GFAP isoforms control intermediate filament network dynamics, cell morphology, and focal adhesions

Glial fibrillary acidic protein (GFAP) is the characteristic intermediate filament (IF) protein in astrocytes. Expression of its main isoforms, GFAPα and GFAPδ, varies in astrocytes and astrocytoma implying a potential regulatory role in astrocyte physiology and pathology. An IF-network is a dynamic structure and has been functionally linked to cell motility, proliferation, and morphology. There is a constant exchange of IF-proteins with the network. To study differences in the dynamic properties of GFAPα and GFAPδ, we performed fluorescence recovery after photobleaching experiments on astrocytoma cells with fluorescently tagged GFAPs. Here, we show for the first time that the exchange of GFP–GFAPδ was significantly slower than the exchange of GFP–GFAPα with the IF-network. Furthermore, a collapsed IF-network, induced by GFAPδ expression, led to a further decrease in fluorescence recovery of both GFP–GFAPα and GFP–GFAPδ. This altered IF-network also changed cell morphology and the focal adhesion size, but did not alter cell migration or proliferation. Our study provides further insight into the modulation of the dynamic properties and functional consequences of the IF-network composition. Electronic supplementary material The online version of this article (doi:10.1007/s00018-016-2239-5) contains supplementary material, which is available to authorized users.


Introduction
Intermediate filaments (IFs) are part of the cytoskeleton. Together with actin filaments and microtubules, they form an integrated system that regulates many cellular processes, such as cell morphology, cell signaling, cell migration, and proliferation [1][2][3][4]. The main IF protein expressed in astrocytes is glial fibrillary acidic protein (GFAP). The ten different GFAP isoforms, of which GFAPa is the canonical isoform, are formed by alternative splicing [5,6]. The function of GFAP and its isoforms is still elusive, but there is emerging evidence that at least one isoform, GFAPd, alters the properties of the IF network. GFAPd differs from GFAPa only in its C-terminal tail, and in non-pathological human brains, and is expressed in specific types of astrocytes, including the adult neural stem cells in the human subventricular zone and subpial astrocytes [7][8][9]. The GFAPd protein has a unique 41 amino acids long C-terminal tail [6,7] and is one amino acid shorter than the canonical GFAPa protein [10]. In pathological conditions, GFAPd is expressed in certain types of reactive gliosis and glial tumors [11][12][13][14][15]. The tail of GFAPd disables the protein to form homodimers making it impossible to selfassemble [16]. GFAPd is able to form heterodimers with other type III IF proteins and can, therefore, be integrated in an IF network. Depending on the level of expression and the concentration of other IFs present, GFAPd is either tolerated in the network or it causes the whole IF network to collapse in the perinuclear region [7,17]. Assembly experiments in a cell free environment showed that GFAP networks start to collapse when there is more than 10 % of GFAPd protein present in the network [17].
In the cell, IF proteins are present in a soluble form in the cytoplasm and in filamentous structures that form an important part of the cell's cytoskeleton [18,19]. These IF networks are highly motile structures that are constantly rearranged. The proteins within the filaments are also dynamic, since there is an active exchange between the filamentous and non-filamentous pool of IF proteins [20][21][22][23]. IF networks that are already formed can be actively disassembled by phosphorylation of IF proteins, whereas the lack of dephosphorylation will hamper new IF network assembly [24,25]. It has been shown that phosphorylation of GFAP at the N-terminal head domain by kinases, such as Aurora B or CF kinase, is important for proper dissociation from the filaments during cytokinesis [25][26][27][28].
Previously, we showed both in vitro and in vivo that physiological levels of GFAPd are well tolerated in a GFAPa network [7,8,15,17], although it has also been shown in a cell free system and in vitro that a high expression of GFAPd can lead to an IF network collapse [7,16,17]. These collapses resemble aggregates of GFAP proteins, which occur when cells are transfected with mutant R416W GFAP [29]. This is one of the mutations in GFAP that causes Alexander disease (AxD), a fatal neurodegenerative disease characterized by leukodystrophy, macrocephaly, and psychomotor retardation [30]. A pathological hallmark of this disease is the presence of Rosenthal fibers, which are astrocytic aggregates that are comprised of GFAP, ubiquitinated proteins, and stress proteins, such as heat shock proteins like aBcrystallin (CRYAB) and heat shock protein 27 (HSP27) [29,[31][32][33][34], but also IF-associated proteins like plectin [35]. AxD mutations in GFAP and the subsequent collapse of the network influence astrocyte viability, morphology, and glutamate transport, and aggregates or accumulations of mutant AxD GFAP have a profound effect on astrocyte biology and physiology [36][37][38][39].
There is increasing evidence that GFAPd changes IF properties [17,40]. Here, we studied in more detail the differences in dynamic exchange of GFAPa and GFAPd with the IF network in vitro. Furthermore, we analyzed the functional consequences of GFAPa or GFAPd expression by assessing the effect of the altered GFAP network on the dynamics and localization of other IF proteins in the cells, on cell morphology, and focal adhesions. We also determined the effect on cell migration and proliferation, since it has been reported that GFAP affects these processes [4,[41][42][43]. Investigating the dynamic properties of the different GFAP isoforms and the effects of the different GFAP networks on cellular functions will contribute to our understanding of the consequences of modulation of the GFAP network for astrocyte physiology and pathology.

Isolation of primary human astrocytes
Primary human adult astrocytes were obtained from freshly dissected postmortem subcortical white matter of a 79-year-old female control donor (NBB 2010-038), with a postmortem delay of \18 hours (h) and a cerebrospinal fluid pH of 6.30. The tissue was obtained from the Netherlands Brain Bank (NBB), which performs brain autopsies with short postmortem intervals. The brain donors gave informed consent for using the tissue and for accessing the extensive neuropathological and clinical information for scientific research, which is in compliance with ethical and legal guidelines [44]. The tissue was collected in 25-mL cold Hibernate A (Invitrogen), and mechanically dissociated into small pieces. The tissue was digested with 0.2 % trypsin (Invitrogen) and 0.1 % DNA-seI (Invitrogen) at 37°C, while shaking for 30 min (min). Next, 2-mL FBS was added to the mixture, and subsequently, the cells were collected by centrifugation. The pellet was taken up in DMEM without phenol red containing 10 % FBS, 2.5 % Hepes, and 1 % P/S (all Invitrogen), and the suspension was filtered through a 60-lm mesh screen. Then, Percoll (Amersham/GE Healthcare) was added (half of the cell suspension volume), and this mixture was centrifuged at 3220 relative centrifugal force (rcf) at 4°C for 30 min to separate cells, debris, and myelin. The second layer (glial cell containing fraction) was collected and washed with complete DMEM (containing 10 % FBS, 1 % P/S, 2.5 % Hepes, and 1 % gentamycin, all Invitrogen). After centrifugation, the pellet was taken up in complete DMEM, and cells were seeded in a 6-cm uncoated culture dish. Microglia adhered to the dish, and after 6 h at 37°C/5 % CO 2 , the medium, containing astrocytes, was taken off, centrifuged, and the microglia depleted pellet was seeded onto poly-L-lysinecoated wells [PLL, Sigma-Aldrich, 15 lg/mL in PBS, 1 h at room temperature (RT)] in DMEM/Ham's F12 Glu-taMAX medium containing 5 % FBS, 1 % P/S, and 0.25 % Fungizone (all Invitrogen).

Plasmid construction, transient transfection, and virus production
Expression vectors were prepared by cloning human GFAPa and human GFAPd [7,17,29] full length cDNA sequences into the pIRES2EGFP (Clontech). For the GFAPd constructs, the eGFP sequence was replaced by mCherry.
GFP tagged GFAP constructs were created by cloning human GFAPa and GFAPd cDNA sequences [7] in frame after the eGFP sequence in peGFP using BAMHI and HindIII as restriction sites (Clontech) to create N-terminal eGFP tagged GFAPs. The N-terminal side was chosen for the eGFP tag, since GFAPa and GFAPd differ in their C-terminal tail. All plasmids were sequenced.
Subsequently, to produce lentiviral vectors, the constructs were subcloned into a pRRL lenti backbone. Lentiviruses were produced as described before [45,46] with some alterations. In short, 10 9 10 6 HEK 293T cells were plated in a 15-cm culture dish and transfected with a total of 90 lg of the envelope (pMD2.G), packaging (pCMV-dR8.74) and p156RRL plasmid, containing different expression cassettes per dish, using PEI. In total, 90 lg of DNA was mixed with PEI (67.5 ng/lL), incubated for 15 min at RT, and added dropwise to the cell culture. The culture medium was replaced 16 h after transfection, and the medium containing viral particles was collected 24 h after transfection. Supernatants were ultracentrifuged at 22,000 rpm (rotor SW28, Beckman-Coulter) for 2.5 h. The resulting pellet was resuspended in phosphate-buffered saline (PBS) (pH 7.4), aliquoted and stored at -80°C until further use.
To measure viral titers, a dilution series across five orders of magnitude of the viral stock solutions was made and HEK293T cells were transduced. After 2 days of incubation at 37°C, the number of transduced fluorescent cells at the different viral dilutions was counted, and the viral titer was determined in transducing units (TU)/mL.

Creating stable cell lines
For functional experiments, cell lines expressing GFAP isforms were created. U251 cells were transduced with lentiviral constructs with a multiplicity of infection of 10. Medium was refreshed after 16 h. To maintain a population of transduced cells, cells were sorted on their EGFP or mCherry expression using fluorescent activated cell sorting (FACS ARIA II, BD Bioscience, Franklin Lakes, NJ, USA). In between experiments, U251 cells were stained for GFAP to ensure that more than 70 % of the cells were expressing the construct. The primary human astrocytes were checked for fluorescent reporter expression before any analysis to ensure cells were expressing GFAP isoforms.

MTT assay
To measure cell proliferation, an MTT(3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide) assay was performed. MTT is reduced into a soluble blue formazan product by mitochondrial enzymes in living cells only. Therefore, the amount of formed formazan is proportional to the amount of living cells present [47]. MTT assays were performed by plating cells in non-coated plastic 24-wells plates (Greiner). To quantify cells, medium was replaced by 500-lL serum free medium containing 0.5-mg/mL MTT, which was incubated at 37°C for 2 h. Cells were subsequently lysed in 100 % DMSO, which dissolves the purple formazan resulting in a color change of the DMSO. The amount of purple formazan, and therefore the amount of cells able to metabolize the MTT, was measured using a Varioskan Flash (Thermo scientific, USA), measuring the absorbance at 570 nm. Significance was tested with a Kruskal-Wallis test with Dunn's Multiple Comparison post hoc test on 3 independent experiments. Every measurement in the independent experiments was the average of a biological duplicate.

Phospho-histone H3 quantification
To determine the number of actively proliferating cells, U251 cells expressing GFAP isoforms were plated on noncoated coverslips and fluorescently stained for phosphohistone H3 (PHH3) together with the nuclear dye Hoechst (1:1000 dilution) (Invitrogen). Subsequently, micrographs were taken, the number of PHH3 positive, dividing, nuclei was counted using ImagePro software (version 6.3), and the percentage of dividing cells was calculated by dividing the number of PHH3 positive cells by total number of Hoechst positive nuclei. Per experiment, 5 fields of view were analyzed and averaged, each containing at least 50 cells. Data from separate experiments were corrected for inter-experimental variation as stated below. Significance was tested with a Kruskal-Wallis test with Dunn's post hoc test on data from 3 independent experiments.

Migration assay
To measure cell migration, a scratch assay was performed. U251 cells were plated in a 24-wells plate (100,000 cells per well) coated with 20-lg/mL PLL at 37°C for 1 h. The confluent cell monolayer was scratched with a P10 plastic pipet tip. Pictures were taken using an Axiovert 135 M (Zeiss) with a Sony XCD-X700 camera (Sony) at the time points indicated in the results. To quantify cell migration, the surface area not covered by cells was determined at different time points. The migration was calculated as the percentage of uncovered surface area compared with t = 0. A mean of 9 pictures was measured per condition, in at least 3 separate experiments. Significance was tested with a Kruskal-Wallis test with Dunn's post hoc test.

Single cell motility assay
Single cell motility assays were performed on a Zeiss Axiovert 2000 inverted microscope (Zeiss, Jena, Germany). A single cell suspension was plated on PLL-coated glass dishes with four compartments (CELLview, Greiner bio-one, Alphen a/d Rijn, The Netherlands) and allowed to adhere for at least 8 h. Dishes were kept on the microscope in a pre-heated and humidified incubation chamber (OKO labs) at 37°C and 5 % CO 2 . Pictures were taken every 10 min with an Axi Aqua camera (Q imaging). Cell motility was measured by tracking single cells throughout all frames of the sequence and measuring the average velocity in lm per min using the manual tracking plugin from ImageJ (Rasband, W.S., ImageJ, U. S. National Institutes of Health, Bethesda, Maryland, USA, http:// imagej.nih.gov/ij/, 1997-2012 version 1.46f). Per experiment, at least 20 cells were analyzed per condition in at least 3 independent experiments. Data from separate experiments were corrected for inter-experimental variation as stated in the ''Statistics and factor correction'' section below. A Kruskal-Wallis test with Dunn's post hoc test was performed to test for significance. For the primary human astrocytes, which were not sorted, we checked for GFP and mCherry expression to make sure that we only tracked transduced cells.
Quantitative reverse transcriptase PCR (qPCR) analysis U251 cells were transduced like described before. Medium was refreshed after 16 h. RNA was extracted 7 days after transduction. RNA was extracted from cells using TRIsure (Bioline, London, UK) and precipitated in isopropanol overnight (O/N). Five hundred nanograms of RNA was reverse transcribed into cDNA with a QuantiTect reverse transcription kit (Qiagen), as described before [5]. cDNA was diluted 1:20 before being used as a template in qPCR assays (SYBR Ò Green PCR Master Mix, Applied Biosystems). qPCR conditions were similar as described before [5] and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) and hypoxanthine phosphoribosyltransferase (HPRT) were used as reference genes to normalize gene expression. Data from 4 separate experiments were factor corrected as stated in the ''Statistics and factor correction'' section below and tested for significance using a Kruskal-Wallis test with Dunn's post hoc test. Primer pairs used are listed in Table 1.

Immunocytochemistry
To perform immunocytochemical staining, cells were cultured on uncoated glass coverslips, fixed with 4 % Paraformaldehyde (PFA), washed in PBS, and incubated in SuMi buffer for 10 min. Primary antibodies were diluted in SuMi and incubated at 4°C on a shaker O/N. Cells were washed 3 times in PBS and, subsequently, incubated with secondary antibodies and Hoechst 33258 (1:1000 dilution) (Invitrogen) diluted in SuMi at RT for 1 h. The antibodies used are listed in Table 2. All secondary antibodies were from Jackson Immune Research and diluted 1:1400 in SuMi. Cells were washed again in PBS, before the coverslips with cells were mounted on slides with Mowiol

Cell morphology measurements
Phase contrast pictures were taken from the U251 cells transduced with GFAP isoforms and mCherry (control) on a Zeiss Axiovert 2000 with an Exi Aqua camera (Q Imaging, Surrey, Canada). Cell outlines were manually drawn using Image J. Area and perimeter were measured in square units. Form factor was calculated as 4p ðAreaÞ ðPerimeterÞ 2 , where perfectly round cells will have a form factor of 1 [43,48]. Five independent experiments were performed with 40 cells analyzed per experiment. Data have been factor corrected for inter-experimental variation as stated

GCGAGGAGAACCGAAACCAG CTTCACCACGATGTTCCTCTTG
Endogenous GFAPa and GFAPd primers are located in the 3 0 UTR part, which is absent in the plasmid cDNA. The primer sequences used were published before [15] GFAP glial fibrillary acidic protein  33 9 GFAPa in a network, 30 9 GFAPd in a network, 8 9 GFAPa in a collapse, and 8 9 GFAPd in a collapse.

FRAP analysis
ROIs were positioned in post bleach pictures manually. Per time point, the position of the ROI was corrected for cell movement, and average fluorescence was measured using Image J. The average fluorescence was plotted in time to obtain fluorescence recovery curves for every ROI. The half time was calculated by interpolating the time at 50 % of the fluorescence of the maximum fluorescence at 30 min. The immobile fraction was calculated by comparing the fluorescence in the bleached area after recovery (F ? ) with the fluorescence before bleaching (F i ) and just after bleaching (F 0 ). The immobile percentage is defined as (1 -(F ? -F 0 )/(F i -F 0 )) 9 100. For F ? , the average of the last three time points was used for analysis.

Statistics and factor correction
Data obtained from independent experiments were corrected with a factor correction program (version 10.5 2012) [49] when stated. Kruskal-Wallis, Kolmogorov-Smirnov, or Mann-Whitney tests were performed to test for significance. Differences were considered significant at p \ 0.05. All statistical tests were performed using Graphpad Prism 5 (version 5.04) (Graphpad Software Inc., La Jolla, CA, USA).

Focal adhesion analysis
To analyze focal adhesions, coverslips were coated with laminin in 24-wells plates. Cells were plated at 10,000 cells per well. After attaching for 2 days, cells were fixed with 4 % PFA, stained with rabbit anti-phosphorylated-paxillin (pY118; Life Technologies), and counterstained with Hoechst and Phalloidin-488. Fixed and immunostained samples were imaged on an inverted microscope (Zeiss Axiovert 200) with a Confocal Spinning-disk Unit (Yokogawa CSU X-1) and an emCCD camera (Andor iXon 897). Excitation was accurately controlled by 405 nm (Crystalasers), 514 nm (Cobolt), and 642 nm (Spectra Physics) lasers through an acousto-optic tunable filter (AA Optoelectronics) coupled into the CSU with a polarization-maintaining optical fiber. Images were acquired with Andor IQ 2 software, and further processing and analysis were performed in specifically designed software (Matlab, Mathworks).
Focal adhesions were automatically detected from the 642-nm fluorescence images. Focal adhesion size could be at least 0.4 9 0.4 lm and at most 4 9 4 lm. In an image of 512 9 512 pixels low pass frequency with a cut-off at 2 pixels filtering removed large features. With a subsequent threshold of 3 times the standard deviation of the image, we obtained and characterized the focal adhesions. Focal adhesion size differences were tested using a 2-tailed Kolmogorov-Smirnov test. Differences were considered significant if p \ 0.05. Per condition, a different amount of cells and focal adhesion were measured, but at least exceeded 30 different cells and a total of more than 700 focal adhesions.

Results
GFAPd perturbs the GFAP network in astrocytes Cellular models for studying GFAP isoform function were established by expression of GFAP isoforms in U251 cells and primary human astrocytes. The mRNAs of the different isoforms were highly expressed in the transduced U251 cell lines (Sup. Fig. 1a, b) and primary human astrocytes (Sup. Fig. 1d, e). Western blot analysis showed a clear expression of either GFAPa or d (Sup. Fig. 1c), while the control (Ctrl) only showed a band with the pan GFAP antibody, representing endogenous GFAPa. The endogenous expression of the other GFAP isoforms was determined with qPCR and is presented as a percentage of the canonical GFAPa expression. GFAPa and GFAPd were the most abundant isoforms expressed, followed by GFAPj. Other isoforms were expressed at a very low level (Sup. Fig. 1f).
First, we confirmed whether GFAPd-induced cytoplasmic collapses of GFAP also occurred in the transduced U251 cells and primary human astrocytes, as we have shown before in SW13 human adrenal carcinoma cells and U343 astrocytoma cells [5,17]. We investigated the IF network morphology by immunostainings for GFAP, and we observed that the different GFAP isoforms gave similar results in both the primary human astrocytes (Fig. 1a) and the human U251 astrocytoma cells (Fig. 1b). The endogenous GFAP network was stained in the control condition, in which cells were transduced with mCherry. The GFAP network was present throughout the cytoplasm up to close proximity of the cell periphery, which is visualized by actin staining. Expression of recombinant GFAPa in human astrocytes, or in U251 cells, resulted in a GFAP network which was morphologically similar to astrocytic IF networks, indicating that the recombinant GFAPa was incorporated into the endogenous IF network of the cells. In contrast, expression of GFAPd led to a collapse of the GFAP network mostly in a perinuclear fashion (Fig. 1a, b). This perturbing effect of GFAPd on the GFAP network in these cells is concentration dependent and is a gradual process (data not shown), as has also been described before [16,17].

Vimentin and nestin co-collapse with GFAPd, while actin and microtubules stay intact
To assess whether GFAPd induces a collapse of the complete cytoplasmic IF network, we also studied the distribution of vimentin (Fig. 2a) and nestin (Fig. 2b), which were both present in the IF network in primary human astrocytes. Transduction with GFAPa had no effect on the intracellular location of these proteins, but transduction with GFAPd led to a condensation of vimentin and nestin, mostly around the nucleus. The same effect was observed in the U251 cell line (not shown). The actin and microtubule cytoskeletal networks did not co-collapse with the IF network upon GFAPd expression (Fig. 2a, b, Sup.  Fig. 2a, b). The staining intensity of actin was very variable between cells and between conditions, but the overall morphology of the actin network showed no actin collapse induced by GFAPd.
To examine whether the additional expression of GFAPa or GFAPd would lead to effects on transcription of other IFs, we analyzed the change in mRNA expression of vimentin, nestin, and endogenous GFAPa and GFAPd in U251 cells using quantitative PCR. GFAPa mRNA (Sup. Fig. 3a), GFAPd mRNA (Sup. Fig. 3b), vimentin mRNA (Sup. Fig. 3d), and vimentin protein (Sup. Fig. 3e) levels were not significantly changed due to an increase in either GFAPa or GFAPd. However, we noticed a small but significant upregulation of nestin transcript in GFAPa expressing cells (p = 0.03) (Sup. Fig. 3c).

GFP tagged GFAP incorporates into the endogenous IF network
Next, we assessed the dynamic properties of GFAP isoforms using N-terminally GFP tagged GFAP fusion proteins for live cell imaging. The transduced cells express GFP-labeled GFAP isoforms, and thereby, transduced cells can be directly identified based on the reporter GFP. Both GFP-GFAPa and GFP-GFAPd did incorporate into the endogenous IF network of U251 cells (Sup. Fig. 4a, b). Since a high expression of tagged GFAP also led to a collapse of the network, a relatively low expression was needed to image a non-collapsed network, which was established by analysis 24 h after transient transfection using U251 cells for a reasonable transfection efficiency. 24 h after transfection, the GFP-GFAPa transfected cells showed a spread out network (Sup. Fig. 5a), and GFP-GFAPd expressing cells showed a mixture of cells with either a spread out network (Sup. Fig. 5b) or a collapsed IF network (Sup. Fig. 5c). The endogenous IF network is visualized by the vimentin staining. About 30 % of the GFP-GFAPd expressing cells in our cell culture condition showed a collapse at this time point, although the exact percentage varied between experiments and was dependent on the transfection efficiency. To visualize the dynamics of the collapsing network over time, U343MG astrocytoma cells transfected with GFP-GFAPd were imaged for 48 h, starting 4 h after transfection. U343MG cells were used here, since they are less motile than U251MG cells, which enabled us to image the IF network for a long period of time. First, GFP-GFAPd was distributed throughout the IF network in the whole cell (arrow in Fig. 3a t = 12 h). GFP-GFAPd started to condensate around the nucleus, as the expression of GFP-GFAPd increased over time (Fig. 3a and Sup. Movies 1 and 2). Sometimes small aggregates or condensations were seen, which joined the already collapsed GFAP proteins (arrowheads Fig. 3a t = 30 and Sup. Movies 1 and 2).  During and after the process of the accumulation, cells were still migrating, and the collapsed IF network was a motile structure in both stationary and moving cells. These experiments showed that GFP-GFAPd, in small amounts, is incorporated into the endogenous IF network before it causes a collapse of the IF network. Images from fixed cells stained for GFP, GFAP, and vimentin showed that the squiggles (short filaments) of GFP-GFAPd sometimes colocalized with vimentin and were not part of larger filaments (Fig. 3b, arrow).

Dynamic properties of GFP-GFAPa are different from GFP-GFAPd
To assess the dynamic properties of GFAPa and GFAPd, FRAP experiments were performed on U251 astrocytoma cells in which small boxed regions of fluorescent cells were photobleached, and recovery of fluorescence was measured over time. Cells were transfected with the GFP-GFAP isoforms, and the FRAP analysis was started 24 h later. Cells with non-collapsed and collapsed networks were measured and analyzed separately. Regions of fluorescent GFAP networks were bleached, and the fluorescence recovery was imaged up to 30 min after bleaching. A typical example is shown in Fig. 4a. The median of all FRAP recovery experiments for GFAPa and GFAPd in an extended network and GFAPd in a collapsed network are shown in Fig. 4d. The half time t (the time needed to recover to 50 % of the final fluorescence) and the immobile fraction (the percentage of fluorescence which is not recovered) were calculated. The median t value of GFAPd in extended, non-collapsed networks was 2.3 min, and this was significantly (p \ 0.05) longer than the median t of GFAPa, which was 1.1 min (Fig. 4b). Since high GFAPd expression leads to a collapse of the IF network, we next investigated the dynamic properties of GFAPd in a collapsed network. Although not significant, there was a clear trend visible that GFAPd had a longer t (median = 3.8 min) when in a collapsed network compared with GFAPd in a spread out network (Fig. 4b), indicating a slower on/off rate from the IF network. We also calculated the immobile fraction of the GFP-GFAPs, which is the percentage of fluorescence that did not recover from the FRAP curves. We observed that the immobile fraction did not significantly differ between GFAPa (median = 40.3 %) or GFAPd in extended network (median = 49.1 %). There was, however, a significant difference between the immobile fraction of GFAPd when the majority of the GFAP was in a collapsed network (median = 56.4 %) compared with the immobile fraction of GFAPa and GFAPd when GFAP was in a spread out network (Fig. 4c).
To assess whether the change in dynamic properties between GFAPd when in a network and when it has collapsed was due to the collapse of the network and not GFAPd itself, we performed experiments in which we also measured the dynamics of GFAPa in a collapsed network. To study this, we transfected cells with untagged GFAPd to induce the collapse, and co-transfected either GFP-GFAPa or GFP-GFAPd to visualize the fluorescent isoform dynamics. Both GFP-GFAPa (Sup. Fig. 5a) and GFP-GFAPd (Sup. Fig. 5b) were incorporated into the collapse as can be seen by staining for vimentin, which is highly expressed in U251 cells and shows the endogenous IF network. A representative example of a bleached collapsed network and the subsequent recovery is shown in Fig. 5a. The median t for GFAPa (4.4 min) and GFAPd (5.5 min) did not differ significantly (p = 0.8) (Fig. 5b). The t measured for GFAPa was remarkably increased due to the collapsed network (Fig. 5b). The fluorescence had not yet reached plateau after 30 min of recovery. Taken together, this indicates that GFAPa and GFAPd have different exchange dynamics (as measured by t ) and the dynamics show a trend for higher t when the IF network is collapsed.

A collapsed IF network changes cell morphology
Next, we aimed at defining whether the IF network collapse and changes in dynamics also result in a change in morphology, since IF expression has been linked to cell morphology [40,43]. Morphological parameters of living U251 cells expressing different GFAP isoforms were determined by measuring the cell surface area and the perimeter. From these parameters, we calculated the form factor, as described in the methods section. Forty cells were analyzed per experiment in five independent experiments. GFAPd expression caused the cells to become rounder (0.56 ± 0.006; mean ± SEM) in comparison with the control (0.49 ± 0.005, p = 0.009) (Fig. 6a). We also observed significant differences in perimeter (Fig. 6b) and area (Fig. 6c) between cells expressing GFAPd and GFAPa.
To assess whether the morphological changes we observed were due to changes in cell-extracellular matrix interaction, the sizes of focal adhesions were assessed by phosphorylated paxillin stainings in cells plated on a laminin substrate. In both GFAPa and GFAPd expressing cells, the size of the focal adhesion was increased (mCherry: 1.13 lm 2 ± 0.02; GFAPa: 1.34 lm 2 ± 0.03; GFAPd: 1.37 lm 2 ± 0.03 (Fig. 6e, f). There was, however, no transcriptional regulation of the main lamininbinding integrins expressed in these cells (data not shown).

GFAP overexpression does not affect cell motility or cell proliferation
Focal adhesions are tightly regulated during cell migration, and GFAP has been linked to in vitro migration by others [4,43]. To check for changes in cell motility, we analyzed single cell motility and scratch wound healing speed. Single cell motility assays were performed on U251 cells expressing GFAPa, GFAPd, or mCherry. Cells were seeded on PLL coated glass coverslips, and thirty cells were analyzed per experiment in 3 independent experiments. There were no statistically significant differences in average velocity between U251 cells expressing GFAPd (0.53 lm/min ± 0.09), GFAPa (0.44 lm/min ± 0.08), or control (0.50 lm/min ± 0.02) (p = 0.9) (Fig. 7a). Similar results were found in a wound healing assay where a monolayer of cells was scratched and the wound healing speed was measured over time (Fig. 7g). Wound healing speed of U251 cells expressing GFAPa (52.9 % ± 9.3), GFAPd (54.7 % ± 6.2), or the control (50 % ± 7.3) was not different at 12.5 h (p = 0.7). To study the effects of GFAP isoform expression on primary cell motility instead of on tumor cells, single cell motility assays were also done on primary human astrocytes. Again, we found no statistically significant difference between cell motility of cells expressing GFAPa (0.39 ± 0.03), GFAPd (0.37 ± 0.02), and control cells (0.43 ± 0.009) (p = 0.15) (Fig. 7b).

Discussion
Astrocytes and astrocytoma cells tightly regulate the expression of at least 10 different GFAP isoforms [6]. GFAPa and GFAPd are the two most highly expressed isoforms. The functional consequences of changes in the IF network in astrocytes and astrocytoma cells are still elusive. In this study, we have investigated the effect of GFAPd on cell proliferation, migration, motility, and shape but also the intracellular effects on IF network dynamics. We here show that GFAPa and GFAPd have different intrinsic dynamic properties, but that expression of GFAPd does not affect cell proliferation or migration when there is a collapse of the IF network.

Dynamic properties
GFAPd is assembly compromised by itself [16] and causes a collapse of the whole IF network. We here show that even before a collapse, GFAPd changes the dynamics of the the IF network. Dynamic properties of IF proteins differ between IF family members [50][51][52][53]. These differences are likely due to phosphorylation or structural properties that affect the assembly. The latter is for instance reflected in higher FRAP recovery half time values by the nuclear lamins (*140 min) [53,54], which forms IgG-like folds during assembly [55]. In contrast, GFAP and vimentin, which are both located in the cytoplasm, have half times between 1 and 5 min [51,52]. In this study, we showed that GFAPd incorporates and dissociates slower from an IF b Fig. 4 Different dynamics between GFAPa and GFAPd. a FRAP experiments consisted of bleaching ROI and measuring the fluorescence recovery up to 30 min after the bleach in U251MG cells.
Within 150 s, most of the fluorescence was recovered, but recovery was never complete. b FRAP experiments were performed for GFAPa, GFAPd, and GFAPd in a collapsed network, and half times were calculated. There is a significant difference in half time between GFAPa (median = 1.1 min) and GFAPd (median = 2.3 min) (p \ 0.05; n = 33, n = 30), and between GFAPa and GFAPd in a collapsed network (median = 3.8 min) (p = 0.000; n = 32). There was no significant difference in the half time of GFAPd in spread out or collapsed network, although there was a trend that a collapse decreased the half time of GFAPd. c The immobile fractions were not significantly different between GFAPa ( Non-parametric tests were performed on the data extracted from the FRAP measurements, so non-overlapping curves do not equal significant differences in this graph network than GFAPa. Since the C-terminal tail is the only difference between GFAPa and GFAPd, the differences in dynamics must be due to this specific sequence. IF protein motility and exchange are mediated through interactions with microtubules and actin to regulate transport [50], and IF protein phosphorylation to regulate filament stability [25,56,57]. In the part of the C-terminus, where GFAPa and GFAPd differ, the same amount of putative phosphorylation sites are present (7 residues). However, the position of these residues is different, possibly leading to a different availability of phosphorylatable residues in the protein tertiary structure. In the tail of GFAPd is a coil 2B binding site, which lacks in the GFAPa tail [16]. It has been proposed by Nielsen and Jorgensen [16] that this domain results in a gain of coiled-coil binding activity of GFAPd, which might explain the slower dissociation of GFAPd from an IF network. Earlier experiments have shown that ablation of the whole C-terminal tail of IF type III proteins inhibits IF assembly [38,58], but phosphorylation of vimentin at the C-terminal side did not affect its assembly [59]. Mutant GFAP isoforms, with mutated phosphorylation sites, will help to elucidate the effects of GFAP phosphorylation on network dynamics.

Effect on proliferation, motility, and migration
In this study, we showed that a GFAPd-induced collapse of the IF network results in a reorganization of the whole IF network. This is in contrast with a knockout or knockdown of GFAP, which does not severely affect vimentin or nestin localization [40]. GFAP, as well as other IFs, has been linked to changes in proliferation and astrocyte motility, but it has to be noted that the studies on the role of GFAP in cell proliferation are inconclusive. A GFAP knockdown has been shown to lead to an increase in proliferation in some studies [4,60], but not in others [61]. On the other hand, astrocyte cultures of transgenic mice overexpressing human GFAPa showed a decrease in proliferation [62]. Since GFAPd is expressed in cycling cells in the human brain [8,12,14], we expected an effect of GFAPd on cell proliferation. Unexpectedly, we observed that not the expression of GFAPd, but an enhanced expression of GFAPa resulted in a significant increase in proliferation of primary astrocytes. We detected a similar trend in the U251 astrocytoma cells. Our data show that the increase in proliferation is caused by the mere increase in GFAPa and that there is no direct role of GFAPd in cell proliferation in cells with a collapsed IF network. Lowering the ratio of GFA-Pa:GFAPd making sure that there is still an intact network also did not change cell proliferation in vitro [40]. Several studies have found that a knockdown or knockout of GFAP increases cell motility [4,43,63]. In addition, we recently showed that a specific knockdown of GFAPa leads to a reduced motility in astrocytoma cells [40].Here, we show that an overexpression of GFAPa or GFAPd has no effect on cell motility, even if the IF network is collapsed by high GFAPd levels. Regulation of motility by GFAP is rather complex. Mutations in the rod domain of GFAP that causes collapses of the network, and thus mimics our GFAPd condition, have been shown to increase cell motility. In contrast, mutations in the tail domain of GFAP had no effect on motility [64,65]. GFAPa and GFAPd only differ in the C-terminal tail, thus this might explain why we do not see differences in cell motility.

Effect on ECM interaction
GFAP has been linked to cell morphological changes in astrocytes in vitro, where a correlation was observed between the level of GFAP expression and the number of cell protrusions [4,[41][42][43]61]. In reactive gliosis, the production of GFAP as well as other IF proteins is upregulated. This results in a more pronounced, GFAPpositive, IF network [66][67][68]. Here, we show that the collapse of the IF network due to GFAPd expression resulted in more round cells with longer focal adhesions in vitro. This change resembles the morphological change in astrocytes devoid of IFs in vitro [4,41,43,61]. Thus, an intact IF network is important for the formation or stabilization of processes of cells. In our earlier study, we showed, however, that a pan-GFAP knockdown in U373 cells did not result in a rounder morphology [40]. This is probably due to the presence of an intact vimentin and nestin IF network, which is lacking in the cells with a GFAPd-induced collapse.
The change we observed in cell morphology in cells with a GFAPd-induced collapsed IF network shows that the IF network distribution per se can affect the shape of the cell. The way GFAPd alters shape is likely to occur through integrins. Integrins are the main linkers between the extracellular matrix and the cytoskeleton in vivo and, together with other proteins, they form the focal adhesions. Focal adhesions can be present in different sizes and maturation states, ranging from small structures, of less than 1 lm, to larger focal adhesions. Focal adhesion size is dependent on actomyosin generated tension [69,70]. Indeed, interactions between the IF protein vimentin and mature focal adhesions have been shown to be essential for proper cell spreading [71]. Vimentin can regulate adhesion and focal contact size under shear stress [72] and controls cell adhesion strength through plectin and b3 integrins [73]. Thus, GFAP could have a similar function in adhesion of astrocytes to the ECM and in this way alter cell shape. Our data confirm a direct effect of the IF constellation on focal adhesions and cell-matrix interactions, as both GFAP isoforms increase focal adhesion size. Since only GFAPd expressing cells showed altered morphology, this implies a difference in the effect GFAPa and GFAPd has on the functionality of focal adhesions. This might partly be due to differences in production of ECM molecules, such as laminin, as we have shown before [40]. Although the exact interaction of GFAP with focal adhesions is still elusive, interactions between other IFs and focal adhesions have been described, as is nicely reviewed by Leube et al. [74].

Conclusion
To summarize, we have shown that a GFAPd-induced collapse of the IF network has a profound effect on IF network morphology, increases focal adhesion size, and changes the IF network dynamics, without altering astrocyte motility or proliferation. Although GFAPd is expressed in more proliferative cell types with a higher migration potential, GFAPd itself does not directly Fig. 7 Migration and proliferation is not affected by GFAPd. a Single cell motility was measured as the average velocity in lm/min of a single cell, in a sequence of images, which were taken overnight. Average velocity did not differ significantly between GFAPa, GFAPd, and control cells (p = 0.9) (n = 3) in U251 cells or in b primary astrocytes (p = 0.15) (n = 3). c Proliferation was measured by metabolic conversion of MTT, which was measured by absorbance of light. The absorbance at t = 24 h was put at 100 %. There was no significant difference between GFAPd, GFAPa, and control (p = 0.56) (n = 3) at 48 h in U251 cells or d primary astrocytes (p = 0.56) (n = 3). e Cells were stained for PHH3, indicating dividing cells. The average percentage of dividing cells per condition at one time point is not significantly different in cells with GFAPa, GFAPd, and control in U251 cells (p = 0.21) (n = 4) (f). In primary astrocytes, there was a significant difference in proliferation between GFAPa and GFAPd (p = 0.04) (n = 3). g Scratch assays were done by scratching a monolayer of cells and measuring the area uncovered by cells, over time. The area of the scratch at t = 0 was put at 100 %. The bars represent the area where there are no cells at different time points. There was no significant difference (p = 0.7) (n = 3) between GFAPa, GFAPd, and the control after 12.5 h. All graphs show mean with SEM Fig. 6 Morphology of U251 cells with different IF networks. a GFAPd expressing cells showed a more round morphology in comparison to the control (p = 0.009) as is measured by the form factor of these cells. Significant differences between GFAPa and GFAPd were found in b perimeter (p = 0.02) and c area (p = 0.03). Bars show mean and SEM (n = 5). d Phase contrast pictures showing the more round morphology of the GFAPd expressing cells in comparison with the control vector and GFAPa. Scale bars represent 100 lm. e U251 cells expressing different GFAP isoforms are stained for actin with phalloidin (green) and phosphorylated paxillin (red). Scale bars represent 10 lm. The lower panel shows focal adhesions (overlap of phalloidin and phosphorylated paxillin). f Quantification of focal adhesion shows larger focal adhesion size in GFAPa (p = 0.0001) and GFAPd (p = 0.001) expressing cells compared with the control mCherry (mCherry: 1.13 lm 2 ± 0.02 n = 735; GFAPa: 1.34 lm 2 ± 0.03 n = 994; GFAPd: 1.37 lm 2 ± 0.03 n = 828 b influence proliferation or migration when there is a collapse of the IF network. The changes in IF network dynamics could hold clues to GFAP isoform specific functions.