Observation of keratin particles showing fast bidirectional movement colocalized with microtubules

Over the past quarter of a century, analysis of the substructure of cells has been built substantially on the use of specific antibodies. However,immunofluorescence labelling of cell components requires the cell to be chemically or physically damaged so that cell structures can be exposed to the antibodies; it also causes the unavoidable generation of artifacts. The use of green fluorescent protein (GFP) hybrid constructs now enables live-cell imaging of specific components, offering a new perspective of cell structures and processes.

Keratins are members of the intermediate filament family of proteins that form an intricate network running through the cytoplasm of eukaryotic cells,and, together with microfilaments and microtubules, they make up the cytoskeleton. They are characterized by strikingly tissue-specific expression patterns from early embryogenesis onwards, suggesting that these major structural proteins are important in defining tissue structure and potential function.

Understanding the assembly and maintenance of intermediate filament networks in vivo is still a challenge to cell biology. Intermediate filament assembly in vitro is not energy dependent and does not require any associated proteins. A consensus hypothetical assembly sequence is now emerging from many laboratories, on the basis of in vitro assembly studies(Steinert et al., 1993a; Steinert et al., 1993b; Herrmann et al., 1999), as follows: monomeric subunits first dimerize in parallel and in register; dimers then assemble into one of the several possible antiparallel (mostly staggered)alignments, giving rise to tetramers, which are the major soluble subunits identifiable in vivo (Quinlan et al.,1984; Soellner et al.,1985; Chou et al.,1993). In vitro, tetramers associate laterally into unit length filaments (ULFs) of around 60 nm long and 16 nm wide, which then anneal longitudinally to form loosely packed immature filaments. Further filament maturation then occurs without change in mass per length as filaments become more compact, reducing their diameter to about 10 nm(Herrmann and Aebi, 1998; Herrmann et al., 1999). All of these stages have been deduced from tracking in vitro polymerization, but so far they have not all been identified in living cells.

The identification of mutations in keratin genes is helping to define the structural role of intermediate filaments as an increasing number of human diseases are being linked to mutations in keratins and other intermediate filaments – see, for example, The Human Intermediate Filament Database(http://www.interfil.org). Pathogenic mutations appear to compromise the filament network assembly characteristics and may therefore shed some light on the function of specific subdomains. In particular, mutations in keratins, the epithelial intermediate filament structural proteins, have been shown to cause a range of genodermatoses (for reviews, see Irvine and McLean, 1999; Coulombe and Omary, 2002). All intermediate filament proteins share a similar structure consisting of a central α-helical rod domain and N- (head) and C-terminal (tail) domains, and most pathogenic keratin mutations lie in the conserved peptide motifs located at either end of the rod domain. This concurs with in vitro experiments showing that these helix boundary sequences are important for efficient polymerization and that mutations here will often interfere with dimer formation or impair dimer stability(Herrmann et al., 2000; Wu et al., 2000; Wang et al., 2000a; Mehrani et al., 2001). In addition to the rod end mutation `hot-spots', mutation cluster sites also occur in other places: in the head domain adjacent to the amino terminal end of the rod domain, in the non-helical linker domain in the middle of the rod domain, and slightly further inside the rod ends. These secondary cluster sites, however, are typically associated with milder disease phenotypes than those associated with the hot-spots of the helix boundary motif sites. The differences in disease severity associated with different mutation sites on the keratin proteins is most apparent in the case of keratin (K5) and K14(Lane, 1994), although there are some exceptions to this (Liovic et al., 2001). K5 and K14 are the major intermediate filament components of epithelial basal cells throughout the body and are important structural components of the epidermis. Mutations in K5 and K14 cause epidermal basal cell lysis after mild trauma to the skin (epidermolysis bullosa simplex, or EBS). Severe cases of EBS, the so-called Dowling–Meara form of the disease, show clusters of blisters at any site in the body, and are associated with mutations in the helix boundary peptides of K5 or K14. Mutations outside the helix boundary peptides in K5 or K14 are generally associated with milder (Weber-Cockayne) EBS forms, in which blistering is limited to hands and feet. The Dowling-Meara mutations, but not Weber-Cockayne mutations, cause substantial abnormalities in keratin assembly in vivo, and electron-dense keratin aggregates are visible by electron microscopy in some Dowling-Meara epidermal cells(Anton-Lamprecht and Schnyder,1982; Haneke and Anton-Lamprecht, 1982). These aggregates are diagnostic of this form of EBS and were the structures in which keratin abnormalities were first shown (Ishida-Yamamoto et al.,1991). The nature and origin of these aggregates in relation to keratin assembly processes is unclear, as is their role in the development of EBS lesions. There is therefore a need for closer analysis of keratin assembly and disassembly dynamics in normal and mutant cells so that new routes to intervention therapy for these skin fragility diseases can be identified.

The overall emerging picture of intermediate filament function has thus been one of stable filament systems, which provide physical resilience to cells. All the more striking, therefore, are several recent studies on live cells using GFP-tagged intermediate filament proteins vimentin(Prahlad et al., 1998; Yoon et al., 1998; Ho et al., 1998) or keratins K8, K18 or K13 (Windoffer and Leube,1999; Strnad et al.,2001; Yoon et al.,2001; Windoffer and Leube,2001), which have revealed these filament networks to be flexible and dynamic structures. The first papers using this type of analysis of intermediate filaments showed that filament movement appears to be at least partly dependent on an intact microtubule filament system(Prahlad et al., 1998; Yoon et al., 1998; Windoffer and Leube, 1999). Flux in the system was detectable as continuous lateral subunit exchange along the filament length, taking place between the filaments and a `soluble'protein pool (Yoon et al.,1998; Yoon et al.,2001). Significant differences were observed between vimentin and keratin intermediate filaments. For example, the subunit exchange rate calculated from fluorescence recovery after photobleaching (FRAP) experiments is almost 20 times faster for vimentin than for keratin intermediate filaments(Yoon et al., 2001). Some studies have also described large fragments of filaments (squiggles), which may constitute part of the assembly process(Prahlad et al., 1998; Yoon et al., 2001).

In this paper we present observations from a live-cell-imaging study of epithelial cells transfected with a fluorescently tagged keratin construct[enhanced green fluorescent protein (EGFP)-K5] and grown as confluent monolayer cultures. We show that keratin in these cells exists not only as intermediate filaments but also in the form of small dynamic particles. Keratin particles are found throughout the cell. Their movement is aligned with microtubules and inhibited by nocodazole, and it can be both plus-end-and minus-end-directed. A population of particulate keratin structures can also be detected in untransfected cells using scanning and transmission electron microscopy (EM) combined with immunogold labelling. The data suggest that keratin particles may be a constitutive part of keratin intermediate filament network dynamics, representing an as yet uncharacterized stage in keratin filament turnover in living cells.

The full-length wild-type K5 cDNA was re-cloned from a pcDNA3 vector (clone provided by C.S. Shemanko) into pEGFP-C2 (Clontech, Basingstoke, UK) using EcoRI and HindIII restriction enzymes (EGFP-K5 clone provided by D. Gibbs). The mutant pEGFP-K5 N176S construct was made using the Quik-Change Site-Directed Mutagenesis kit (Stratagene, Amsterdam, Netherlands)and the following forward and reverse primers: pN176Sf 5′-GCAGATCAAGACCCTCAGCAATAAGTTTGCCTCC-3′ and pN176Sr 5′-GGAGGCAAACTTATTGCTGAGGGTCTTGATCTG-3′. The enhanced yellow fluorescent protein (EYFP)–tubulin construct was obtained by cloning a PCR-amplified full-length cDNA sequence of α-tubulin into the XhoI and BamHI sites of the pEYFP–C1 vector(Clontech). Before use, the newly generated constructs were transfected into epithelial cells, fixed 3 days after transfection and stained to ensure that the transfected tubulin and keratin were being expressed, integrated and colocalized with the endogenous microtubule or keratin filament network (data not shown).

Immortalized human keratinocytes (NEB-1)(Morley et al., 1995) were cultured in Dulbecco's modified Eagle's medium (DMEM) with 25% Ham's F12 medium, 10% fetal calf serum (FCS) and growth factors hydrocortisone (0.4μg/ml), cholera toxin (10^–10 M), transferrin (5 μg/ml),lyothyronine (2×10^–11 M), adenine(1.9×10^–4 M) and insulin (5 μg/ml). PtK2 cells(potoroo kidney simple epithelial cells) were cultured in a 1:1 mixture of DMEM and Ham's F12 growth medium supplemented with 10% FCS and 1% L-glutamine. Human lens cells (H36CE1) were cultured in DMEM supplemented with 10% FCS. All cell lines were grown (without feeder cells) at 37°C and 5%CO₂, either directly in plastic tissue culture grade dishes or on 13 mm or 40 mm diameter coverslips (BDH, Lutterworth, UK) for transfection experiments. Cells for transfection were cultured on coverslips until 50%confluent, then transfected using the FuGene 6 (Boehringer Mannheim, Lewes,UK) transfection reagent (according to the manufacturer's instructions) plus 0.5-1 μg of plasmid DNA per 2 ml of serum-free medium. In double-transfection experiments, both plasmid DNAs were mixed at 1:1 ratio and incubated with the FuGene 6 reagent. Cells were incubated with the transfection reagent for 24 hours at 37°C, after which the culture medium was replaced with fresh medium and cells were further incubated for another 48 hours. By 3 days after transfection, cells grown on the coverslips were confluent and the newly expressed fluorescent proteins from the transfections were fully incorporated into the cytoskeleton.

The K14pcDNA3 vector (clone provided by C. S. Shemanko), encoding full-length wild-type keratin 14, was mixed in a 1:1 ratio with either the EGFP-K5 or the wild-type K5pcDNA3 construct. The mixture was then used to transfect (as above) human lens cells grown on 13 mm coverslips. Any keratin intermediate filament network formed as a result of this transfection must be formed de novo, as lens cells do not express keratins. Cells were fixed with methanol/acetone (1:1) at 1, 2 and 3 days post transfection. The K5/K14 pcDNA3 expression products were detected by immunofluorescence, using as primary antibody mouse monoclonal anti-K14 (LL001)(Purkis et al., 1990) with FITC-labelled sheep anti-mouse serum (Sigma) as the secondary antibody.

Confluent coverslips of live cells were assembled in a perfusion open-closed chamber (POC chamber, H. Saur, Reutlingen, Germany). A heater was used to keep the chamber temperature constant at 37°C during imaging. Images (512×512 dpi, 2×2 binning) were collected using a 100× 1.4 NA oil immersion objective on a Nikon inverted TE200 Eclipse epifluorescence microscope equipped with a fully motorized Z stage (Applied Precision, Issacruach, USA) and linked to a Micromax CCD camera (Roper Scientific, Trenton, USA), part of the DeltaVision imaging system (Applied Precision). Stacks of optical sections spaced 200-500 nm apart were collected at each time-point (20-30 second time-lapse intervals) and subsequently deconvolved and analysed using softWorX software (Applied Precision). The softWorX `edge enhance' tool was used for analysing images acquired in double-labelling experiments. For Figs 4 and 9, stacks of optical sections were spaced 100 nm apart. In the case of Fig. 4/Movie 5, the time lapse between time-points was shortened to 5 seconds.

Cells expressing the fluorescent constructs were selected with the `mark and visit' tool of softWorX 2.50. Before the treatment, cells were imaged for 5 minutes with a 20 or 30 second time lapse. To depolymerize microtubules, the cell culture medium in the perfusion chamber was supplemented with nocodazole to a final concentration of 10 μg/ml. In heat-shock experiments, the temperature in the POC chamber was raised to 45°C and maintained constant for 15 minutes. The temperature was then lowered back to 37°C. In both cases (nocodazole treatment and heat shock) images were taken at fixed time-points 30 minutes, 1 hour and 2 hours after the treatment.

PtK2 cells grown on 13 mm coverslips were fixed in a solution containing 4%paraformaldehyde, 0.1% glutaraldehyde and 0.3% Triton X-100 in cytoskeleton buffer (CB) at pH 6.1 (CB buffer contains 10 mM MES, 150 mM NaCl, 5 mM EGTA, 5 mM MgCl₂, 5 mM glucose). This fixative has previously been successfully used for EM studies (Small,1988; Mies et al.,1998). Cells were fixed for 15 minutes and washed with CB buffer(3×10 minutes). Sodium borohydride (500 μg/ml in CB buffer,incubations on ice, 3×10 minutes), followed by washing in CB buffer, was used to quench background staining due to glutaraldehyde. Primary antibodies used were a mouse monoclonal antibody against α-tubulin (Amersham,Little Chalfont, UK) and CAM5.2 (Becton-Dickinson, Oxford, UK), a mouse monoclonal antibody recognizing K7/K8. The incubations were carried for 1 hour at room temperature, or overnight at 4°C. After extensive washing with PBS(6×10 minutes if overnight incubation), FITC-labelled sheep-anti mouse immunoglobulin was applied as secondary antibody (1 hour incubation). Coverslips were then washed again with PBS and mounted with Hydromount(National Diagnostics, Hessle, UK) supplemented with Dabco (2.5%) as antibleaching reagent. Actin filaments were stained with rhodamine-labelled phalloidin (Molecular Probes, Leiden, Netherlands).

The same fixative as above was used to prepare untransfected PtK2 cells for scanning and transmission EM with immunogold labelling. Cells were grown either on small squares of silicone wafer coated with L-polylysine (for scanning EM) or on CELLocate coverslips (Eppendorf) with a 175 μm square(for transmission EM). For keratin immunolocalization, cells were incubated overnight at 4°C in CAM5.2 antibody, washed in PBS (6×10 minutes),incubated for 4 hours with rabbit anti-mouse secondary antibody, washed again with PBS (5×10 minutes) and incubated overnight with a Protein A–8 nm gold conjugate. After a series of washes with PBS (4×10 minutes),cells were postfixed with 2% glutaraldehyde in PBS (30 minutes) and washed with PBS (2×15 minutes).

For scanning EM, cells were then dehydrated through graded alcohols and critical-point dried, coated with 2 nm of chromium (Cr) and viewed with a Hitachi S-4700 field emission scanning electron microscope. Images were acquired at a running voltage of 15 kV using the SE (secondary electrons) and YAGBSE (backscattered electrons) detectors.

For transmission EM, coverslips with immunostained cells were fixed for 30 minutes in 1% osmium tetroxide (OsO₄) solution, washed in PBS and dehydrated through graded alcohols, and soaked in propylene oxide (2×15 minutes) before overnight infiltration in Durcopan resin (Sigma) and propylene oxide (1:1) followed by Durcopan resin. Coverslips were then inverted onto gelatine capsules filled with resin and left to polymerize at 40°C. At this point, coverslips were snapped off the capsules, exposing the cell monolayer at the surface of the resin blocks. Sections of about 70 μm thickness were mounted onto copper grids and stained with 1% uranyl acetate and lead citrate before viewing in a JEOL 1200 EX transmission electron microscope equipped with the Fuji FDL5000 digital plate reader.

To ascertain the ability of the transfected EGFP-K5 protein to form a functional keratin intermediate filament network, two sets of lens epithelial cells (H36CE1) were transfected in parallel with either the EGFP-K5pcDNA3 and K14pcDNA3, or K5pcDNA3 and K14pcDNA3, constructs. Lens cells do not otherwise express keratins, so any formation of a keratin filament network would have to be de novo. Fig. 1 shows representative images of keratin filament formation in lens cells transfected with EGFP-K5/K14pcDNA3 constructs at 1-3 days after transfection, in comparison with cells transfected with keratin without EGFP. In both cases filaments began forming around the nucleus by day 1, and proceeded to extend into a keratin filament meshwork by days 2-3. The EGFP-K5 transfectants appeared a little slower at extending their networks than the K5 without EGFP transfectants, but the networks were indistinguishable by day 3. Thus, the EGFP-K5 protein is capable of forming an extended keratin network with K14 in the absence of any wild-type K5.

When keratinocytes (NEB-1 cells) were transfected with the EGFP-K5 construct, fluorescent keratin was produced which assembled into filaments in the endogenous filament network. Other than by their fluorescence, the filaments and the filament network were indistinguishable from those in untransfected cells. Transfected cells were filmed when they reached confluence, three days after transfection. We were able to confirm the characteristic undulatory motion of keratin filament bundles as reported by others (Windoffer and Leube,1999). Fluorescent keratin was also detected as small, highly mobile particles, which hitherto had not been detected by immunofluorescence. We could distinguish between a smaller, highly dynamic class of keratin particles (referred to henceforth in the text as fast or F particles), and a second, larger and less mobile class of particles that appeared to be associated with the keratin filament network (referred to hereafter as slow,static or S particles). To ascertain that these particles were not an artifact of EGFP per se, cells were also transfected with EGFP alone. The expressed EGFP fluorophore produced a diffuse fluorescent staining all over the cytoplasm, but no aggregates or particulate structures were observed.

Although scattered throughout the cell, the keratin particles were small and encountered infrequently; in many cases their presence was only detectable by analysing the high-resolution movies. The behaviour of keratin F particles was therefore tracked and analysed by time-lapse filming. Fast F particles were seen to move around the cell with a saltatory type of movement, as shown in Fig. 2 (Movie 1) and Fig. 3 (Movie 2). They moved in a non-random way, following trajectories with a characteristic zig-zag appearance, and alternating between leaps and pauses during which their direction may have changed. The size of keratin particles was determined to be between 30 and 70 nm by measuring them with the `measure distances' tool from softWorX and dividing this value by a correction factor of 6 to account for the resolution limitations of an epifluorescence microscope. The correction factor was calculated by comparing the difference between the similarly determined diameter of microtubules (labelled with the EYFP–tubulin construct) and their known actual diameter of about 25 nm. The speed of individual leaps of keratin particles was calculated to be greater than 100 nm/s (n=16, calculated by dividing the distance travelled by the particle in a single leap by the time elapsed between the frames), suggesting the involvement of motor proteins. The length of the pauses appeared to be random, as they varied from as little as 15 seconds to over 3 minutes (see Fig. 3).

Another feature of F particle movement is the close association of the particles with keratin intermediate filament bundles during the `pause'periods. Although the movement of the F particles does not appear to occur along keratin filaments, nevertheless the keratin network is used as a transient stop to the particles during the saltatory movement. This was particularly visible not only in Fig. 3 (Movie 2), but also, to a certain extent, in Fig. 2 (Movie 1).

By contrast, the S particles moved in a more restricted way. They appeared to be tethered, or linked, to the keratin filament network, and moved only slightly with an irregular oscillating type of motion around a fixed point(see Movie 2).

Keratinocytes (NEB-1 cells) expressing the EGFP-K5 construct were treated with the microtubule-depolymerizing drug nocodazole (10 μg/ml). Fig. 4 shows the dynamic movement of a keratin F particle before the addition of nocodazole to the cell culture medium. In the corresponding movie (Movie 3), several other F particles and a few S particles are also visible. After the drug is added to the cell culture medium (Fig. 4E-F; Movie 4), the wave-like movement of keratin intermediate filaments slows down (compare Movies 3 (before) and 4 (after the addition of nocodazole)), and only S keratin particles remain visible. These results suggest that the saltatory motion of the F particles is dependent on intact microtubules, and the speed at which they normally move suggests the involvement of microtubule motor proteins.

To visualize both keratin and tubulin cytoskeletal components in vivo simultaneously, PtK2 cells were transfected with an EGFP-K5 keratin and an EYFP-tubulin construct. Acquired images(Fig. 5) were deconvolved and subsequently further processed using the edge enhancement algorithm of the softWorX image analysis software, which presents both keratin intermediate filaments and microtubules as tubular rather than linear structures. This significantly improved the visualization of keratin particles in images obtained from double-transfection experiments. Fig. 5 (Movie 5) shows two keratin particles (labelled green, see arrows) apparently following an identical course and briefly stopping at the same `pause' points. Their trajectory is colocalized (yellow signal) with the course of a microtubule (in red) that is situated slightly outside the focal plane of the keratin particle. In other cases the particles also exhibited changes in the direction of their translocation through the cell, following zig-zag courses superimposed over intersecting lengths of microtubule tracks (data not shown). This reinforces the possibility that F particles are using motor proteins running on microtubule tracks to achieve fast dispersal through the cell.

Imaging of keratin particles was constantly hindered by their low numbers and high speed. Previous work in this laboratory has shown that filaments incorporating pathogenic mutant keratins can be selectively destabilized by a transient shift in temperature (Morley et al., 1995). We hypothesized that this phenomenon could be used to increase the non-filamentous pool of keratin in the cell and thus possibly to amplify the number of particles, which would facilitate their observation. We selected a K5 mutation with a strong negative effect, N176S, and engineered an EGFP construct with the mutation. This K5 N176S mutation was previously identified in a patient with a severe, Dowling-Meara, form of epidermolysis bullosa simplex (Stephens et al.,1997). PtK2 cells were then transfected with either the EGFP-K5 or EGFP-K5 N176S construct. Both the wild-type(Fig. 6A) and the mutant K5 constructs (Fig. 6C) were incorporated into the endogenous keratin network. When transfected cells were subjected to heat shock, cells expressing the mutant keratin gradually developed changes in their phenotype up to a peak by 1 hour after their exposure to a thermal stress (15 minutes at 45°C), showing a significant increase in the number of visible keratin particles(Fig. 6D; Movie 6). Here,again, the two types of keratin particles – fast-moving F particles and slower S particles – were distinguishable (see Movie 6). The S particles changed their positions only slightly, shifting with an oscillatory type of motion around a central point and during longer time intervals. Cells expressing the wild-type keratin construct remained unaffected and showed no increase in particle numbers (Fig. 6B).

To determine the directionality of the movement of the F keratin particles,we used the centrosome as a reference point for minus end anchorage of microtubules. In double-labelling experiments cells were transfected with the EGFP-K5 (or EGFP-K5 N176S) and EYFP–tubulin constructs, and keratin particle number was boosted by a 15 minute heat shock (at 45°C) before imaging.

Figs 7 and 8 show keratin particles moving along microtubule tracks. In Fig. 7, a keratin particle positioned at the periphery of the centrosome area attaches to a microtubule(Fig. 7B) and slides along it away from the centrosome (Fig. 7C-D; see Movie 7). In Fig. 8, keratin particles are seen moving towards the centrosome,following microtubule tracks (see Movie 8). Thus, the movement of keratin F particles can occur in both directions (plus-end and minus-end directed),implying the involvement of more than one motor protein.

Our results suggest that keratin particles are able to interact both with the keratin filament and the microtubule networks. We reasoned that we should also be able to detect keratin particles in fixed cells, if a fixative that allows good structural preservation of all three cytoskeletal components(keratin filaments, microtubules and microfilaments) was used. Different fixation procedures were tried, and a solution containing paraformaldehyde,glutaraldehyde and Triton X-100 (Small,1988; Mies et al.,1998) gave by far the best results with the antibodies used. Using this fixation schedule the keratin filaments acquired a grainy or particulate quality (see Fig. 9A,B).

As this fixation procedure has been previously used for transmission EM(Small, 1988), we examined these keratin filaments by electron microscopy of untransfected PtK2 cells that had been immunogold-labelled for K8 before resin embedding. As the fixative contains a small amount of anionic detergent (Triton-X 100) to permeabilize the cells, we used this procedure to fix and label keratin filaments for scanning EM.

The cytoskeleton could then be visualized through the resulting holes in the membrane. The results are shown in Fig. 10. Both scanning and transmission EM show small localized thickenings, as well as clear particles attached to the keratin filaments, both of which clearly contain keratin. The particles are globular in shape and have a size range of about 30-60 nm,consistent with the F particles seen in living cells.

We have obtained evidence that non-filamentous keratin protein is rapidly translocated throughout epithelial cell cytoplasm in the form of small, fast keratin particles (F particles). We have recorded these particles using GFP-tagged keratin K5, in both keratinocytes (which naturally express K5 with K14) and in simple epithelial cells (in which the K5 is supernumerary). Similar keratin particles of the same order of magnitude in size can also be detected in cells that have not been transfected with EGFP-K5, suggesting that these structures are a natural feature of keratin biology in epithelial cells and not only an artifact of overexpression of keratin or of defective behaviour of EGFP fusion proteins.

These small particles have not been described previously. In both previous reports dealing with keratin intermediate filament dynamics, spreading subconfluent cells were used. Yoon et al.(Yoon et al., 2001) compared keratin K8/K18 and vimentin intermediate filaments. Although the authors used PtK2 cells (as we did), fast keratin particles were not reported. Instead,larger short filament fragments located at the cell periphery (squiggles) were seen – structures that had also been identified previously in vimentin networks (Yoon et al., 1998). Windoffer et al. (Windoffer et al., 1999) looked at A431 cells and, by tracking EGFP-K13, they observed structures formed by the constant fusing,extending and branching of keratin filaments during their inwardly oriented wave-like translocation. Fast-moving keratin particles were not reported in this study either, although after careful examination we could see several of them in one of their published time-lapse films(Windoffer and Leube, 1999)(Movie 3, `without nocodazole'). The fact that the A431 cells used in that study were stably transfected with EGFP-K13 also illustrates that the keratin particles are not a short-term artifact of transfection.

Working with confluent cell cultures enabled us to observe two distinct types of keratin particles: F particles, which are very dynamic and have a saltatory pattern of movement, and S particles, which appear to be tethered to the keratin filaments and have only a limited, oscillatory motion. The movement of the F particles consists of alternating fast `leaps' with interspersed `pauses' of variable length, resulting in a characteristic zig-zag trajectory through the cytoplasm. This motion bears a striking similarity to fast axonal transport, the translocation of neurofilaments along axons (Wang et al., 2000b). Transport of the F particles occurs at high speed and is aligned with microtubules. Fast-moving particles in interphase cells have been described for vimentin intermediate filaments(Prahlad et al., 1998) and for neurofilaments (Prahlad et al.,2000). Prahlad et al. (Prahlad et al., 1998) looked at vimentin particles in spreading BHK-21 cells: the particles were most pronounced at the cell periphery, where they were seen to fuse into vimentin `squiggles' and short fibrils. In our experiments, `squiggles' were not observed. As we were looking at the cytoskeleton in cultured cells grown as confluent monolayers, cell translocation was minimal. It is possible that the intermediate filament(keratin and vimentin) structures referred to as `squiggles' are specifically associated with bulk transport or rearrangements of the intermediate filament protein network around the cell edge in spreading or mobile cells.

Both our findings (for keratin) and those for vimentin and neurofilaments(cited above) show that fast transport of intermediate filament particles depends on, and is associated with, microtubules. The vimentin particles were seen in early stages of spreading and were suggested to interact with the microtubule plus-end-directed motor protein kinesin(Prahlad et al., 1998). Most recently, Helfand et al. (Helfand et al.,2002) have suggested that some structural forms of vimentin can also be associated with subunits of dynein and dynactin, implying that bidirectional transport of vimentin is possible. Our results on keratin intermediate filaments concur with those on vimentin, as we have shown that the keratin F particles are able to move both towards (minus-end directed) and away from (plus-end directed) the microtubule organizing centre of the centrosome. Neurofilament particles have been shown to be associated with at least one type of kinesin motor protein in vivo(Prahlad et al., 2000), but bidirectionality in their movement is yet to be shown.

Keratin filaments may have an active role in this bidirectional transport of keratin particles. In our time-lapse films, filaments appear to function as anchor points during the `pause' phase, a detail previously unreported for other intermediate filament fragments and particles. The zig-zag trajectory also seems to be an intrinsic characteristic of the keratin F particles. One possible model accounting for the observed movement would be that keratin particles `hitchhike' on microtubule motors to traverse micrometre-long distances in a matter of seconds, and the keratin filaments are used as temporary `anchoring' points while waiting for motor protein(s) to pass close enough for the F particles to adhere to them. The dense microtubule network would also enable the keratin particles to switch between different microtubule tracks, thus giving rise to the zig-zag travel pattern. Such movement between two cytoskeleton systems would obviously be dependent on maintaining structural links between the systems, as effected by `cytolinker'proteins such as plectin (Svitkina et al.,1996; Svitkina et al.,1998) and possibly other plakins.

What are these keratin particles and why were they not detected before?Keratin aggregates have often been observed during large-scale intermediate filament network rearrangements in cells. As well as the diagnostic keratin aggregates of EBS, keratin aggregate formation was reported in cells undergoing mitosis (Lane et al.,1982), during the assembly and disassembly of the keratin intermediate network in live dividing cells(Windoffer and Leube, 2001),and during keratin filament breakdown in epithelial cells induced by the treatment with okadaic acid (Strnad et al., 2001) or keratin antibody microinjection(Klymkowsky et al., 1983). Many of these rearrangements require phosphorylation, which destabilizes filaments and drives the equilibrium towards depolymerization, leading to aggregate formation (Inagaki et al.,1996; Omary et al.,1998; Feng et al.,1999; Sahlgren et al.,2001). The use of phosphatase inhibitors on live cells –pervanadate (Windoffer and Leube,2001) and okadaic acid by(Strnad et al., 2001) –has confirmed a correlation between increased keratin phosphorylation levels,and filament breakdown and aggregate formation. Although these previously described keratin aggregates are all much larger structures than the keratin particles we describe here, the aggregates may derive from condensations of smaller particles over time. There are also several technical reasons that can help to explain why they have been overlooked until now. For example, a single still image cannot identify an object that is characterized only by its movement. Additionally, methanol and acetone, the inorganic solvents most often used to fix cells for intermediate filament visualization, are weak fixatives that do not preserve more delicate structures – such as membranes, microfilaments, microtubules and possibly other structures interacting with them – to the same extent. Therefore, small and soluble structures may simply be washed out during processing.

Most of the data on stages of intermediate filament assembly have been derived from in vitro polymerization studies, and their counterparts in live cells still need to be identified. Considering the keratin particle size, the bidirectionality of their dispersal system, the fact that the visible keratin particle pool was significantly increased in stressed cells expressing mutant keratin and that particles were found in nontransfected cells as well, it is highly possible that keratin particles may be a constitutive part of the regulation and maintenance of the keratin intermediate filament network in cells. We are currently limited by available technology in defining their origin and function, but one look at the three-dimensional density of cytoskeleton filaments in the cytoplasm of these cells(Fig. 10A), and the frequency of intersections between them, emphasizes the potential of a system that could disperse small keratin particles rapidly along and across such structures. The small fast keratin particles may be oligomeric assemblies of keratin, which may represent the live-cell counterpart of the `unit length filaments' found in vitro. However, they may also represent hyperphosphorylated oligomers,which are transiently in an unfavourable configuration for incorporation into filaments (Omary et al.,1998), and while some may be destined for degradation(Toivola et al., 2002), others may eventually get incorporated into filaments. Although the size of the particles currently makes it difficult to identify them conclusively by other means than live-cell imaging, this study does emphasize the need for detailed analysis of keratin dynamics so that new routes to intervention therapy for EBS and related skin fragility diseases can be identified.

Very special thanks for their advice, support and the best of company during the long hours of EM work to John James, Calum Thomson, Martin Kierans and Euan James at the Center for High resolution Imaging and Processing in Cell and Molecular Biology (CHIPs, www.dundee.ac.uk/biocentre/chips),School of Life Sciences, University of Dundee. We also thank Paul Andrews and Jason Swedlow in our institute for their assistance with the use of the DeltaVision imaging system, and Pat Ogden and Mariella D'Alessandro for their help with cell culture. We also thank C. S. Shemanko and D. Gibbs for the pcDNA3K5 and PEGFP-K5 constructs, which they generated while in this lab. This work was supported by the Wellcome Trust (055090/A/98/Z, to E.B.L.) and the Ministry of Science and Technology of the Republic of Slovenia(Z3-3310-0381-01, to M.L.).