The keratin-filament cycle of assembly and disassembly

The cytoplasmic cytoskeleton of mammalian cells is composed of three major filament networks – actin filaments, microtubules and intermediate filaments (IFs). This scaffolding is not simply a static system conferring stability on cells, but is highly dynamic and capable of rapid reorganization in response to various extracellular and intracellular stimuli. Adjustment of actin filaments and microtubules is accomplished through differential regulation of polymerization at either end, depending on the availability of nucleoside-triphosphate-bound soluble subunits and regulatory factors. One prominent mechanism of filament remodelling is treadmilling, which exploits the structural asymmetry of filament ends (Amann and Pollard, 2000; Margolis and Wilson, 1981; Pollard et al., 2000). Thus, subunits dissociate from the minus end and are added to the plus end. IFs are profoundly different from actin filaments and microtubules because they lack polarity as a result of the symmetric composition of their tetrameric subunits. Furthermore, spontaneous self-assembly of tetramers into unit-length filaments (ULFs), followed by compaction and longitudinal annealing, occurs – at least in vitro – without nucleoside triphosphates and additional cofactors (Herrmann et al., 2007; Kim and Coulombe, 2007). How these in vitro observations relate to the in vivo situation and how assembly is regulated is not understood.

IFs are ubiquitous cytoskeletal components that are particularly abundant in epithelial cells. Epithelial IFs are composed of desmosome-anchored keratins, providing a mechanically resilient, complex scaffold (Sivaramakrishnan et al., 2008) that responds quickly to various stimuli through specific structural adaptations. Metabolic, thermal and mechanical stress therefore results in considerable keratin IF (KF) reorganization (Magin et al., 2007; Pekny and Lane, 2007). Recent observations further revealed that the keratin system affects many basic cellular processes, such as growth, proliferation, organelle transport, malignant transformation and stress response, further accentuating the dynamic properties of the keratin network (Magin et al., 2007). Thus, interference with KF dynamics in human disease and transgenic mice leads to reduced resilience of epithelia to mechanical and other challenges with deleterious functional consequences (Arin, 2009; Vijayaraj et al., 2007). For this reason, elucidating the still-unresolved molecular mechanisms of KF-network biogenesis and turnover is a pressing issue.

At present, two conceptually different hypotheses are being discussed to explain IF assembly in living cells. The first model, which was recently termed dynamic co-translation, stresses the integration of newly synthesized IF polypeptides at multiple sites throughout the entire network (Chang et al., 2006; Goldman et al., 2008). An alternative model suggests that keratin particles are preferentially generated at distinct loci in the cell periphery, subsequently integrating into the peripheral keratin cytoskeleton (Windoffer et al., 2006; Windoffer et al., 2004). Although both scenarios provide concepts explaining how the network ‘grows’, they do not address the problem of how the fully assembled network changes over time. This is crucial for adaptation to varying environmental conditions, for example, during development and under mechanical stress.

The goal of our study was to fill the existing gaps in these two hypotheses and to evaluate their contribution to KF-network dynamics. Using live-cell imaging, we discovered a protein-biosynthesis-independent multistep disassembly and assembly cycle that is continuously active and allows rapid keratin-network adaptation, for instance in migrating cells.

It was observed by time-lapse fluorescence recording of cell lines producing fluorescent epithelial KFs that KF formation occurs preferentially at free cell edges and adjoining cell borders of stationary cells (Kolsch et al., 2009; Windoffer et al., 2004). After scratch wounding, abundant KF precursors (KFPs) were seen at the leading edge of approaching cells, thereby extending the IF cytoskeleton towards the gap (Fig. 1A; supplementary material Movie 1). This was also noted in single migrating cells (Fig. 1B-E; supplementary material Movie 2). KFPs grow by elongation and fusion until integration into the peripheral KF network (for details, see Windoffer et al., 2004; Woll et al., 2005).

Considering the abundance of keratin particles in the cell periphery of wounded and migrating cells, we wanted to know whether protein biosynthesis is sufficient to account for keratin assembly, as predicted by the dynamic co-translation model (Chang et al., 2006). Treatment with the translation inhibitors cycloheximide and puromycin, however, did not prevent KFP formation (Fig. 2; supplementary material Movies 3 and 4). KFPs can therefore assemble from a pre-existing keratin pool that is either of significant size or continuously replenished.

Newly formed KFPs are continuously transported toward the nucleus, relying primarily on intact actin filaments (Kolsch et al., 2009; Woll et al., 2005). To further track KFs after integration into the peripheral network, image-processing algorithms were developed and applied to motility analyses. First, global cellular movement was compensated for. Then, the medians of the translation vectors were computed for each grid point over time, assuming long-term local stability of keratin movement. This revealed coordinated and continuous centripetal keratin movement all the way to the nucleus at an average speed of ~250 nm/minute, with local heterogeneities between 100 and 600 nm/minutes (Fig. 3; supplementary material Movie 5). The motion stopped upon reaching the juxtanuclear region.

Given the movement of the entire keratin system toward the nucleus, one would expect an incremental build up of keratin fluorescence near the nucleus; this is obviously not the case. To test the hypothesis that KFs disassemble, the fluorescence of selected KF subsets was monitored (Fig. 4; supplementary material Movies 6 and 7). First, regions of interest (ROIs) were defined between prominent peripheral KF bundles. The enclosed KFs translocated continuously towards the nucleus while fusing, generating fewer and thicker bundles (Fig. 4A-F; supplementary material Movie 6). As a result, the ROI size shrank (Fig. 4H). Fluorescence quantification revealed gradual loss that was fast during the first 1-2 hours (~40% loss/hour), affecting the younger peripheral filaments, and slow later (<10% loss/hour), affecting the older thickened filament bundles (Fig. 4I), which eventually stabilize to form the juxtanuclear static cage. This difference is best appreciated when the substantial initial fluorescence loss in the blue, green and pink ROIs is compared with the lower initial loss in the red ROI, which was already more central when measurements began (Fig. 4). The apparent fluorescence loss is not caused by bleaching, because overall cellular fluorescence did not decrease. Furthermore, degradation can also be excluded as a major contributing factor, because the fluorescence decrease in entire cells was less than 5% per hour in the presence of puromycin (n=4).

To directly monitor disassembly of selected filament bundles, short filament pieces were ‘isolated’ by bleaching the surrounding areas (Fig. 4J-N; supplementary material Movie 7). These experiments showed that fluorescence decreased continuously and was not associated with filament fragmentation (i.e. particle formation). We therefore conclude that KFs disintegrate on their way to the nucleus by the release of small, non-filamentous oligomers that cannot be resolved as single particles, but appear as a diffuse weak signal throughout the cytoplasm.

To examine the nature of the non-filamentous pool of disassembled KFs, fluorescence recovery after photobleaching (FRAP) analyses were performed. Filament-free regions were selected within bleached areas to measure the mobile fraction of non-filamentous keratins. The mobile fraction was 52.2±12.4% in PK18-5 cells (n=5). The diffusion coefficient of this mobile fraction was determined to be 0.88±0.08 μm²/second for human keratin 18 (HK18)-YFP fusion protein (n=9). Furthermore, the diffusion coefficient was not significantly altered in the presence of cytochalasin and nocodazole [1.06±0.31 μm²/second for HK18-YFP (n=8); P=0.1052]. In control experiments, the diffusion coefficient for YFP alone was 27.10±7.15 μm²/second (n=20), which is in close agreement with the values for GFP measured by fluorescence correlation spectroscopy (FCS) (Dross et al., 2009). These findings reveal the existence of a mobile and diffusible cytoplasmic keratin pool whose precise composition and multimerization stage, however, still need to be determined by more sensitive and accurate methods, such as FCS.

To demonstrate that assembling peripheral KFs are made from material that is derived from disassembled KFs, peripheral cytoplasmic zones of HK18-YFP-producing cells were bleached (Fig. 5A). The mean fluorescence recovery in the peripheral cytoplasm was 72±24% after 15 minutes (n=4). During the same time interval, 7.3±2.8% of the fluorescence was lost in the central, non-bleached region, presumably due to translocation of non-filamentous material to the cell periphery.

Next, an entire cell half was bleached (Fig. 5B). The appearance of fluorescence predominantly in the periphery of the bleached half and its subsequent movement to the cell interior are clearly seen in supplementary material Movie 8. Fluorescence intensities were measured within and outside of the bleached areas, and also in the neighbouring unbleached control cells. As expected, fluorescence increased gradually in the bleached red ROI, with ~60% recovery after 1 hour (Fig. 5B). The recovery rate in this ROI is slower than that determined in the previous experiment (Fig. 5A), because of its more central location and the lower availability of material due to the much more extended bleaching. Remarkably, fluorescence in the unbleached part of the cell decreased during the 60 minute observation period and approached levels similar to those measured in the bleached area, thus providing evidence of continuing KF-network turnover. In comparison, the fluorescence intensity in a corresponding ROI of the adjacent control cell remained almost the same, except for a slight increase that is most probably due to de novo protein biosynthesis (see also Fig. 4I).

Photoactivation experiments were designed to obtain positive proof that filament-bound keratin molecules are released and used for reassembly of new filaments without the need for de novo biosynthesis. After photoactivating human keratin 8 (HK8) hybrids in entire cell halves, new fluorescence appeared in the non-activated parts, the source of which can only be the filament-bound photoactivated molecules from the other cell half (Fig. 5C-D). Within 4 hours, an almost even distribution was noted throughout the cytoplasm (not shown). This is in rough agreement with the bleaching experiments, in which complete exchange of the KF network was observed within 4 hours (supplementary material Movie 8) (data not shown). Furthermore, photoactivated filaments concentrated around the nucleus, whereas inward-moving filaments re-appeared in the cell periphery (see Fig. 5D for quantifications). In support of this, photoactivating only perinuclear filaments resulted in the ‘spreading’ of fluorescence to the cell periphery (Fig. 5E-F).

Combining the current observations with previous reports on keratin dynamics, key steps of the KF assembly and disassembly cycle have begun to emerge.

Nucleation of KFPs occurs preferentially in the cell periphery (Fig. 1 and Fig. 5B; supplementary material Movies 1, 2, 8) (Windoffer et al., 2006; Windoffer et al., 2004; Yoon et al., 2001). Cells with increased KF-network reorganization, such as migrating cells, show a considerable increase in KFP formation (Fig. 1) (Windoffer et al., 2006; Woll et al., 2005). Focal adhesions are important regulators of keratin nucleation (Windoffer et al., 2006) by direct and/or indirect molecular association [(Steinbock et al., 2000); for other IFs, see the following references (Kreis et al., 2005; Nikolic et al., 1996; Spurny et al., 2008; Steinbock et al., 2000; Sun et al., 2008a; Sun et al., 2008b; Tsuruta and Jones, 2003)] and/or by signalling (e.g. Omary et al., 1998; Woll et al., 2007).

KFPs elongate by continuous integration of soluble oligomers at either end and mutual end-on fusion (Fig. 1B-D) (Windoffer et al., 2006; Windoffer et al., 2004; Woll et al., 2005) [for other IFs, see the following references (Colakoglu and Brown, 2009; Kirmse et al., 2007)].

End-on integration of KFPs into the peripheral KF network terminates elongation and thereby introduces branch points into the filament network (Windoffer et al., 2006; Windoffer et al., 2004; Woll et al., 2005).

The thickness of the KF bundles increases towards the nucleus because of the lateral association of filament bundles (Figs 3, 4 and 5; supplementary material Movies 5, 6 and 8). Bundling is regulated by IF-associated polypeptides (Listwan and Rothnagel, 2004) and intrinsic properties of KFs (Lee and Coulombe, 2009).

An intact actin-filament system is required for inward-directed KFP transport. The actin retrograde flow and ruffling activity (Small and Resch, 2005) might provide the driving force for centripetal KF movement (Kolsch et al., 2009; Windoffer et al., 2006; Windoffer and Leube, 2001; Woll et al., 2005). Microtubules might also be needed, especially for the subsequent inward motion of the entire network (for a review, see Helfand et al., 2004).

Inward-moving KFs dissolve without the appearance of distinct particles (Fig. 4) either by degradation of KF polypeptides and/or by disassembly into reusable non-filamentous subunits. Although keratin ubiquitylation has been implicated in KF-network turnover (Ku and Omary, 2000) and occurs in stress and disease (Jaitovich et al., 2008; Zatloukal et al., 2007), the second mechanism appears to be the major mode because pre-existing filaments are the source of protein-biosynthesis-independent peripheral filament formation. Of note, the static juxtanuclear filaments are quite stable, as is the case for desmosome-anchored KFs (Strnad et al., 2002).

A rather small though significant mobile fraction of fluorescent keratins with a fast recovery rate was identified in this study. This diffusible keratin pool corresponds, at least in part, to the detergent-soluble pool, which consists primarily of tetramers and also higher order oligomers (Bachant and Klymkowsky, 1996; Chou et al., 1993; Soellner et al., 1985), as indicated by a diffusion coefficient that was significantly lower than that of YFP alone. The identification of diffusing keratins shows that active transport machinery is not needed to deliver disassembled keratins to the cell periphery for KF reassembly. This is also supported by the finding that KFP formation persists, although at reduced levels, in the presence of actin filament and microtubule disruptors (Kolsch et al., 2009; Woll et al., 2005). The soluble pool is also available for lateral exchange into existing filaments (Miller et al., 1991), which might explain the slight recovery of bleached filaments in our recordings (e.g. supplementary material Movies 7 and 8).

The revealed keratin cycle is different from dynamic co-translation (Chang et al., 2006), because it is independent of protein biosynthesis, does not require microtubules and occurs in a spatially defined manner. Our data suggest that the turnover cycle accounts for the vast majority of newly formed filaments. In contrast to the estimated ~25% hourly turnover rate of the keratin cycle, biosynthesis appears to be less than 5% per hour, as inferred from the overall fluorescence increase (e.g. Fig. 5B; Fig. 3N; data not shown). It will be of interest to determine the different contributions of the keratin cycle and dynamic co-translation mechanisms to network plasticity in relation to IF type, cell specialization and specific situation.

The keratin cycle maintains a stationary network at steady state. Preferential assembly at the leading edge of migrating cells leads to an imbalance and subsequent extension of the network towards the lamellipodium. The self-sufficient keratin cycle can be accelerated or slowed by external factors, depending on specific requirements. In support of this, both assembly and disassembly of wild-type and mutant keratins are slowed by p38 mitogen-activated protein kinase inhibitors (Woll et al., 2007). This observation indicates the presence of a basic mechanism whereby signalling affects network reorganization. It exemplifies how keratin-network cycling can be modulated by various types of stress, most notably mechanical stress. Furthermore, regional regulation of the keratin cycle facilitates network restructuring in specific cell topologies. This is needed, for example, in wounded tissues, in which migrating cells not only have to maintain a functional transcellular network but also need to modulate their network architecture during wound closure.

Preparation of expression vectors pHK18-YFP and pHK8-CFP, which contain cDNA encoding fusion proteins HK18-YFP and HK8-CFP, respectively, has been described previously (Strnad et al., 2002; Windoffer and Leube, 2004). Plasmid clone paGFP-N1, encompassing a CMV-promoter-driven cDNA encoding photoactivatable green fluorescent protein (paGFP), was kindly provided by George Patterson (NIH, Bethesda, MD) (Patterson and Lippincott-Schwartz, 2002). To prepare cDNA encoding HK8-paGFP, HK8 cDNA was excised with BamHI and EcoRI from plasmid pHK8-CFP and inserted into the paGFP-N1 vector.

The human hepatocellular-carcinoma-derived epithelial cell line PLC and the stable HK18-YFP-producing subclone PK18-5 were described previously (Strnad et al., 2002). Spontaneously immortalized mouse mammary epithelial cells of line EpH4 and stable transfectants synthesizing HK18-YFP have been presented elsewhere (Kolsch et al., 2009).

For assessment of drug-induced keratin alterations, cells were seeded in glass-bottom Petri dishes (MatTek) and treated with 17 μM cycloheximide (Windoffer et al., 2000) (Sigma) or 1-2 μg/ml puromycin (Sigma).

Scratch-wounding assays were performed in completely confluent cell monolayers producing homogenously labelled KF networks using a microinjection needle. KF-network reorganization during wound closure was monitored by recording time-lapse fluorescence images.

Cells were imaged by epifluorescence and confocal laser-scanning microscopy as described (Windoffer and Leube, 2004). Conversion of paGFP was done by irradiation with five successive scans of 405 nm light at 100% laser intensity.

Two types of FRAP experiments were performed to either characterize fast processes concerning non-filamentous keratins or examine the comparatively slow KF dynamics.

To determine the mobile fraction of non-filamentous keratins, a Zeiss 710 confocal laser-scanning microscope was used. Square regions of ~8×15 μm were bleached in the peripheral cytoplasm using the 488 nm laser at high intensity. Twenty frames were recorded within the next 10-20 seconds. Within the respective bleached regions, completely filament-free irregularly shaped ROIs of 1-2 μm width were chosen and fluorescence recovery in these ROIs was analyzed with the help of Zeiss Zen software.

To characterize the slower KF dynamics, a Leica TCS SP5 confocal laser-scanning microscope was employed. Complete fluorescence was measured in single cells and stacks of images containing 4096 grey values (12 bit) were recorded with the 488 nm laser excitation line prior to bleaching. Step size was between 300 and 500 nm. For bleaching, the 488 nm laser line was applied to different ROIs at high intensity until the fluorescence was below 10% of the pre-bleach level. Starting directly after bleaching, stacks were recorded at regular intervals using the same parameters as for pre-bleach image recording.

To measure fluorescence intensities in 3D stacks of time series, maximum intensity projections were first generated using ImageProPlus software (Media Cybernetics). The projected images were used for preparing figures and movies. For quantification, ROIs were selected from these projected images. Then, the sum of the grey values was determined in the corresponding ROIs of each image of the z-stack. The values determined for the ROIs in all planes of a stack were then added to calculate the complete 3D fluorescence intensity of a ROI. For comparison of ROI intensities at different time points, it was often necessary to re-define ROIs to account for shifting, size changes and deformation. To correct measurements for background noise, the mean of the grey level per pixel was measured in cell-free regions and subtracted from the grey values determined for pixels measured inside the ROI. Measurements were analyzed with the help of Excel spreadsheets.

To represent movement of KFPs, kymograms were prepared from time-lapse fluorescence recordings. Regions were selected extending from the cell edge to the central cytoplasm. Amira software (Visage Imaging) was used to project the enclosed fluorescence and plot it over time.

Confocal 4D data sets were recorded at 12 bit in cells producing fluorescent keratins. Projections of each time point were used for tracking. ROIs containing a distinct population of filaments were manually defined and Amira software was used to adapt the edges of the chosen ROIs to altered shape in the following frames by visual inspection. From selected time points, the fluorescence intensity of all ROIs was determined by adding their grey values in all 11 planes using ImagePro Plus macros.

In some instances, global cellular movement was compensated by an image-intensity-based registration method (Würflinger et al., 2008) using the normalised sum of squared intensity differences (SSDs). Robustness was increased by embedding the registration in a multiscale Gaussian pyramid framework. Visual assessment of the results revealed no misalignments. Based on the aligned image series, analyses could be restricted to intracellular keratin motion. A z-axis projection of each volume was superimposed by a grid of ROIs at 10-pixel spacing. The overlapping ROIs were of 41-pixel width. Each ROI was rigidly registered to the previous frame by SSD criterion using an exhaustive search for translations of ±20 pixels. SSD was modified by a weighted window function on the ROIs, which dropped linearly from the centre to zero. Thus, the non-rigid character of the data was accounted for and enough structural information was captured by the ROIs, similarly to the work of Wilkie and Vrscay (Wilkie and Vrscay, 2005). Assuming long-term local stability of the keratin flow, the medians of the translation vectors were computed for each grid point over time.

We thank Ursula Wilhelm for expert technical assistance and George Patterson (NIH, Bethesda, USA) for the plasmid encoding paGFP. We also thank Stéphanie Portet (University of Manitoba), Bernd Hoffmann(IBN, Jülich) and Claudia Krusche (RWTH, Aachen) for help analyzing the data. This work was supported by the German Research Council (LE 566/10, WI 731/6-1).