Molecular Parameters of Type IV α-Internexin and Type IV-Type III α-Internexin-Vimentin Copolymer Intermediate Filaments*

During neuronal development, a dynamic replacement mechanism occurs in which the type VI nestin and type III vimentin intermediate filament proteins are replaced by a series of type IV proteins beginning with α-internexin. We have explored molecular details of how the type III to type IV replacement process may occur. First, we have demonstrated by cross-linking experiments that bacterially expressed forms of α-internexin and vimentin form heterodimer molecules in vitro that assemble into copolymer intermediate filaments. We show using a urea disassembly assay that α-internexin molecules are likely to be more stable than those of vimentin. Second, by analyses of the induced cross-links, we have determined the axial lengths of α-internexin homodimer and α-internexin-vimentin heterodimer molecules and their modes of alignments in filaments. We report that these dimensions are the same as those reported earlier for vimentin homopolymer molecules and, by implication, are also the same for the other neuronal type IV proteins. These data suggest that during neuronal development, α-internexin molecules are readily assimilated onto the pre-existing vimentin cytoskeletal intermediate filament network because the axial lengths and axial alignments of their molecules are the same. Furthermore, the dynamic replacement process may be driven by a positive equilibrium due to the increased stability of the α-internexin network.

Intermediate filaments (IF) 1 are highly dynamic components of the cytoskeletons of most eukaryotic cells (1)(2)(3). Although a great deal is known about their protein complexity, composition, and expression characteristics during development and differentiation, comparatively less is known about IF structure and how this relates to their dynamism and functions in cells. Established data report that IF are built in an hierarchical manner. First, two compatible protein chains form a dimer molecule in which four well conserved ␣-helical sequences along a central rod domain of each chain associate in parallel and in register to make a segmented coiled-coil molecule (4 -7). Second, a pair of molecules align in an antiparallel fashion (8,9) in one of several basic modes (10 -12) as follows: A 11 , halfstaggered with the 1B segments largely overlapped; A 22 , halfstaggered with the 2B segments largely overlapped; and A 12 , in which the entire molecules are largely overlapped. Determination of the exact alignments of the A 11 and A 22 modes has revealed a fourth mode termed A CN , in which the end of one molecule is overlapped by about 1 nm with the beginning of the next parallel molecule in the same axial row. Third, quantitative mass measurements suggest that 12-24 molecules, most commonly 16 in native IF (13)(14)(15), associate in three dimensions in as yet unknown ways to form an IF.
However, whereas most or all cytoplasmic IF appear morphologically similar in vivo or when reconstituted in vitro, there are complex variations in at least the first two of these hierarchical steps. In step one, the coiled-coil molecule may be either an obligatory heterodimer (all type I/II keratin chains) (6, 7), a homodimer (types III and some IV chains) (16 -20), or a facultative heterodimer within a chain type, in which for example two different chains may participate (e.g. type III; Refs. 16 and 21-24) or be required (e.g. type IV; ref. 25,26). In the second step, we have found that the exact axial alignments of the three nearest-neighbor assembly modes identified above vary between the different IF types. All cytokeratin molecules seem to be aligned the same way (10, 11) but differently from vimentin molecules (12). The consequence of this is that these chain types cannot and do not copolymerize in vitro, or in vivo, so that certain cells that co-express keratins and vimentin, for example, have elaborated different cytoplasmic IF networks (27). A related complexity arises in cells and tissues during development. Many embryonic or immature differentiating cell types express first one type of IF chain and then replace it in a dynamic exchange/replacement process (28) as differentiation proceeds. A cogent example of this are neuronal cells (reviewed in Ref. 29). Neuroectodermal stem cells initially express the type VI IF chain nestin (30) alone or with type III vimentin (central nervous system) (31,32), or peripherin (peripheral or regenerating nerves) (33)(34)(35), or desmin (neuromuscular cells) (36). When the stem cells become postmitotic, transcription of types VI and III IF genes diminishes, and they are replaced by the first of a series of type IV neuronal IF chains, ␣-internexin (37)(38)(39)(40), which in turn is largely replaced by the other type IV chains including the NF-L, NF-M, and finally the NF-H chains as mature neurons are formed (25,41). To date there is considerable evidence from co-expression and co-transfection experiments for copolymerization of these different chain types onto the same IF networks or even individual IF in living cells (Refs. 21-26; reviewed in Ref. 30). However, certain molecular details of these changes in IF composition remain unclear. It has not been determined yet whether type III vimentin or peripherin and type IV ␣-internexin copolymerize into the same IF and, if so, at which level of IF hierarchy: does this occur at the dimer (that is homo-versus heterodimer) or tetramer (that is different pairs of homodimers) (4,(42)(43)(44), or higher (45)?
As a first step toward addressing the issue of type replace-ment during development, we have previously demonstrated that different keratin chain pairs can easily substitute for one another during differentiation because keratin molecules possess the same dimensions and align themselves in the same way (11). The related question thus arises as to how the type III vimentin network in neuronal cells can be substituted by the series of type IV chains during neuronal development. To address this question, we have explored in vitro the co-assembly properties of the type IV IF chain ␣-internexin with type III vimentin. We show by cross-linking experiments for the first time that these chain types can indeed co-assemble to form heterodimers and that heterodimers or homodimers of them may then participate to form IF. This then allowed us to perform further cross-linking experiments to establish that their precise molecular alignments are in fact the same.

MATERIALS AND METHODS
Expression and Purification of Proteins-A vector containing the full-length coding sequence of rat ␣-internexin in the pET11a system was a generous gift of Dr. Ron Liem (46). The vector expresses a product with an additional 14-amino acid sequence on the head domain of ␣-internexin derived from the T7 10 promoter. Following transformation into the host Escherichia coli B strain BL 21/DE3 pLys(S) (Novagen), cultures (0.5-1-liter volume) were grown in LB medium supplemented with 50 g/ml ampicillin to A 600 nm of 0.6, and protein expression was induced by the addition of 1 mM isopropyl-1-thio-␤-Dgalactopyranoside for 3 h. The ␣-internexin was recovered and purified from bacterial pellets (10,000 ϫ g for 30 min) as described (47). The yield was about 20 mg/liter of bacterial growth. In one experiment, 35 S-labeled methionine (1000 Ci/mmol) was added during protein expression (0.5 Ci/ml of medium) to yield protein of 86 dpm/g. The purified protein was stored at Ϫ70°C in a buffer containing 10 mM Tris-HCl (pH 7.6), 1 mM dithiothreitol, 1 mM EDTA, and 0.1% SDS, freed of SDS by ion-pair extraction (48), and redissolved in a desired buffer (see below) containing 9.5 M urea immediately prior to use.
Purified bacterially expressed human vimentin was a generous gift of Dr. Robert Goldman and was stored and prepared for use as above.
IF Assembly in Vitro-The conditions used for assembly of ␣-internexin were essentially identical to those described previously (37,39,49), at 0.2-0.5 mg/ml using a buffer containing 10 mM triethanolamine HCl in the pH range of 6.7-8.0, 0.17 M NaCl, 1 mM dithiothreitol, and 1 mM EDTA. Vimentin assembly was done at 0.2-0.5 mg/ml in a similar triethanolamine HCl buffer but at pH 7.4 -8.0 (16,19). ␣-Internexin and vimentin co-assembly experiments were done in similar buffers at pH 7.0 -8.0. Solutions of expressed proteins in 9.5 or 6 M urea either singly or mixed as required were dialyzed twice against 1000 volumes of the buffer of choice. In all assembly experiments, protein exposure to concentrated urea solutions was kept to a minimum (Յ4 h). Samples were examined by negative staining with uranyl acetate on glow-discharged carbon-coated grids (50).
Cross-linking Procedures-Cross-linking was performed using the periodate-cleavable bifunctional cross-linking reagent disulfosuccinimidyl tartrate (DST) as described previously (10 -12). Before reaction, the pH of the triethanolamine HCl buffers was raised to 8.0. Cleavage was done by making desired protein solutions to 0.1 M sodium periodate and reaction for 1 h at room temperature.
In trial cross-linking experiments to determine molecular alignments, the amount of DST used ranged from 0.1 to 1 mM, but we found that 0.15 mM gave the most reproducible results; at higher concentrations, it appeared that the numerous lysine residues of the tail domain of ␣-internexin produced multiple random and nonspecific cross-linked products with rod domain lysines. Using an iodoacetamide titration assay (11,12), only about 5% of the ⑀-NH 2 groups of lysines were modified under these conditions. Thus we were able to achieve a very high degree of cross-linking specificity, but the molar yield of crosslinked peptides was quite low. Reactions were done for 30 min, stopped by quenching in 0.1 M NH 4 HCO 3 (final concentration), and dried. We performed cross-linking reactions on homopolymeric ␣-internexin IF, oligomers of ␣-internexin (primarily homodimers and homopolymers estimated to be 16-mers (49)) formed in the absence of 0.17 M NaCl, and IF formed in mixtures of 70% ␣-internexin and 30% vimentin in 0.17 M NaCl. Following cross-linking, the products were resolved on 3-mm thick slab gels. The homodimer and homopolymer bands were eluted, freed of glycine salts, and dried.
In cross-linking experiments performed in concentrated urea buffers to examine molecular stabilities, we used 0.1 M DST instead to overcome the trace amounts of NH 4 ϩ ions in the urea solutions. In this case, to minimize cross-linking of collision complexes, the protein concentrations used were reduced 10-fold to 25-40 g/ml. In trial experiments, we found that the extensive degree of protein modification that occurred under these conditions changed the mobilities of the monomer and dimer bands by Ͻ10%. ␣-Internexin or vimentin were equilibrated in the pH 8.0 buffer (50 l) of desired urea concentration for 1 h at room temperature before addition of the DST, reacted and terminated as above, and made to 2% SDS in sample buffer before electrophoresis on 3.75-7.5% gradient PAGE gels.
To examine formation of heterodimers in mixing experiments, copolymer IF were assembled as described above, pelleted at 100,000 ϫ g for 30 min in a Beckman Airfuge, and amounts in pellets, that is assembly efficiencies, were determined either spectrophotometrically or by amino acid analysis after acid hydrolysis. The pellets were redissolved in 10 mM triethanolamine HCl buffer (pH 8.0) containing 6 M urea (10 -40 g/ml) for 1 h at room temperature before addition of the DST to 0.1 M. Alternatively, mixtures in 9.5 M urea were dialyzed into 6 M urea buffer for 2 h at room temperature before cross-linking as above. In these two cases, the terminated reactions were resolved on shallow 5-7.5% gradient SDS-PAGE gels. Coomassie-stained gels were scanned with a densitometer. Quantitative data were obtained after normalization of staining intensities of 0.1-1-g amounts of ␣-internexin and vimentin on control gels. Alternatively, gels were examined after Western blotting using monoclonal antibodies against either vimentin (Boehringer Mannhiem) or ␣-internexin (a kind gift of Dr. Liem) and developed by enhanced chemiluminescence (Amersham Pharmacia Biotech).
Analysis of DST Cross-linked Peptides-The dried DST cross-linked products and untreated ␣-internexin or vimentin were redissolved (1 mg/ml) in 70% aqueous formic acid and reacted with a 1:1 weight of protein:CNBr overnight. Following drying, the reactions were digested to completion with trypsin (Sigma, bovine, sequencing grade) using 2% enzyme for 6 h at 37°C, at which time another 1% was added for a further 12 h. Peptides were resolved by reverse-phase HPLC as before (10 -12). Potential DST cross-linked peaks were identified by comparisons of profiles before and after cross-linking and then harvested preparatively. When necessary, peaks were purified by a second run. An aliquot of each candidate peak was cleaved with periodate, and the products resolved by HPLC. When two peaks were obtained, they were each subjected to Edman sequencing in a Porton LF-3000 sequencer. When only one peak was obtained, another uncleaved aliquot was used to measure the amount by amino acid analysis following acid hydrolysis, and a third aliquot was used for sequencing. In this way, it could be determined whether the single-cleaved product peak contained only one uncross-linked peptide, or two peptides that were cross-linked together but could not be resolved by HPLC, or contained a peptide cross-linked to itself. By a combination of these methods, the lysine residues involved in cross-links could be unequivocally identified.
Isolation and Characterization of Stable ␣-Helix-enriched Particles-Copolymer IF (0.2-0.4 mg/ml) were digested by dripping during a 30-s period into a solution of trypsin (Sigma, sequencing grade) in a buffer of 50 mM triethanolamine HCl (pH 8.0) to achieve a net protein:enzyme ratio of 100:1, and the digestion continued for 10 -17 min at room temperature (8,9,51,52). Tetramer and dimer ␣-helix-enriched particles were recovered by precipitation at pH 5.2 with 0.1 volume of 3 M sodium acetate, collected by centrifugation, and resolved by chromatography on a 20 ϫ 1-cm column of Sepharose 6B equilibrated in a buffer of 50 mM sodium tetraborate (pH 9.2). The fractions eluted in peak 3 consisting of 2B dimers were collected by reprecipitation at pH 5.2. Aliquots were cross-linked with 0.1 M DST as above, and the products were resolved on 5-10% gradient SDS-PAGE gels. Aliquots were also chromatographed by FPLC on a 5 ϫ 1-cm column of MonoQ equilibrated in the borate buffer and eluted with 50 ml of a 0 -0.15 M KCl gradient. Samples of each of the three peaks were sequenced.
Determination of Axial Parameters in ␣-Internexin and ␣-Internexin/ Vimentin Copolymer IF-Each of the cross-links could be assigned by simple inspection to one of four modes (see Figs. 6 and 7); one group arose from intramolecular cross-links across the ␣-internexin homodimer or ␣-internexin-vimentin heterodimer. Each of the others could be assigned to one of three groups that arose from intermolecular cross-links established previously for epidermal keratin (10, 11) and vimentin (12) IF. These data were used in a least squares analysis to refine the six parameters (A 12 , A 11 , and A 22 , and the linker lengths L1, L12, and L2) that define the axial lattice structure of the ␣-internexin IF (10 -12).

Assembly of ␣-Internexin in IF in Vitro and Copolymerization
with Vimentin-Vimentin IF assembled in vitro in buffers of pH 7.4 to 8.0 were long (Ͼ20 m) smooth-walled structures uniformly 10 nm in diameter (Fig. 1A). Optimal assembly of ␣-internexin occurred at pH 6.7, but IF were much shorter (0.5-2 m) and somewhat irregular in width (8 -15 nm) and appearance (Fig. 1B). Also, irrespective of the protein concentration used for assembly, some particles were only 60 -100 nm long, described previously as "half-thickness unit-length" molecular aggregates that contain 16 chains (49). Above pH 7.0, such particles predominated (Fig. 1C, at pH 8.0). These characteristics for ␣-internexin IF have been reported previously (37,39,49). Mixtures of ␣-internexin with 25 ( Fig. 1D), 50, or 75% vimentin formed IF that were longer at pH 7.0 ( Fig. 1D) than at pH 8.0 (Fig. 1E) and appeared more like vimentin IF alone, since there were very few short particles and most were regularly shaped. Similar images for ␣-internexin and vimentin IF were obtained when assembled from either 9.5 or 6 M urea (data not shown). Together, these studies indicate that the bacterially expressed proteins are indeed co-assembly capable and, moreover, encouraged us to explore further the hierarchical level of IF structure at which the type III vimentin and type IV ␣-internexin co-assembled.
Cross-linking with DST Suggests ␣-Internexin Forms Stable Heterodimers with Vimentin-Two high molecular weight bands of protein were observed when vimentin ( Fig. 2A) or ␣-internexin (Fig. 2B) at 25 or 250 g/ml (data not shown) in pH 8.0 buffer were assembled in the presence of 0.17 M NaCl, followed by cross-linking with 0.1 M DST. Irrespective of the protein concentration, in the absence of salt, vimentin formed primarily tetramers and some higher oligomers, as expected (16,19,22,53,54) (Fig. 2A). In contrast, ␣-internexin formed a high molecular weight band, corresponding to the smaller molecular weight band seen in the presence of salt, that consists of a 16-chain oligomer (49) (Fig. 2B, arrow), as well as dimers and only traces of tetramers. When the two assembly reactions were performed at 25-40 g/ml in the presence of various urea concentrations and then cross-linked after 1 h, we observed that vimentin tetramers were dissociated to dimers above 4 M urea and then to monomeric chains by 8.75 M urea (approximate 50% dissociation point) ( Fig. 2A). However, the ␣-internexin dimers were not dissociated to monomers until about 9.25 M urea instead (Fig. 2B).
To assess further these different properties, ␣-internexin and vimentin in 9.5 M urea buffer were mixed in proportions ranging between 0 and 100%. Following assembly in the presence of 0.17 M NaCl in pH 7.0 buffer, IF were pelleted in the Airfuge, redissolved in 6 M urea, under which conditions they are dissociated into dimers (Fig. 2), cross-linked with 0.1 M DST, and resolved by SDS-PAGE. The shallow gradient used allowed separation of three dimer bands of 100 -120 kDa which, by Western blotting with specific antibodies, contained ␣-internexin-␣-internexin homodimers (uppermost band), ␣-internexin-vimentin heterodimers, or vimentin-vimentin homodimers (Fig. 3A). Notably, quantitation of the amounts of each protein observed in Coomassie-stained gels (Fig. 3B) revealed that in the presence of molar excesses of ␣-internexin, significant amounts of ␣-internexin-vimentin heterodimers were formed; particularly in mixtures of 70% ␣-internexin and 30% vimentin, the resulting IF contained only traces of vimen- tin homodimers but about equal amounts of ␣-internexin-vimentin heterodimers and ␣-internexin homodimers. Note in the control experiments of Fig. 2 that the 6 M urea solutions could not effectively dissociate either the ␣-internexin or vimentin homodimers to permit chain exchange in dimers. Also, in experiments using tracer amounts (0.1%) of 35 S-labeled ␣-internexin, no label was incorporated into the central heterodimer band when mixed with vimentin in 6 M urea for up to 16 h (not shown). These experiments were repeated by dialysis of the initial protein mixtures in 9.5 to 6 M urea instead, followed by cross-linking and resolution by SDS-PAGE. Essentially identical data were obtained (not shown). Thus together, these data provide robust support for the notion that ␣-internexin and vimentin form heterodimers that are about as stable as ␣-internexin homodimers and measurably more stable than vimentin homodimers.
As an independent analysis of these observations, we counted the numbers of potential ionic interactions between e-g, g-a, and d-e pairs of charged residues on the heptad repeats across these two homodimer and heterodimer molecules, as well as for other pairs of neuronally expressed IF chains (Table I). We determined the numbers for scenarios in which the two chains are aligned in exact axial register or offset by one or two heptads. The data reveal first that the largest number of favorable ion-pair interactions always occurs when the two chains are in exact axial register, as expected (55). Significantly, the highest two scores were found for ␣-internexin homodimers and ␣-internexin-vimentin heterodimers. Scores for ␣-internexin-neurofilament M and ␣-internexin-neurofilament H heterodimers were the next highest, followed by the score for vimentin homodimers. These analyses offer a potential explanation for the observed urea stability data seen in Figs. 2 and 3.
As a further control for these mixing experiments, we individually dialyzed the solutions of ␣-internexin or vimentin in 9.5 M urea to 6 M urea for 1 h, then prepared 0 -100% mixtures, and assembled them into IF. Following cross-linking as above, we did not find the intermediate ␣-internexin-vimentin heterodimer band by SDS-PAGE (data not shown), but instead, the copolymer IF contained amounts of the two proteins in the same proportions as included in the mixtures (Fig. 3C).
Limited Proteolysis Methods to Isolate Stable ␣-Helix-enriched Particles Confirmed That ␣-Internexin and Vimentin Can Form Copolymer IF from Either Heterodimers or Homodimers-We have demonstrated previously that it is possible to isolate stable ␣-helix-enriched oligomeric particles from keratin (8,9,51) and vimentin/desmin (54) IF by limited proteolysis procedures. These particles are derived from the longer 1B and/or 2B coiled-coil rod domain segments of the molecules because the head and tail domains, and linker segments of the rod domain, are considerably more sensitive to cleavage by proteolysis. When IF co-assembled in a 70:30% mixture of ␣-internexin and vimentin from 9.5 M urea were subjected to limited trypsin digestion, three time-dependent peaks of protein of different apparent sizes were recovered from a Sepharose 6B column (Fig. 4A). Previous studies have indicated that peak 3 material contains dimeric ␣-helix-enriched particles about 20 nm long that are derived from the 2B rod domain segments of the constituent chains (8,9,53,54). Indeed, amino acid sequencing for 15 Edman degradation cycles of protein eluted in peak 3 yielded two sequences FANLNEQAARSTEAI and FADLSEAANRNNDAL that correspond exactly to residues 290 -304 and 293-307 of rat ␣-internexin and human vimentin, respectively; these are equivalent to positions 5-19 of their 2B rod domains. When denatured and resolved on SDS-PAGE gels, two bands of about 12 and 13 kDa were obtained, thus indicating that this peak 3 peptide material consisted of most of the 2B rod domain regions of the two chains (data not shown). We further noted that peak 3 material could be resolved on a MonoQ FPLC column into three peaks labeled A, B, and C (Fig. 4B). Following sequencing again, peak A contains homodimeric 2B rod domain sequences of ␣-internexin only; the minor peak C contained the more acidic homodimeric 2B sequences from vimentin only; and the central peak B contained equimolar amounts from both chains. These peaks re- solved because of significant charge differences between the 2B rod domain sequences of ␣-internexin and vimentin. When copolymer IF were assembled from mixtures made in 6 M urea instead, followed by digestion and fractionation on the Sepharose column, an identical elution profile was obtained (data not shown). However, when the peak 3 material was resolved by FPLC on the MonoQ column, only peaks A and C were obtained (Fig. 4C), containing homodimeric ␣-internexin and vimentin 2B fragments, respectively.
Cross-linking of ␣-Internexin and ␣-Internexin-Vimentin Copolymer IF with DST-We then undertook cross-linking reactions in an effort to determine the exact alignments of nearest neighbor molecules in ␣-internexin homopolymeric and ␣-internexin-vimentin copolymer IF. In this case, cross-linking was done with 0.15 mM DST in order to maximize specificity and avoid interference between rod and tail domain lysines.
In the first set of experiments with ␣-internexin homopolymeric IF, Fig. 5 shows HPLC chromatograms of CNBr/tryptic peptides recovered from cross-linking reactions of the dimer formed in 2 M urea (Fig. 5B) and 16-mer species formed in low salt buffer (Fig. 5C). Although the molar yields of cross-links were low, a total of 4 and 14 peaks, respectively, which disappeared on treatment with periodate and thus were candidate cross-linked species, were isolated and characterized by a combination of amino acid analysis and protein sequencing (Table  II). The four peaks found in Fig. 5B were also present in Fig. 5C and clearly resulted from intramolecular cross-links between the two chains of the coiled-coil ␣-internexin dimer molecule. All 11 lysines of the rod domain of ␣-internexin participated in cross-links, and most were used multiple times with different partners. Interestingly, a number of these occupy exactly the same positions as were found in previous cross-linking studies of K1/K10, K5/K14, and vimentin IF (10 -12), which adds validity to the significance to the present data set. Moreover, all 11 (10 unique species) seen in Fig. 5C could be confidently assigned to one of three modes of intermolecular alignment seen previously for these other types of IF as follows: the A 11 (two cross-links, and see Fig. 6, second model from right) and A 22 (three cross-links; Fig. 6, second model from left) modes which suggest staggered antiparallel alignments; and A 12 , in which the two molecules are antiparallel and in near registration (five cross-links, Fig. 6, center model).
We then repeated these experiments with intact copolymer IF assembled from a 70:30 mixture of ␣-internexin:vimentin from 9.5 M urea. Under these conditions, about half of the molecules are ␣-internexin homodimers, and another half are ␣-internexin-vimentin heterodimers (Figs. 3 and 4). Since the amount of vimentin-vimentin cross-links was expected to be very low, and the positions of the cross-links on the HPLC profile in homopolymeric ␣-internexin were already known (Fig. 5), it was possible to identify new peaks that were candidates for cross-links between heterodimers. In this way, 29 peaks were obtained by HPLC (not shown) and sequenced (Table II), of which 3 were between ␣-internexin molecules, 5 were intramolecular across the heterodimer, and 4, 7, and 10 could be assigned to the A 11 , A 22 , and A 12 modes, respectively (Fig. 7).
Calculation of Rod Axial Alignments and Linker Segments-These data were used in a least squares analysis to refine the six parameters (A 12 , A 11 , and A 22 , and the linker lengths L1, L12, and L2) that define the axial lattice structure of the ␣-internexin. In principle, the actual positions of the crosslinks determine the number of unique equations that can be derived. Providing that there are enough of these, the parameters can be refined successfully. In the case of ␣-internexin homopolymers, only four parameters refined satisfactorily (A 12 , A 11 , L1, and L2) and two did not (A 22 and L12) (Table III). Interestingly, however, the nature of the equations is such that The abbreviations used are: INX, ␣-internexin; VIM, vimentin; NFL, NFH, NFH, neurofilament light, medium, and heavy chains, respectively. Sequences are rat ␣-internexin (Swiss-Prot accession number P23565); human vimentin (P08670); mouse neurofilament chains (P08551, P08553, and P19246, respectively). The scores are for e-g, a-g, and d-e ionic pairs across a coiled-coil pair.

FIG. 4. Recovery of ␣-helix-enriched particles from ␣-internexin-vimentin heterodimers.
Separation of particles on Sepharose 6B after limited trypsin digestion of copolymer ␣-internexin-vimentin IF assembled in a 70/30% mixture. Protein from peak 3 containing 2B dimer particles from a 9.5 M (B) or 6 M (C) urea assembly experiment was further resolved by FPLC on a MonoQ column. By amino acid sequencing the peaks contained the following: A, ␣-internexin 2B homodimers; B, ␣-internexin-vimentin 2B heterodimers; and C, vimentin 2B homodimers. the difference between A 22 and L12 is refinable and has a value similar to that seen previously for vimentin. Using the 16 unique data elements from the copolymer IF, however, all six parameters refined successfully (Table IV). Significantly, there is a high correspondence in these refined parameters and those previously determined for homodimeric vimentin. They differ from those determined for vimentin only in those parameters which were previously relatively poorly defined in vimentin, because of the nature and disposition of the limited cross-link data then available. DISCUSSION   NLQEAEEWYKSK a The ␣-internexin (INX) and vimentin (VIM) peptide sequences are identical. In the cases where one sequence was unequivocally identified as vimentin, it is likely the second was ␣-internexin, since there were very few vimentin-vimentin homodimer molecules in the IF used for this reaction (Fig. 2). In those cases where ␣-internexin was unequivocally identified as one of the sequences, the second was also likely to have derived from ␣-internexin, that is a homodimer cross-link. Note that methionine (M) was identified as homoserine.

␣-Internexin and Vimentin Promiscuously Form
FIG. 5. HPLC fractionation of DST cross-linked CNBr/tryptic peptides. The profiles are as follows: A, unreacted ␣-internexin; B, ␣-internexin dimers formed in 2 M urea solutions; and C, ␣-internexin 16-mer formed at pH 6.7 in low salt buffer. Comparisons of profile A with profiles B and C revealed shifted peaks that were chosen as candidate cross-linked peptides that were recovered for analyses. Numbers correspond to those of Table II. The broken line denotes the acetonitrile gradient. this co-assembly may occur at different levels of IF structural hierarchy. When assembled from mixtures made in 9.5 M urea solutions, wherein the proteins exist almost exclusively as single chains (Fig. 2), heterodimer formation is favored (Figs. 3  and 4) and IF are formed in high yield. In 6 M urea solutions, the chains form homodimers instead (Figs. 3 and 4), which also favored the subsequent formation of IF. Control experiments revealed that once homodimers are assembled in 6 M urea, little or no chain exchange can occur, until or unless the urea concentration is raised to Ϸ9 -9.5 M. Similar data were established previously for keratin IF (8,9). Accordingly, these data mean that copolymer ␣-internexin-vimentin IF may be assembled from either heterodimer or homodimer molecules. In mixtures formed in either 6 or 9.5 M urea containing Յ90% ␣-internexin, long (Ͼ5 m) smooth-walled IF of uniform width are formed (Fig. 1), that is by all available criteria, the IF are morphologically identical. Thus there is no a priori reason why normal appearing copolymer IF could not consist of mixtures of both homodimers and heterodimers in vitro and in vivo.
These data complement and extend earlier in vitro studies on the assembly proclivities of types III and IV proteins (43,44). In co-assembly experiments with the type III protein peripherin and ␣-internexin, it was not possible to ascertain whether the two proteins engaged in heterodimer formation because the authors were unable to resolve homodimers from heterodimers on the SDS gel systems employed (44). Furthermore, the crosslinking conditions with the copper-phenanthroline reagent could not provide evidence for heterodimeric interactions (44). However, the SDS gels and cross-linking experiments did reveal heterodimer formation of peripherin with the other larger type IV NF-M or NF-H chains (44). Accordingly, combinations of our present work and the studies with peripherin allow the general conclusion that all type III and all type IV IF chains are capable of heterodimer formation and co-assembly into IF.
Molecular Packing of Coiled-coil Molecules in Type IV IF Is the Same as Type III-The only way by which such promiscuous co-assembly into IF could occur is if in fact each of the ␣-internexin or vimentin homodimer and ␣-internexin-vimentin heterodimer molecules has the same dimensions that permit co-alignment through several levels of structural hierarchy to form IF. Indeed, extensive cross-linking studies performed here all point to the conclusion that the molecular alignment parameters are all the same within experimental error. The limitations of the present data sets are imposed on us primarily by the availability of numbers of informative cross-links, which in turn is a reflection of the juxtaposition of available lysines for cross-linking. In the case of ␣-internexin homodimers, it was not possible to refine adequately all six parameters clearly (Table III), but robust data were obtained in the case of the ␣-internexin-vimentin copolymer IF (Table IV).
As the alignments are likely to be the same, we then reevaluated the unique cross-link data (22 cross-links) from vimentin, ␣-internexin, and the ␣-internexin-vimentin copolymer IF in a comprehensive analysis to refine the axial parameters relating to the molecular disposition for all type III and all type IV IF (Table V). The values are determined with small standard errors (0.13-0.34) consistent with all of the data originating from a common set. Values of the parameters determined for the epidermal keratins using cross-link data, as well as those modeled for hard ␣-keratin from x-ray and other data, are given for purposes of comparison.
It is interesting to compare these parameters and to assess common and differing features. First, the parameters for epidermal keratin, vimentin, and ␣-internexin lead to a head-totail overlap of about 7-10 residues between the beginning of segment 1A and the end of segment 2B of molecules aligned in the same axial row. These regions are the most highly conserved ones in IF chains as a whole. They also correspond to the regions in which many mutations leading to keratinopathies are located or closely juxtaposed (2,3,56). It has been predicted, nonetheless, that in hard ␣-keratin this head-to-tail overlap does not exist in vivo after disulfide bond formation occurs (57). Second, a feature common to IF (except type V), however, is the predicted molecular length, i.e. 308 residues in a coiled-coil conformation, a value equivalent to 45.6 nm. Likewise, the combined length of the linkers L1, L12, and L2 is equivalent to about 32 residues in a coiled-coil conformation  Fig. 6. In this case green designates the vimentin chain in the ␣-internexin-vimentin heterodimer in each of the alignment modes.
erodimers Are More Stable Than Those of Vimentin, Implications for Neuronal Development-Although ␣-internexin and vimentin chains appear to be able to co-assemble into morphologically identical IF and in equivalent yields, we have detected subtle but significant differences in the apparent stabilities of homo-and heterodimer molecules and in their states of assembly in low ionic strength buffers (Fig. 2). First, in low ionic strength buffer at pH 8.0 where efficient and specific crosslinking experiments are possible, vimentin forms primarily tetramers ( Fig. 2A; Refs. 19, 22, 53, and 54), but ␣-internexin forms primarily a 16-mer oligomer ( Fig. 2B; Ref. 49). However, by ultracentrifugation methods, ␣-internexin forms a more heterogeneous population of species as the pH is lowered toward known optimal IF assembly conditions near 7.0 (49). 2 Such pH-dependent association phenomena have not been reported for vimentin (19,22,54). These data therefore allow the speculation that in living cells the state of assembly of an ␣-internexin-containing IF may be significantly affected by the microenvironment. Second, we noted that ␣-internexin homodimers and ␣-internexin-vimentin heterodimers are measurably more resistant to urea dissolution than vimentin homodimers (Fig.  2), and indeed, when mixed together, ␣-internexin prefers to form heterodimers with vimentin (Fig. 3). A proximal explana-TABLE III Unique ␣-internexin cross-links and the equations used in the least squares analysis The equations are derived for example as follows: 1A-17/2B-112 is given by (1 ϫ A 12 ) ϩ (0 ϫ A 11 ) ϩ (0 ϫ A 22 ) Ϫ (0 ϫ L1) ϩ (0 ϫ L12) ϩ (0 ϫ L2) ϭ 7 residues.  tion for this may be the ability to form a larger number of pairs of ionic charges along the rod domains of the dimers (Table I).
The potential cell biological consequences of this in vivo are significant. A variety of data have revealed that vimentin expression precedes that of ␣-internexin in developing neuronal cells (31,32). Therefore it could be argued that the proposed equilibrium dynamic exchange/replacement of type III vimentin by type IV ␣-internexin (28) may be driven by a net increase in the stability of ␣-internexin-containing molecules and in this way more assertively drive the gradual replacement process during development. Moreover, this transition may result in a more stable IF network in the axons of maturing neurons.