and 2₁‐C, respectively.

FIGURE 4.

CD analysis of vimentin fragments. For each fragment, the left panel shows the CD spectrum at 20°C before the melting experiment. Thin vertical lines indicate a wavelength of 222 nm. Data for the constructs with N‐and C‐terminal caps are plotted with solid and dashed lines respectively; the data for the uncapped 1B₁ fragment are shown as a dotted line. The right panel shows the melting curves recorded at 222 nm. Vertical lines indicate 20°C. Thicker vertical lines indicate the calculated T _m values.

2.6. Thermal stability

Next, CD melting curves were recorded at 222 nm with a temperature range of 1–95°C (Figure 4, right panels). In addition, re‐folding upon cooling down was also studied (Figure S5). Most constructs showed a distinct single sigmoidal transition upon heating (see Figure S4 for theoretical fitting) suggestive of a cooperative CC unfolding. The corresponding calculated T _m values are provided in Table 3. Here, the most stable construct was 1B₂‐C (T _m = 79.7°C), while the least stable was N‐1A (T _m = 25.7°C). At higher temperatures, most curves reached a plateau at about −4 × 10³ deg × cm² × dmol⁻¹. Similar values were reported in the past for a “premolten globule‐like” subclass of intrinsically disordered proteins. ²⁰

TABLE 3.

Melting temperatures

Fragment	T m of N‐capped construct (°C)	T m of C‐capped construct (°C)
1A	25.7 ± 0.1	‐
1B1 a	47.07 ± 0.04	57.06 ± 0.05
1B2	‐	79.7 ± 0.2
21	43.7 ± 0.06	17.6 ± 0.4 45.5 ± 0.6
22	46.63 ± 0.03	‐
23	51.1 ± 0.5	74.8 ± 1.0

^a T _m for uncapped 1B₁ fragment was 31.06 ± 0.06°C.

Of the three 1B₁ constructs studied, the least stable was the uncapped one (T _m = 31.1°C), followed by the N‐capped (T _m = 47.1°C) and the C‐capped (T _m = 57.1°C). This series exemplifies the extent of additional stabilization delivered by either capping motif. Here, we also wanted to explore the capacity of in silico simulations to quantify such stabilization. To this end, we have additionally run MD for capped fragments N‐1B₁ and 1B₁‐C. For C‐terminal capping, these computations indeed indicate a clear stabilization, as judged by both increases in the average α‐helicity and a decrease of the CC radius (Figure S6). This correlates with a 26°C increase in experimental thermal stability. For N‐terminal capping, no univocal indication of improved stability was obtained in silico (Figure S6), despite a significant 16°C increase in T _m observed experimentally.

While our CD melting experiments yielded reliable T _m values in most cases, three of the total ten melting curves deviated substantially from a single sigmoidal transition. 2₁‐C could be interpreted as having two distinct transitions at 17.6 and 45.5°C, respectively with low cooperativity. In comparison, only one transition at 43.7°C was observed for the corresponding N‐terminally capped construct. Both constructs containing fragment 2₁ showed a rather low α‐helicity even at 20°C (Figure 4). Next, both constructs containing fragment 2₃ displayed an aberrant behavior: while the CD melting profile followed the usual sigmoidal shape at a lower temperature, the signal plateaued abruptly at a certain point (Figure 4). Here, the T _m had to be estimated by fitting to the lower part of the curve only; the T _m values for both N‐2₃ and 2₃‐C are thus less accurate. Interestingly, while 2₃‐C denatured completely at 95°C upon heating, subsequent cooling down of the sample showed no change in ellipticity indicating that the unfolding was irreversible (Figure S5). In stark contrast, N‐2₃ seemed to maintain some ellipticity at temperatures as high as 95°C and could be refolded when the temperature decreased. While we have no convincing explanation for such behavior, it is possible that the N‐2₃ fragment retains some secondary structure at high temperatures.

Finally, we have compared the experimental T _m values with in silico calculations. According to CD melting, the least stable fragments were 1A and 2₁. Computationally, the same two fragments were estimated to have the lowest free energy solvation gains of all fragments (Figure 5).

FIGURE 5.

Overview of melting temperatures obtained using CD measurements. Data for the N‐capped fragments are shown as filled circles, data for C‐capped fragments as open circles. T _m for the second transition observed for 2₁‐C is shown as an open triangle. T _m of the uncapped 1B₁ fragment is shown as a square. All T _m were estimated using the Boltzmann fit. A plot of ΔG ⁱ estimates for unfused fragments (Table 2) is overlaid as a gray dashed line.

Moreover, the constructs 1B₂‐C and both constructs containing the 2₃ fragments were the most stable in CD measurements. The same trend is preserved in the ΔG ⁱ estimates, although here the most stable fragment by theory is 2₃, even though the highest experimental T _m was measured for the 1B₂‐C construct.

3. DISCUSSION

As outlined in the Introduction, the CC rod domain is a well‐conserved signature element of all IF proteins. Furthermore, the dynamics of the IF dimer, including the unzipping of some CC segments, is expected to play an important role in the filament assembly process, and thus of considerable interest. ⁵ , ⁶ , ²¹ However, obtaining experimental data on the dynamics of the complete dimer is a challenging task. Performing an MD simulation is a possible alternative, although it would require substantial computational resources due to the large length of the dimer (~50 nm). Here, we have addressed the dynamics and stability of six isolated 50‐residue fragments, as a proxy of the corresponding regions within the complete vimentin dimer. In silico evaluation of these fragments could be done directly through MD. At the same time, we have experimentally studied the thermal stability of recombinant, terminally capped fragments, as a measure of their capacity to unravel.

3.1. Use of capping motifs to stabilize CC fragments

To stabilize the relatively short vimentin fragments, we have employed small N‐terminal capping motif Gp7 and C‐terminal capping motif Eb1 as previously used by Taylor et al. ¹⁷ , ¹⁸ , ¹⁹ toward X‐ray studies of myosin fragments. Once such fragments were fused to either of the capping motifs, the resulting chimeras could be crystallized. More recently, we could also employ these two capping motifs for the crystallization of lamin fragments. ⁷ These reports suggested that both the Gp7 and Eb1 motifs promote correct parallel CC dimers, which prompted their use in the current study. However, our new observations reveal that the efficiency of these capping motifs in solution is less than perfect. After each of the six 50‐residue fragments of the vimentin rod was fused to either the N‐ or C‐terminal cap, only nine of the total 12 fragments revealed homogenous solutions of dimers, judging by SEC‐MALS measurements.

In particular, an uncapped coil1A of vimentin was previously shown to form a rather weak CC dimer. ⁵ In the current study, we observed that the N‐terminally capped 1A fragment formed the correct dimers, but the C‐terminally capped one was trimeric. This observation points to a limited potency of the Gp7 and Eb1 capping motifs toward the correct oligomer assembly, on top of the potency of the vimentin fragment itself. Indeed, our SEC‐MALS data on unfused N‐ and C‐terminal capping motifs indicate that they are both predominantly monomeric (Figure S3).

Moreover, the N‐1B₂ construct (containing vimentin residues 199–249) yielded two distinct SEC peaks, which corresponded to trimers and dimers, respectively. In the past, crystallization of the entire coil1B of vimentin (residues 144–251) ²² as well as of a somewhat shorter fragment (153–238) ²³ yielded the correct parallel CC dimers, even in the absence of any capping motifs. However, crystals of vimentin fragments 161–238 and 161–243 revealed CC trimers instead. ²³ Given these observations, we conclude that while the used capping motifs do help to promote the correct CC formation, in future studies they should ideally be substituted by more potent dimer‐forming motifs. In particular, the application of the Xrcc4 domain has been described. ¹⁹ Another option is to use a capping motif with an engineered disulfide bridge across the CC interface, such as the modified Gp7 motif described in Reference 7. If the corresponding chimeras reveal a disulfide bridge (which could readily be checked on a non‐reducing SDS‐PAGE), this warrants the correct dimeric in‐register structure.

3.2. Varied CC stability across the rod domain

In the past, the thermal stability of only one rod segment, coil1A of human vimentin (residues 103–138), was characterized in detail using CD spectroscopy. ⁵ A low melting temperature of 32°C was obtained for 1 mg/ml solution. This observation correlated with the previously published crystal structure of this fragment, ²⁴ which surprisingly revealed a crystal packing that lacked any CC dimers. A related observation concerns the N‐terminal part of vimentin coil2. Here, analytical ultracentrifugation studies of two related fragments indicated a monomeric conformation. ²⁵ Furthermore, a construct comprising residues 261–335 behaved as a monomer in 1 mg/mL solution but formed aberrant tetrameric assemblies in the crystals rather than dimers. ²⁶

Importantly, our current work provides a first systematic evaluation of the CC stability profile along the IF rod domain, We show that both the coil1A segment and the N‐terminal portion of coil2 are only marginally stable under physiological conditions, in line with previous observations, while other regions of the rod have distinctly higher stability. Here, our MD‐based in silico analyses of isolated fragments and thermal melting experiments yield a largely consistent picture. The only noticeable discrepancy between the in silico and experimental data are between the two fragments comprising coil1B: while the predicted stability for 1B₂ was somewhat less than for 1B₁, the fusion construct 1B₂‐C had a melting temperature close to 80°C, the highest of all constructs measured (Table 3). Naturally, due to cooperativity, the melting of various regions of the complete rod domain is expected to occur at higher temperatures than for isolated 50‐residue fragments. While the addition of the capping motifs should diminish this effect, it is clear that the obtained T _m values for fragments (Table 3) do not reflect the exact melting temperatures in the context of the full‐length dimer, but rather indicate the relative stability of these regions.

3.3. Structural plasticity of coil1A and the N‐terminal part of coil2 is needed for IF assembly process

The capacity of the elementary IF dimer to change its three‐dimensional conformation has previously been hypothesized to be important for filament assembly. Indeed, due to the elongated and flexible nature of the dimer, major structural adjustments are to be expected as dimers associate into tetramers, assemble into the ULFs and ultimately form elongated filaments. In particular, it was suggested that longitudinal assembly, which involves the interaction of the opposite ends of the rod, could proceed via CC unzipping at either side. This suggestion is linked to the pronounced effect of certain point mutations within the full‐length vimentin on filament assembly. In particular, the Y117L mutation results in arresting the assembly at the ULF stage. ⁵ At the same time, this mutation has a drastic effect on an isolated coil1B dimer, increasing its melting temperature from 32 to 76°C. ⁵ More recently, it was shown that the Y400L mutation located at the opposite end of the rod domain has the same apparent effect on vimentin assembly. ²¹ Just like Y117L, the Y400L mutation replaces a tyrosine in a core position of the CC with a leucine residue, which is expected to have a local stabilizing effect on the CC.

Furthermore, our new data convincingly indicate that the N‐terminal part of coil2 is another weak region of the dimer. Here, the 2₁‐C construct revealed a complex melting curve, with the first apparent transition occurring already at 18°C and the second at 46°C, while the N‐2₁ construct was seen to melt at 44°C (Figure 4). In agreement with that, fragment 2₁ was shown to lose a significant part of its α‐helical structure in the course of the MD simulation (Figure 2b). This fragment also had the lowest free energy of the interface (ΔG ⁱ , Table 1) of all fragments. As discussed before, ¹⁵ the N‐terminal part of coil2 features a ~30‐residue long parallel α‐helical bundle based on hendecad repeats. The lability of this part is likely to be caused by the presence of suboptimal residues A280, N283, and A287 in core positions of the hendecad repeat, as well as by the aromatic cluster composed of three bulky residues Y276, W290, and Y291. Past residue 302, coil2 contains a regular heptad periodicity save a single stutter at position 350. The higher thermal stability of capped constructs carrying fragments 2₂ and 2₃ apparently results from optimal hydrophobic core interactions in that region.

Interestingly, hydrogen‐deuterium exchange experiments on vimentin in various assembly states by Premchadar et al. ²¹ have indicated that the N‐terminal portion of coil2 becomes significantly more stabilized in assembled filaments compared to soluble tetramers. The only two other regions where a similar stabilization was observed are the opposite ends of the rod domains. These authors propose the interesting hypothesis that vimentin assembly involves a radical reorganization of both coil1A and the N‐terminal part of coil2, whereby these regions switch from isolated dimers to a mixed heterotetrameric CC assembly.

4. MATERIALS AND METHODS

4.1. Structural modeling

The AlphaFold algorithm version 2.1.0 (weights from October 27, 2021) was run locally on a workstation equipped with a i9‐11900K processor with 128G RAM and a GeForce GTX 1080Ti GPU under Ubuntu 21.10. ColabFold with Mmseqs2 for generating multiple sequence alignments and a recycle value of 24 were used. The complete amino‐acid sequence of human vimentin (Uniprot entry P08670) was used as input. This sequence was duplicated on input in order to produce a dimeric model.

4.2. MD simulation

MD simulations for uncapped ~50 residue vimentin fragments were done using GROMACS version 2021.4 ²⁷ with a 2 fs time step, the AMBER99SB force field, and the TIP4P water model. Prior to an MD run, each fragment was solvated in an 8 × 8 × 8 nm³ box at neutral pH and with ionic strength corresponding to 150 mM NaCl. An energy minimization step was performed to yield a maximum gradient of 1,000 kJ.mol⁻¹ nm⁻¹ followed by a 100 ps temperature equilibration step using the canonical NVT ensemble at 293 K. Next, a 100 ps pressure equilibration step was conducted with the NPT ensemble. Finally, 10–100 ns simulations were run using a Verlet cut‐off of 1 nm for electrostatic and Van Der Waals forces, particle mesh Ewald for longer range interactions, the LINear Constraint Solver for bond constraints, the Berendsen thermostat for temperature coupling and the Parrinello–Rahman pressure coupler (see Reference 27 for parameter definitions).

After an MD simulation for each fragment (performed in triplicate), an analysis of trajectories was done using the Python library MDAnalysis. A frame was captured every 10 ps of the simulation (10,000 frames for a 100 ns simulation). The number of parameters was determined for each frame, and overall statistics were calculated. In particular, α‐helicity was calculated using a python version of GROMACS's helicity calculation. Root mean square deviation from the starting coordinates and inter‐residue distances were measured using the built‐in functions of MDAnalysis. The CC radius was determined using a Python version of Twister. ¹⁴ The free energy of solvation of the dimeric interface (∆G ⁱ ) was calculated using PISA. ¹⁶

4.3. Recombinant protein expression and purification

Expression constructs were obtained through the sequence‐ and ligation‐independent cloning. Vimentin fragments were inserted in a linearized pETSUK2 ²⁸ vector which already carried either the Gp7 or Eb1 cap. ¹⁷ , ¹⁸ The C‐terminal residue of Gp7 and the N‐terminal residue of Eb1 were chosen such that the phase of the CC heptad pattern between the vimentin fragment and the cap would be preserved. To this end, for vimentin fragments that were starting at heptad position a or d, Gp7_1‐49 or Gp7_1‐45 were selected as N‐terminal caps, respectively. For fragments ending with heptad positions a or d, Eb1_27‐63 or Eb1_30‐63 were used as C‐terminal caps, respectively.

Expression and purification of the protein fragments were performed as previously described. ⁷ In brief, cell pellets were resuspended and sonicated in 40 mM Tris‐HCl, 250 mM NaCl, 12.5 mM imidazole, 10 mM MgCl₂, 1% (v/v) Triton X, 100 U Cryonase Inhibitor cocktail (Sigma) and 0.350 μM βME. The supernatant fraction of the lysate was run through an immobilized metal ion affinity chromatography (IMAC) column in a low‐imidazole buffer (40 mM Tris‐HCl, 250 mM NaCl, 12.5 mM imidazole, pH 7.5). The target protein was eluted using a high‐imidazole buffer (40 mM Tris‐HCl, 250 mM NaCl, 500 mM imidazole, pH 7.5). The cleavage of the 6xHis SUMO tag was performed with SUMO protease during sample dialysis in a low‐imidazole buffer. Of note, SUMO protease cleavage was more efficient for Gp7‐capped constructs, which start with an N‐terminal glycine of the capping motif. A subtractive IMAC step was performed to separate the cleaved protein from the affinity tag. Combined protein fractions were concentrated up to 10 mg/mL using ultrafiltration devices (Amicon Ultra, 0.5 ml, 3 K, Merck). For CD measurements, the samples were additionally purified by SEC. Hundred microliter of 10 mg/ml sample was injected into the Superdex Increase 200 10/300 column (GE Healthcare) column and eluted with the SEC buffer (10 mM Tris‐HCl, 100 mM NaCl, pH 7.5) at 0.5 ml/min on an Äkta purifier (GE Healthcare). All purification steps were assessed by SDS‐PAGE.

4.4. Size exclusion chromatography coupled to multi‐angle light scattering

Hundred microliter sIMAC sample (1 mg/ml) was analyzed in the SEC buffer using the liquid chromatography system Shimadzu LC‐20 AD equipped with MALS, UV, and dRI detectors (Heleos, Wyatt Technology). Superdex Increase 200 10/300 column was used. Prior to sample injections, one blank injection and calibration with 1 mg/ml bovine serum albumin were executed at a flow rate of 0.5 ml/min. Data were evaluated in the ASTRA® 5 software (Wyatt Technology, Dernbach, Germany).

4.5. Circular dichroism

SEC‐purified protein samples were diluted to 0.2 mg/ml in the SEC buffer. Quartz cuvettes with an optical path of 2 mm (Hellma Benelux BVBA) and a JASCO J‐1500 CD Spectrometer were used. Four hundred fifty microliter of diluted sample was required per experiment. A baseline correction for the SEC buffer was performed for every measurement. Before and after the melting experiments, CD spectra were recorded at both 20 and 4°C between 180 and 300 nm wavelength. For melting experiments, the data interval was set to 0.1°C with a constant wavelength of 222 nm. The samples were heated from 1 to 95°C and cooled back to 1°C.

The mean residue ellipticity ([ϴ]_MRW,λ) was calculated as ${\left[\theta \right]}_{\mathrm{MRW},\lambda }=\mathrm{MRW}\times \frac{{\theta }_{\lambda }}{10\times d\times c}$, where ϴ _λ is the observed ellipticity at wavelength λ, d is the optical path in cm, c is concentration in g/ml, and MRW is the mean residual weight per peptide bond. MRW equals $\frac{M}{N-1}$, where M is the molecular weight of the protein in Da, and N is the number of residues.

T _m was estimated by fitting to Boltzmann's sigmoidal equation for a single‐stage transition (1) using a Python script as described in References 29, 30, 31:

$$\theta \left(T\right)={\alpha }_N+{\beta }_{N}T+\left({\alpha }_D+{\beta }_{D}T\right)\left(\frac{\exp \left(\frac{-∆H_m(1-\frac{T}{T_m}}{\mathit{RT}}\right)}{1+\exp \left(\frac{-∆H_m(1-\frac{T}{T_m}}{\mathit{RT}}\right)}\right),$$ (1)

where $\theta \left(T\right)$ is the measured ellipticity at 222 nm; ${\alpha }_N$ and ${\alpha }_D$ are the ellipticity for native and denatured protein, respectively; ${\beta }_N$ and ${\beta }_D$ are the slopes of pre‐ and post‐unfolding regimes, respectively; $T$ is the temperature in K, R is the gas constant (8.31432 J/[mol K]); $∆H_m$ is the change of enthalpy upon protein denaturation and $T_m$ is the temperature when 50% of the protein is in denatured state.

To evaluate the curves where two consecutive transitions were suspected, the Equation (1) was modified to accommodate two melting temperatures and the corresponding $∆H_m$ values:

$$\theta \left(T\right)=\begin{array}{l}{\alpha }_N+{\beta }_{N}T+\left({\alpha }_D+{\beta }_{D}T\right)(n_1\bullet \frac{\exp \left(\frac{-∆H_{m1}\left(1-\frac{T}{T_{m1}}\right)}{\mathit{RT}}\right)}{1+\exp \left(\frac{-∆H_{m1}\left(1-\frac{T}{T_{m1}}\right)}{\mathit{RT}}\right)}+\left(1-n_1\right)\bullet \frac{\exp \left(\frac{-∆H_{m2}\left(1-\frac{T}{T_{m2}}\right)}{\mathit{RT}}\right)}{1+\exp \left(\frac{-∆H_{m2}\left(1-\frac{T}{T_{m2}}\right)}{\mathit{RT}}\right)}).\end{array}$$ (2)

AUTHOR CONTRIBUTIONS

Anastasia V. Lilina: Conceptualization (equal); data curation (equal); formal analysis (equal); investigation (equal); supervision (equal); visualization (equal); writing – original draft (equal); writing – review and editing (equal). Simon Leekens: Data curation (equal); formal analysis (equal); investigation (equal); visualization (equal); writing – original draft (equal); writing – review and editing (equal). Hani M. Hashim: Formal analysis (supporting); software (lead); writing – review and editing (supporting). Pieter‐Jan Vermeire: Data curation (supporting); investigation (supporting); writing – review and editing (supporting). Jeremy N. Harvey: Data curation (supporting); writing – review and editing (supporting). Sergei V. Strelkov: Funding acquisition (lead); conceptualization (equal); supervision (equal); writing – review and editing (equal).

FUNDING INFORMATION

This research was supported by the KU Leuven CELSA Alliance (grant 18/044).

CONFLICT OF INTEREST

The authors have no conflict of interest to disclose.

Supporting information

Figure S1. Percentage a‐helicity of each fragment after MD simulations of different time spans. The helicity is measured using the last 1 ns (1000 frames) of each time interval.

Figure S2. Molecular dynamics simulations for vimentin fragments. The left column shows a ribbon diagram of the fragment at the start of simulation, after an energy minimization step. The right column shows the last frame of the 100 ns simulation.

Figure S3. SEC‐MALS data for isolated N‐ and C‐capping motifs. Elution profiles are shown together with the MALS‐based molecular mass profile across each peak.

Figure S4. Circular dichroism melting curves (blue lines) and corresponding sigmoidal fits used (orange lines) for all constructs.

Figure S5. Re‐folding of vimentin fragments. Left panels: circular dichroism (CD) spectra of each fragment before (colored) and after melting (gray) at 20°C. Data for constructs with N‐ and C‐terminal caps are plotted with solid and dashed lines, respectively; the data for the uncapped 1B1 fragment are shown as a dotted line. Right panels: temperature dependence of the CD signal at 222 nm corresponding to heating/melting (colored lines) and cooling down/re‐folding (gray lines).

Figure S6. Molecular dynamics runs with vimentin fragments N‐1B1, uncapped 1B1 and 1B1‐C. (a) Plots of α‐helicity per residue in the course of simulation. Three 100 ns simulations for each of six fragments were analyzed. Dark blue means both strands were α‐helical at that residue, light blue means only one of them was, and white indicates a total loss of helicity. (b). Average α‐helicity over the course of a 100 ns simulation performed in triplicate is shown by a red line. Boxes in the boxplot extend from the 25th to 75th percentile. (c) Per‐fragment average CC radius.

Lilina AV, Leekens S, Hashim HM, Vermeire P‐J, Harvey JN, Strelkov SV. Stability profile of vimentin rod domain. Protein Science. 2022;31(12):e4505. 10.1002/pro.4505

DATA AVAILABILITY STATEMENT

Any and all data reported in this study are available upon request. Please contact the corresponding author.