Growth, lifetime, directional movement and myosin-dependent motility of mutant keratin granules in cultured cells

Intermediate filament polypeptides (IFPs) are prominent components of cytoplasmic aggregates, which are pathognomonic for multiple diseases. Recent observations in cultured cells suggest that they are dynamic and subject to regulated turnover. The emerging concept is that multiple factors contribute to motility and turnover of IFP-containing aggregates. To understand their relative contribution, quantitative tools are needed. The current study addresses this need using epithelial cells producing mutant keratin IFPs that have been identified as the cause of the hereditary blister-forming skin disease epidermolysis bullosa simplex. Digital image analysis of individual granules allowed mapping of their complete life cycle, with information on multiple characteristics at any given time-point. The deduced signet features revealed rapid granule fusion and directed transport from the periphery towards the cell centre, and a limited, ~ 30 min lifetime with a slow, continuous growth phase followed by fast disassembly. As paradigmatic proof-of-principle, we demonstrate that inhibition of myosin II selectively reduces granule movement, linking keratin granule motility to retrograde cortical acto-myosin flow. The newly developed methods and established parameters will help in the characterization of known and the identification of novel regulators of IFP-containing aggregates.

Keratin granule fusion occurs randomly but fusion rates correlate with granule size and density. We first wanted to find out, whether position, size or density of keratin granules affect their fusion rate.
62.68% of the 820 granule tracks deduced from Movie 1 were of the merging type. The number of fusions ranged between 1 and 10 ( Fig. 3A). Combining track analyses from 6 recordings of 6 different cells showed that fusion occurred 2.22 ± 0.69-times per track (Fig. 3B). Mapping of the merging events revealed that they occur mainly in the middle and inner granule-bearing region (Fig. 3C). Cells with a high overall fluorescence, reflecting a large number of large granules, had a high rate of granule fusion (Fig. 3D). The latter observation may be due to a higher probability of granule approximation.

Mutant keratin granules have a limited lifetime encompassing a slow growth and rapid dissolution phase.
To find out, whether fusion affected the lifetime of keratin granules, the duration of solo and main merging tracks was compared (Fig. 4A,B). The average lifetime of non-fusing granules turned out to be shorter (18.49 ± 5.92 min) than that of fusing granules (25.82 ± 6.34 min). To understand granule growth kinetics in the absence of fusion, granule fluorescence was examined in non-merging tracks. Movie 5 and Fig. 4C present a prototypical track and the corresponding changes in fluorescence of a single keratin granule. As to be expected, the increase in granule fluorescence corresponded well with the size of the growing granule (Movie 5). Changing fluorescence was therefore taken as a measure of granule growth. The slow initial growth of the nascent granule was followed by an exponential growth phase that levelled off at a distinct maximum. The ensuing disassembly of the granule occurred rapidly with almost linear kinetics. Plotting the positions of initial granule detection together with the positions of last granule detection showed that both events are spatially separated (Fig. 3C). Granule appearance was restricted to the peripheral 2/3 of the granule-bearing zone, granule disappearance to the inner 1/3. Comparing growth and dissolution of granules further demonstrated that granule growth was limited to the peripheral part of the granule-containing zone whereas shrinking occurred only in the inner part of that zone (Fig. 4D).
To see whether the change in fluorescence observed for the selected single granule was representative, normalized granule fluorescence was compared for 137 solo tracks (Fig. 4E). The average granule fluorescence determined for 5 cells, that were imaged with identical settings, varied by a factor of 3. A slow growth phase and a much faster dissolution phase of keratin granules, however, were readily apparent independent of the average fluorescence of granules, whose average lifetime was 25.56 ± 6.22 min (Fig. 4F,G). The scheme summarizes the different analysis steps of keratin granule dynamics beginning with the recording of changing fluorescence patterns in living cells producing fluorescence-tagged EYFP-K14R 125 C, followed by tracking of single granules, classification of tracks and determination of different granule properties at the single granule level and final multimodal statistical evaluation of entire granule cohorts in a given cell. The manuscript further describes how these methods can be used for the analysis of granule dynamics in different cells and its response to specific drugs. The described approach can be extended to examine keratin granule dynamics in genetically modified cells and can be extrapolated to other intermediate filament polypeptides, diseases and cell types.  Fig. S1B). The average local track speed of granules along solo tracks was very similar to that of granules moving along main merging tracks (0.46 ± 0.06 µm/min vs. 0.44 ± 0.08 µm/min; Fig. 5A,B). To obtain a measure for the persistence of movement, we determined the local advance speed, which is defined as the speed determined for the shortest distance between the first and last positon of each 8 frame window ( Supplementary Fig. S1B). As expected, the values were lower than those for the local track speed but, again, did not differ between solo and main tracks (0.35 ± 0.07 µm/min vs. 0.39 ± 0.09 µm/min; Fig. 5C,D).
To determine whether the findings were representative, we examined the track and advance speed of granules in 6 cells (Fig. 5E). The overall values (mean local track speed for solo tracks: 0.39 ± 0.09 µm/min; mean local track speed for main tracks: 0.39 ± 0.08 µm/min; mean local advance speed for solo tracks: 0.28 ± 0.12 µm/min; mean local advance speed for main tracks: 0.27 ± 0.09 µm/min) were slightly different from the single cell data. Closer inspection revealed considerable heterogeneity and indicated the presence of a group with fast-and another group with slow-moving granules. Figure 5F,G depicts colour-coded tracks representing local track speed and local advance speed of main tracks. Both parameters are high during most of the granule's lifetime and slow down considerably at the end. The side-by-side comparison further suggested that the local advance speed rapidly slowed down at the end of granule lifetime and slightly before the local track speed slowed down. In accordance, the ratios of local advance speed to local track speed were stable for most of the granule's lifetime and decreased at the end of its lifetime (Fig. 5H). The phenomenon was also apparent in graphs, in which the fluorescence of granules was normalized to the time point of the highest fluorescence level (Fig. 5I-L). The decrease in granule fluorescence (after time point 0 min) was coincident with a transient increase in local track speed. At the same time the local advance speed was drastically reduced. The graphs in Fig. 5I-L further demonstrate that both local track and local advance speed decreased substantially as granules grew (before time point 0 min).
Together, the observations indicated that mutant keratin granules move directionally during most of their lifetime but lose directionality while still remaining motile just prior to dissolution. Support for this idea was also derived from inspection of single tracks at high resolution, revealing erratic granule movement at the end of granule lifetime (Movies 3, 4). The selected track in Fig. 5M-O (Movie 6) exemplifies the described alterations in local track and advance speed and their changing ratios at the end of granule lifetime.
To show that the relationship between granule speed and granule size is a general property, we determined the ratios between local track speed and granule fluorescence in time-lapse recordings of 5 different cells (Fig. 5P). The results show that increased granule fluorescence correlates with reduced velocity, indicating that granule size and speed are inversely related.
Mutant keratin granules move by directed transport. The above analyses of local track and advance speed indicated that keratin granules move coordinately toward the cell centre. To distinguish between directed, free and restricted mode of mutant keratin granule transport, we performed mean square displacement (MSD) analyses. The coefficient derived therefrom is the alpha (α) value, which indicates free diffusion if it equals 1, whereas a value of α > 1 indicates movement by active transport, and a value of α < 1 points to motion constrained in space 20 . Analysing the data of Movie 1, the alpha value was determined for each time point of each granule and was projected onto the track data. Figure 6A shows that the tracks display alpha values > 1 almost www.nature.com/scientificreports/ over their entire length, indicating directed transport. Shortly before the tracks end, values of < 1 indicated the end of directed transport and restricted diffusion for the last frames instead.
To distinguish between inward movement and movement perpendicular to this axis, the movement along the x-axis (defined as the axis between start-and end-point of this track) and y-axis (perpendicular to the x-axis) were analysed separately (Fig. 6B,C). Movement along the x-axis showed alpha values above 1 for most of the granule lifetime, which became only less than 1 at the end. In contrast, the movement along the y-axis showed alpha values below 1 during most of the granule lifetime.
Together, we concluded that directed movement of granules occurs only in one direction, i.e. centripetally towards the cell interior. The findings further suggested that specific transport mechanisms are likely involved in keratin granule translocation.
Myosin II contributes to the speed and directionality of keratin granule movement. In trying to identify likely candidate motor proteins that contribute to the directed transport of mutant keratin granules, we reasoned that microtubule-dependent motors confer typically much faster transport rates than those observed in our study and therefore focused on actin-dependent myosin motors. A prime candidate is nonmuscle myosin II, which is known to support inward-directed transport processes and has been implicated in the regulation of keratin granule dynamics 21 . To this end, EYFP-K14R 125 C-expressing MCF-7 cells were treated with the myosin II-specific inhibitor para-nitroblebbistatin 22,23 . 30 images were recorded at 20 s intervals prior to drug addition and 50 images immediately afterwards at the same frequency without interruption. In this way, alterations in granule motility could be assessed in the same cells. The recordings were then subjected to the image analyses described above taking only solo tracks into account.  Fig. 7A,B, respectively (Movie 7). The decrease of track lengths after addition of para-nitroblebbistatin was obvious. The track speed changes were then analysed in 17 blebbistatin-treated cells and compared to those in 11 control cells that were only treated with the DMSO solvent. No significant changes were observed in the DMSO-treated controls whereas a highly significant decrease of mean granule speed from 0.33 to 0.22 µm/min was detected in blebbistatin-treated cells (Fig. 7C). Furthermore, the persistence, defined as the shortest distance between start and end point of a track divided by the actual distance travelled, was also significantly reduced from 0.74 to 0.66 after inhibitor addition (Fig. 7D). Averaged MSD curves of tracks in a single cell before and after addition of blebbistatin are depicted in Fig. 7E,F. Both the individual and weighted curves are considerably flattened in the presence of blebbistatin, indicating a decrease in total motility of mutant keratin granules. The alpha value calculated from the MSD analyses, however, did not change in the presence of blebbistatin or DMSO (Fig. 7G), remaining above 1 in all conditions. Similarly, the alpha value calculated along the x-axis was not altered (not shown). But the alpha along the y-axis was significantly, although only to a minor degree, increased after blebbistatin addition (Fig. 7H).
To further substantiate the inhibitor experiments, cells were treated with ML-7 as an alternative upstream myosin inhibitor (Supplementary Fig. S2). Analysis of the speed of granule movement showed a significant reduction from 0.28 to 0.17 µm/min. Together, the observations indicated that inhibition of myosin II significantly slowed down the inward movement of mutant keratin granules.

Movement of mutant keratin granules coincides with cortical acto-myosin flow.
In view of the partial myosin-dependency of keratin granule movement, we compared motion of granules with that of myosin. To this end, MCF-7 cells stably expressing EYFP-K14R 125 C were transiently transfected with fluorescent proteintagged non-muscle myosin IIB. The keratin granule-bearing zone coincided with the peripheral enrichment zone of non-muscle myosin IIB (Fig. 8A). Time-lapse fluorescence recording revealed that the constant myosin flow corresponded with the inward flow of keratin granules (Movies 8 and 9). Both types of motility were equally affected by para-nitroblebbistatin treatment (Fig. 8A, B; Movie 9). To understand why residual inward-directed motility persisted in the presence of blebbistatin, we examined actin dynamics using LifeAct in comparison to keratin granule mobility. In untreated cells, keratin granule motility corresponded well to the inward movement of actin often co-distributing with transverse actin arcs (Fig. 8C,D; Movie 10). As to be expected [24][25][26] , blebbistatin dissolved transverse arcs but did not completely abolish retrograde actin flow, which still coincided with the remaining granule movement. We therefore concluded that the inward movement of mutant keratin granules is linked to retrograde actin flow.

Discussion
Ex vivo analyses are not sufficient to understand and characterize the mechanisms and consequences of pathological keratin aggregate formation. Thus, recombinant K14R 125 H forms prototypical intermediate filaments under optimal in vitro conditions 27 . Production of the same or similar keratin 14 mutants in cultured cells, however, leads to abundant keratin granule formation with little remaining filamentous keratins in the perinuclear region [28][29][30][31] . Time-lapse imaging revealed that these keratin granules are highly dynamic, displaying a high degree of motility and a limited lifetime. These properties offer an inroad into the investigation of factors that modulate keratin aggregation. Progress has been hampered so far by the multifactorial regulation of keratin granule dynamics. Quantitative parameters are therefore needed to determine the contribution of www.nature.com/scientificreports/ individual factors. Microscopic recording of fluorescence-tagged keratin mutants provides temporal and spatial information that can be dissected by image analysis to obtain such parameters. By pursuing this approach, we were able to digitally track keratin granule formation, growth, movement and disassembly at the single granule level. The results demonstrate that the determined parameters do not reflect random processes but must be the consequence of coordinated regulation and can therefore be described in quantitative terms. Specifically, we found that granule formation is restricted to less than 10 µm from the plasma membrane, that granule growth is continuous up to a certain plateau size for about 30 min, that granule mobility is consistently directed toward the cell centre, that the speed of granule movement differs between cells but always slowly declines as granules grow, and that granule disassembly is faster than assembly and is limited to a distinct transition zone between the thin cell periphery and thicker cell centre.
Probably the most remarkable property of keratin granules is their coordinated motility. It carries the signet features of an actin-dependent transport mechanism with respect to speed, directionality and continuity. In accordance, previous studies had shown that actin filament disruption by cytochalasin or latrunculin prevented keratin granule motility whereas disruption of microtubules by nocodazole did not 9 . Whether and to which degree actin-dependent motor proteins are involved in keratin granule motility, however, could not be resolved. Using the standardized imaging conditions and methods of data analysis we addressed this aspect by treating K14R 125 C-expressing cells with the myosin inhibitors para-nitroblebbistatin and ML-7. This revealed a significant www.nature.com/scientificreports/ reduction, though not complete inhibition of granule motility with comparatively minor effects on directionality. The findings indicated that either drug inhibition was not complete or that other transport systems contribute to keratin granule movement. Thus, kinesin-1 has also been shown to be involved in keratin motility 32 and may act in concert with non-muscle myosin II. The kinesin-1-dependent type of motility, however, appears to be different from the restricted inward-directed motility observed for the mutant keratin granules. Furthermore, alternative scenarios may be underlying the effect of myosin II-dependent keratin granule transport: Non-muscle myosin II either directly or indirectly affects keratin granule movement. In support of the first model physical association between keratins and non-muscle myosin IIA has been documented for keratins K8/K18 and K6 with functional implications for cell migration 21,33 . A possible mechanism for the second scenario is that actin retrograde flow is the primary stimulus for inward-directed keratin granule motility. The observation that the myosin inhibitors used in this study are known to abolish transverse actin arcs but only partially (~ 50%) abolish retrograde actin flow provides such a link [24][25][26] . It is fully supported by the blebbistatin-induced dissolution of transverse actin arcs and the ~ 33% reduction of keratin granule speed in the present study. A challenge in elucidating keratin aggregate formation is the influence of context-dependent modulators and the highly variable effects of different mutations. Thus, EBS-associated mutations have been shown to affect keratin network formation differently, resulting in different functional phenotypes of cell adhesion and mechanical resilience 29,[34][35][36] . Environmental factors including mechanical stress are known to affect mutant keratin network organization and mechanotransduction 34,37,38 . This high degree of variability and the underlying multifactorial nature of keratin aggregate formation necessitate means to assess keratin networks in quantitative terms. The presented methods offer a way to screen for modifiers of keratin granule dynamics (Fig. 1B).
While our findings are relevant for dissecting mechanisms of keratin granule-forming diseases, they also improve our understanding of normal keratin network regulation and dynamics. Despite the inability of K14R 125 mutants to support normal keratin network formation in transfected epithelial cells, keratin granule formation and dissolution share key features with the normal keratin turnover, including induction of keratin polymerization in the cell periphery and inward-directed transport accompanied by continuing polymerization 10,39 . We therefore propose that myosin-dependency of motility applies also to the observed inward transport of growing keratin filament precursors. Remarkably, keratin filament precursors are released from the transport machinery for network incorporation at a transition zone which appears to correlate with the zone, where the mutant keratins are disassembled into subunits that can be re-used for another round of assembly and disassembly 10,40 .
The developed methods will help to clarify similarities and differences to other situations of keratin granule formation, which occur for example during mitosis, apoptosis, phosphatase inhibition or increased stress 41 . More generally, it will be important to compare key features of keratin aggregate formation and turnover with other aggregates formed in other diseases involving other cytoplasmic IFPs.

Materials and methods
Cells. The MCF-7 cell line producing EYFP-K14R 125 C has been described 9 . Cells were grown in uncoated flasks or on glass bottom dishes (12 mm glass-diameter, thickness 1.5#; MatTek) in Dulbecco's Modified Eagle's Medium (DMEM; Sigma Aldrich) supplemented with 10% fetal calf serum (FCS; Pan Biotech) in a humidified atmosphere with 5% CO 2 at 37 °C. Cells were passaged once a week at a ratio of 1:5-1:30 or at 1:30-1:40 for live cell imaging. For passaging, cells were washed with and subsequently incubated for 10 min in phosphate buffered saline (PBS) without Ca 2+ or Mg 2+ (Sigma Aldrich). They were then incubated with accutase (Sigma Aldrich) until cells detached. They were then re-suspended in trypsin neutralizer solution (Thermo Fisher), centrifuged and seeded.
Time-lapse microscopy. MCF-7 cells were seeded in glass-bottom dishes (12 mm glass-diameter, thickness 1.5#, MatTek) two days prior to imaging and transfected one day before imaging. The cells were transfected with 5 µg of plasmid DNA and 1.5 µl Xfect polymer in a total volume of 100 µl Xfect reaction buffer (Takara) per 2 ml of cell culture medium in 35-mm diameter dishes. The LifeAct-RFP plasmid was kindly provided by Bernd Hoffmann (Forschungszentrum Jülich 42 ) and the mCherry-MyosinIIB-N-18 plasmid was purchased from Addgene (#55107; kindly provided by Michael Davidson). For live-cell imaging, cells were incubated in 25 mM HEPES-buffered DMEM without phenol red (Life technologies) supplemented with 2% fetal calf serum (Pan biotech). Fluorescence recordings were performed with a laser scanning confocal microscope (LSM 710; Carl Zeiss) using Zen black 2.1 SP3 software (Carl Zeiss). The microscope was equipped with an Airyscan detector, oil immersion objective (63 × /1.40-N.A. DIC M27) and a focus-shift correction system (DefiniteFocus; all from Carl Zeiss). In order to perform live-cell imaging, the microscope incubation chamber was pre-warmed to 37 °C and humidified. For detection of EYFP the argon-ion laser was used at 514 nm and 0.3% power together with a BP 420-480 + BP 495-620 filter (Movie 7) or the argon-ion laser was used at 488 nm and 0.1-0.3% power and signal emission was acquired at a range of 505-555 nm (Movies 8,9,10). For detection of mCherry in living cells a 543 nm HeNe-laser (module LGK 7786 P) was used at 5.0% power and signal emission was acquired at a range of 600-700 nm. For detection of RFP a 543 nm HeNe-laser (module LGK 7786 P) was used at 10% power with signal emission being acquired at a range of 600-700 nm. The experiments were performed at an optical resolution corresponding to approximately 200-300 nm resolution in x/y length for the YFP signal. To improve accuracy, the scan-pixel-size was set to 77 nm. Microscopy images were processed in Fiji distribution of ImageJ software package 43,44 and further processed for visualization by Adobe Photoshop CS3 and Microsoft Office Professional Plus 2016. The scheme in Fig. 1  www.nature.com/scientificreports/ red (Sigma Aldrich) supplemented with 2% FCS (Pan Biotech). Live cell imaging was performed on a Zeiss LSM710 Duo microscope at 37 °C. The 488 nm line of an argon/krypton laser was used for fluorescence recording with a 63 × /1.40-N.A. DIC M27 oil immersion objective. The emitted light was monitored between 500 and 540 nm (green signal), with a pinhole set at 1-2 AU (airy unit) and a laser intensity of 0.2%. Cells were imaged in control medium. Subsequently, the medium was aspirated using a vacuum pump and replaced by medium supplemented with the respective inhibitor for further imaging without interruption of recording. Para-nitroblebbistatin was purchased from Optopharma. Concentrations of 5, 10, 20 and 50 µM para-nitroblebbistatin were tested. While 5 µM inhibited keratin granule motility only mildly, 50 µM showed signs of toxicity, completely inhibiting intracellular motility and dynamics. Based on the test results and the recommendation of the manufacturer to not exceed 20 µM because of unspecific side effects, the reported efficient inhibition of myosin 2 function in cultured cells using 10-20 µM para-nitroblebbistatin 22 , the limited solubility of blebbistatin (4-25 µM in aqueous solution containing 0.1-2% DMSO for blebbistatin 22,23 ), and the concern for toxicity of high paranitroblebbistatin concentrations we decided to use 20 µM as the standard concentration. ML-7 was purchased from Tocris and used at a standard concentration of 20 µM which has been shown to inhibit MLCK efficiently in MCF-7 cells and is known to avoid unspecific inhibition of protein kinase A and C 45,46 . Automated image analysis. For best comparison high quality recordings of isolated cells were selected that had a comparable overall cell morphology (round shape, flat cell body). Image sequences were either recorded over a long period (between 270 and 365 frames per movie with 3 frames/min) allowing the analysis of complete tracks, or in a before-after fashion to study the impact of modulators. In both instances, the analysis followed the scheme shown in Fig. 1B. First, granules were tracked with the help of the FIJI plugin TrackMate 47 . For segmentation, the LoG detector was applied with the following parameters: blob diameter = 1 µm; threshold = 33 (in some experiments 1, due to different recording settings); no mean filtering; and sub-pixel localization active. For tracking, the LAP tracker was used with the following parameters: Frame to frame linking = 1 µm; track segment gap closing with maximal distance of 1 µm and maximal frame gap of 2; track segment splitting disabled; track segment merging 1 µm. After initial track calculation, the tracks were filtered for mean quality set to "auto". For further inspection numerical data (track-IDs; x, y, t; fluorescence) were exported and MAT-LAB (The MathWorks, Inc.) custom scripts were used. Since merging of tracks occurred regularly, tracks were automatically classified as solo tracks which did not merge, and merging tracks that were subdivided into main tracks and sub tracks ( Supplementary Fig. S1A). Mean local track speed, local advance speed and fluorescence of granules were calculated for the different track types. To consider local changes of speed, a gliding mean of speed was calculated by computing the local track and advance speeds for each time point taking 8 image recordings from t − 3 to t + 4 into account (Fig. S1B).
To further examine the motion properties of granules, MSD analysis was performed using the MATLAB class @msdanalyzer 48 . The coefficient alpha was determined in the following way: The MSD function was calculated for each individual trajectory and its log-log representation was fitted with a linear function such that if the MSD curve could be modelled by ρ(r) = ⟨r2⟩ = Γta, then log(⟨r2⟩) = f(log(t)) was fitted with Γ + α log(t). We fitted individual MSD curves and discarded those for which the R2 coefficient, reflecting the quality of the fit, was < 0.8 48 . For MSD analysis along or perpendicular to the track direction, tracks were rotated in a way that start-and end-point were located horizontally. In that instance, x values reflect back and forth movements while y-values reflect up and down movements, i.e. movements perpendicular to the track axis. MSD analysis then was performed for x and y values separately.
To test significance of before and after parameters the signrank function applying the Wilcoxon matched-pairs signed-ranks test 49 was used. Graphs were prepared using GraphPadPrism v7 Software. Values are given with standard deviation of the mean, i.e. ± SD. Significance with a p ≤ 0.001 was labelled ***, p ≤ 0.01 was labelled ** and p ≤ 0.05 was labelled *.

Data availability
The automated image analysis protocols generated during the current study are available in the GitHub repository, https ://githu b.com/rwind offer /KER_TRACK .