Cofilin nuclear–cytoplasmic shuttling affects cofilin–actin rod formation during stress

Cofilin and its family members are ubiquitously expressed proteins best characterized as actin binding and modulating proteins. These proteins have now been shown to be involved in multiple facets of cellular biology independent of their function with respect to actin tread-milling. These functions are as diverse as involvement in membrane and lipid metabolism (Han et al., 2007), mitochondrial-dependent apoptosis (Klamt et al., 2009) and the regulation of transcription and chromatin structure (Obrdlik and Percipalle, 2011). The improper regulation of the cofilin family members have been implicated in disease mechanisms (Bernstein and Bamburg, 2010). Cofilin1, the non-muscle specific isoform, is of particular interest to the field of neurodegeneration. We have previously reported the involvement of nuclear cofilin–actin rods in a Huntington’s disease (HD) cell model and changes in the cofilin protein expression profile in HD patient lymphocytes (Munsie et al., 2011) and cytoplasmic cofilin–actin rods are involved in the progression of Alzheimer’s disease (AD) (Maloney and Bamburg, 2007; Minamide et al., 2000). Cofilin–actin stress rods can form in either the nucleus or cytoplasm in response to stress (Minamide et al., 2000) as a potential means to stop actin tread-milling and in turn free up ATP for other more immediate cellular uses (Bernstein et al., 2006). Improper regulation of the actin cytoskeleton is becoming a focus in neurodegenerative diseases, as actin dynamics are critical for maintenance of healthy synapses and dendrites (Hotulainen and Hoogenraad, 2010). Thus, understanding actin dynamics under disease and aging related stresses will give important insights into disease mechanisms for neurodegeneration.

During our previous studies, we observed that cofilin could enter the nucleus upon cell stress. Cofilin is known to be actively imported into the nucleus through a classic importin or karyopherin alpha/beta (α/β)-dependent NLS (26-PEEVKKRKKAV-36) (Abe et al., 1993; Iida et al., 1992). Importin-α/β-dependent NLSs are defined by a stretch of basic amino acids that are recognized by the adapter protein importin α and subsequently imported into the nucleus through the nuclear pore complex (NPC) with the assistance of the nuclear transport receptor importin β (Lange et al., 2007). Directionality of nuclear transport is maintained through the small Ras family GTP-binding protein, Ran. Ran–GTP, which is maintained at high levels in the nucleus, binds the nuclear import factors facilitating cargo release in the nuclear compartment. This process is reversed for nuclear export proteins where Ran–GTP binding encourages cargo interaction in the nucleus. RanGAP-mediated nucleotide exchange to Ran–GDP in the cytoplasm disassociates the complex (Lange et al., 2007). Cofilin is an 18 kDa protein that should be able to freely diffuse across the NPC, which has a passive diffusion cut-off point of ∼40 kDa in yeast (Fahrenkrog et al., 2001), but likely higher in mammalian cells (Seibel et al., 2007). Cofilin has been shown to be required for the nuclear transport of actin, although how it is mediating this nuclear entry is unknown (Dopie et al., 2012; Iida et al., 1992). Cofilin recently has been shown to have a role in the nucleus with respect to RNA polymerase II transcription elongation and is likely required for the function of F-actin in the nucleus (Obrdlik and Percipalle, 2011). During certain stress situations, such as DMSO treatment and heat shock, the majority of cofilin in the cell can be found in the nucleus, and at nuclear rods, indicating that the NLS may have a stress-dependent function aside from transporting actin into the nucleus or steady state functions with respect to RNA polymerase activity (Nishida et al., 1987; Obrdlik and Percipalle, 2011; Ono et al., 1993). During our live-cell imaging of heat shock in real time, we noted the rapid exit of cofilin from the nucleus after extended stress. We hypothesized that cofilin may have an active nuclear export signal (NES). The classic CRM-1/exportin-dependent NES has a conserved consensus of Lx_2–3Lx_2–3Lx_1–2L, where L is typically a leucine but can be any hydrophobic amino acid and x is any amino acid (Bogerd et al., 1996). This consensus is recognized by the nuclear export protein, CRM1, which relies on the Ran gradient to transport its cargo across the NPC, releasing the cargo in the cytoplasm (Fornerod et al., 1997). This pathway can be disrupted by leptomycin B, which covalently links to, and inactivates mammalian CRM1 at a single cysteine residue (Kudo et al., 1998). During this study we examined the amino acid sequence of cofilin and defined a conserved CRM1-dependent NES sequence as well as redefined the cofilin NLS as a bipartite NLS.

Cofilin–actin rods are thought to be transiently neuroprotective in cellular models (Bernstein et al., 2006), but improper formation and clearance is thought to be neurotoxic in models of AD and HD (Minamide et al., 2000; Munsie et al., 2011). Since aberrant rod formation may be involved in the progression of AD and HD, it has been hypothesized that altering rod formation or breaking down persistent rods could be therapeutic (Bamburg et al., 2010). By mutating key residues in the cofilin NES and NLS, coupled with endogenous cofilin knock down by short hairpin (sh) RNA, we investigated the actin-binding activity of cofilin mutants in a cell-based model. We made minimal mutations to the cofilin NLS that resulted in loss or decreased rod formation during stress and observed how cells responded to stress when they were unable to form cofilin–actin rods. Importantly, cells expressing cofilin mutants that had abolished rod formation had an impaired ability to respond to stress but a cofilin mutant that could still form rods was able to respond to stress. These data highlight the importance of cofilin activity in response to stress and also imply that altering, but not completely abolishing the rod-forming capabilities of cofilin by targeting the NLS may be tolerated by the cell and may thus define cofilin as a new drug target for some neurodegenerative diseases.

Under steady state conditions, cofilin is diffuse throughout the cell with varying levels in the nucleus (Fig. 1A). During heat shock and other stresses cofilin has been shown to rapidly accumulate in the nucleus (Iida et al., 1992), which we could recapitulate in our cell model system (Fig. 1B). During live-cell imaging of cells stably expressing mCerulean-cofilin and undergoing heat shock stress, cofilin accumulates in the nucleus as predicted and nuclear rods form. If the stress stimulus is not removed following the cofilin rod stress response, cofilin is observed to rapidly exit the nucleus (Fig. 1C; supplementary material Movie 1). The rapid nature of this nuclear export led us to hypothesize the presence of an active nuclear export signal in cofilin. We identified a single sequence in cofilin that fits the putative CRM1-dependent NES consensus (11-VIKVFNDMKV-20) (Bogerd et al., 1996; Fornerod et al., 1997). This sequence is conserved between different species of cofilin and between cofilin1 (non-muscle) and cofilin2 (muscle enriched; Fig. 2A). To initially test this sequence for NES activity, the putative NES sequence, VIKVFNDMKV, was fused in frame to EYFP to see if it would mediate the localization of EYFP to the cytoplasm, compared with the 26 kDa EYFP which can freely diffuse into the nucleus (Fig. 2B). Under normal conditions, EYFP appears diffuse throughout the cell (Fig. 2B). Imaging revealed that the putative cofilin NES sequence can function as an NES since EYFP fused to this sequence appeared predominantly cytoplasmic (Fig. 2B). When cells expressing VIKVFNDMKV–EYFP were treated with the CRM1 inhibitor, leptomycin B (Kudo et al., 1998; Wolff et al., 1997), the fusion protein was observed to be able to diffuse into the nucleus, but no longer exit the nucleus, indicating that this sequence can function as a CRM1-dependent NES (Fig. 2B; supplementary material Movie 2). To test if this identified sequence works as an NES in the context of the full length cofilin, we fused cofilin to EYFP and mutated one of the consensus hydrophobic amino acids in the NES, valine 20 to alanine (V20A). The V20A mutation caused a significant increase in detectable nuclear cofilin when percent nuclear fluorescent signal analysis was performed (Fig. 2C,D; supplementary material Fig. S1A,B and Table S1). This mutation also inhibited the ability of EYFP–cofilin to form rods in either the nucleus or cytoplasm during heat shock stress (Fig. 2C; supplementary material Table S1); however, when immunofluorescence for actin was performed, we show that rods can still form endogenously, in cells expressing the V20A mutant during stress, indicating that overexpression of this protein does not dominantly inhibit endogenous rod formation (Fig. 2E). When cells were treated with leptomycin B and challenged with heat shock stress fewer rods formed endogenously (Fig. 2F,G), and cofilin remained nuclear after stress was removed from the cells as compared to untreated control cells (Fig. 2F,H). The endogenous data recapitulates some of what we observed with overexpressing the cofilin V20A mutant. Cofilin and actin have been shown to be the only required proteins for rod formation in vitro (Minamide et al., 2010) indicating that likely only cofilin–actin is affected in the in vivo leptomycin B rod forming assay. Although CRM1 has many target proteins for nuclear export, we believe that leptomycin B, through CRM1, is specifically affecting cofilin with respect to rod formation since actin nuclear export is mediated through profilin and exportin 6 and is not leptomycin B-sensitive (Dopie et al., 2012; Park et al., 2011; Stüven et al., 2003). From these data, we conclude that cofilin has a CRM1-dependent NES and that NES mutant V20A does not have the ability to form or be incorporated into cofilin–actin rods during stress.

Cofilin has a previously defined putative nuclear localization signal (NLS) at amino acids 26-PEEVKKRKKAV-36 (Iida et al., 1992), which is a monopartite, importin-α/β-dependent NLS. To test whether inhibiting the ability of cofilin to enter the nucleus would also inhibit the formation of cofilin–actin nuclear rods in our system, we mutated critical amino acids lysine 31 and arginine 32 to alanines, KR31-32AA, in the context of the full-length EYFP-cofilin construct, similar to what has been done previously (Iida et al., 1992). Contrary to previous studies by others, we observed no change in the ability of cofilin to enter the nucleus and cofilin KR31-32AA maintained its ability to form rods during stress (Fig. 3A,B; supplementary material Table S1), although fewer cells expressing this construct formed rods compared with the wild-type cofilin sequence (supplementary material Table S1). During stress, a large proportion of cofilin is localized to the nucleus (Fig. 1A,C) and nuclear import via cofilin has also been shown to be critical for actin nuclear entry (Dopie et al., 2012; Iida et al., 1992). These data suggest that either cofilin may enter the nucleus alone by simple diffusion, or that additional amino acids are required for optimal active nuclear entry.

There is a second type of NLS that is recognized by importin α/β, known as a bipartite NLS. The bipartite NLS contains two basic rich regions, generally separated by 10–12 amino acids, with the consensus (K/R)(R/K)x_10–12(K/R)_3–5 (Robbins et al., 1991)_. Seven amino acids upstream of the KKRKK sequence in cofilin there are two basic amino acids, RK. The sequence (21-RKSSTPEEVKKRKK-34) in cofilin loosely fits the bipartite NLS consensus. Both basic parts of the bipartite NLS in cofilin are evolutionarily conserved between different species of cofilin and between cofilin1 and cofilin2 (Fig. 3C). In order to determine if either the monopartite NLS (cofilin amino acids 26–34) or the bipartite NLS (cofilin amino acids 21–34) were functional, these sequences from cofilin were taken out of context and fused between GFP and β-galactosidase (β-Gal). β-Gal is 112 kDa, tetramerizes, and lacks an endogenous NLS. The triple fusion of GFP–putative-NLS–β-Gal allowed us to assess the activity of the cofilin NLS sequence by its ability to heterologously mediate GFP–β-Gal entry (>550 kDa as tetramer) into the nucleus (Sorg and Stamminger, 1999). As a positive control, we expressed GFP–SV40NLS–β-Gal as the SV40 NLS is a well-defined monopartite NLS that is known to mediate β-Gal nuclear entry (Kalderon et al., 1984). While the positive control is active in our cell lines, the putative cofilin monopartite NLS sequence was not able to mediate β-Gal nuclear entry (Fig. 3D,E), indicating that amino acids 26–36 in cofilin do not function as an NLS in this classic assay. The extended putative cofilin bipartite NLS was able to mediate the nuclear entry of GFP–β-Gal indicating that it can function as an active NLS (Fig. 3D,E). When both critical basic regions were mutated to alanines in the triple fusion GFP–(AASSTPEEVKAAKK)–β-Gal, the sequence could no longer mediate nuclear entry of β-Gal (Fig. 3D,E). Method and controls for percent nuclear fluorescent signal analysis are shown in supplementary material Fig. S2. Additional testing of these constructs in mouse NIH 3T3 fibroblasts confirmed that the NLS activity was not cell type specific (supplementary material Fig. S1C). To test whether the bipartite NLS sequence functions in the context of the full length cofilin protein, we mutated both basic regions to make the fusion EYFP–cofilin RK21-22AA KR31-32AA. Under steady state conditions, EYFP–cofilin RK21-22AA KR31-32AA had less percent nuclear fluorescent signal compared to wild type EYFP–cofilin (Fig. 3F,G; supplementary material Table S1). As with the cofilin NES mutant, the cofilin NLS mutant is no longer capable of forming nuclear or cytoplasmic rods (supplementary material Table S1). The NLS mutant in the context of full length cofilin was also functional in NIH 3T3 fibroblasts indicating this is not a cell type specific effect (supplementary material Fig. S1B and Table S1).

To test if the decrease in nuclear signal of EYFP–cofilin KR21-22AA RK31-32AA, the NLS mutant (Fig. 3F,G), is in part mediated by active nuclear export via the NES, and not simply an exclusion from the nucleus due to the cumulative size of EYFP fused to cofilin (44 kDa in total), we treated cells expressing EYFP–cofilin or EYFP–cofilin RK21-22AA KR31-32AA with leptomycin B. We also created a full-length, NLS mutant cofilin construct fused between GFP and β-Gal. Due to the high molecular weight of β-Gal in addition to cofilin and GFP (156 kDa as monomer, 624 kDa as tetramer), this construct far exceeds the predicted diffusion limit of the nuclear pore and is thus too large to diffuse into the nucleus upon leptomycin B treatment, even with an inactive NES. After leptomycin B treatment both EYFP–cofilin and EYFP–cofilin RK21-22AA KR31-32AA had an increase in nuclear fluorescence, however, the GFP–cofilin RK21-22AA KR31-32AA β-Gal construct did not (Fig. 3H,I; supplementary material Table S2), indicating the NLS and NES are functional in the context of the EYFP–cofilin protein and that the EYFP–cofilin is not being excluded from the nucleus due to size when the NLS is mutated (Fig. 3H,I).

In attempt to further verify these findings, we tested our cofilin NLS and NES mutants fused to the 22 amino acid tag, 3xFLAG (DYKDHD-G-DYKDHD-I-DHKDDDDK). We expressed these constructs and performed immunofluorescence using anti-FLAG monoclonal antibody. Under steady state conditions, we found no difference in percent nuclear fluorescent signal regardless of mutation (supplementary material Fig. S3 and Table S3). However, during heat shock stress, we did observe a reduced ability of FLAG-tagged cofilin NLS mutant to enter the nucleus compared to those constructs where the NLS remained intact (supplementary material Fig. S3 and Table S3). We interpreted these results cautiously, since the cofilin–3xFLAG construct showed signs of functional impairment, forming almost no rods during heat shock, whereas EYFP–cofilin forms rods in a similar manner to endogenous cofilin (Munsie et al., 2011) (supplementary material Tables S1, S3). This may have been due to the significant increase in acidic charge (11 of 22 residues) due to the 3xFLAG fusion to cofilin or due to differences and artifacts in immunofluorescent imaging versus direct live-cell imaging (Schnell et al., 2012). The differences are not due to tag location since cofilin behaves the same when tagged with EYFP at either the amino or carboxyl terminus (data not shown).

It has been hypothesized that being able to inhibit rod formation or break down rods could be a novel therapeutic approach for HD or AD (Bamburg et al., 2010); however, the absolute requirement of cofilin–actin rod formation during cell stress response is unknown. We hypothesized that regulating cofilin nuclear import and export may be an avenue to modulate rod formation and that we could use these mutants to test our hypothesis in vivo. Therefore, we had to ascertain if these mutants could still bind actin and therefore still be able to perform other critical cellular functions. Using an in vitro F-actin co-sedimentation assay, we show that both V20A and RK 21-22AA KR31-32AA have some ability to bind F-actin, however neither bound as strongly as wild-type cofilin (supplementary material Fig. S4). To measure the cofilin–actin interaction in vivo, we employed fluorescent lifetime imaging microscopy (FLIM) to accurately measure Förster resonance energy transfer (FRET) between cofilin and actin in a live-cell environment. FRET can be used to measure protein–protein interactions at the nanometer scale in live cells by calculating efficiency of the resonant transfer of energy from a donor to an acceptor fluorophore. For mCerulean and EYFP donor–acceptor pairs, optimal FRET efficiency occurs at 8 nm or less, for a maximal possible FRET efficiency of ∼30%, as dictated by the degree of spectral overlap between mCerulean emission and EYFP excitation. FLIM measures this transfer of energy by measuring the change in the lifetime of the donor fluorophore in the absence and presence of the acceptor fluorophore (Llères et al., 2007; Llères et al., 2009). FRET determination by this method is independent of donor/acceptor fluorophore concentrations, as only the donor lifetime is being measured, thus is optimal for two molecule FRET, without concern of spectral overlap as with other methods. FRET efficiency is calculated using the formula E_FRET = 1−(average lifetime D.A/average lifetime donor), where average lifetime D.A indicates the average lifetime of the donor in the presence of the indicated acceptor. We used mCerulean-actin as the donor and EYFP alone as a negative control acceptor, EYFP–cofilin as a positive control acceptor and EYFP–cofilin NES and NLS mutants as the experimental acceptors (Fig. 4). The lifetime of mCerulean alone is approximately 2.9 nanoseconds (ns) (Rizzo et al., 2004), and when mCerulean was tagged to actin this construct had an average lifetime of approximately 2.7 ns. In the presence of molar excess of EYFP alone (negative control) the lifetime of mCerulean–actin was slightly decreased giving a FRET efficiency of 2.6%. In the presence of EYFP–cofilin (positive control), which is known to interact with mCerulean–actin to allow FRET, the mCerulean–actin lifetime was decreased and FRET efficiency was 9.0%. In the presence of the cofilin NLS mutant there was a FRET efficiency of 9.6%, similar to EYFP–cofilin indicating that this mutant retains the ability to interact with actin in vivo. In the presence of the cofilin NES mutant the FRET efficiency was 2.6%, which was the same as EYFP alone, indicating that this cofilin mutant has lost its ability to interact with actin in our live-cell-based assay (Fig. 4B). Based on this data, we decided to further assay the requirement of rod formation using the cofilin NLS mutant.

Although the cofilin NLS mutant, RK21-22AA KR31-32AA, could still bind actin in vivo, we wanted to test if a minimal mutation to the cofilin NLS would have the same effect on nuclear localization, actin binding and rod formation as RK21-22AA KR31-32AA. We created two new mutations in the first basic region of the bipartite NLS, resulting in EYFP–cofilin R21A and EYFP–cofilin K22A. We did not focus on KR31-32AA as we already had found mutating these residues did not affect cofilin nuclear localization and this mutation did not abolish rod formation (Fig. 3A; supplementary material Table S1). We measured the ability of the R21A and K22A cofilin mutants ability to bind F-actin in vitro and found that these mutants maintained some F-actin binding ability (supplementary material Fig. S4A,B). As expected, both the R21A and K22A cofilin mutant fusion proteins retained the ability to bind actin in vivo by measuring FLIM–FRET (Fig. 5A,B). Quantification of percent nuclear fluorescent signal showed that each of these mutations affected the ability of cofilin to enter the nucleus (Fig. 5C) but not to the same extent as mutating both halves of the NLS (supplementary material Table S1). We subjected cells expressing the R21A and K22A constructs to cofilin/actin rod-inducing stress. Under conditions of heat shock or ATP depletion, the R21A mutant did not form any cofilin rods (Fig. 5D; supplementary material Table S1), however the K22A mutant still formed rods in 2.8% of cells after a one hour heat shock, although fewer than wild type EYFP–cofilin which formed rods in 37% of cells (Fig. 5D; supplementary material Table S1). We additionally performed an in vitro F-actin co-sedimentation assay at low speed centrifugation, to determine actin bundling activity of our cofilin mutants and found that only wild-type cofilin and cofilin K22A had actin bundling activity, whereas all other mutants, NLS and NES had no F-actin bundling activity in vitro (supplementary material Fig. S4C,D,F) which may correspond to rod forming ability in vivo. This assay could just be detecting F-actin binding and further work would need to be done to show any bundling or rod forming activity, however during low speed spins wild-type cofilin and K22A cofilin were the only proteins to co-sediment with F-actin consistently, although varying amounts of F-actin in the pellet and supernatant were observed during these low speed experiments.

To reduce effects of endogenous wild-type cofilin, we created a system where cofilin mutants could be expressed on the same plasmid as an shRNA to endogenous cofilin allowing simultaneous knockdown of the endogenous protein and overexpression of our cofilin mutants. The EYFP–cofilin contains a silent mutation in the cofilin DNA sequence which makes it immune to shRNA knockdown (supplementary material Fig. S5A). We tested these constructs for knockdown of endogenous cofilin and found that the shRNA worked effectively on endogenous cofilin but allowed our EYFP constructs to be expressed (supplementary material Fig. S5B–E). Knocking down endogenous cofilin had no effect on our EYFP–cofilin and cofilin mutants ability to enter the nucleus or form rods (supplementary material Table S1). When observing cells undergoing heat shock with cofilin knocked down we noted that cells would rapidly shrink, a sign of poor cell health or apoptosis (Fig. 6A; supplementary material Movie 3) whereas cells where wild-type EYFP–cofilin was re-expressed tolerated cell stress by forming rods and generally maintained their cell shape and size (Fig. 6A; supplementary material Movie 4). During heat shock stress, rods formed by endogenous cofilin and actin almost exclusively form in the nucleus; however, when overexpressing cofilin we often get rods forming in the nucleus and cytoplasm in response to heat shock likely due to a higher local cofilin–actin ratio in the cell, as shown in supplementary material Movie 4. The ability of cofilin to form rods in general appears to be protective, and it seems that the shuttling capabilities of cofilin is required for both nuclear and cytoplasmic rods, likely through a similar mechanism. Since cell shrinkage is a sign of apoptosis (Friis et al., 2005), we wanted to investigate if these cells with knocked down levels of cofilin and undergoing heat shock stress were shrinking due to apoptosis or shrinking due to another reason, since cofilin knockdown can actually inhibit some forms of oxidant-induced apoptosis (Klamt et al., 2009). To analyze this we used immunofluorescence against two proteins that change localization during apoptosis: cytochrome c and activated Bax (Annis et al., 2001), in cells expressing our cofilin-shRNA plasmid before and after heat shock. For the Bax immunofluorescence, we used an antibody that recognizes an epitope to Bax that is only exposed when it undergoes a conformation shift during apoptosis (Eskes et al., 1998). After heat shock we found a shift in localization from mitochondria (supplementary material Fig. S6A) to diffuse cytochrome c staining (supplementary material Fig. S6B) indicating apoptosis, as well as an increase and change in localization of Bax in cells expressing cofilin-shRNA–EYFP (supplementary material Fig. S6A,B). These cells are also visibly smaller than surrounding cells not expressing cofilin-shRNA–EYFP, indicating that these shrinking cells are likely undergoing apoptosis under stress conditions. We therefore used cell size after stress to measure cell health in our cofilin-knockdown–EYFP–cofilin-overexpression system.

To objectively quantify our assays with robust numbers and avoid investigator bias, we used an image based assay coupled with open-source software, CellProfiler^TM, to determine cell health based on cell morphology by automated image scoring (Lamprecht et al., 2007). The software can automatically identify objects from an image, measure and sort them based on pre-defined guidelines (pipelines). This method allows the unbiased quantification of large numbers of cells. An example of the pipeline we created is shown in supplementary material Fig. S7. We imaged live cells expressing these constructs both prior to stress and directly after heat shock stress at 20× magnification. Using cell size, approximately 50% of cells had a ‘healthy phenotype’ based on cell size prior to cell stress (Fig. 6B). All of our knockdown constructs had baseline toxicity, possibly due to the knockdown of cofilin and the overexpression of the EYFP–cofilin constructs from the CMV promoter. There were no differences in any of our constructs at steady state, likely due to the fact that some aspects of cofilin1 knockdown can be rescued by other family members including ADF (actin depolymerizing factor) (Hotulainen et al., 2005). Immediately after stress there was a significant increase in the unhealthy cell population when EYFP alone, V20A, RK 21-22AA KR31-32AA or R21A were expressed, however there was less of a toxic effect when K22A was expressed compared to re-expression of wild-type cofilin (Fig. 6C). This indicates that although cofilin mutants that cannot form rods can still bind actin, and under steady state do not have an obvious effect on cell health, they cannot rescue cells during stress. However, the K22A mutant maintained some ability to form rods and had some actin bundling activity in vitro and could mediate a better stress response than the other cofilin mutants (Fig. 5D, Fig. 6C; supplementary material Fig. S4D). This indicates that the cofilin protein is likely integral to cell health during stress and the rod forming function may be required for cell survival under stress conditions.

At 18 kDa, cofilin can passively diffuse in and out of the nucleus, however a large proportion of cofilin localizes to the nucleus during stress (Iida et al., 1992; Nishida et al., 1987) and recently cofilin has been shown to be required for actin nuclear import (Dopie et al., 2012). Actin is known to be transported out of the nucleus when bound to profilin through the exportin 6 pathway, and does not require cofilin for export (Dopie et al., 2012; Stüven et al., 2003). Although actin is now known to be active in the nucleus and therefore many actin binding proteins may be as well, cofilin seems to have its own nuclear functions aside from actin transport. Cofilin has been shown to be in complex with actin and phosphorylated RNA polymerase (pol) II and depleting cofilin from the cell affects active transcription (Obrdlik and Percipalle, 2011). It is hypothesized that cofilin is involved in regulating actin polymerization in the nucleus to accomplish this function; when cofilin is depleted from the cell actin no longer enters the nucleus and there is a decrease in pol II transcription, however simply forcing actin into the nucleus by tagging it with an NLS when cofilin is knocked down does not rescue these defects (Dopie et al., 2012). We further show that during certain stress events a large proportion of cellular cofilin can be found in the nucleus and is found at nuclear cofilin–actin rods. Small proteins, i.e. the 7 kDa HIV-1 Rev protein, can have active nuclear transport signals as the facilitated diffusion mediated by these signals through nuclear transport factors is far faster than passive diffusion (Mohr et al., 2009). We initially noted that cofilin rapidly exits the nucleus during prolonged cell stress and hypothesized the existence of an NES in cofilin. By visually searching the protein amino acid sequence we found a single sequence that fit the CRM1-dependent NES consensus. We found that this sequence was functional as an NES and was responsive to leptomycin B, indicating that export is through the classic CRM1 pathway. Using live-cell imaging FLIM–FRET, we found that when this sequence was mutated at a single valine, V20A, cofilin lost its ability to bind actin in vivo. Subsequently, other amino acid substitutions were assayed at position 20, but with similar loss of actin association (data not shown). While any actin-binding mutant of cofilin is likely to have pleiotropic effects on cofilin or cytoskeletal functions, the coincident phenotypes of NES inactivation and abrogated actin binding with V20A may suggest that actin association may competitively inhibit the NES by competing CRM1 and actin interactions.

Recently, actin was shown to compete with importins for binding to JMY, a regulator of transcription and actin filament assembly. In the presence of actin monomers, the nuclear import signal of JMY is inhibited (Bradley Zuchero et al., 2012). Other actin associated protein import and export is controlled by actin binding; including MAL, the serum-response factor co-activator. MAL has CRM1-dependent nuclear export activity that requires it to be bound to G-actin for nuclear export (Vartiainen et al., 2007). MAL has a bipartite NLS with a long intervening sequence that is activated when the G-actin pool is depleted (Pawłowski et al., 2010). The nuclear–cytoplasmic trafficking of these proteins is dependent on the polymerization state of actin. We propose that under cell stress events, the enhanced association of cofilin with F-actin in the nucleus at rods may prevent CRM1 interaction, thus allowing increased nuclear cofilin levels by active import of cofilin and the prevention of CRM1-mediated nuclear export of cofilin. The release of cofilin from actin rods would permit CRM1 interaction and thus reduce nuclear cofilin levels by active nuclear export (see model in supplementary material Fig. S8). This model would connect the higher order structural state of nuclear actin with cofilin sub-cellular localization, and hence compartmental activity.

When the putative cofilin NLS (26-PEEVKKRKKAV-36) was mutated, we found no effect on nuclear localization under steady state conditions or stress conditions, and cofilin maintained its ability to form some rods in the nucleus, contrary to previously reported data (Iida et al., 1992). Discrepancies between data may reflect the different mutations: we chose to use alanine substitution with hopes of not disrupting the secondary structure of the protein, but still removing some critical basic charge, and that we performed direct imaging rather than indirect imaging methods. Upon closer examination of the amino acid sequence we found that seven amino acids upstream of the basic rich ‘KKRKK’ sequence there was a conserved ‘RK’ sequence. The sequence (21-RKSSTPEEVKKRKK-34) is consistent with a bipartite NLS sequence. The intervening sequence between the two basic rich sequences in a bipartite NLS is generally 10–12 amino acids long. In the cofilin NLS, this intervening sequence is shorter than predicted, however, it has been reported that the number of amino acids between the basic regions of a bipartite NLS can vary greatly, being either longer (Lange et al., 2010) or shorter (Taniguchi et al., 2002) than the earlier defined consensus. When we mutated both basic regions of the bipartite NLS (AASSTPEEVKAAKK) the nuclear localization signal was not functional, either out of context or in context of full length cofilin. We treated cells expressing the NLS mutant or wild-type cofilin with leptomycin B which induced increased nuclear accumulation of both constructs. If leptomycin B treatment were to cause a generalized cell stress, the increased nuclear accumulation of EYFP–cofilin could be attributed to its normal stress related function. However, since the NLS mutant also became more nuclear in response to leptomycin B, we can conclude that the compound is working more directly, by blocking the activity of the cofilin NES. This further supports our conclusion that cofilin has an active NES that is CRM1 dependent, and indicates that EYFP–cofilin RK21-22AA KR31-32AA exclusion from the nucleus under steady state conditions is due to active nuclear export and not simply protein size. However, tagging EYFP to cofilin does bring the size of the protein closer to the diffusion limit of the nuclear pore. To try and determine if the NLS and NES are constitutively active and not just stress-dependent signals, we tagged full length cofilin NLS or NES mutants to the 3xFLAG tag, which is much smaller than EYFP, and assessed nuclear localization of cofilin–3xFLAG using immunofluorescence. No changes were seen in the nuclear localization of cofilin under steady state conditions by immunofluorescence (supplementary material Fig. S3), however we did observe that when the NLS is mutated in this context, cofilin did not enter the nucleus during stress. The wild-type cofilin–3xFLAG vector did not form rods as efficiently as endogenous or EYFP tagged cofilin (supplementary material Tables S1, S3). To date, the EYFP–cofilin construct has behaved the most appropriately in comparison to endogenous cofilin with respect to nuclear localization and rod formation during stress.

The ability of cofilin to enter the nucleus in the presence of leptomycin B was different when the cofilin NLS mutant was fused to EYFP as opposed to β-Gal, indicating that EYFP–cofilin, although larger and possibly having different dynamics than endogenous cofilin, is still being shuttled out of the nucleus by some active transport and not simple diffusion. Although it is likely that active shuttling of cofilin is required during stress (Fig. 1); the changes in localization of EYFP–cofilin NLS and NES mutants under no stress conditions suggests that there may be active shuttling of cofilin into and out of the nucleus under steady state conditions as well. With the newly defined role of cofilin in RNA polymerase activity (Obrdlik and Percipalle, 2011), it will be interesting to determine if the cofilin NLS is required for this function under steady state conditions; either bringing actin into the nucleus or if a specific pool of nuclear cofilin is required to regulate the polymerization of actin at actively transcribing genes. However, we focused our studies on the requirements of active transport of cofilin in response to stress.

The RK21-22AA KR31-32AA mutation did not impair the ability of cofilin to bind actin in vivo. This mutant did inhibit rod formation and its expression under stress conditions affected cell health. Single mutations to the first basic regions in the bipartite NLS, R21A and K22A affected nuclear localization of cofilin, however not to the same extent as mutating both halves of the NLS, but also maintained their ability to bind actin. There were differences in these mutants abilities to bind F-actin in vitro and therefore may have different affinities for F-actin versus G-actin or may differ in their abilities to sever F-actin. While the R21A mutant could not form rods under any stress condition tested, the K22A mutant could still form rods, although to a lesser extent than wild-type cofilin, and was the only mutant that had F-actin bundling/binding activity in vitro when co-sedimentation was performed at low speeds. When analyzing cell health in cells expressing these constructs, the R21A mutant had fewer healthy cells after stress compared to the K22A mutant or wild-type cofilin, indicating that the ability of cofilin to properly bind actin and form cofilin–actin rods may be critical to cell health under stress conditions. Cofilin–actin rods form rapidly in response to many ATP depleting stresses (Minamide et al., 2000; Nishida et al., 1987), stresses that likely affect all cells in the body under different conditions. This may be particularly relevant in neurodegeneration since ATP levels are critical to neuronal health, as are actin dynamics (Hotulainen and Hoogenraad, 2010). This may be additionally relevant to neurons undergoing stresses associated with the aging process (Gibson et al., 2010).

Aberrant rod formation has been described in AD and HD, in the form of persistent rods, or improper rod formation and dynamics (Bamburg et al., 2010; Minamide et al., 2000; Munsie et al., 2011). Persistent rods could lead to trapped cofilin and altered cytoskeletal dynamics after stress leading to neurite and dendrite dystrophy or dysfunction. Alternatively, improper rod formation could mean that cells do not have the required ATP levels available to respond to stress, leading to cell dysfunction or death. Notably, cofilin–actin inclusions have been visualized in AD brains (Minamide et al., 2000). The involvement of both actin and actin binding proteins (Lim et al., 2007; Maloney and Bamburg, 2007; Mila et al., 2009) as well as microtubule related proteins (Atwal et al., 2011; Godin et al., 2010) in neurodegeneration is becoming more evident, and cell stress has been hypothesized to be involved in these disorders (Gibson et al., 2010; Keller, 2006). Thus, being able to modulate the stress response in relation to the cytoskeleton may be a novel therapeutic approach for neurodegeneration and was the driving force behind this work. To this end, we have defined the import and export signals of cofilin, and have further defined two cofilin mutants, discovered by altering cofilin shuttling activity that affect rod formation and cell health during stress. The cofilin R21A mutant binds actin in vivo, affects nuclear localization and does not form rods, and thus would be a good model to discern the requirement of rod formation in vivo. As more is revealed about the necessity of cofilin–actin rod formation for cell health and survival, the K22A mutant will offer valuable insight into the level which modulating rod formation can be tolerated since this mutant has all the same features of the R21A mutant except it has increased in vitro F-actin bundling activity as well as some limited capacity to form rods. If rod formation is intimately tied into the proper ability of cofilin to bind actin, it may not be possible to target rod formation directly through cofilin. However, if the R21A or K22A mutations are tolerated in stress conditions or cell types tested, then compounds that specifically inhibit or alter cofilin nuclear shuttling may be therapeutic without affecting other essential actin dynamics. The cofilin NLS is bipartite with a unique intervening sequence, thus it may be possible to target this nuclear import activity without affecting other proteins NLSs. Notably, there is a conserved serine, serine, threonine (SST) motif in the intervening sequence of the cofilin bipartite NLS (Fig. 3C). This SST motif in the cofilin NLS may be a target of post-translational modification and is the subject of further study.

Mouse STHdh^Q7/Q7 cell line (a kind gift from M. E. MacDonald, Massachusetts General Hospital, derived from the mouse striatum, were grown in Dulbecco’s modified Eagle’s medium (Invitrogen) with 10% fetal bovine serum (Invitrogen) at 33°C with 5% CO₂. Striatal cells were clonally selected and grown under G418 drug selection at 33°C to ensure temperature sensitive selection and maintain SV40 Tag expression. Derivation and characterization of stable cell lines used were previously described (Munsie et al., 2011).

Mouse fibroblast NIH 3T3 cells were grown in Dulbecco’s modified Eagle’s medium (Invitrogen) with 10% fetal bovine serum (Invitrogen) at 37°C with 5% CO₂.

EYFP-human cofilin plasmid used was previously described (Munsie et al., 2011). All primers were synthesized and all sequencing was performed at McMaster MOBIX Facility. All primer sequences and detailed cloning protocols or constructs are available upon request.

Primers to mutate the NES (V20A) and the NLS (KR31-32AA) were made and mutations achieved according to manufacturer’s protocol using the QuickChange kit (Stratagene). Primers to mutate the first part of the bipartite NLS: RK21-22AA, R21A or K22A, were made using inverse PCR on the EYFP-cofilin or e-EYFP-cofilin KR31-32AA plasmids. Cofilin shRNA plasmids were purchased (HuSH, Origene) in pRFP-C-RS vector. Cofilin shRNA F1575659 was used for all experiments. To make silent mutations in the EYFP-cofilin plasmid so that it would not be susceptible to the shRNA inverse PCR was used. Inverse PCR was used to remove the RFP cDNA from the cofilin shRNA vector. Primers to express EYFP-cofilin (with silent mutations) or EYFP were made with 5′ MluI and 3′ PmeI overhangs and cloning was performed, using PCR product from all EYFP-cofilin plasmids or EYFP alone plasmid (previously described), between MluI/PmeI sites into the modified cofilin shRNA constructs. Primers to express human cofilin1 were made with 5′ Acc651 and 3′ XbaI overhangs and cloning was performed, using PCR product from our previously constructed cofilin and cofilin mutant plasmids, between Acc651/XbaI sites of 3xFLAG-CMV-14 (Sigma) to create the cofilin-3xFLAG plasmids.

mCerulean–human β-actin was a kind gift from Kotaro Oka, (Keio University, Japan).

Plasmids were transfected using the cationic-polymer-based in vitro transfection reagent Turbofect^TM (Fermentas) according to manufacturer’s instructions for different sizes of culture dishes (i.e. 2.0 μg DNA for 25 mm dishes). Expression time was 24 hours unless otherwise indicated in the methods.

The synthetic NES was made with 5′ BspEI and 3′ Acc651 overhangs and cloned using the annealed product into pEYFPC1 (BD Biosciences/Clontech) to create synthetic NES–EYFP.

The assay for nuclear export was performed as previously described (Xia et al., 2003).

The assay for NLS activity was based on the pHM830 GFP–β-galactosidase triple fusion plasmid construct and conducted as described (Sorg and Stamminger, 1999). Synthetic DNA oligonucleotides encoding the cofilin NLS, bipartite NLS and bipartite NLS mutations were made with 5′ SacII and 3′ XbaI overhangs (MOBIX, McMaster University) and cloned using the annealed products into GFP–MCS–β-Gal (pHM830 vector) to create GFP–‘synthetic NLS’–β-Gal.

To make the GFP–cofilin RK21-22AA KR31-32AA–β-Gal construct, primers to express human cofilin1 were made with 5′ SacII and 3′ XbaI overhangs and cloning was performed as described previously.

Cells were cultured in 25 mM live-cell culture dishes at ∼75–85% confluence and transfected as described. If leptomycin B was required it was added for 4 hours prior to heat shock. Medium was exchanged for HEPES-buffered (20 mM pH 7.4) medium for all live-cell imaging and heat-shock experiments. Plates were wrapped with Parafilm and placed in a pre-warmed water bath at 42.5°C for 60 minutes. Cells were fixed in 4% paraformaldehyde for 45 minutes at room temperature. Cells were permeabilized by the addition of pre-cooled methanol and placed at −20°C for 5 minutes. Cells were blocked in 1% FBS in PBS for 45 minutes. Primary antibody anti-cofilin (mAB22 a kind gift from J. Bamburg, 1:250 dilution), anti-3xFLAG (Sigma; 1:500 dilution), and anti-cytochrome c and anti-Bax (6A7) (kind gifts from D. Andrews; 1:500 dilution) were applied in antibody solution (1% FBS blocker + 0.02% Tween 20) for 2 hours. Secondary antibodies conjugated to Alexa Fluor 488 or 594 (Molecular Probes) were used for 30 minutes at room temperature in antibody solution.

All 60× widefield fluorescence microscope images were captured on a Nikon TE200 epi-fluorescence inverted microscope (Nikon, USA) equipped with a 60× oil immersion plan apochromat NA1.4 objective and a Hamamatsu Orca ER digital camera (Hamamatsu Photonics, Japan). All 20× widefield fluorescence microscope images were captured on an Evos digital LED inverted microscope equipped with a 20× air plan fluor NA1.2 objective (AMG, Seattle, USA).

To determine the percent nuclear fluorescence, cells were imaged at 60× magnification and then analyzed using Compix Simple PCI (Hamamatsu, Japan). To ensure unbiased data collection, the cells were co-transfected with pmCherry-C1 (Clontech/Invitrogen), and cells were selected under the red channel before being imaged in the green channel, in a method of unbiased imaging data acquisition (Voss et al., 2005). Nuclear and total cell intensities were collected by manually defining the nuclear (nucleus was defined using either brightfield or Hoechst staining) region and the outline of the entire cell. For each cell the total intensity was measured and compared to the total intensity of the nucleus. Areas of equal size to represent the image background (supplementary material Fig. S2) were obtained. The intensities of the defined regions were then measured using the Simple PCI measurement tool. Percent nuclear fluorescence was calculated using the equation: % nuclear fluorescence = [(nuclear fluorescence−background)/(total cell fluorescence−background)]/100. For quantification of immunofluorescent images when looking at endogenous protein (Fig. 2H), background subtraction was not performed since every cell in the image was fluorescent and there was not enough background area available.

Live-cell visualization was performed using the Delta T4 heated stage, lid, and objective system (Bioptechs Inc., Butler, PA, USA). Cells were seeded and treated in 0.17 mm delta T dishes (Bioptechs Inc.). Cells were heated to 42.0°C using the heated stage and objective and visualized at 100× oil immersion NA1.3 objective with an objective heater. As soon as dish temperatures reached 42.0°C, fluorescence images were recorded once every 60 seconds for the duration of the session.

Cells were transfected and plasmids expressed for 36 hours in 5 cm dishes, washed with PBS and collected using a rubber scraper. Cells were pelleted, incubated and re-suspended in NP-40 lysis buffer with protease inhibitor cocktail. Cells were spun at 14,000 g for 10 minutes and the supernatant (protein fraction) was collected.

Equal amounts of protein were loaded on 12% SDS–polyacrylamide gel and electroblotted to a poly-vinyl difluoride (PVDF) membrane. Membranes were blocked with 5% non-fat dry milk in TBST for 1 hour followed by 1 hour incubation at room temperature with anti cofilin mAB22 (a kind gift from J. Bamburg), anti-EYFP (Clontech/Invitrogen) or anti-actin (Sigma). After incubation with appropriate HRP-conjugated secondary antibody (Sigma) bands were visualized by enhanced chemiluminescence using the MicroChemi system (DNR Bio-Imaging Systems, Israel). Quantification of western blot bands was performed using the NIH ImageJ software, and pixel intensity quantification, normalized to actin or EYFP–cofilin.

Statistical analysis was performed using the SigmaPlot Software 11.0 (Systat Software Inc.) For single comparisons Student’s t-tests were performed if data passed normality assumptions. If data did not pass normality assumptions it was tested by the Mann–Whitney method.

FLIM was conducted using an inverted confocal laser-scanning microscope (Leica TCS SP5) with a 63× glycerol immersion NA 1.4 Plan apochromat objective. The SP5 is run using the LAS Advanced Fluorescence software from Leica. Two-photon excitation of samples was performed using a tunable multi-photon (MP) Chameleon laser (Coherent, USA), mode-locked to deliver sub-femtosecond pulses at a rate of 80 MHz with an output power of 1.8 W for a peak wavelength of 820 nm. Laser output was restricted to 12.5%.

EYFP and mCerulean fluorophores were used as FRET pairs. Excitation of the mCerulean donor using the MP laser was found to be optimal at 820 nm. Collection of mCerulean fluorescence emission was gated to 480 nm (±20 nm). All live-cell imaging and FLIM was performed in Hank’s saline HEPES buffer, pH 7.4.

Photons from the donor fluorophores were collected, and counted using a Becker-Hickl photon counter and TCSPC software (SPC-830). The laser power was adjusted in order to give a photon collection count of ∼10e5 photons/sec, where all FLIM measurements were conducted over a 60 second collection time. The lifetimes of all the pixels in the field of view (256×256) were calculated by the SPCImage analysis software (Becker & Hickl GmbH, Germany) to generate mono-exponential decay curves. Binning and thresholding values (bin = 3, threshold = 10) were kept constant to ensure consistency of lifetime measurements over multiple trials.

FRET analysis was performed using the Becker & Hickl FLIM plug-in for ImageJ software (McMaster Biophotonics Facility, www.macbiophotonics.ca). The lifetime of every pixel in the image was calculated to give a mean lifetime value (τ) for each cell when fit to a mono exponential decay curve. Pixels with lifetimes outside the range of 1750–3250 were excluded from further analysis, and visually identified via the intensity-weighted lifetime image to exclude areas of artifactual lifetimes due to very low or very high intensities. FRET efficiency for each image was determined using the equation: E_FRET = 1−(average lifetime D.A/average lifetime D), where average lifetime D.A indicates the average lifetime of mCerulean–actin in the presence of the indicated acceptor and average lifetime D indicates the overall average lifetime of mCerulean–actin alone with no acceptor present.

Automated Image analysis for cell health was done using the CellProfiler cell image analysis software (http://www.cellprofiler.org). We created a pipeline to classify healthy versus unhealthy cells based on size. Cells were transfected with the shRNA constructs of interest and plasmids were allowed to express for 48 hours. Images at 20× magnification were taken and cells were subjected to a 1 hour heat shock. Images were then taken directly after heat shock. Images were loaded into cell profiler and analyzed for number of healthy versus unhealthy cells based on size and morphology.

Primers to express human cofilin1 were made (McMaster Mobix facility) with 5′ BamHI and 3′ EcoRI overhangs and cloning was performed, using PCR product from all cofilin mutant plasmids previously made, between BamHI/EcoRI cut sites of pGEX 5X-1 (GE Healthcare) to create cofilin-GST-cofilin expression plasmid.

Wild-type and mutant cofilin constructs were then expressed as glutathione S-transferase (GST) fusion proteins in E. coli BL21 cells. Cells were grown in 200 ml of LB medium to an optical density of 0.7 at 600 nm, and expression was induced with isopropyl-thio-beta-D-galactoside (IPTG; 0.2 mM). Cells were harvested 6 h after induction, washed with 200 ml of 150 mM NaCl, resuspended in 20 ml of PBS and lysed by sonication. GST fusion proteins were enriched using glutathione–Sepharose 4B (GE Healthcare), washed three times with PBS and then eluted using elution buffer (50 mM Tris-HCl, 10 mM reduced glutathione). The samples were dialyzed into 50 mM Tris-HCl pH 6.8 overnight using Slide-a-Lyzer mini dialysis units (Thermo Scientific).

F-actin co-sedimentation experiment was performed using the actin binding protein kit (non-muscle actin, catalog number BK013) from Cytoskeleton Inc. (Denver, USA). Changes to protocol include using the provided actin at 10 µM and cofilin mutants in excess at 10–15 µM. Actin buffer was made in 5 mM Tris-HCl pH 6.8, 0.2 mM CaCl_2. Cofilin and actin were incubated for 40 minutes and fast spins for F-actin binding assay were performed in a Beckman ultracentrifuge using the TLA100 rotor at 150,000 g. Slow spins for F-actin bundling assay were performed in a table top microcentrifuge (VWR, Galaxy 16D) at 14,000 g. Equal amounts of supernatant or pellet fractions were loaded onto an SDS gel and western blotting was performed as described using antibodies to GST (Sigma) or actin (Sigma).