Characterization of interactions between the neurofilament triplet proteins by the yeast two-hybrid system.

In the adult axon, the neurofilaments (NFs) are heteropolymers formed from the low (NFL), middle (NFM), and high (NFH) molecular weight neurofilament triplet proteins (NFTPs). All three proteins have the basic intermediate filament protein tripartite structure, which consists of a short amino-terminal head region, an alpha-helical rod region of approximately310 amino acids, and a carboxyl-terminal tail region of variable length. In vitro polymerization studies have shown that only NFL can assemble into homopolymeric 10-nm filaments. The assembly of intermediate filaments, including the NFs, begins with the formation of a coiled-coil dimer involving the alpha-helical rod domains of two molecules. In order to determine whether homodimers or heterodimers of NFTPs are the preferred intermediates in the assembly of NFs, we have used the yeast two-hybrid system to study the interactions between the different NFTPs. By monitoring the activity of the lacZ reporter gene product, we are able to show that the interactions of NFL with NFL, NFM, or NFH are stronger than the interactions of NFM with NFM or NFH and the interaction of NFH with NFH. These results imply that NFM and NFH are more likely to form heterodimers with NFL than homodimers and are consistent with the inability of NFM and NFH to self-polymerize in vitro and in vivo.

In the adult axon, the neurofilaments (NFs) are heteropolymers formed from the low (NFL), middle (NFM), and high (NFH) molecular weight neurofilament triplet proteins (NFTPs). All three proteins have the basic intermediate filament protein tripartite structure, which consists of a short amino-terminal head region, an ␣-helical rod region of ϳ310 amino acids, and a carboxylterminal tail region of variable length. In vitro polymerization studies have shown that only NFL can assemble into homopolymeric 10-nm filaments. The assembly of intermediate filaments, including the NFs, begins with the formation of a coiled-coil dimer involving the ␣-helical rod domains of two molecules. In order to determine whether homodimers or heterodimers of NFTPs are the preferred intermediates in the assembly of NFs, we have used the yeast two-hybrid system to study the interactions between the different NFTPs. By monitoring the activity of the lacZ reporter gene product, we are able to show that the interactions of NFL with NFL, NFM, or NFH are stronger than the interactions of NFM with NFM or NFH and the interaction of NFH with NFH. These results imply that NFM and NFH are more likely to form heterodimers with NFL than homodimers and are consistent with the inability of NFM and NFH to self-polymerize in vitro and in vivo.
Neurofilaments (NFs) 1 are a class of intermediate filaments (IFs) that are exclusively expressed in mature neurons in both the central and peripheral nervous systems. They have been suggested to play an important role in the modulation of axonal caliber (1,2) and are composed of the neurofilament triplet proteins (NFTPs) named NFH, NFM, and NFL for the high, middle, and low molecular weight neurofilament subunits, respectively. Similar to other IF proteins, the NFTPs conform to a tripartite structure consisting of a conserved central ␣-helical "rod" domain of ϳ310 amino acids and variable amino-and carboxyl-terminal head and tail domains (3). The ␣-helical rod domains of the NFTPs contain long stretches of heptad repeats that mediate the formation of coiled-coil dimers, the common first step in intermediate filament assembly.
The assembly of neurofilaments has been extensively studied both in vitro and in vivo. In vitro, purified NFL can selfassemble into filaments of normal morphology, whereas NFM and NFH only form irregular filamentous aggregates. In the presence of NFL, these two proteins co-assemble into heteropolymeric filaments (4,5). In transfected cells, each of the NFTPs is able to co-assemble with the endogenous vimentin network (6,7). In contrast to the in vitro polymerization studies, transfection studies have shown that in the absence of an endogenous vimentin network, none of the NFTPs is able to self-polymerize into a filamentous network. NFL co-assembled with either NFM or NFH into filamentous arrays in the transfected cells, but NFM and NFH could not co-assemble into filaments (8 -10).
Like other intermediate filaments, neurofilaments are thought to assemble through three intermediates, dimers, tetramers, and octamers (11)(12)(13). For keratins, which are heteropolymers of a type I and a type II keratin, the first step of assembly is the formation of a heterodimer. The type III IF proteins (vimentin, desmin, glial fibrillary acidic protein, and peripherin), which self-assemble in vitro as well as in vivo, form homodimers in the first step of polymerization, although heterodimers between different type III IF proteins can also be formed (14 -15). As to the NFTPs, there is evidence for the existence of tetramers that contain mixtures of NFL and NFM (13); however, there is not yet any conclusive evidence as to the identity of the first-formed dimer. This dimer can be a homodimer, a heterodimer, or both. In the case of NFL, which can self-assemble in vitro, one would expect that it is able to form homodimers. In the formation of copolymers of NFL and NFM (or NFH), either homodimers of NFM (or NFH) or heterodimers of NFL/NFM (or NFL/NFH) could be formed, which could then interact with NFL homodimers to form filaments.
Although there is no direct evidence indicating that NFL/ NFM or NFL/NFH heterodimers can be formed, the results of an antibody labeling study of bovine neurofilaments suggested the possible formation of heterodimers of NFL with NFM or NFH (16). This study showed that an epitope in the NFL tail, which was accessible in NFL homopolymers, was masked in native filaments. This phenomenon can be explained by the hypothesis that NFL/NFM and NFL/NFH heterodimers are the initial building blocks in native NFs. The elucidation of this first step of NFTP polymerization is important because it will help explain the detailed arrangement of the protein chains within the neurofilament.
To determine if the NFTPs preferentially form homodimers or heterodimers, one would like to be able to compare the relative affinities of an NFTP monomer for its own kind and for other NFTP monomers. However, these studies are confounded by the difficulty of stopping the polymerization at the dimer-* This work was supported by National Institutes of Health Grant NS15182 (to R. K. H. L.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
§ Supported by Training Grants EY07105 and AG00189 as a National Institutes of Health predoctoral trainee.
* ization step. In order to circumvent this problem, we have employed the yeast two-hybrid system to measure the relative strengths of the NFTPs to interact with their own kind and with each other, as well as with vimentin.
␤-Galactosidase Assays-Interactions of proteins in the two-hybrid system can be monitored both qualitatively and quantitatively. Filter lift assays were performed for the qualitative measurement of ␤-galactosidase activity. Transformed yeast colonies were patched onto filters layered over minus Trp-Leu synthetic dropout synthetic agar plates overnight at 30°C. Filters were submerged in liquid nitrogen for 10 s and placed on other filters that were presoaked in 5-bromo-4-chloro-3indolyl ␤-D-galactoside solution containing 100 mM phosphate buffer (pH 7.0), 10 mM KCl, and 1 mM MgSO 4 . Filters were then incubated at 30°C and periodically examined for the appearance of blue colonies. Quantitative assays were performed on liquid cultures. Cultures of 2.0 ml were grown to mid-log phase in minus Trp-Leu media. Cells were spun and resuspended into 0.5 ml of 100 mM phosphate buffer (pH 7.0), 10 mM KCl, 1 mM MgSO 4 , and 50 mM ␤-mercaptoethanol, and 50 l of CHCl 3 and SDS were added to permeabilize the cells. 100 l of Chlorophenol red-␤-D-galactopyranoside solution (4 mg/ml in 0.1 M phosphate buffer) was used as the substrate. The reaction mixtures were incubated at 30°C for 1-2 h and finally quenched by adding 0.25 ml of 1 M Na 2 CO 3 . A ␤-galactosidase unit is defined as 10 3 ϫ A 574 /min of reaction for 1 ml of culture at 1 A 600 unit (19). The reported results were means and standard deviations of four independent transformants.

RESULTS AND DISCUSSION
In this study, we have utilized the yeast two-hybrid system to characterize the interactions between the NFTPs. The mechanism of the system has been described by Fields and Song in detail (20) and takes advantage of the properties of the GAL4 protein of the yeast Saccharomyces cerevisiae. In essence, the two domains necessary for transcriptional activation are separated into a GAL4 DNA binding domain and an activation domain. Various proteins are then expressed either as a hybrid with the GAL4 DNA binding domain or with the GAL4 transcriptional activation domain. The ability of two proteins to interact can then be determined by co-transformation into a yeast strain, which contains an integrated copy of a GAL1-LacZ indicator gene. If the two proteins interact, the two domains of GAL4 are brought into close physical proximity and allow activation of transcription from the lacZ gene, whose gene product can be readily detected. The cDNAs encoding amino acids 24 -542 of NFL, as well as the full-length NFH and NFM, were cloned into the pGBT9 and pGAD424 vectors. Since the assembly of NFL from which the amino-terminal 23 amino acids are deleted was comparable with that of the full-length NFL in transfected fibroblasts (21), we assumed that the interaction strengths of the NFL 24 -542 hybrids would also be comparable with the full-length NFL hybrids. Different combinations of the NFTP hybrid constructs were co-transformed into the yeast strain SFY526 and selected on the minus Trp-Leu synthetic dropout synthetic plates for 4 days. Interactions between the hybrid proteins were determined by filter assays measuring the ␤-galactosidase activities of transformed yeast colonies, and the results are summarized in Table I. None of the NFTP-containing chimeric proteins showed nonspecific activation of the lacZ reporter gene. ␤-Galactosidase activity could readily be detected when GAL4 transcriptional activation hybrids of NFL 24 -542 were co-transformed with GAL4 DNA binding hybrids of NFL 24 -542 , NFM, and NFH, indicating the interactions of NFL with NFL, NFM, and NFH. Surprisingly, even though both NFH and NFM contain rod domains that potentially interact, they did not interact with themselves or with each other in the two-hybrid system.
When the hybrid combinations for NFL, NFM, and NFH were reversed, i.e. GAL4 transcriptional activation hybrids of NFM or NFH were co-transformed with GAL4 DNA binding hybrids of NFL 24 -542 , no ␤-galactosidase activity could be detected on filters. In order to determine whether this inconsistency was due to the lack of expression or improper folding of the activation domain-fused NFM and NFH, we also tested for their interactions with vimentin. We reasoned that since each of the NFTPs co-assembles with vimentin in transfected fibroblasts (9,10), it was possible that these proteins would interact to form a heterodimer with vimentin in the yeast two-hybrid system. As shown in Table I, all of the NFTPs hybrids interacted with vimentin, although stronger interactions were detected when vimentin was expressed as a hybrid with the GAL4 DNA binding domain. The discrepancy of the results due to

Interactions between NFTPs with each other and with vimentin on
␤-galatosidase filter assays The numbers of plus signs represent the relative rates that the transformed yeast colonies turned blue after incubation at 30°C on filters; ϩϩϩϩ, Ͻ1 h; ϩϩϩ, 1-3 h; ϩϩ, 3-8 h; ϩ; Ͼ8 h. Yeast cotransformed with pGBT-p53 and pGAD-T-antigen, which started to turn blue in less than 1 h on the filter assay (designated ϩϩϩϩ), was used as a positive control. At least four independent transformants were tested for each pair.
switching the vectors may indicate that the GAD-vimentin hybrid is less well expressed than the GBT-vimentin hybrid. A lower level of expression of GAD-vimentin could also account for the apparently stronger interactions between GBT-vimentin and the GAD-NFTP hybrids than between GBT-vimentin and GAD-vimentin. In any case, our results imply that interactions of NFL with NFL, NFM, and NFH can occur, and in addition each of the NFTPs can interact with vimentin. The lack of interactions between NFM and NFH with themselves and the interactions observed between the GAD-NFL hybrid with the GBT-NFM and GBT-NFH hybrid are consistent with the tentative conclusion that NFL/NFM and NFL/NFH heterodimers are formed in preference to NFM/NFM and NFH/ NFH homodimers; however, this conclusion is tempered, because GBT-NFL did not interact with GAD-NFM or GAD-NFH.
One problem with the previous results may be that the large size of the NFH and NFM tails in the hybrids may interfere with the activation of transcription of the lacZ reporter gene resulting in the lack of detection of their self-interactions by the ␤-galactosidase assay. To eliminate this possibility, we constructed tailless NFTPs in pGBT9 and pGAD424 to characterize all the possible interactions in dimer formation between the NFTPs. The tailless NFL (L⌬) and NFH (H⌬) constructs contained cDNA of rat NFL and NFH from amino acids 1 to 415 and the tailless NFM construct (M⌬) coded for amino acids 1-421. All the different combinations of the tailless constructs were co-transformed into yeast cells, and the interactions were monitored both by the filter assays and the liquid ␤-galactosidase assays. Yeast that had been co-transformed with combinations of hybrids containing NFTPs turned blue quickly after being permeabilized on filters, indicating that the tailless constructs interacted more strongly than the full-length hybrid proteins (Table II). Since we were not able to determine the relative strengths of interaction between the different NFTPs from the qualitative filter assays, we did the quantification using liquid assays with Chlorophenol red-␤-D-galactopyranoside as the substrate. The results are illustrated in Fig. 1. Except for the pair of pGBT-L 1-415 and pGAD-M 1-421 , co-transformation of hybrids of tailless NFL with any of the NFTPs yielded higher ␤-galactosidase activity than M⌬/M⌬, H⌬/H⌬, or M⌬/H⌬. It is interesting to notice that combinations with tailless NFL fused to the GAL4 DNA binding domain (GBT-L⌬) showed much lower ␤-galactosidase activities than those with tailless NFL fused to the GAL4 activation domain (GAD-L⌬). This phenomenon is consistent with the results shown in Table  I, in which full-length NFTPs were used for hybrid construction. The cloning of NFL in the pGBT9 vector may thus account for the low ␤-galactosidase activity of GAD-M 1-421 with GBT-L 1-415 (Fig. 1), as well as the lack of detection of interaction between GBT-L 24 -524 and GAD-NFM or GAD-NFH (Table I). These results suggest that NFL is capable of forming homodimers as well as heterodimers with NFH and NFM. Furthermore, the formation of NFL/NFH and probably NFL/NFM heterodimers appears to be preferred to the formation of homodimers of NFH and NFM.
To determine whether the head domains of the NFTPs have an effect on the interactions observed in these assays, we removed a portion of the head domain from the tailless NFL construct and then tested for the resulting partial headless/ tailless NFL hybrid (⌬L⌬) for its interaction with the other tailless NFTPs constructs. The results of the quantitative assays of different combinations are shown in Fig. 2A. From comparisons with the results shown in Fig. 1, it is apparent that the partial removal of part of the NFL head domain did not prevent NFL from interacting with the NFTPs. These results support the hypothesis that the formation of NFTP dimers is mostly mediated by the rod domains although the remaining head domain of NFL could still play a role in these interactions. We also deleted part of the head domain from the tailless NFM construct to see if the resulting partial headless/tailless NFM (⌬M⌬) would be able to interact with the other NFTPs constructs. The results show that the ⌬M⌬ has a similar interaction profile with the other NFTPs (Fig. 2B) as the tailless NFM ( Fig. 1) and that lower ␤-galactosidase activities were observed when the ⌬M⌬ construct is fused to the GAL4 activation domain than when it is fused to the binding domain. However, the interactions between tailless NFL (L⌬) with ⌬M⌬ are stronger than the self-interaction of ⌬M⌬ regardless of which vector the NFM construct was cloned (compare ⌬M⌬/L⌬ and L⌬/⌬M⌬ with ⌬M⌬/⌬M⌬ in Fig. 2). In a recent study of the interactions between different domains of vimentin using the yeast two-hybrid system, Meng et al. (22) showed that vimentin dimers and tetramers both exist in the transformed yeast cells. This result raises the possibility that tetramers are also formed in our system and that part of the ␤-galactosidase activities of yeast co-transformed with NFL and NFM (or NFL and NFH) is due to the interactions between NFL homodimers and NFM (or NFH) homodimers. However, our results show that the self-interactions of NFM and NFH are weak (Figs. 1 and 2). Therefore, if tetramers are formed in this system, they are more likely to consist of either two NFL/ NFM (or NFL/NFH) heterodimers or an NFL/NFL homodimer and an NFL/NFM (or NFL/NFH) heterodimer rather than NFL/NFL and NFM/NFM (or NFH/NFH) homodimers.
Overall, the results from the two-hybrid system show that the interactions between NFL and NFL, NFL and NFH, and  1. Quantitative analysis of interactions between tailless neurofilament triplet proteins. SFY526 cells co-transformed with plasmids derived from pGBT9 (top line) and pGAD424 (bottom line) were grown to mid-log phase in minus Trp-Leu media. Cells were permeabilized and assayed for ␤-galactosidase activity as described in the text. H⌬, M⌬, and L⌬ indicate the tailless constructs NFH 1-415 , NFM 1-421 , and NFL 1-415 , respectively. The backgrounds were measured as the ␤-galactosidase activity, which was obtained from the co-transformation of pGAD424 and the corresponding pGBT9 NF derivatives and were subtracted from each set of data. also very likely NFL and NFM are stronger than the interactions of NFM with NFM, NFH with NFH, or NFH with NFM. We therefore would predict that in the presence of all three NFTPs, these higher interaction affinities would cause NFL to self-polymerize and co-polymerize with NFM and NFH into filaments, whereas NFM and NFH would be less likely to form homopolymers or NFM/NFH heteropolymers than NFM/NFL or NFL/NFH heteropolymers. These results are therefore con-sistent with the ability of NFL to self-polymerize in vitro but are in apparent contradiction with the transfection results which show that NFL alone does not assemble into filaments in transfected cells (9,10). However, the data are consistent with the results that showed that NFL is present only in the Triton X-100-insoluble fraction of transfected cells, even in the absence of a filamentous network, whereas relatively higher amounts of NFM and NFH are present in the soluble fraction (9). Thus NFL is able to form higher order filamentous structures, but not a filamentous network, which would require NFM and NFH (in the form of NFL/NFM and NFL/NFH heterodimers) to interact with the NFL homodimers.
Our results also show that the NFTPs are able to interact with vimentin in the two-hybrid system and imply that in transfected cultured cells they could co-polymerize with vimentin via the formation of heterodimers. Finally, our results suggest that during the assembly process of neurofilaments, both homodimers and heterodimers of the various NFTPs are present, but for NFM and NFH, heterodimer formation with NFL would be more preferred to formation of homodimers.