The endless story of the glial fibrillary acidic protein

Intermediate filaments (IF), along with microtubules and microfilaments, make up the cytoskeleton of most eukaryotic cells. On the basis of amino acid sequence homologies and intron positions, IFs can be subdivided into at least five types (Steinert and Roop, 1988; Bloemendal and Pieper, 1989; Steinert et al., 1985). IFs are specifically expressed in restricted tissues, except for the type V lamins, which are universally expressed in the nucleus of all eukaryotic cells.

All IFs share a common tripartite structure consisting of a highly conserved, α-helical rod domain flanked by non-helical end domains of variable sizes and sequences (Geisler and Weber, 1982; Steinert and Roop, 1988; Conway and Parry, 1988; Parry and Steinert, 1992). The rod domain, with its heptad substructure of the form (a-b-c-d-e-f-g)_n, where positions a and d are generally occupied by apolar residues, favors the formation of coiled-coils between two a-helices. In addition, the periodic distribution of acidic and basic residues in coils 1b and 2 (Crewther et al., 1983; Steinert and Roop, 1988; Parry and Steinert, 1992; McLachlax and Stewart, 1982) has been proposed to provide interhelical ionic interactions that further stabilize the coiled-coil interactions and may help determine both the relative axial stagger and the polarity between two adjacent helices (O’Shea et al., 1991). The rod domain by itself can form only dimers or tetramers (Kaufmann et al., 1985; Quinlan et al., 1989). Further assembly of these oligomers into higher-order structures requires the contribution of the non-helical end domains.

Studies, in vitro and in vivo, of the possible contributions of the N-terminal head and/or C-terminal tail domains to IF formation have often led to contradictory conclusions. Most of these studies involve removal of the head and/or the tail domain by specific proteases or by deletion mutations of the cDNAs. There seems to be a general agreement that most, if not all, of the N-terminal head is required for IF self-assembly (Kaufmann et al., 1985; Raats et al., 1990; van den Heuvel et al., 1987; Traub and Vorgias, 1983; Quinlan et al., 1989). However, when co-assembly instead of self-assembly is considered, the role of the head domain is still unclear. For most IFs, the head domain can be deleted without disabling their coassembly with vimentin, a type III IF that exists in mesenchymal tissues and most cultured cells, as long as the deletions do not involve coil 1a (Chin et al., 1991; Albers and Fuchs, 1989; Wong and Cleveland, 1990). Whether or not an incomplete N-terminal deletion mutant will disrupt the vimentin network is partly determined by the relative expression levels between the mutants and the endogenous vimentin. For those IFs that normally form heteropolymers, more of the head domain seems to be required to co-assemble with their normal counterparts than with vimentin (Lee et al., 1993).

The role of the tail domain is even more confusing. The finding that a naturally occurring type I human keratin, K19, which has a very short tail of only 13 amino acids (Bader et al., 1986; Stasiak et al., 1989), can still form a normal filamentous network with a type II keratin argues for the dispensable role of the tail domain in co-assembly. From a study using retrovirus-mediated expression of paired human type I and type II keratin mutants in NIH3T3 cells, Lu and Lane (1990) concluded that stable filament formation requires the presence of both N- and C-terminal domains on at least one of the two interacting keratins. In other words, two tail-less keratins might coassemble but could not form filaments.

For IFs other than keratins, it has been shown that most of the C-terminal end of two neurofilament proteins, NF-M and NF-L, can be deleted without affecting their ability to coassemble with either vimentin or wild-type NF-L in transfected cells (Ching and Liem, 1993; Lee et al., 1993). Nonetheless, deletion of more than 90% of the tail domains of either NF-M or NF-L, especially when deletion involves the highly conserved C-terminal end of the rod domain, results in disruption of the endogenous vimentin or the stably transfected NF-L network (Chin et al., 1991; Wong and Cleveland, 1990; Gill et al., 1990). Therefore, in the case of neurofilaments, not all of the tail domain can be deleted without affecting their ability to co-assemble with a full-length IF.

Purified tail-less Xenopus K8 and human K19 co-assembled in vitro into normal-looking filaments at pH 7.0 (Hatzfeld and Weber, 1990). In addition, a transfection study using an expression vector driven by the human β-actin promoter to express various combinations of human keratins in murine 3T3-L1 cells showed that both K19 and the tail-less K18 mutant were capable of forming normal filamentous networks with the tail-less K8 mutant (Bader et al., 1991). The results of these in vitro and in vivo assembly studies suggest that the tail domain is not only dispensable for co-assembly with a fulllength IF but also for co-assembly with another tail-less IF, which is contradictory to the findings of Lu and Lane (1990).

Although, unlike keratins and neurofilaments, type III IFs normally form homopolymers, they can readily form heteropolymers with vimentin. By transfecting different hamster desmin mutants into SV40-transformed hamster lens cells and HeLa S3 cells, van den Heuvel et al. (1987) demonstrated that the desmin C-terminal tail is dispensable and can be exchanged with the tail domain of hamster vimentin without affecting its ability to co-assemble with endogenous vimentin and form a normal filamentous network.

An in vitro self-assembly study by Quinlan et al. (1989), using Escherichia coli-expressed mouse glial fibrillary acidic protein (GFAP) showed that the complete absence of the C-terminal tail is not compatible with filament self-assembly even if the highly conserved C-terminal end of the rod domain is preserved. In contrast, Eckelt et al. (1992) showed that tail-less Xenopus vimentin can form normal-looking filaments in an in vitro assembly experiment although abnormal structures, including short fibrillar structures and spheroidal aggregates, were observed when this tail-less vimentin was expressed de novo in vimentin(−) BMGE+H cells. Similar aggregates were also observed when completely tail-less desmin was transfected in MCF 7 cells, which contain keratins but no other cytoplasmic IFs (Raats et al., 1991). However, McCormick et al. (1993) came to a different conclusion with tail-less human vimentin, which forms a filamentous network in MCF 7 cells albeit atypically concentrated at the protruding edges of the cell. Therefore, it seems that the role of the tail domain is still far from clear in type III IFs.

In an effort to clarify the roles of the head and tail domains of type III IFs, we constructed several N- and C-terminal deletion mutants of GFAP and transfected them into vimentin (+) or (−) cell lines to study their co-assembly and selfassembly properties. Our results show that only the C-terminal one fifth of the tail domain of GFAP can be deleted without affecting its self-assembly and an intact head domain is required for normal self-assembly. For efficient co-assembly with vimentin, the intact rod domain is essential.

A tail-less GFAP fragment was obtained while we were sequencing rat GFAP cDNA, which we isolated from a rat brain cDNA library (Chen and Liem, unpublished data). After digestion with EcoRI to isolate this fragment from the M13mp18 vector, this fragment was blunt-ended with the large Klenow fragment of E. coli DNA polymerase I (New England BioLab, Inc., Beverly, MA). HndIII linkers (BMB, Indianapolis, IN) were attached to both ends of this fragment before it was inserted into the unique HindIII site in the eukaryotic expression vector, pRSVi-HindIII (a kind gift from Drs B. Forman and H. Samuels, New York University, Medical Center, New York, NY) (Chin and Liem, 1989). The sequences of the resulting expression vector (RSV-rGF-N1100), and all the other mutants described below were confirmed by the dideoxy chain termination method described by Sanger et al. (1980).

The entire protein-encoding region of human GFAP cDNA was synthesized by polymerase chain reaction (PCR) (Perkin-Elmers) using mRNA isolated from a human astrocytoma cell line, U251MG (Westermark, 1973), as template. The 5′- and 3′-end primers, hGF-5H and hGF-3T (Table 1), respectively, were obtained from Operon Technologies Inc. (Alameda, CA). The resulting human GFAP (hGF) cDNA was digested with HindIII and cloned in pRSVi-HindIII. Using the resulting construct, RSV-hGF, as template and a set of primers, hGF-5H and hGF-3R, that corresponds to the 3′-terminal sequence of rod domain (Table 1), we synthesized a tail-less human GFAP, hGF-5H3R (Fig. 1) by PCR. This fragment was then cloned into the unique HindIII site of pRSVi-HindIII and became RSV-hGF-5H3R. The hGF-3T primer and another primer corresponding to the 5′-end of the rod domain, hGF-5R (Table 1), were used to synthesize head-less human GFAP, hGF-5R3T (Fig. 1), by PCR. Since the 5′-primer we designed contains an in-frame ATG and a HindIII site, we were able to clone this fragment into the HindIII site of pRSVi-HindIII and express this head-less human GFAP in eukaryotic cells.

A 1.16 kb HindIII-PstI fragment was isolated from RSV-hGF and blunt-ended with T4 DNA polymerase (New England Biolab, Inc., Beverly, MA). This fragment was then ligated to pRSVi-HindIII vector after phosphorylated HindIII linkers, CAAGCTTG, were attached to its blunt ends to form RSV-hGF-N1166 (Fig. 1). Similarly, a 1.26 kb HindIII-MseI fragment was isolated from RSV-hGF, blunt-ended with Klenow fragment of E. coli DNA polymerase, attached with HindIII linkers, CAAGCTTG, and ligated with pRSVi-HindIII vector to form RSV-hGF-N1265 (Fig. 1).

To prepare RSV-hGF-1134/1234, we used primer hGF-5H and a 30mer, hGF-RDG (Table 1). hGF-RDG is a chimeric primer with 15 nucleotides at its 3′-end complementary to the C-terminal end of the rod domain, i.e. nucleotides 1120–1134, and 5′ sequence complementary to nucleotides 1234–1248 of human GFAP cDNA. These primers were used first to synthesize a double-stranded intermediate template consisting of the head and rod domains of human GFAP through nucleotide 1134 plus an extension of 15 nucleotides corresponding to nucleotides 1234–1248 (Fig. 2). This intermediate template was then denatured and hybridized with denatured head-less human GFAP, RSV-hGF-5R3T. The resulting hetero-hybrid was amplified with PCR using primers hGF-5H and hGF-3T. After digestion with HindIII, the amplified products were cloned in pRSVi-HindIII and sequenced to confirm the deletion of the segment between the C-terminal end of rod and nucleotide 1234.

The head-less rat GFAP, RSV-T7-rGF-148C (Fig. 1), was created by modifying the AccI site at nucleotide position 148. The Klenow fragment of E. coli DNA polymerase I was used to blunt-end the AccI-HindIII fragment of pRSVi-GFA (Chen and Liem, unpublished data). After NcoI linkers, CCCATGGG, had been attached to both its ends by T4 DNA ligase (BMB), this fragment was inserted in the,NcoI site of pRSVi-T7-NcoI (Chin et al., 1991). The ATG in the NcoI recognition sequence provided the translation initiation codon for the expression of this head-less rat GFAP in eukaryotic cells. To construct RSV-T7-rGF-97C (Fig. 1), pRSVi-GFA was linearized with KpnI, blunt-ended with T4 DNA polymerase and ligated with phosphorylated NcoI linkers, GCCATGGC. Since there is a NcoI site at nucleotide position 2055 of pRSVi-GFA, a 1.958 kb NCOI-NCOI fragment was isolated after overdigestion with excess NcoI and subsequently inserted into the unique NcoI site of pRSVi-T7-NcoI vector. RSV-T7-rGF-71C (Fig. 1) was made by ligating the NcoI-digested pRSVi-T7-NcoI with the 1.984 kb NcoI-NcoI fragment of pRSVi-GFA after attaching phosphorylated NcoI linkers, AGCCATGGCT, to its Klenow-blunted ends.

A 1.93 kb /Aid II-.NcoI fragment of pRSVi-GFA was blunt-ended with Klenow fragment of E. coli DNA polymerase and ligated with the large fragment of NcoI-digested pRSVi-GFA, which had been blunt-ended with Klenow enzyme to form RSV-rGF-71/126 (Fig. 1).

SW13.Cl2.Vim⁻ (SW13.Vm⁻) and SW13.Cl1.Vim⁺ (SW13.Vm⁺) (Sama et al., 1990) were kind gifts from Dr Robert M. Evans (University of Colorado Health Sciences Center, Denver, CO). These cell lines were grown in monolayer culture in a 1:1 (v/v) mixture of Ham’s F12:Dulbecco’s MEM (DMEM/F12, GIBCO BRL, Gathersburg, MD) containing 5% fetal bovine serum (FBS) 200 units/ml penicillin, 200 μqg/ml streptomycin sulfate and kept at 37°C in 7% CO₂.

All the transfections in this study were done by the calcium phosphate precipitation procedure (Graham and van der Eb, 1973) with a 15% DMSO shock suggested by Parker and Stark (1979). Confluent cultures were split 36 hours prior to transfection and replated, about 2.5×10⁶ cells per plate, on 100 mm tissue culture plates (Becton Dickinson and Comp., Lincoln Park, NJ). For immunofluorescence studies, 0.05 mg/ml poly-L-lysine-coated coverslips were placed on the bottom of tissue culture plate before adding cells to them. Unless otherwise specified, 30 L g of DNA per plate was used for transfection and cells were exposed overnight to the DNA-calcium phosphate complex and shocked with 15% DMSO for 2 minutes the next morning.

Unless otherwise specified, all the immunocytochemistry experiments were performed on 12 mm circular coverslips (VWR Scientific Inc., San Francisco, CA). Cells were fixed in ice-cold acetone for 3 minutes and then washed with calcium- and magnesium-free phosphate buffered saline (CM-PBS). Non-specific binding was blocked by incubation with 5% FBS, diluted in CM-PBS, for 30 minutes at room temperature. The blocking solution was replaced with polyclonal rabbit anti-GFAP antibody (pABGF) (Wang et al., 1984), diluted, 1:200, in CM-PBS and monoclonal anti-vimentin antibody (mABVm) (Sigma, St Louis, MO), diluted 1:150, and incubated for 1 hour at room temperature. In most cases fluorescein (FITC)-conjugated goat anti-rabbit immunoglobulin (Cappel, Durham, NC), diluted 1:200, and rhodamine (TRITC)-conjugated goat anti-mouse immunoglobulin (Cappel), diluted 1:200, in CM-PBS, were used as secondary antibodies and incubated for 30 minutes at room temperature. For some experiments, we used TRITC-conjugated goat anti-rabbit antibody and FITC-conjugated goat anti-mouse antibody instead. These coverslips were then mounted in 16% polyvinyl alcohol (Sigma), 33% glycerol in CM-PBS and examined with a Nikon Optiphot microscope equipped for epifluorescence. Confocal microscopy study was done with a Zeiss Axiovert 35 microscope equipped with a Bio-Rad MRC-600 confocal attachment.

A 30 or 50 μ g sample of DNA was used to transfect a 100 mm plate of 80% confluent cells. Detergent-soluble and-insoluble proteins were recovered 55 to 60 hours after transfection following the procedures described by Ching and Liem (1993). Briefly, cells were lysed with lysis buffer containing 1% Triton X-100, 150 mM NaCl, 1 mM EDTA, 10 mM Tris-HCl (pH 7.5) and various proteinase inhibitors including 0.5 mM PMSF, 10 μqg/ml aprotinin and 2 μg/ml leupeptin. The lysates were then centrifuged at 430,000 g for 1 hour at 4°C. The pellets were first digested with 0.2 mg/ml DNase I, at 4°C for 15 minutes, then washed with buffer containing 10 mM Tris-HCl (pH 7.5), 150 mM NaCl and the three proteinase inhibitors. The resulting pellets were dissolved in SDS-sample buffer. Proportional amounts of the detergent-soluble and-insoluble fractions were electrophoresed in 10% SDS-PAGE and subsequently transferred to nitrocellulose papers for western blotting. The blots were first incubated in 5% BSA/PBS for 5 hours and then shift to pAbGF/PBS (1:500), at 4°C, overnight. ¹²⁵I-labelled Protein A (NEN, Wilmington, DE) in a dilution of 1:1000 was added to the blots after they were washed several times in 0.1% Triton X-100/PBS. After a one-hour incubation at room temperature, the blots were washed in 0.1% Triton X-100 overnight at room temperature. The resulting blots were exposed to XAR-5 film (Eastman-Kodak Co., Rochester, NY) for various times with or without enhancement at −80°C. Densitometric scanning was done using an Apple OneScanner with Ofoto scanning software (Apple Comp. Inc., Cupertino, CA) and analyzed by the NIH Image 1.43 Gel Analysis Program (National Institute of Health, Bethesda, MD).

To study the role of the tail domain of GFAP in de novo filament formation without the interference of pre-existing IFs, we used a human adrenocarcinoma-derived cell line, SW13.Cl2.Vim⁻ (SW13.Vm⁻), in which no known cytoplasmic IFs normally exist, as the host cell. When the wild-type human GFAP expression vector, RSV-hGF, was transfected into SW13.Vm⁻ cells, a delicate filamentous network was always observed (Fig. 3A), regardless of the amount of DNA we used for transfection and the time elapsed after the transfection.

To see if tail-less GFAP would behave the same way, we first examined a clone of tail-less GFAP that was obtained while we were sequencing rat GFAP cDNA (Chen and Liem, unpublished data). This clone was modified as described in Materials and Methods and inserted into the unique Hindlll site of pRSVi-HndIII vector. In addition to missing the entire C-terminal tail domain of the resulting construct, RSV-rGF-N1100 (Fig. 1), the last five amino acids of its α-helical domain were also changed into KAEAC, instead of the highly conserved, wild-type LLEGE (Fig. 1). Transient transfection of RSV-rGF-N1100 into Sw13.Vm⁻ resulted in a distinct immunofluorescence pattern after staining with polyclonal anti-GFAP antiserum. The transfected cells displayed dots or aggregates of various sizes and shapes (Fig. 3B). The distribution of these dots in most of the cells was both intranuclear and cytoplasmic, although some cells seemed to have predominantly, if not solely, intranuclear dots while others seemed to be mainly cytoplasmic. These dots or aggregates were noticed as early as 18 hours after transfection and in all three levels, 10, 30 and 50 μg, of DNA that we used for transfection. we could not find a wild-type filamentous network in any of the transfected cells under all the conditions we tried.

We then transfected RSV-rGF-N1100 into SW13.Cl1.Vim⁺ (SW13.Vm⁺), a cytoplasmic vimentin-positive variant of SW13 cells, to see if this tail-less construct can co-assemble with other type III IFs, such as vimentin. Although filamentous co-assembly of the tail-less rat GFAP with endogenous vimentin can be found in a few cells, dots or aggregates were found in nearly all transfected cells (Fig. 4A). The distribution of these aggregates included both the nuclear region and the periphery of cells. There is co-local-ization of anti-GFAP and anti-vimentin staining at these dots (Fig. 4E), suggesting the ability of this tail-less GFAP mutant to interact with and disrupt the endogenous vimentin filaments. However, the disruption of endogenous vimentin was not efficient, since some of the transfected cells preserved a filamentous vimentin network despite the presence of these dots (not shown).

The C-terminal end of coil 2B contains a highly conserved motif (Conway and Parry, 1988; Hatzfeld and Weber, 1991; Geisler and Weber, 1982). Deletion mutagenesis and point mutations involving this motif have substantiated the indispensable role of this motif in filament assembly, including vimentin (McCormick et al., 1993) and desmin (Raats et al., 1991, 1992). In addition to the deletion of the entire tail domain, the last few amino acids of RSV-rGF-N1100 had also been altered from the highly conserved KLLEGEE sequence in the C terminus of the α-helical rod domain. Therefore, a tailless fragment of human GFAP that preserved the KLLEGEE sequence was synthesized by PCR and cloned into the pRSVi-Hindlll vector. We used human GFAP as template for PCR rather than rat GFAP, since we were transfecting these constructs into human cell lines and we wanted to rule out the possibility that there may be unknown species differences between human and rat GFAP, which could confuse the results. This construct, RSV-hGF-5H3R (Fig. 1), was transfected into SW13.Vm⁻ (Fig. 3C) and SW13.Vm⁺ (Fig. 4B). The results were similar to those of the transfection studies using RSV-rGF-N1100 in SW13.vim⁻ cells, i.e. all transfected cells showed polymorphous aggregates that were either intranuclear or cytoplasmic. The transient transfections of SW13.Vm⁺ with RSV-hGF-5H3R produced results different from those of RSV-rGF-N1100. In about half of the transfected cell, the expressed tail-less, KLLEGEE-preserved human GFAP coassembled with endogenous vimentin and formed normal filamentous networks (Fig. 4B and F). Dots or amorphous aggregates disrupting the endogenous vimentin network were found in the other half of the transfected cells (not shown). Occasionally we found the preservation of the endogenous vimentin network despite the co-existence of amorphous anti-GFAP positive aggregates (Fig. 5A).

As mentioned above, in a few of the RSV-rGF-N1100- or RSV-hGF-5H3R-transfected cells, these aggregates or dots were restricted to the nuclear area. To clarify whether these dots are intranuclear or peri-nuclear, we examined these cells with confocal microscopy. As shown in Fig. 5B, these aggregates are distributed randomly in the nucleus. There was no obvious pattern of distribution and neither was there any definite association with the nuclear membrane.

We were interested in determining how much of the tail domain of GFAP would be necessary to form cytoplasmic filaments in the absence of vimentin. On the basis of the findings of Albers and Fuchs (1987) and the existence of two naturally occurring, so-called ‘tail-less’ keratins in which there are 13 to 15 amino acid extensions beyond the C-terminal end of the rod domain, we designed a C-terminal deletion that had a 16 amino acid extension beyond the end of the rod domain. This mutant, RSV-hGF-N1166, formed cytoplasmic, thick ribbon-like structures when expressed in SW13.Vm⁻ cells (Fig. 3D). There were no aggregates or dots and all the ribbon-like structures were clearly cytoplasmic. These ribbon-like structures were often found to surround the nucleus, as illustrated in Fig. 3D.

RSV-hGF-N1166 co-assembled with the endogenous vimentin to form a normal filamentous network when it was transfected into SW13.Vm⁺ cells (Fig. 4C and G). We did not find ribbon-like structures or disruption of the vimentin network.

The finding that an extension of 16 amino acids beyond the rod domain did not lead to normal self-assembly in vivo prompted us to design another deletion mutant that preserved a larger portion of the tail. We decided to focus on the RDG motif close to the C-terminal end of the tail, which is highly conserved among all type III intermediate filaments, keratin 8 and keratin 18 (Krauss and Franke, 1990; Franke, 1987; Leonard et al., 1988). Certain point mutations in this motif have been demonstrated to have significant deleterious effects in vimentin filament network formation (McCormick et al., 1993). We therefore designed a deletion mutant that ends right after this motif. The resulting RSV-hGF-N1265 (Fig. 1) formed a delicate filamentous network similar to the wild-type human GFAP when expressed in SW13.Vm⁻ cells. Some of the transfected cells still showed focal coils or thick filaments, which may imply that the absence of the last one-fifth of the tail domain has a slight effect on the efficiency of RSV-hGF-N1265 in forming a normal filamentous network (Fig. 3E).

To see if the inclusion of this RDG motif is sufficient for normal self-assembly, we created another mutant in which an intervening sequence between the C-terminal end of the rod domain and the start of RDG motif had been deleted, and the RDG-containing motif along with the rest of the tail domain was brought next to the C-terminal end of the rod domain. The resulting construct RSV-hGF-1134/1234 (Figs 1 and 2) behaved like RSV-hGF-N1166 when transfected into SW13.Vm⁻ cells and formed thick, ribbon-like, cytoplasmic filaments (Fig. 3F). Clearly, the presence of the RDG-contain-ing motif is not sufficient for normal self-assembly of GFAP. Similar to the results with RSV-hGF-N1166, RSV-hGF-1134/1234 co-assembled with endogenous vimentin and mostly formed a normal filamentous network when expressed in the SW13.Vm⁺ cells (not shown). Occasionally, we observed transfected cells with mainly thick filaments that were stained positive for both GFAP and vimentin (Fig. 4D and H). These thick filaments were not as thick as the ribbons in RSV-hGF-1134/1234-transfected SW13.Vm⁻ cells and were longer and more curvilinear, suggesting a balance between the disrupting effects of the mutant protein and the wild-type, filamentous network of vimentin.

Western blot analysis of whole cell extracts of these C-terminal deletion mutant-transfected SW13.Vm⁻ cells showed that each mutant encoded a protein of the expected size (Fig. 6). There is a correlation between their ability to form filaments and their relative proportion in the Triton-insoluble part. RSV-hGF-N1265 and wild-type rat GFAP can self-assemble into normal filaments and most of their protein stayed in the Tritoninsoluble fraction while all the other C-terminal deletion mutants encoded more soluble proteins (Table 2). Degradation products were noticed for most C-terminal deletion mutants except RSV-hGF-N1265, implying a possible mechanism in preventing protein degradation by forming a normal filamentous network. Similar degradation products were also reported for some of the C-terminal deletion mutants of desmin (Raats et al., 1991).

Since there is a convenient AccI site near the begining of the rod domain of GFAP, as defined by Geisler and Weber (1983), we used this site to clone this head-less fragment into the unique AcoI site of the pRSVi-T7-NcoI vector. This construct, RSV-T7-rGF-148C (Fig. 1), was first transfected into SW13.Vm⁻ cells to study its ability to self-assemble. In striking contrast to the results of wild-type rat GFAP, which forms a curvilinear filamentous network in SW13.Vm⁻ cells (Fig. 7A), the expression of this head-less rat GFAP in SW13.Vm⁻ resulted in diffuse, fine, cytoplasmic staining (Fig. 7B). It is impossible to tell at the light microscopic level whether this diffuse staining is due to a collection of very small aggregates or tiny, short filaments. Western blot analysis of the whole cell extract of RSV-T7-rGF-148C-transfected cells showed very little Triton-insoluble GFAP (Fig. 6 and Table 2), implying that most of the expressed GFAP existed as oligomeric or lower structures. This diffuse, amorphous material characteristically occupied the entire cytoplasm, but never encroached on the nuclear region. These cells, however, were quite healthy, since we noticed pairs of transfected cells apposing each other, suggesting that they were daughter cells from a recent replication (Fig. 7B).

Several phenotypes were observed when RSV-T7-rGF-148C was transfected into SW13.Vm⁺ cells. At one end of the spectrum was the diffuse, fine, cytoplasmic staining similar to that observed in SW13.Vm⁻ cells. There was no significant coassembly of head-less GFAP with the endogenous vimentin in these cells (not shown). At the other end of the spectrum, we observed filamentous staining co-localized with the endogenous vimentin network (not shown). Fig. 8A and E illustrate an example of a phenotype in between these two extremes. The expressed head-less GFAP did not show any filament formation (Fig. 8A) and seemed to be partially disrupting the endogenous vimentin network (Fig. 8E). Apparently, this headless rat GFAP was able to co-assemble with the endogenous vimentin and even disrupt its filament formation, albeit fairly inefficiently.

The head-less fragment in RSV-T7-rGF-148C has in addition a few amino acids missing from the N terminus of the rod domain of GFAP (Geisler and Weber, 1983; Reeves et al., 1989). Therefore, we synthesized another head-less fragment that included the entire rod and tail domains by PCR, using human GFAP cDNA as template (see Materials and Methods). Similar to the results for RSV-T7-rGF-148C, the resulting construct, RSV-hGF-5R3T (Fig. 1), also showed diffuse, fine, cytoplasmic structures when transfected into SW13.Vm⁻ (Fig. 7C). The expression of this construct in SW13.Vm⁺ cell also led to two major phenotypes: diffuse cytoplasmic staining and filamentous staining reminiscent of the endogenous vimentin network. Unlike the results of RSV-T7-rGF-148C, there were more cells with filamentous staining (Fig. 8B and F) in the RSV-hGF-5R3T-transfected SW13.Vm⁺ cells.

In an effort to find out how much of the head domain can be deleted without affecting the ability of GFAP to self-assemble into a normal filamentous network, we transfected RSV-T7-rGF-97C and RSV-T7-rGF-71C (Fig. 1), which preserved about one-fourth and one-half of the head domain, respectively, into SW13.Vm⁻ cells. These two mutants behaved in very much the same way as RSV-rGF-148C and RSV-hGF-5R3T in SW13.Vm⁻ cells (Fig. 7D and E). Diffusely distributed, fine, amorphous, cytoplasmic structures were the common phenotypes in SW13.Vm⁻ cells. In contrast to RSV-T7-rGF-148C and RSV-hGF-5R3T, both RSV-T7-rGF-97C and RSV-T7-rGF-71C co-assembled well with the endogenous vimentin, since most of the transfected SW13.Vm⁺ cells showed a filamentous network co-staining with GFAP and vimentin (for example, Fig. 8C and 8G). Another difference was the occasional big clumps of cytoplasmic aggregates noticed in RSV-rGF-71C-transfected SW13.Vm⁻ cells (Fig. 7E).

There is a nonapeptide in the N-terminal half of the head domain that is highly conserved among type III intermediate filaments, except GFAP (Herrmann et al., 1992). However, GFAP does have a serine- and arginine-rich region in the corresponding position of the head domain and an oligopeptide corresponding to this region, residues 3–19, affected the in vitro assembly of GFAP isolated from bovine spinal cord (Nakamura et al., 1992). To test the significance of this region, we designed a mutant in which this region was moved closer to the start of the rod domain by deletion of the intervening sequence. The resulting construct, RSV-rGF-71/126 (Fig. 1), formed fine, amorphous, cytoplasmic structures indistinguishable from the other N-terminal deletion constructs, when transfected into the SW13.Vm^- cells (Fig. 7F). Occasional large clumps of aggregates were also noticed. RSV-rGF-71/126 coassembled well with the endogenous vimentin without significant disruption of the endogenous vimentin network (Fig. 8D and 8H).

The western blots of whole cell extracts of these N-terminal deletion mutant-transfected SW13.Vm⁻ cells confirmed our immunocytochemical findings. Proteins encoded by these mutants predominantly existed in the Triton-soluble part (Fig. 6 and Table 2). Although these results did not rule out the possibility that there were tetramers or octamers, they certainly precluded the conclusion that these mutants were capable of forming wild-type filaments in SW13.Vm⁻ cells.

In this study, we examined the role of the non-helical end domains in GFAP assembly. Our results with GFAP argue against the assumption that the complete absence of the tail domain is compatible with self-assembly, although the most C-terminal one-fifth of the tail domain can be deleted without affecting its ability to self-assemble. In addition, a preserved KLLEGEE sequence at the C-terminal end of the rod domain is required for GFAP to co-polymerize efficiently with en dogenous vimentin in SW13.Vm⁺ cells. Our results with completely tail-less GFAP agree with the in vitro assembly studies using purified fusion protein in which 32 residues of λcII protein were placed in front of mouse GFAP (Quinlan et al., 1989) and the in vivo transfection studies of tail-less vimentin by Eckelt et al. (1992), except that we did not notice any short fibrillar structures. Transfections of completely tail-less hamster desmin into MCF 7 cells, which express various keratins but no vimentin, also led to cytoplasmic aggregate formation (Raats et al., 1991). In contrast to our results and those of Eckelt et al. (1992), tail-less desmin does not form intranuclear aggregates.

We cannot provide an explanation for the discrepancies between the behavior of our completely tail-less GFAP in SW13.Vim⁻ cells and the normal filamentous network made of tail-less vimentin in the MCF 7 cells reported by McCormick et al. (1993). The possible interfering role of the endogenous keratin network in the MCF 7 cells that were used in their transfection study is unlikely, since we observed the same dots and aggregates in transfected MCF 7 cells with both RSV-rGF-N1100 and RSV-hGF-5H3R (Chen and Liem, unpublished results).

A study using tail-less Xenopus vimentin, which was created by engineering a translational stop codon right after the isoleucine of the conserved sequence KLLEGEENRI, showed that tail-less vimentin can co-assemble with endogenous vimentin without noticeable perturbation of the filamentous network (Eckelt et al., 1992). The RSV-hGF-5H3R that we used in this experiment ended before the asparagine (Fig. 1) and is, therefore, three amino acids shorter than this tail-less construct. When expressed in vimentin-positive cells, the tailless GFAP formed aggregates in part of the transfected cells even though it co-assembled with vimentin. Similar aggregates were observed when tail-less desmin in which the entire KLLEGEENRI sequence was preserved was expressed in HeLa cells (Raats et al., 1991). Therefore, the observed differences in assembly characteristics may be due to the relative ratio between the endogenous vimentin and the expressed tailless mutant rather than the presence of the three additional amino acids or the preservation of the fairly conserved arginine (Conway and Parry, 1988; Geisler and Weber, 1982; Hatzfeld and Weber, 1991).

The conserved RDG-containing sequence in the tail domain of type III IFs, KTIETRDGE, has been demonstrated to play a role in type III IF filament assembly. Certain point mutations in this motif have been shown to result in aberrant vimentin filaments in both in vitro assembly and in vivo transfection studies (McCormick et al., 1993). A synthetic nonapeptide corresponding to this motif has been reported to bind to the last 27 amino acids of desmin and to affect the packing of desmin in assembly in vitro (Birkenberger and Ip, 1990). This motif has also been implicated to be site-specifically associated with the highly conserved C-terminal motif of the rod domain by an anti-idiotypic antibody approach (Kouklis et al., 1991). A desmin mutant in which the intervening sequence between the C terminus of the rod domain and the start of the RDG-con-taining motif has been deleted forms a normal filamentous network when expressed in MCF 7 cells (Raats et al., 1991). It seems therefore that the RDG-containing motif is necessary and sufficient for desmin filament formation. In contrast, our results with a similarly designed GFAP mutant indicate that the RDG-containing motif may be necessary but certainly not sufficient for self-assembly of GFAP. Furthermore, although both tail-less GFAP and desmin form amorphous aggregates in the transfected vimentin-negative cells, the distribution of aggregates of the tail-less GFAP were both intranuclear and cytoplasmic while those of tail-less desmin were only cytoplasmic. The different types of behavior would lead us to conclude that there may be intrinsic differences between these filament proteins that lead to dissimilar requirements of the RDG-containing motif.

In the transfection studies with some of the tail-less GFAP constructs, i.e. RSV-rGF-N1100 or RSV-hGF-5H3R, we noticed that the dots or punctate staining were located intranu-clearly in the transfected cells. Similar intranuclear aggregates were observed with tail-less vimentin-transfected BMGE+H cells (Eckelt et al., 1992) and some C-terminal deletion mutant of NF-L-transfected mouse L cells (Gill et al., 1990). However, in the case of the NF-L mutant, these intranuclear dots did not stain with anti-vimentin antibody and there was deformity of the nucleus. No intranuclear filamentous structures as described by Bader et al. (1991) in their experiments with tailless keratin were observed in any of our transfected cells. Since we were able to find pairs of transfected cells that were occasionally mirror images of each other, we believe that the intranuclear aggregates did not inhibit cell replication and were the end results of certain repeatable processes instead of an intermediate phase of a random process.

Why would these cytoplasmic IFs end up in the nucleus after C-terminal truncation? The well-known nuclear localization signal of lamins (McKeon, 1991; Silver, 1991), VTKKRKLE, which prevents ectopic lamin filament formation by efficiently transporting newly synthesized lamins into the nucleus, does not exist in the GFAP sequence. Entrapment of these cytoplasmic aggregates into the nucleus during mitosis could be considered, but the fact that we can find pairs of cells with predominantly cytoplasmic aggregates (not shown) and pairs of cells with exclusively intranuclear aggregates does not favor this mechanism. A cytoplasmic retention signal as proposed for protein kinase Cα (Schmitz et al., 1991) or the precursor of NF-кB (Parker and Stark, 1979) is unlikely, since the tail domains of IFs are highly variable. Nevertheless, there remains the possibility that there could be more than one type of retention signal. Our results with RSV-hGF-N1166 and RSV-hGF-1134/1234 favor the possibility that these tail-less mutants are transported into the nucleus by a mechanism such as bulk flow because of their inability to assemble into filaments and are subsequently trapped inside the nucleus. In contrast, if these mutant products are able to form any filamentous structures they would be retained in the cytoplasm.

In vitro assembly experiments with thrombin-treated chicken desmin have demonstrated that desmin fragments lacking only the basic and arginine-rich headpiece cannot selfassemble and cannot be incorporated into filaments unless they are mixed with full-length desmin prior to the stage of protofilament formation (Kaufmann et al., 1985). In vivo transfection studies with head-less IFs in cells that contain no co-polymer-izable IFs support the indispensable role of the head domain in the self-assembly of IFs (this study; and Raats et al., 1990). However, the role of the head domain in co-assembly with other IFs has not yet been settled. In one study, an N-terminal deletion mutant of NF-L lacking the entire head domain plus 10 amino acids of coil 1a was still capable of co-assembly with vimentin in Ltk⁻ cells (Chin et al., 1991). In contrast, a similar transfection study using mouse L cells concluded that a deletion of more than 90% of the head domain of NF-L resulted in non-filamentous, cytoplasmic aggregates that did not contain vimentin (Gill et al., 1990). The importance of the head domain in co-assembly seems to vary with different IFs. Truncation of 75% of the NF-M head domain fails to complement NF-L assembly when they are co-transfected into SW13.Vm⁻ cells (Lee et al., 1993), although it can co-assemble with vimentin in mouse L cells (Wong and Cleveland, 1990). Our results with completely and partially head-less GFAP in co-assembly experiments with vimentin agree with those obtained with desmin (Raats et al., 1990). A priori, N-terminal deletion mutants of both GFAP and desmin can always co-assemble with vimentin and more extensive deletions of the head domain result in more disruption of the endogenous vimentin networks.

Analyses of the sequences of the head and tail domains of IF have shown that there are subdomains, termed H1 and H2, immediately adjacent to the begining and the end of the rod domain, respectively, that share a high degree of sequence homology among members of the same type of IF (Steinert and Roop, 1988; Conway and Parry, 1988). Competition experiments with synthetic peptides have suggested that the H1 (and possibly H2) subdomain sequences unique to type II keratins play pivotal roles in the registration of neighboring molecules at the earliest steps of keratin filament assembly (Steinert and Parry, 1993). The H1 subdomain in type III IFs is about 20 residues long (Conway and Parry, 1988) and is preserved in the partially head-less mutant RSV-T7-rGF-71C. If the H1 subdomain in type III IF plays the same role in filament assembly as in keratins, the diffuse-staining phenotype in the head-less mutants that do not preserve the H1 subdomain would be predicted (such as RSV-T7-rGF-148C, RSV-hGF-5R3T,RSV-T7-rGF-97C and RSV-rGF-71/126). However, the finding that head-less mutants preserving the H1 subdomain (i.e. RSV-T7-rGF-71C) also showed diffuse staining in SW13.Vim⁻ cells implies that more than the H1 subdomain is required at the early steps of type III IF filament assembly.

A stretch of nine amino acids, SSYRRIFGG, in the early segment of the non-helical head domain of vimentin (Herrmann et al., 1992) and desmin (Raats et al., 1990) has been shown to be important for filament formation in vitro (Herrmann et al., 1992) and in vivo (Raats et al., 1990). When deletion involved the SSYRRIFGG sequence of desmin, the transfected MCF 7 cells that expressed no endogenous vimentin started to show cytoplasmic aggregates or dots. As the deletion involved more than 60% of the head domain, a characteristic staining pattern similar to what we observed from expressing N-terminal mutants in SW13.Vm⁻ cells was found. Since GFAP does not have the SSYRRIFGG sequence, we studied the importance of a comparable serine- and arginine-rich region at its N terminus in filament assembly. Deletion of this entire sequence (RSV-T7-rGF-71C) prevented self-assembly of the mutant GFAP, indicating that this sequence may function like the SSYRRIFGG region in desmin. To determine if this sequence is sufficient for filament formation, we constructed a mutant protein in which this sequence was brought closer to the rod domain by deleting 20 intervening amino acids. This mutant protein was also unable to assemble in the absence of vimentin. These results are consistent with an in vitro study on Xenopus vimentin in which the SSYRRIFGG sequence was brought closer to the rod domain by deletion of 55 intervening amino acids, resulting in a mutant protein incapable of forming normal filaments (Herrmann et al., 1992).

Several of our deletion mutants, including the two completely tail-less mutants and all the N-terminal deletion mutants, did not form any discernible filamentous structures in SW13.Vm^- cells at the resolution of light microscope. Western blot analysis of whole cell extracts of the corresponding transiently transfected cells showed that there was more protein in the Triton X-100-soluble fraction. Although we have not performed EM studies on these aberrant structures, these results suggest that they were oligomers or aggregates of them. The EM studies on similar structures of similarly designed mutants of either desmin (Raats et al., 1992) or vimentin (Eckelt et al., 1992) have demonstrated that these dots or diffuse cytoplasmic stainings were made of amorphous structures, which support our conclusion that these mutants could not form filaments.

In summary, an intact rod domain with preserved KLLEGEE sequence at its C-terminal end is necessary for GFAP to co-assemble with vimentin and up to one-fifth of the C-terminal end of the tail domain can be deleted without affecting its self-assembly ability. Our results with the N-terminal deletion mutants do not support the presence of any dispensable segment in the head domain of GFAP for its selfassembly. Both the RDG-containing motif in the tail domain and the Ser- and Arg-rich motif in the head domain may be essential but are certainly not sufficient for self-assembly.

We thank Dr K. Dunn (Department of Pathology, Columbia University) for helping us with confocal microscopic studies of the intranuclear aggregates. This work was supported by NIH grant NS 29224.