Altering integrin engagement regulates membrane localization of K_ir2.1 channels

The activity of inwardly rectifying K⁺ channels, such as K_ir2.1 (encoded by KCNJ2) (Hibino et al., 2010), sets the resting membrane potential of various excitable cells (e.g. cardiomyocytes, neurons, smooth and skeletal muscle cells) to a highly negative (hyperpolarized) value. During action potential firing, a hyperpolarized membrane potential enables the voltage-gated Na⁺ channels (e.g. cardiac sodium channel, Na_v1.5, encoded by SCN5A) to transiently depolarize the cell membrane to positive potentials, followed by the return to the original resting state. In the heart, the K_ir2.1-mediated I_K1 current is upregulated during development (Wahler, 1992; Huynh et al., 1992), but downregulated in certain pathologies. These include myocardial infarction, heart failure (Beuckelmann et al., 1993; Nattel et al., 2007) and Anderson–Tawil syndrome (Bendahhou et al., 2003), a congenital disease characterized by periodic paralysis and polymorphic tachycardias (Miake et al., 2003; Hibino et al., 2010). Such electrophysiological changes in the heart are often associated with profound alterations in cardiomyocyte size and shape, and the extracellular matrix (ECM), as well as cytoskeletal tension (Jacot et al., 2010; Peter et al., 2016). Cardiomyocytes sense physical changes in their environment through integrins (Ross and Borg, 2001), heterodimeric receptors for ECM proteins that relay information to focal adhesions (FAs), force-sensitive subcellular structures connected to actin cytoskeleton (Kanchanawong et al., 2010; Geiger et al., 2009). In different systems, dynamic changes in FA proteins, as well as current flow through ion channels, including K_ir2.1, are known to govern important cellular processes, including survival, proliferation, differentiation, migration and matrix remodeling (Gautrot et al., 2014; Bates, 2015; Zaritsky et al., 2000; Chittajallu et al., 2002). However, the relationships between ion channels, FAs and cellular electrophysiology are not well understood.

Previous studies have shown that integrins and integrin-associated signaling can regulate the function of several voltage-gated ion channels through interactions with their regulatory subunits, phosphorylation, oxidative state, gene expression and trafficking (Davis et al., 2002; Becchetti et al., 2010; Arcangeli and Becchetti, 2006). In particular, trafficking has been identified as a mechanism to regulate the number of ion channels in the cardiomyocyte membrane (Boycott et al., 2013; Schumacher and Martens, 2010; Balse and Boycott, 2017). FA proteins could play a role in trafficking of K_ir2.1, since both integrins (Bridgewater et al., 2012) and K_ir2.1 channels (Ma et al., 2011) are trafficked from the Golgi to sub-membrane regions along microtubules, which are locally stabilized at FAs via phosphorylation of focal adhesion kinase (FAK, also known as PTK2) (Palazzo et al., 2004). Both integrins (Bridgewater et al., 2012) and K_ir2.1 channels (Varkevisser et al., 2013) are removed from the membrane via clathrin- or caveolae-mediated endocytotic processes that are dynamin-dependent [blocked by a dynamin 2 inhibitor (Hansen and Nichols, 2009)] and regulated by FA-associated proteins, including FAK and Src (Wang et al., 2011).

In this work, we sought to explore how changes in FA assembly within excitable cells affect K_ir2.1 membrane localization, I_K1 amplitude, and action potential characteristics. We utilized a monoclonal line of HEK293 cells engineered to express fluorescently tagged K_ir2.1 to visualize channels, a monoclonal line of HEK293 cells (‘Ex293’) engineered to express K_ir2.1, the cardiac Na⁺ channel Na_v1.5 and the gap junctional channel connexin-43 (Cx43; also known as GJA1) as a well-defined excitable cell source (Kirkton and Bursac, 2011; McSpadden et al., 2009), and neonatal rat cardiomyocytes as native excitable cells. We applied micropatterning of ECM proteins, pharmacological and environmental manipulations to alter FA size and distribution in individual cells. Combining these techniques with single-cell electrophysiology and quantitative image analysis, we provide evidence that altering FAs and integrin engagement affects K_ir2.1 membrane localization, I_K1 amplitude, and action potential morphology in excitable cells. Furthermore, by studying K_ir2.1 turnover dynamics, we show that these channels are uniformly transported to the membrane, where they preferentially accumulate near FA-rich sites via the local inhibition of dynamin-mediated endocytosis of K_ir2.1. Overall, our studies reveal new links between the electrophysiology of ion channels and FA biology, coupling action potential dynamics to changes in the cellular environment and cytoskeletal mechanics.

We lentivirally transduced HEK293 cells to express K_ir2.1 channels fused with tdTomato (K_ir2.1–tdTomato) and used whole-cell voltage clamp to confirm channel functionality by recording I_K1. The whole-cell voltage clamp recordings provide a method to specify various membrane voltage levels to a cell and quantify corresponding ion flow through membrane-bound channels (Fig. S1A,B). With expressed K_ir2.1 being the only K_ir2.x isoform present in these cells, the amplitude of the I_K1 current was a direct measure of the level of active K_ir2.1 channels on the plasma membrane. When the K_ir2.1–tdTomato-expressing cells were plated on fibronectin- or poly-L-lysine-coated dishes (Fig. 1A), a significantly larger I_K1 density (current amplitude divided by membrane area) was recorded on fibronectin (Fig. 1B,C), suggesting that integrin engagement affects K_ir2.1 channels.

Integrin engagement in a cell is reflected by the sizes and distribution of its FAs (Schoenwaelder and Burridge, 1999). Varying the cell attachment area and shape is a common means to alter integrin engagement and FA assembly (Chen et al., 2003), which we achieved by micropatterning Ex293 cells on fibronectin islands of two defined sizes (961 µm² or 1600 µm²), each in three distinct shapes (circular, square or star; Fig. S1D–G). To relate resulting changes in integrin engagement with the activity of membrane-bound K_ir2.1 channels, cells were immunostained for FA markers and imaged using confocal microscopy to determine FA coverage (the total area of FAs in microns) and I_K1 density was measured by whole-cell voltage clamp 6 h after plating (Fig. S2A). Since FAs are complex structures, we stained multiple FA-associated proteins, including integrins, paxillin, talin and vinculin, all of which have been shown to have distinct roles in FA assembly and dynamics (Wozniak et al., 2004; Hoffman et al., 2011). Quantification of the area of active β1-integrin (Fig. S2B,C), paxillin (Fig. S2D,E), talin (Fig. S2F,G) and vinculin (Fig. 1D,E; Fig. S2H) staining revealed predominant localization of FAs in the cell periphery and an increase in FA coverage with cell size, as previously reported (Kilian et al., 2010; Chen et al., 2003). The cell shape also affected FA coverage with a significant increase found in star-shaped cells. The whole-cell I_K1 density at −90 mV increased in larger cells and was unaffected by the cell shape (Fig. 1F,G; Fig. S2I). Averaged over all shapes, a 1.66-fold larger micropatterned cell area yielded a 1.80±0.23-fold increase in FA coverage as marked by vinculin and a 1.18±0.05-fold increase in I_K1 density, suggesting a correlation between these variables. Since all FA proteins showed similar distributions, vinculin and active β1-integrin were chosen as representative markers of FAs for subsequent studies.

Since FA assembly is regulated by variety of cell signaling pathways (Burridge et al., 1997), we sought to probe the effects of integrin activation, FAK signaling and cell contractility on the correlation between FA coverage and I_K1 density. To do so, cells were plated onto 961 µm² star-shaped islands, pharmacological agents were applied 3–6 h post plating, and FA coverage and I_K1 density were quantified at 0, 2, 4 and 6 h post drug application (Fig. S3A). First, we activated integrins using manganese chloride (MnCl₂, 1 mM), which increases integrin affinity for the ECM and causes an increase in both FA area and cell traction forces (Lin et al., 2013; Bazzoni et al., 1995). The MnCl₂ treatment increased both FA assembly (Fig. S3B,C, to 159±27% of initial coverage; mean±s.e.m.) and I_K1 density (Fig. S3D–F, to 166±23% of initial value). Furthermore, we disrupted FA assembly using the FAK inhibitor (FAKi) PF-573228 (100 µM), which specifically blocks the ATP-binding site at Tyr³⁹⁷ and alters the FA dynamics by blocking FAK–Src interactions (Slack-Davis et al., 2007). FA disassembly was indicated by a loss of vinculin immunostaining (to 30±2.0% of initial coverage after 6 h; Fig. S3G,H), and was accompanied by the progressive loss of I_K1 current density (to 49±7.2% of initial value after 6 h; Fig. S3I–K). Finally, we decreased contractility in the patterned cells using the myosin II inhibitor blebbistatin (10 µM), which has previously been shown to cause a reduction in FA size and loss of vinculin at FAs (Pasapera et al., 2010). Exposure to blebbistatin caused FA disassembly (Fig. S3L,M, to 25±2.5% of initial coverage) and a reduction in I_K1 density (Fig. S3N–P, to 40±5.1% of initial value). The FA disassembly and loss of I_K1 were fully reversible as shown when 4 h blebbistatin treatment was followed by a 4 h drug washout (Fig. S4A–D). Moreover, the effect of blebbistatin on I_K1 density was found to be dose dependent (Fig. S4E,F). Collectively, these results suggest that interventions that enhance FA assembly increase I_K1 membrane density, while FA disassembly is associated with lower I_K1 density, in a reversible and dose-dependent manner.

Since different ECM proteins can engage specific types of integrins to distinctly regulate FA assembly (Friedland et al., 2009; Geiger and Yamada, 2011), we sought to determine whether the observed correlation between FA coverage and I_K1 density is dependent on specific ECM–integrin interactions. We thus compared star-shaped cells cultured on micropatterned islands of 30 µg/ml fibronectin or 30 µg/ml laminin, and found that cells on laminin exhibited a 2.01±0.25-fold (mean±s.e.m.) increase in FA coverage (Fig. S5A,C) and consequently had a 1.76±0.20-fold higher I_K1 density (Fig. S5B,D,E). These ECM-dependent differences may have resulted from the different types of engaged integrins, different numbers of integrin-binding sites, or both. Tension in the plasma membrane has also been shown to control integrin signaling (Bottcher and Fassler, 2014) as well as vesicle trafficking (Ferguson et al., 2017). We thus utilized hypotonic swelling to increase membrane tension (Boulant et al., 2011) by placing micropatterned cells in solutions of decreasing osmolarity for 4 h, followed by immunostaining and whole-cell patch clamp recording (Fig. S5F). Consistent with previous studies (Hirakawa et al., 2004), reduction of extracellular osmolarity resulted in increased FA coverage up to 1.76±0.20 fold (Fig. S5G,I) and a corresponding increase in I_K1 membrane density up to 1.51±0.19 fold (Fig. S5H,J,K). We next investigated whether the observed changes in I_K1 density induced by manipulation of integrin engagement and FA assembly are directly correlated with the altered abundance of K_ir2.1 channels at the cell membrane. For the case of FAK inhibition where I_K1 density was gradually decreased with increased drug exposure (Fig. S3F–I), immunoblots for surface-biotinylated K_ir2.1 showed corresponding loss of channels from the membrane (Fig. S5L,M), suggesting the regulation of channel numbers at the membrane as the primary means of regulating I_K1.

Collectively, these results show that various means of manipulating FA assembly via extracellular and intracellular cues lead to concordant changes in the presence of K_ir2.1 channels at the membrane and, consequently, I_K1 density. Quantitatively, when I_K1 density was plotted against FA coverage for all studied conditions, we found strong positive linear correlation, with a Pearson's coefficient of R=0.8704 (r²=0.7576, Fig. 1H). This conserved correlation across a diverse set of treatments further suggested that a common process may control both FA coverage and K_ir2.1 channels at the membrane.

FAs are potent regulators of a number of cellular processes, including the activation of diverse signaling pathways as well as cell contractility (Parsons et al., 2010). Thus, the observed correlation between global FA coverage and I_K1 density could be due to a cell-wide phenomenon, such as increased activity of a mobile kinase. Alternatively, the relationship could be due to a phenomenon within or proximal to FAs, potentially affecting the local recruitment of the channels. To begin to distinguish between these possibilities, we tested whether ECM-coated microspheres, which locally engage integrins, can recruit K_ir2.1 to the cell membrane (as visualized by fused tdTomato). Polystyrene microspheres (∼10 µm diameter) were coated with either poly-L-lysine, to enable electrostatic integrin-independent adhesion to cells, or fibronectin, to enable integrin-dependent adhesion. Coated microspheres were then added to cells cultured on fibronectin (Fig. 2A) and live confocal images of the membrane stained with CellMask (green) and K_ir2.1–tdTomato (red) were taken at the plane of the microspheres (Fig. 2B). At the sites where microspheres adhered to cells (appearing as circles in the confocal images), we observed enhanced recruitment of K_ir2.1–tdTomato to the microspheres for the fibronectin compared to what was seen with the poly-L-lysine coating (Fig. 2B). As a measure of K_ir2.1 membrane recruitment, we quantified the ratio of the fluorescence intensity of K_ir2.1–tdTomato to the membrane staining and found a significant increase for fibronectin-coated microspheres (Fig. 2C). These results clearly show that engaging integrins leads to the local accumulation of K_ir2.1 within the plasma membrane.

We next assessed membrane localization of K_ir2.1–tdTomato in relation to that of FAs (visualized by immunostaining for vinculin) in micropatterned Ex293 cells of different shapes and sizes (Fig. 2D). To determine the average spatial distributions of FAs and K_ir2.1, we constructed heat maps by stacking and averaging fluorescent images from large numbers of individual micropatterned cells (Thery et al., 2006). To simplify visualization of the K_ir2.1–tdTomato near FAs at the cell periphery, we masked out the large internal pools of K_ir2.1 surrounding the nucleus (Fig. S6A), previously shown in heterologous expression systems to predominantly associate with the Golgi complex (Stockklausner et al., 2001; Hofherr et al., 2005). From vinculin heat maps (Fig. 2D, left), we observed that FA density varied along the periphery of the cells with more FAs being present at the ‘corners’ than ‘edges’ of the square- and star-shaped cells (Chen et al., 2003; Grosberg et al., 2011; Mandal et al., 2014). Similarly, the distribution of tdTomato fluorescence at the periphery of individual cells (Fig. 2D, middle) and in corresponding heat maps (Fig. 2D, right) suggested that K_ir2.1 channels were uniformly distributed in membranes of circular cells but were more abundant at corners than edges in square- and star-shaped cells, concordant with the spatial distribution of FAs. This observation was further corroborated by quantifying local (edge and corner) abundances of FAs (protein coverage, Fig. 2E and Fig. S6) and K_ir2.1 (tdTomato intensity, Fig. 2F), as both showed similar cell shape and size dependences. We then labeled the membrane of star-shaped K_ir2.1–tdTomato HEK293 cells with CellMask Green (Fig. 2G) and quantified the corner versus edge fluorescence intensity ratio for tdTomato and membrane staining. Compared to a 2.9-fold increase in tdTomato intensity at corner versus edges, the membrane staining intensity was increased only 1.3-fold (Fig. 2H), demonstrating that K_ir2.1 channels are preferentially enriched within the membrane at cell corners. Furthermore, we expressed EGFP fused to the N-terminus of paxillin (EGFP–paxillin) in K_ir2.1-tdTomato cells to directly visualize the potential colocalization of K_ir2.1 and FAs in live cells. Confocal imaging suggested that channels accumulate in proximity to, but not within, FAs (Fig. 2I). Overall, the ECM-coated microsphere experiments, quantitative analyses of local FA and K_ir2.1 abundances, and the assessment of K_ir2.1 and paxillin co-localization suggest that K_ir2.1 localization in the cell membrane is stimulated by local integrin engagement, but likely without the physical interaction of the channel with FAs.

We assessed the functional consequences of K_ir2.1 recruitment proximal to FAs by performing cell-attached patch clamp at corners and edges of micropatterned Ex293 cells (Fig. 3A,B). The recorded local I_K1 strongly correlated with the mean K_ir2.1–tdTomato intensity, with a Pearson's coefficient of R=0.963 (r²=0.9287, Fig. 3C), suggesting that the quantified tdTomato labeled active K_ir2.1 channels reside on the membrane. Next, we sought to manipulate FA assembly in a micropatterning-independent manner and probe the local correlations between FA coverage and I_K1 at the cell edges and corners. To do so, we applied 100 µM FAKi for 2 h, which shifted the FA distribution in star-shaped cells from being enriched at the corners to being relatively uniform along the cell boundary (Fig. 3D, left). This change in FA distribution coincided with the loss of the K_ir2.1 enrichment at corners and the emergence of an apparently homogenous channel distribution (Fig. 3D, right). Quantitative image analysis and cell-attached patch clamp revealed that the uniform FA redistribution along the cell boundary (Fig. 3E) resulted in the decrease of I_K1 amplitude at the corners and increase at the edges (Fig. 3F). Plotting the local I_K1 amplitude versus the corresponding local FA coverage across different conditions showed a strong linear correlation (r²=0.8485, Fig. 3G), confirming that active K_ir2.1 channels are enriched near FAs and suggesting that a locally regulated process mediates the functional relationship between FA assembly and I_K1.

In excitable cells, I_K1 plays important roles in regulating the resting membrane potential (RMP) and cell excitability by influencing both Na⁺ current (I_Na) availability and action potential (AP) threshold (Kléber and Rudy, 2004; Miake et al., 2003). Additionally, the outward portion of I_K1 influences the terminal phase of action potential repolarization and AP duration (APD). The use of monoclonal excitable Ex293 cells in our studies enabled us to further assess potential roles of FA-mediated K_ir2.1 regulation in AP shape and excitability. By performing voltage- and current-clamp recordings in star-shaped Ex293 cells of different size, we found that larger cells had higher outward peak I_K1 density (Fig. 3H), shorter APD (Fig. 3I,J) and similar RMP (Fig. 3I,K) compared to smaller cells. Furthermore, compared to cells of the same size and shape, cultured on fibronectin islands, Ex293 cells cultured on laminin exhibited a higher outward peak I_K1 density (Fig. 3L), reduced APD (Fig. 3M,N) and hyperpolarized RMP (Fig. 3M,O). Overall, these results suggest that intracellular and extracellular signals regulating integrin engagement and FA assembly may also have important roles in controlling the electrophysiological properties of excitable cells and tissues.

Since previous studies have shown that Na_v1.5 and K_ir2.1 have a parallelism in expression (Milstein et al., 2012; Utrilla et al., 2017; Ponce-Balbuena et al., 2018), we next explored whether the cell shape and size affect membrane localization and activity of Na_v1.5 channels by immunostaining and electrophysiological analysis, respectively. We found that, unlike K_ir2.1, Na_v1.5 exhibited no preferential localization in the corners versus edges of star-shaped Ex293 cells (Fig. S7A,B). Interestingly, the whole-cell I_Na density (Fig. S7C,D) and action potential upstroke (Fig. S7E) were still increased in larger Ex293 cells, collectively suggesting that while the numbers of active Na_v1.5 channels on cell membrane may be related to the numbers of active K_ir2.1 channels, the membrane distribution of Na_v1.5 does not appear to be affected by changes in either local integrin engagement or K_ir2.1 distribution. Furthermore, similar to Na_v1.5, immunostained Cx43 in star-shaped Ex293 cells exhibited a uniform membrane distribution (Fig. S7F,G), indicating the lack of local regulation by engaged integrins or K_ir2.1.

Since K_ir2.1 channels are trafficked to the membrane (Leonoudakis et al., 2004; Balse et al., 2012), we evaluated the possibility that FAs regulate K_ir2.1 localization and distribution by modulating channel trafficking. To probe FAs, we imunofluorescently labeled active β1-integrin, representing the transmembrane component of the FA. Active β1-integrin immunostaining led to a small amount of labeling of the nucleus that was not observed in any of the other FA markers. Therefore, we regarded this nuclear labeling as artifactual and excluded it from the visualization and quantification (using masks from DAPI stainings). For this set of experiments, we first halted microtubule (MT)-dependent forward and reverse trafficking of K_ir2.1 (Loewen et al., 2009) in micropatterned HEK cells by using the microtubule-destabilizing agent Nocodazole (Noco, 12 µM). This treatment yielded a transient increase in FA coverage within 2 h, but no statistically significant difference in FA coverage after 6 h (as determined by active β1-integrin labeling, Fig. 4A,B). No statistically significant increase was observed in I_K1 density (Fig. 4C). Given a previously demonstrated relationship between MT disruption and enhanced FA assembly (Enomoto, 1996), we simultaneously treated cells with 12 µM nocodazole and 100 µM FAKi (Noco+FAKi), which resulted in the gradual disassembly of FAs (Fig. 4A,B) and reduction of I_K1 density (Fig. 4C). Previous work has shown that MT-dependent trafficking is required for the initial delivery of K_ir2.1 to the membrane and recycling of K_ir2.1 (Loewen et al., 2009). Our results make evident that MTs do not govern the selective accumulation of K_ir2.1 near FAs and suggest that the membrane localization of the channel may be locally regulated by FAs.

Thus, we next examined the potential roles of FAs in the insertion and retention of channels at the membrane as means to govern the local accumulation of K_ir2.1. Specifically, we investigated whether dynamin-dependent endocytosis mediates the relationship between FAs and K_ir2.1. Dynamin is a protein required for the final membrane scission to complete endocytosis that is inhibited by a small molecule, Dynasore (Dyn; Macia et al., 2006). A 6-h treatment of Ex293 cells with 25 μM Dyn yielded a progressive increase in both active β1-integrin coverage (Fig. 4D,E; P<0.001) and I_K1 density (Fig. 4F; P<0.001), an expected result since dynamin mediates internalization of integrins (Alanko et al., 2015) and K_ir2.1 (Varkevisser et al., 2013). To probe the role of dynamin in FAKi-induced removal of integrin and K_ir2.1 from the membrane, FAKi-treated Ex293 cells were co-treated with Dyn, which prevented the FAKi-induced loss of active β1-integrin (Fig. 4D,E) and I_K1 (Fig. 4F), and maintained the correlation between FA coverage and I_K1 density (Fig. 4G). Given that adding Dyn to FAKi was the first intervention to prevent decrease in I_K1, we immunostained for vinculin (Fig. 4H) and found that the Dyn+FAKi treatment yielded progressive loss of vinculin coverage (Fig. 4I), indicating that FA disassembly was still occurring under these conditions without loss of active β1-integrin coverage. These results suggested that dynamin-dependent stabilization of active integrins on the membrane results in K_ir2.1 localization near FAs leading to a local increase in I_K1. Engaging integrins prevents endocytosis of K_ir2.1 channels, while removing active integrins from the membrane is associated with internalization of K_ir2.1 channels and an I_K1 decrease.

Next, we sought to determine the turnover rate of K_ir2.1 channels at cell corners, where FAs are large, and edges, where FAs are small through fluorescence recovery after photobleaching (FRAP) (Fig. 5A). After local bleaching of K_ir2.1–tdTomato in star-shaped HEK293 cells, fluorescence recovered at a slower rate at the cell corners (τ_c=158±35 s) than edges (τ_e=70±10 s, Fig. 5B,C). Cell corners also exhibited a smaller mobile fraction, M_f, of K_ir2.1 channels compared to cell edges (Fig. 5B,C), indicating the stabilization of K_ir2.1 in the membrane at the FA-rich sites. To test the role of dynamin activity in K_ir2.1 turnover dynamics, we performed FRAP experiments in K_ir2.1–tdTomato-expressing cells that had been treated with 25 µM Dyn for 4 h (Fig. 5D). Since inhibition of endocytosis by Dyn leads to the membrane accumulation of K_ir2.1 (Fig. 4F), as reflected in an increased I_K1 density, this experiment preferentially probed the rate of forward transport of the channel to the membrane. Endocytosis inhibition by Dyn thus revealed that channel forward transport rates are the same at cell corners and edges (Fig. 5E,F). The mobile channel fraction remained smaller at corners than edges (Fig. 5E,F), likely due to the preserved spatial distribution of FAs (Fig. 4D). Differences in the initial data points in the FRAP curves were due to variable levels of bleaching. Together, these data are consistent with the model where K_ir2.1 channels are spatially uniformly delivered to the cell membrane, then non-uniformly endocytosed in an FA-dependent and dynamin-dependent manner. This spatially non-uniform endocytosis causes a heterogeneous distribution of K_ir2.1 channels along the membrane with more channels localizing near larger FA complexes with increased integrin engagement.

We next explored whether the strong correlation between integrin engagement and I_K1 density observed in engineered HEK293 cells is also found in native excitable cells – neonatal rat ventricular myocytes (NRVMs). The NRVMs micropatterned on circular-, square- and star-shaped fibronectin islands displayed similarly distributed but more abundant FAs (Fig. 6A) than those in HEK293 cells, with larger NRVMs exhibiting both larger total FA coverage (Fig. 6B) and endogenous I_K1 density (Fig. 6C). Averaged over all shapes, a 1.66-fold larger micropatterned NRVM area yielded a 1.35-fold increase in FA coverage marked by vinculin and a 1.59-fold higher I_K1 density. Furthermore, we applied selected perturbations of FA assembly in NRVMs and recorded the resulting changes in endogenous I_K1 density (Fig. 6D–F). Application of 10 µM Blebbistatin for 6 h resulted in a breakdown of actin stress fibers as well as decrease in FA coverage and I_K1 density to 71% and 47% of original values, respectively. Similarly, 100 µM FAKi yielded a loss of actin stress fibers, accompanied by a decrease in FA coverage and I_K1 density to 64% and 57% of original values. NRVMs were then plated on star-shaped islands of laminin, and found to have increased FA coverage by 33% and I_K1 density by 71%. Similar to engineered HEK293 cells, plotting the endogenous I_K1 density against the total FA coverage for all studied conditions in NRVMs yielded a positive correlation (r²=0.6568, Fig. 6G). Moreover, when 1600 µm² star-shaped NRVMs were transduced with K_ir2.1–tdTomato, the channels localized at a higher density in cell corners than edges (Fig. 6H), consistent with the higher local FA coverage (Fig. 6I). As in HEK293 cells, the local I_K1 amplitude in these NRVMs was found to be higher in cell corners than edges (Fig. 6J). Together, these results indicate that, similar to what is found for Ex293 cells, interventions that alter integrin engagement and FA assembly concordantly alter K_ir2.1 and I_K1 in neonatal cardiomyocytes, as an example of native excitable cells.

Finally, we explored the spatial relationships between K_ir2.1 channels and β1D integrins (the β1 isoform found in cardiac tissue) in adult rat ventricles through immunostaining and image analysis. Specifically, we assessed how the two proteins localize in both the t-tubule (Kostin et al., 1998; Vaidyanathan et al., 2013; Israeli-Rosenberg et al., 2015; Okada et al., 2013) and costamere (Clark et al., 2001; Samarel, 2005) regions of cardiomyocyte membrane (Fig. 6K,L). In agreement with the finding in HEK293 cells (Fig. 2I), the K_ir2.1 and β1D immunofluorescence did not overlap, but were consistently localized proximal to each other throughout the cardiomyocyte at an average distance of 0.494±0.021 µm (Fig. 6M,N). This result supported the possibility that signals for integrin engagement may also regulate membrane localization of K_ir2.1 in adult cardiomyocytes.

Our studies suggest a fundamental link between FA assembly and membrane localization of K_ir2.1 channels (summarized in Fig. 7). By employing cell micropatterning techniques and targeted pharmacological interventions, we show that increasing integrin engagement (and FA coverage) in cells proportionally increases the abundance of functional K_ir2.1 channels at the cell membrane (Fig. 1). The microsphere experiments confirmed that the de novo engagement of integrins can recruit K_ir2.1 channels to accumulate at the sites of newly formed FAs (Fig. 2). Local variations in FA coverage strongly correlated with local membrane abundance of K_ir2.1, such as at corners versus edges of micropatterned cells (Fig. 2), indicating a local mechanism for channel recruitment leading to changes in local and global I_K1 (Fig. 3). In contrast, the membrane distribution of overexpressed Na_v1.5 did not appear to be correlated with local integrin engagement (Fig. S7A). Use of pharmacological inhibitors to probe the involvement of key mechanisms of channel trafficking revealed that the internalization of K_ir2.1 during FA disassembly is MT-independent but requires dynamin (Fig. 4). FRAP analysis of K_ir2.1 turnover suggested that the channel is uniformly transported to the membrane, but is less readily endocytosed from the membrane near sites of strong integrin engagement (large FA complexes, Fig. 5). Overall, we propose that, through this mechanism, a spatially heterogeneous distribution of FAs in a cell contributes to a spatially heterogeneous distribution of K_ir2.1 channels at the cell membrane, while altering the global levels of integrin engagement can serve to alter whole-cell I_K1 density and AP shape, both in engineered (Fig. 3) and primary (Fig. 6) excitable cells. The close spatial localization of K_ir2.1 channels and β1D integrins in rat ventricles (Fig. 6) further indicated the potential relevance of these findings for adult cardiomyocytes in vivo.

Mechanistically, our results suggest that K_ir2.1 membrane localization is at least in part regulated by a local diffusible signal generated by integrin engagement, as the absence of colocalization of K_ir2.1 with paxillin (Fig. 2I) argues against direct interactions of the channel with FA proteins. In principle, this diffusible signal could either affect dynamin activity, K_ir2.1 activity, or the interaction of dynamin and K_ir2.1. The direct regulation of dynamin seems unlikely, as dynamin plays a crucial role in the endocytosis of integrins (Bridgewater et al., 2012) and other membrane receptors (Faelber et al., 2012) within and proximal to FAs. Previous studies have demonstrated that phosphorylation of K_ir2.1 can inhibit its activity (Wischmeyer et al., 1998; Vega et al., 2009). However, the strong correlation between I_K1 and the intensity of K_ir2.1–tdTomato is most consistent with the channel being constitutively active at the membrane and inconsistent with a significant role of post-translational regulation. Therefore, we postulate that integrin-mediated diffusible signals interfere with the interaction between K_ir2.1 and dynamin to prevent removal of channel from the membrane. Filamin A and dystroglycan have been reported to interact with K_ir2.1 as well as dynamin (Noam et al., 2014; Zhan et al., 2005; Sampson et al., 2003). These proteins, which are likely proximal to FAs due to their localization to the actin cortex, could act as K_ir2.1-regulating scaffolding proteins, and their roles in the endocytosis of K_ir2.1 channels remain to be studied.

K_ir2.1 is not a stretch-gated channel, as shown by its insensitivity to gadolinium or lack of immediate response to 10–30 min of hypo-osmotic shock (Ji et al., 1998). However, the application of fluid flow to endothelial cells has been shown to increase K_ir2.1 channel current in a tyrosine kinase-dependent manner (Hoger et al., 2002), suggesting indirect mechanosensitivity of the channel to shear stress. Notably, the application of shear stress also induces integrin engagement and FA assembly (Jalali et al., 2001). Although our studies were performed in static conditions, I_K1 amplitude and K_ir2.1 localization and distribution in membrane were responsive to a variety of intracellular and extracellular modifications of the cytoskeletal mechanics and membrane tension that altered FA coverage and distribution. Specifically, in contrast to the lack of immediate response (Ji et al., 1998), 4 h of hypo-osmotic shock in micropatterned cells resulted in an enhanced FA assembly and a concordant I_K1 increase. Thus, our results provide evidence for K_ir2.1 being an indirectly mechanoresponsive channel via known links between mechanical forces and FA dynamics coupled with apparent roles of integrin engagement and FA assembly in K_ir2.1 endocytosis.

Of significance are also potential implications of our findings for the normal heart development, where a 20–40-fold increase in cardiomyocyte volume (Anversa et al., 1980) and membrane area (Wahler, 1992) is associated with not only a large increase in total I_K1, but also a 2.5-fold increased I_K1 density (Wahler, 1992). Our experiments using monoclonal cell lines and primary cardiomyocytes suggest that increased cardiomyocyte size and associated integrin recruitment may be a contributing factor to the increased I_K1 density, independent of changes in K_ir2.1 gene expression. Moreover, ECM in the developing ventricles changes from being primarily fibronectin-based to being laminin-based (Hirschy et al., 2006; Farhadian et al., 1996), which, from our studies, may further contribute to the age-dependent increase in I_K1. These considerations may also suggest strategies to augment a low I_K1 amplitude in pluripotent stem cell-derived cardiomyocytes (Casini et al., 2017). Finally, various cardiac diseases, such as myocardial infarction (Arslan et al., 2011) or pressure overload (Li et al., 2014), are characterized by excess deposition of fibronectin, collagen I and collagen III, and the reduction of laminin as well as reduced I_K1 density (Nattel et al., 2007), which, based on our studies, could be related to an ECM-dependent reduction in FA coverage and integrin engagement. In general, our results suggest important roles for integrin-modulating signals in I_K1-mediated regulation of cell excitability, AP shape and propagation.

Overall, our studies are consistent with a working model (Fig. 7) where engaged integrins are an important contributor to the membrane localization of K_ir2.1 channels via the local inhibition of dynamin-dependent endocytosis of K_ir2.1. This mechanism renders the I_K1 and cellular electrophysiology indirectly mechanosensitive to various intracellular and extracellular signals affecting FA dynamics. These findings warrant future in-depth studies of how cell–matrix interactions modulate the function of various ion channels across diverse cell types and pathophysiological conditions.

Cells were micropatterned into circular, square and four-armed star shapes in two sizes, 961 µm² and 1600 µm², using microcontact printing (Pedrotty et al., 2008; McSpadden et al., 2012). Micropatterns were drawn using AutoCAD (Autodesk, San Rafael, CA) and printed as high-resolution photomasks (chrome on soda-lime; Advance Reproductions, North Andover, MA). Standard soft lithography techniques were used to prepare master micropatterned silicon wafers, as previously described (McSpadden et al., 2012; Pedrotty et al., 2008). Briefly, a silicon wafer (WaferWorld, West Palm Beach, FL) was coated with a 5 µm layer of photoresist (SU8-5; Microchem, Newton, MA), covered with photomask, illuminated with UV light, and washed to remove un-crosslinked photoresist to generate microfabricated patterns. Poly-dimethylsiloxane (PDMS, Sylgard 184; Dow Corning, Midland, MI) stamps were cast against the microfabricated wafers at 80°C for 2 h, cleaned in 70% ethanol, coated with fibronectin (30 µg/ml in PBS) or laminin (30 µg/ml in PBS) for 1 h, and blow-dried with N₂. The fibronectin micropattern was microcontact-printed from the PDMS stamps onto UV-ozone-treated PDMS-coated 12 mm glass coverslips for 1 h to allow protein transfer. For live-cell imaging, cells were patterned on plasma-treated glass bottom dishes (Fluoro 35 mm, Ibidi, Madison, WI).

For the patch clamp experiments and FA quantification, we utilized the previously described and characterized Ex293 cell line (Kirkton and Bursac, 2011), a monoclonally-derived HEK293 cell line engineered to stably express K_ir2.1, Na_v1.5 and Cx43. In this line, the I_K1 resulting from exogenously overexpressed K_ir2.1 was much larger compared to a negligible endogenous I_K1 present in wild-type HEK293 cells (Kirkton and Bursac, 2011). For live imaging analysis, we used a monoclonal HEK293 line expressing K_ir2.1 fused to tdTomato at the C-terminus of the channel (K_ir2.1–tdTomato). In some studies, these cells were also transfected with EGFP fused to the C-terminus of paxillin (paxillin-EGFP, Addgene 15233; Laukaitis et al., 2001). HEK293 cells were maintained in low-glucose DMEM with L-glutamine and sodium pyruvate (11885-092, Gibco, Waltham, MA) supplemented with 10% fetal bovine serum (HyClone, Logan, UT) and 1% penicillin-streptomyocin (15140122, Gibco, Waltham, MA). For each experiment, cells were plated onto micropatterned glass coverslips at 10,000 cells per cm² and studied within 6–10 h of culture.

All studies conformed to the Guide for the Care and Use of Laboratory Animals, published by the United States National Institutes of Health (Publication No. 85-23, revised 1996) and approved by Duke University Protocol A214-09-07. NRVMs were isolated from 2-day-old Sprague Dawley rats, as previously described (Badie and Bursac, 2009; McSpadden et al., 2012; Jackman et al., 2016) using overnight digestion in 0.1% trypsin and four serial dissociations with 0.1% collagenase. The obtained cell suspension was pre-plated on a tissue culture flask for 45 min to enrich the fraction of cardiomyocytes. Non-adherent cells were collected, centrifuged, re-suspended, counted and plated onto micropatterned fibronectin substrates at density of 40,000 cells/cm² in DMEM/F-12 (11330-057, Gibco) supplemented with 10% fetal bovine serum, 10% horse serum, penicillin (5 U/ml) and vitamin B12 (2 µg/ml). After 24 h, the cell culture medium was changed to DMEM/F12 supplemented with 2% fetal bovine serum, penicillin and vitamin B12. All experiments were performed after 3–4 days of culture.

Glass coverslips with micropatterned cells were transferred to a patch-clamp chamber perfused with Tyrode's solution containing (in mM): 135 NaCl, 5.4 KCl, 1.8 CaCl₂, 1 MgCl₂, 0.33 NaH₂PO₄, 5 HEPES and 5 glucose. Patch pipettes were fabricated with tip resistances of 1–2 MΩ when filled with pipette solution consisting of (in mM): 140 KCl, 10 NaCl, 1 CaCl₂, 2 MgCl₂, 10 EGTA, 10 HEPES and 5 MgATP. Whole-cell voltage-clamp recordings were acquired at room temperature (25°C), using the Multiclamp 700B amplifier (Axon Instruments), filtered with a 10-kHz Bessel filter, digitized at 40 kHz and analyzed using WinWCP software (John Dempster, University of Strathclyde). A steady-state I_K1–V curve was constructed from the current responses to 1 s test potentials (−130 to 50 mV, increments of 10 mV) from a holding potential of −40 mV (McSpadden et al., 2012; Nguyen et al., 2016). A steady-state I_Na–V curve was constructed from the peak current responses to various 500 ms test potentials (−60 to 60 mV, increments of 10 mV) from a holding potential of −80 mV (Kirkton and Bursac, 2011). APs were elicited by injecting a 1 ms current pulse at 1.1× threshold amplitude.

For cell-attached membrane current recordings, patch electrodes were fabricated with tip resistances of 5.0–10.0 MΩ (for HEK293 cells) and 1.5–3.0 MΩ (for NRVMs) when filled with pipette solution consisting of (in mM): 140 KCl, 1 CaCl₂, and 5 HEPES. Bath solution consisted of (in mM): 140 KCl, 1.8 CaCl₂, 5 HEPES, and 0.33 NaH₂PO₄. I_K1 currents were recorded using a similar protocol, filtered with a 2-kHz Bessel filter, digitized at 40 kHz and analyzed using WinWCP software, as previously described (Vaidyanathan et al., 2010).

Micropatterned cells on coverslips were fixed in 4% paraformaldehyde (PFA) for 5 min and permeabilized with 0.1% (v/v) Triton-X at room temperature (10 min for HEK293 cells, 30 min for NRVMs). Primary antibodies were diluted in blocking solution (5:1 solution of 1% BSA and chicken serum) and applied overnight at 4°C and included: anti-Vinculin (1:800, mouse monoclonal, V9131, Sigma), anti-Paxillin (1:300, rabbit monoclonal, ab32084, Abcam), anti-talin (1:300, mouse monoclonal, T3287, Sigma), anti-active β1-integrin (1:300, mouse monoclonal, ab30394, Abcam), anti-Na_v1.5 (1:100, rabbit monoclonal, ASC-005, Alomone) and anti-Cx43 (1:300, rabbit monoclonal, ab11370, Abcam) antibodies. The secondary antibodies were conjugated to Alexa Fluor 488 (1:300, chicken anti-rabbit-IgG, Life Technologies) or Alexa Fluor 594 (1:300, chicken anti-mouse-IgG, Life Technologies) and were applied for 1 h at room temperature. Nuclei were counterstained with DAPI (1:250, Sigma). Alexa Fluor 488–phalloidin (1:150, Thermo Fisher Scientific) was used to visualize actin fibers in cardiomyocytes. Cell Mask Green Plasma Membrane Stain (1:1000, Thermo Fisher Scientific) was used to visualize cell membrane in HEK293 cells.

Adult rat left ventricles were excised from 3-month-old female Sprague Dawley rats, rinsed in 30% sucrose solution followed by PBS, embedded in Tissue-Tek O.C.T. Compound (Sakura Finetek), snap-frozen in liquid nitrogen and placed at −80°C overnight. Samples were then cryosectioned at 5 µm thickness, brought to room temperature, briefly rehydrated, then fixed in 4% PFA for 10 mins and permeabilized overnight at 4°C in a solution containing 0.5% (v/v) Triton-X, 1% (w/v) BSA and 10% (v/v) chicken serum. Primary antibodies were diluted in blocking solution [0.5% (v/v) Triton-X, 1% (w/v) BSA and 3% (v/v) chicken serum] and applied overnight at 4°C and included: anti-K_ir2.1(1:100, rabbit monoclonal, ASC-026, Alomone) and anti-β1D integrin (1:200, mouse monoclonal, MAB1900, EMD Millipore) antibodies. The secondary antibodies included Alexa Fluor 488 (1:300, chicken anti-rabbit-IgG, Life Technologies) and Alexa Fluor 594 (1:300, chicken anti-mouse-IgG, Life Technologies) and were applied for 1 h at room temperature.

All samples were imaged at 40× magnification (HCX PL APO 40×/NA1.25 oil objective; Leica Microsystems, Buffalo Grove, IL), 4× optical zoom, and a pinhole size of 1 Airy Unit (∼1.2 µm slice thickness) using an inverted confocal microscope (Leica, DMI6000CS) and Leica LAS AF 2.6 imaging software. The full width at half maximum of the point spread function was ∼290 nm for the 488 nm excitation wavelength and ∼310 nm for the 561 nm excitation wavelength. Live cells were imaged using the same system, while being maintained at 37°C.

Micropatterned cells stained for vinculin, paxillin, talin or active β1-integrin, as well as live cells containing K_ir2.1–tdTomato and EGFP–paxillin, were imaged by confocal microscopy in the plane of the cell–substrate interface and analyzed using custom ImageJ macros (available from the corresponding author upon request). Masks were created for each analyzed cell by increasing contrast to identify the cell boundaries. To facilitate the visualization of the cell periphery, sub-masks were created at 80% of the cell boundary and the interior pixels within the sub-mask were removed. Images were then cropped to a tightest square image containing the cell and pixel values from different cells of same shape and size were summed into a stack image and pseudocolored to create heatmaps (Gomez and Nelson, 2011). In heatmaps, the removed interior pixels from the sub-masks were marked in black.

For the quantification of FA coverage, FAs were segmented using the ‘water’ algorithm, as previously described (Rothenberg et al., 2015; Zamir et al., 1999). The sums of the areas of all segmented FAs were reported as total FA coverage (in μm²). For integrin staining, structures smaller than 0.5 μm² were excluded. Furthermore, integrin immunostaining yielded significant nuclear staining that did not contain identifiable FA structures; therefore, DAPI-stain​ed images were thresholded and converted into ​binary masks, which were ​applied to images of stained integrin to remove nuclear integrin signal. For all FA images in micropatterned Ex293 cells, the pixel size was 0.0183 μm².

Local FA coverage in micropatterned cells was quantified (in μm²) by the same water method as mentioned above, followed by dividing each cell into a 5×5 grid resulting in 25 regions (Fig. S6A–C). Four of these regions were denoted as ‘corners’ and four other regions were denoted as ‘edges’ depending on the shape of the cell. The FA-positive areas of pixels in the corner and edge regions were calculated. The mean FA area averaged over four corners or four edges was termed the ‘local’ FA coverage. Similar methods were used to measure the mean intensities of K_ir2.1–tdTomato and CellMask Green at cell corners and edges.

Immunostained sections of adult rat ventricles were used to quantify spatial relationships between K_ir2.1 channels (stained in green) and β1D integrins (stained in red). Each analyzed cardiomyocyte was divided into multiple ‘t-tubule regions’ in the cell interior and ‘costamere regions’ at the lateral cell boundaries (Fig. 6K). Custom ImageJ macros (available from the corresponding author upon request) allowed the user to select lines within these regions along which fluorescence intensity profiles were measured in the green and red channels. These lines were selected to encompass the two fluorophores and to be orthogonal to either the registered t-tubule pattern (along the longitudinal cell axis) or lateral membrane (along the transverse cell axis) (Fig. 6L). From the obtained intensity profiles, a custom MATLAB script (available from the corresponding author upon request) then calculated the closest distance between the green and red fluorescence peaks (Fig. 6M).

Polystyrene microsphere beads (10 μm diameter; Polysciences, Inc., Warrington, PA) were washed three times in PBS and incubated in 0.1% (w/v) poly-L-Lysine(Sigma), or 500 µg/ml fibronectin for 1 h at 37°C. Microspheres were then washed in PBS and added (10⁶ beads per cm²) onto K_ir2.1–tdTomato-expressing HEK293 cells already spread on a fibronectin-coated plate. After 4 h, cell membranes were stained with CellMask Green, and imaged at the bead plane within the next hour using a confocal microscope. The mean intensities of CellMask Green and K_ir2.1–tdTomato around a bead were quantified and their ratio was reported as a measure of the channel density at the membrane.

The surface biotinylation assay was performed following previously described methods (Li et al., 2016). Specifically, cells were plated in 10-cm tissue culture-treated dishes at 800,000 cells/dish and FAK inhibitor was added 2 h after plating. At 6 h after treatment, medium was replaced with ice-cold DMEM containing 30 mM HEPES (pH 7.4) for 10 min. The biotin solution was prepared freshly by dissolving EZ-LinkSulfo-NHS-SS Biotin (Pierce) at 0.5 mg/ml in cold HBSS (Gibco) with Ca²⁺ and Mg²⁺ and added to the cells. The biotinylation reaction was performed at 4°C for 30 min with gentle end-over-end mixing. The reaction was terminated by washing the cells three times with ice-cold HBSS (with Ca²⁺ and Mg²⁺) containing 20 mM glycine. Cells were lysed in lysis buffer (RIPA buffer, 20 mM glycine, and complete protease inhibitor) and centrifuged at 12,000 g for 10 min. Protein concentration was measured via a BCA assay (Pierce). 700 µg of protein from each sample was incubated with streptavidin microspheres (Pierce) overnight at 4°C. Microspheres were washed four times with lysis buffer and the streptavidin-bound biotinylated proteins were eluted and resolved by SDS-PAGE followed by western blotting.

Protein extracts were assessed by western blot analysis, as described in our previous studies (Jackman et al., 2016). Briefly, 20 µg of each protein sample was loaded in each lane of a 10% polyacrylamide gel and transferred onto a PVDF membrane. The membrane was then blocked in TBS with 0.05% Tween 20 and 5% milk. To detect K_ir2.1 protein level, anti-Kir2.1 antibody (1:200, ASC-026, Alomone Labs, rabbit monoclonal) was diluted in TBS with 0.05% Tween 20 and 3% BSA (4°C, overnight). As the internal reference, tubulin were blotted by using anti-tubulin antibody (1:10,000, ab6046, Abcam). HRP-conjugated secondary antibodies were applied for 1 h at room temperature. Chemiluminescence was imaged using a Bio-Rad ChemiDoc system and protein levels were quantified by ImageJ.

Mn(II)Cl₂ (1 mM, Sigma), FAK-Inhibitor PF-573228 (100 µM, Tocris) and Blebbistatin (10 µM, Stem Cell Technologies) were used to modulate FA assembly. Nocodazole (12 µM, Sigma) and Dynasore (25 µM, Sigma) were used to inhibit forward and reverse trafficking of K_ir2.1. For Dynasore experiments, FBS in the culture medium was replaced with NuSerum (Thermo Fisher Scientific, Pittsburgh, PA). For all experiments, cells were exposed to drugs for 2, 4 or 6 h prior to patch clamp recordings and immunostaining. These short exposure times were selected to avoid major changes in K_ir2.1 gene or protein expression. For the vehicle control group (VC), cells were exposed to the vehicle solution alone (no drug) for 6 h.

HEK293 cells expressing K_ir2.1–tdTomato were patterned into star shapes. Samples were imaged as previously described (Rothenberg et al., 2018) with a 60× magnification objective (UPlanSApo 60×/NA 1.35 Objective; Olympus, Tokyo, Japan) on an inverted fluorescent microscope (Olympus IX83) illuminated by a xenon arc lamp (Lambda LS) equipped with a 300 W ozone-free xenon bulb (Sutter Instrument, Novato, CA). The images were captured using a sCMOS ORCA-Flash4.0 V2 camera (Hamamatsu Photonics, Hamamatsu, Japan). User-defined regions of interest (ROI) were photobleached using a 515 nm laser (FRAPPA; Andor Technology, Belfast, UK). Fixed cells were used to create a bleaching protocol (10 laser pulses with a dwell time of 1000 μs per pixel) that resulted in 80% reduction in fluorescence intensity in control cells. This threshold was chosen to allow for significant fluorescent recovery, but minimize the chances of photo-damage. As has been observed in previous studies (Canel et al., 2010; Jaskolski et al., 2009), Dynasore treatment reduced the bleaching efficiency. Therefore, the mobile fraction was calculated to explicitly account for this variation in bleaching efficiency.

Pre- and post-photobleaching images of patterned live cells were acquired every 10 s until 5 min after photobleaching using custom filter set comprising a TRITC excitation filter (FF01-560/25; Semrock, Rochester, NY), RFP emission filter (FF01-607/36; Semrock, Rochester, NY) and dichroic mirror (FF410/504/582/669-Di01; Semrock, Rochester, NY). The motorized filter wheels (Lambda 10-3; Sutter Instrument) and automated stage (H117EIX3; Prior Scientific, Rockland, MA), as well as photobleaching and image acquisition were controlled through MetaMorph Advanced software (Olympus).

Data are presented as mean±s.e.m. and were evaluated for statistical significance using a Student's t-test or an analysis of variance (one-way, α-factor of 0.05) followed by Dunnett's post hoc test. When appropriate, an analysis of variance (two-way, α-factor of 0.05) followed by Tukey's post hoc test for multiple comparisons was used. Differences between groups were considered statistically significance for P<0.05.

We acknowledge H. Nguyen, C. Jackman, R. Gorsuch and I. Puranam for technical support and critical discussions of the manuscript.