Cellular Architecture Regulates Collective Calcium Signaling and Cell Contractility

A key feature of multicellular systems is the ability of cells to function collectively in response to external stimuli. However, the mechanisms of intercellular cell signaling and their functional implications in diverse vascular structures are poorly understood. Using a combination of computational modeling and plasma lithography micropatterning, we investigate the roles of structural arrangement of endothelial cells in collective calcium signaling and cell contractility. Under histamine stimulation, endothelial cells in self-assembled and microengineered networks, but not individual cells and monolayers, exhibit calcium oscillations. Micropatterning, pharmacological inhibition, and computational modeling reveal that the calcium oscillation depends on the number of neighboring cells coupled via gap junctional intercellular communication, providing a mechanistic basis of the architecture-dependent calcium signaling. Furthermore, the calcium oscillation attenuates the histamine-induced cytoskeletal reorganization and cell contraction, resulting in differential cell responses in an architecture-dependent manner. Taken together, our results suggest that endothelial cells can sense and respond to chemical stimuli according to the vascular architecture via collective calcium signaling.


Introduction
The calcium ion is a universal second messenger mediating a wide range of dynamic cell functions, such as exocytosis, contraction, transcription, and proliferation [1][2][3]. These biological processes that span time scales from microseconds to hours are regulated by diverse calcium signaling mechanisms. A hallmark of calcium signaling is the transient pulsing dynamics of cytosolic calcium concentration. The calcium dynamics are actively controlled by calcium release from intracellular stores and calcium influx from extracellular media, outflow by pumps and exchangers, and calcium binding proteins, which fine-tune the amplitude and duration of calcium pulses to regulate physiological functions. Calcium oscillation, for instance, can sensitize signal detection at low agonist concentration and its frequency can also encode additional information that selectively controls transcription factor activity [4][5][6].
The calcium dynamics are regulated by multiple intracellular calcium processing mechanisms, such as non-selective cation channels, voltage-dependent calcium channels, store-operated calcium channels, calcium-induced calcium release, and phospholipase C mediated inositol 1,4,5-trisphosphate activity [3]. Intercellular collective calcium signaling mechanisms can also modulate the calcium dynamics. In particular, gap junctional intercellular communication (GJIC) mediates coupling currents between neighboring cells triggered by an imbalance in the membrane potential. GJIC also allows calcium diffusion driven by a difference in the cytosolic calcium concentration [7,8]. For instance, calcium waves can propagate among endothelial cells, and intercellular calcium signaling provides a robust mechanism for mechanosensing in endothelial networks [9,10]. Three-dimensional coupling of β-cells can lead to robust synchronized calcium oscillations and insulin secretion at elevated glucose levels [11]. GJIC also regulates asymmetric neuronal fates [12] and controls collective chemosensing to generate synchronized and coordinated responses under adenosine triphosphate (ATP) stimulation [13,14]. Physiologically, histamine is known to induce calcium signaling and cell contraction in endothelial cells. Interestingly, histamine-induced permeability is confined to venules rather than capillaries [15], suggesting an architecture dependence on the response to histamine. Nevertheless, the effects of structural arrangement, as seen in diverse vascular structures, on calcium signaling and their functional implications are poorly understood.
In this study, we investigate architecture-dependent collective calcium signaling in human umbilical vein endothelial cells (HUVEC). Self-assembled capillary-like networks [16] and microengineered hexagonal cell networks [17] are created to evaluate the effects of cellular architecture on calcium signaling. Endothelial cell networks of arbitrary shapes and widths can be created by plasma lithography cell patterning [18]. The method has been demonstrated for investigating the effects of geometric confinements on various biological systems [19][20][21][22]. Histamine-induced calcium oscillations in the cell networks are compared with individual cells and monolayers to investigate the architecture dependence. Along with micropatterning and pharmacological inhibition, a computational model is developed to elucidate the effects of the number of neighboring cells on collective calcium signaling. The cytoskeletal reorganization and cell contraction induced by histamine are correlated with the calcium oscillation to study the function of collective calcium signaling in different cellular architectures.

Architecture-dependent collective calcium signaling
To explore the effects of cellular architecture, histamine-induced calcium signaling was monitored in individual endothelial cells, monolayers, self-assembled capillary-like networks, and plasma lithography microengineered cell networks (Fig 1). Histamine treatment (5 μM) induced a rapid increase followed by a slow decay of cytosolic calcium concentration in individual cells (Fig 1A and 1E and S1 Video) and in monolayers (Fig 1B and 1F and S2 Video). Mean decay rate, the reciprocal of half decay time, was 0.0087±0.0003 sec -1 for individual cells and 0.0180±0.0018 sec -1 for cells in monolayer. Only a small fraction of cells (~8%) exhibited cytosolic calcium oscillations. This observation is consistent with previous reports that histamine does not induce calcium oscillations in human endothelial cells at high concentrations (over 3 μM) [23]. Surprisingly, the cells in self-assembled capillary-like networks exhibited short initial calcium pulses with a rapid mean decay rate of 0.0380±0.0032 sec -1 . The decay rate was significantly faster than the values in individual cells and monolayers. Importantly, calcium oscillations occurred in a much larger portion (~55%) of cells in self-assembled capillarylike networks (Fig 1C and 1G and S3 Video). The oscillations displayed irregular patterns and did not show any apparent synchronization between neighboring cells.
To confirm the calcium oscillation was triggered by the cellular architecture, instead of the physical or biochemical properties of the extracellular matrix, endothelial cells were patterned into hexagonal cell networks on polystyrene by plasma lithography (S1 Fig). Similar to capillary-like networks, calcium oscillations occurred in a large portion (~61%) of cells and displayed irregular calcium patterns without apparent synchronization (Fig 1D and 1H). The calcium pulses of cells in hexagonal cell networks also had a rapid mean decay rate of 0.0374 ±0.0017 sec -1 . There is no significant difference between capillary-like networks and hexagonal cell networks for the mean decay rate (Fig 2A) and the calcium oscillation occurrence rate ( Fig  2B). The cell response time, calcium increase rate, and amplitude did not depend on the cellular architecture. These results indicate that the calcium dynamics of endothelial cells, in particular the occurrence of calcium oscillation, depends on the cellular architecture.

Architecture-dependent calcium signaling is GJIC dependent
Gap junctions, which electrically and chemically couple endothelial cells, play significant roles in intercellular communication and cell functions [24,25]. To decipher the roles of GJIC in the observed architecture-dependent calcium signaling, endothelial cells in capillary-like networks and hexagonal cell networks were pretreated with a gap junction blocker, 18-β-glycyrrhetinic acid (18β-GA, 40 μM) [26,27]. With histamine treatment, cytosolic calcium levels in 18β-GA treated cells increased abruptly and gradually decreased to the basal level, similar to individual cells and monolayers (S2 Fig). The mean decay rate was 0.0283±0.0011 sec -1 and 0.0306±0.0018 sec -1 for capillarylike networks and hexagonal cell networks, respectively (Fig 2A). The calcium oscillation occurrence rates were reduced significantly in both capillary-like networks and hexagonal cell networks ( Fig  2B). The distributions of the half decay time of calcium pulses for various cell patterns further indicated that the portion of cells with a prolonged calcium response increased with 18β-GA treatment ( Fig 2C). These results suggest the architecture-dependent calcium oscillation is GJIC dependent.

Computational analysis of pluricellular calcium dynamics
We investigate the mechanisms of architecture-dependent calcium oscillation using a computational model [8,28]. This model describes calcium fluxes from intracellular stores and extracellular media to the cytosol while simultaneously considering intercellular fluxes due to Collective Calcium Signaling electrical and biochemical coupling between neighboring cells via GJIC (Fig 3A). The calcium fluxes are regulated by the cytosolic calcium concentration through multiple feedback loops, which capture the dynamics of plasma membrane Ca 2+ ATPase, calcium channels, endoplasmic reticulum uptake, and inositol 1,4,5-trisphosphate activity. A detailed description of the computational model is provided in S1 Text. The model generated calcium pulses and calcium-induced calcium release as observed in endothelial cells. Experiments were first performed to validate the accuracy of the model. In the experiment, endothelial cells were treated with different concentrations of histamine (Fig 3B-3E). The interval between pulses, amplitude of pulses and the calcium oscillation occurrence rate are summarized in Fig 3F, 3G and 3H. The computational model successfully predicted the weak dependence of the interval between pulses on histamine concentration (Fig 3F, 3I, and 3J). The amplitude of the pulses was   Fig 3G). Most importantly, the model successfully captured the prolonged calcium response and low calcium oscillation occurrence rate for cells treated with high (5 μM) concentrations of histamine ( Fig 3H, 3K and 3L). The computational model, therefore, captured the essence of calcium dynamics observed in our experiment, supporting its use for investigating the architecture-dependent calcium signaling.
The computational model was applied to study the effects of calcium diffusion and coupling current on the calcium dynamics. Individual cells did not exhibit calcium oscillations in the simulation ( Fig 4A). When two cells were electrically coupled, a coupling current was generated due to a small difference in the membrane potentials, which tended to equalize the membrane potentials. To stabilize the membrane potential, the coupling current was compensated by other currents, including the calcium current, which perturbed the cytosolic calcium concentration and induced irregular, unsynchronized calcium oscillations (S3 Fig). The electrical coupling formed the basis of calcium oscillation in coupled cells [8,28]. When multiple cells were connected as a linear chain (i.e., each cell was coupled to two immediate neighbors), the model predicted that most cells would exhibit calcium oscillation. Fig 4B shows the behavior of nine cells connected linearly. Similar to the capillary-like and hexagonal cell networks, irregular patterns of calcium pulses were observed and the oscillation did not display apparent synchronization between neighboring cells. Remarkably, a different behavior emerged when the number of coupled cells increased. When the cells were connected in an array and coupled to each other (e.g., monolayer), the calcium oscillations vanished in all of the cells. Fig 4C shows the calcium dynamics when nine cells are coupled. The cells exhibited prolonged calcium responses and did not display any oscillation, similar to the experimental observation. Thus, the computational simulation captures the architecture-dependent calcium signaling observed in the experiment.
Examining the computational model provides insights into the mechanism of the architecture-dependent calcium signaling. In particular, the intracellular calcium dynamics driven by the calcium-induced calcium release were highly sensitive to the cytosolic calcium concentration. The membrane potential and coupling current created the oscillations in coupled cells. On the other hand, the intercellular diffusion of calcium could stabilize the coupled cells by maintaining the cytosolic calcium at the steady state value. This stabilizing capability increased with the number of cells coupled. Fig 4D, 4E and 4F shows the cytosolic calcium dynamics when 3, 4, and 6 cells were coupled. The calcium oscillation occurrence rate decreased as the number of cells increased. Oscillations were not observed when seven or more cells were coupled. Similar to the gap junction blocker experiment, eliminating the coupling in the computational model suppressed calcium oscillations and resumed the calcium dynamics of individual cells.

Collective calcium signaling depends on the number of neighboring cells
The computational model predicts that the calcium dynamics are sensitive to the number of cells coupled. To test the effects of the number of coupled cells on intercellular calcium endoplasmic reticulum Ca 2+ -ATPase (SERCA) and IP 3 receptor (IPR) regulate calcium fluxes between the cytosol and the endoplasmic reticulum (ER). Plasma membrane Ca 2+ -ATPase (PMCA) and calcium channels transport calcium between the cytosol and the extracellular space. With GJIC, the cytosolic calcium concentration can be perturbed by the coupling current due to an imbalance in membrane potential and calcium diffusion due to a concentration gradient. dynamics experimentally, linear cell networks with various widths were designed and patterned via plasma lithography (Fig 5). In particular, linear cell networks with the width of a single cell to multiple cells (20-100 μm) were created. The average number of neighboring cells increased from 2 to 8 in these linear networks. Upon histamine stimulation, the cells in linear networks displayed cytosolic calcium pulses with different decay rates (Fig 5A, 5B and 5C). The mean decay rate decreased as the line width increased from 40 μm to 100 μm (Fig 5D). Importantly, the calcium oscillation occurrence rate decreased gradually as the width of linear patterns increased from 20 μm to 100 μm (Fig 5E). In addition, cells at the edge (not surrounded on all sides) are more likely to undergo calcium oscillations than cells at the inner region (surrounded on all sides) at the presence of 5 μM histamine for capillary-like networks, hexagonal cell networks and 100 μm wide linear cell networks (S4 Fig). Therefore, both experimental and computational models support that the calcium oscillation occurrence rate negatively correlated with the number of neighboring cells in the network.

Collective calcium signaling regulates cytoskeletal reorganization
Histamine mediates changes in actin cytoskeleton, induces intercellular gap formation, increases endothelial permeability, and causes endothelial barrier dysfunction through calcium-mediated cell contraction [29][30][31][32]. The effects of histamine on the actin cytoskeleton were investigated to explore the potential impact of collective calcium signaling on physiological cell functions. Before histamine treatment, dense F-actin structures surrounding the nuclei (but not in the nuclei) were observed in individual cells and monolayers (Fig 6A and 6E). The fluorescence intensity in the nucleus region was significantly lower than the values in the cytoplasm. After histamine treatment for 5 min, the F-actin structures reorganized and stress fibers formed across the cells, including the nucleus region (Fig 6B and 6F). With the stress fiber formation after cytoskeletal reorganization, the fluorescence intensity in the nucleus region was similar to the values in the cytoplasm. The ratio of fluorescence intensities between nucleus and cytoplasm was therefore employed to quantify the actin distribution before and after histamine treatment. Cytoskeletal reorganization was defined quantitatively when the ratio of fluorescence intensities was above 0.7. Red arrows in Fig 6 indicate cells with dense Factin structures surrounding the nuclei (i.e., no cytoskeletal reorganization).
To investigate the influence of collective calcium signaling on cytoskeletal reorganization in cell networks, endothelial cells in linear networks were exposed to histamine for 5 minutes. Unlike individual cells and monolayers, the dense F-actin structures remained unchanged and stress fibers were not observed in linear networks with either Hanks balanced salt solution (HBSS) control (Fig 6I) or histamine (Fig 6J). F-actin reorganization was also studied when both coupled cells and individual cells were present in the same well by controlling the cell seeding density. Without histamine, dense F-actin structures were observed in both coupled cells and individual cells (Fig 6M). With histamine treatment, dense F-actin structures were observed only in coupled cells and cytoskeletal reorganization was observed in individual cells (Fig 6N), supporting the notion that histamine-mediated cytoskeletal reorganization depends on the cellular architecture.
To test if the effect of histamine is modulated by calcium oscillations, pulses of histamine by repetitive pipetting histamine (30 seconds) and HBSS (1 minute) were manually applied to generate calcium pulses in individual cells and monolayers (Fig 6C and 6G). The effects of histamine pulses on the actin cytoskeleton were significantly lower than continuous treatment of histamine for 5 min, despite the fact that the total treatment time were similar (Fig 6H). On the other hand, blocking calcium oscillation in cells in linear patterns via the gap junction blocker, which generate a prolonged calcium response in the network, triggered histamine-mediated cytoskeletal reorganization (Fig 6K). Dense F-actin structures were also significantly reduced by histamine for individual cells in linear cell networks (Fig 6O). There is no significant difference between histamine only and histamine with 18β-GA for individual cells (Fig 6P). These observations support the notion that cytoskeletal reorganization is triggered only by a prolonged calcium response, but not calcium oscillations.

Collective calcium signaling regulates cell contractility
Histamine-mediated cell contraction was investigated in capillary-like networks to further explore the effects and potential physiological functions of collective calcium signaling. With histamine treatment, cells contracted and migrated towards the nodes, resulting in a moderate (~10%) shrinkage of the capillary-like network (Fig 7A and S4 Video). The shrinkage was estimated by the areas of the network before and after histamine treatment. For control with HBSS, no significant changes of the networks were observed (S5 Video). With 1,2-Bis(2-aminophenoxy)ethane-N,N,N',N'-tetraacetic acid tetrakis(acetoxymethyl ester (BAPTA-AM) pretreatment, histamine-induced calcium signaling was inhibited and the network shrinkage rate reduced to~3% (Fig 7B). In contrast, the histamine-induced shrinkage increased to~16% for capillary-like networks treated with a gap junction blocker which generated a prolonged calcium response (Fig 7C). The nucleus displacement was also analyzed by particle image velocimetry analysis to study histamine-induced cell contraction (S5 Fig and S6 Video). The average displacement of nuclei mediated by histamine was decreased by BAPTA and increased by 18β-GA, supporting the contraction is calcium dependent and can be enhanced by a prolonged calcium response (S6 Fig). The nuclei displacement data were in good agreement with the shrinkage data (Fig 7D and S6D Fig).

Discussion
In this study, the significance of cellular architecture in GJIC-mediated collective calcium signaling is identified. Unlike monolayers and individual cells that exhibit prolonged cytosolic calcium response, cells in capillary-like networks and hexagonal cell networks display short calcium pulses and oscillations after histamine treatment. Histamine-mediated calcium oscillations occur only when a small number of (~2-6) cells are physically connected in self-assembled and microengineered networks. Systematic investigation using linear networks and computational modeling suggests that the calcium oscillation occurrence rate negatively correlates with the number of neighboring cells. Patterning cells in linear networks with low cell density and with different widths supports that the calcium oscillation is not due to the elongated morphology of the cells or physical confinement by the patterns. The dependence on the number of coupled cells leads to the architecture-dependent calcium signaling, since most cells in the networks connect to 2-5 cells while cells in monolayers typically connect to over 8 cells. Additionally, pharmacological inhibition, computational modeling and micropatterning revealed that the calcium oscillation is mediated by GJIC. Inhibition of GJIC with a gap junction blocker suppresses the calcium oscillation. With GJIC inhibition, the calcium dynamics of cells in capillary-like networks and hexagonal cell networks resemble the response of individual cells. These results collectively suggest that cellular architecture regulates collective calcium signaling via GJIC.
A computational model is developed to explore the mechanisms of architecture-dependent calcium oscillations. The computational model captures the essence of calcium dynamics, including the transient pulses, the uniform amplitude, the concentration dependence, the irregular pattern, the occurrence rate, and the architecture dependence. The model provides insights into the mechanisms that drive the architecture-dependent calcium oscillation. In particular, the calcium fluxes are highly sensitive to the cytosolic calcium concentration. The coupling current due to electrical coupling perturbs the cytosolic calcium concentration, which generates the oscillatory activity in coupled cells [8]. The electrical coupling provides a potential mechanism on how calcium oscillations are generated in coupled cells, but not in individual cells. Incorporating electrical coupling in the model successfully creates the random distribution of calcium pulses and the irregular pattern observed in the experiment. The oscillation cannot be explained by intercellular calcium diffusion alone, as the oscillations are irregular and unsynchronized between neighboring cells. In contrast, the biochemical coupling through intercellular diffusion of calcium suppresses the oscillation by stabilizing the cytosolic calcium concentration in the cells. As supported by the computational model and linear networks with different widths, the stabilizing capability increases with the number of cells coupled and abolishes calcium oscillations when a large number of cells are connected (e.g., in monolayers). Since the stabilizing capability depends on the number of neighboring cells, the calcium oscillation occurs only when a small number of cells are coupled, providing a basis of the architecture-dependent calcium signaling (S7 and S8 Figs).
Our computational model is based on a set of assumptions and simplifications, which will likely limit the predictive power of the model. Several aspects of the model should be considered. First, the contribution of voltage gated calcium channels to agonist-induced calcium entry in endothelial cells is inconclusive [33][34][35]. Endothelial cells are often considered to be electrically unexcitable and the effect of membrane potential on agonist-induced calcium signaling in endothelial cells remains controversial [36,37]. Second, the permeability of vertebrate gap junction can be modulated by cytosolic ionic composition and connexin phosphorylation. The increase in cytosolic calcium concentration was shown to inhibit electrical coupling and dye transfer among cells [38]. Phosphorylation of connexin by protein kinases also resulted in internalization and degradation of gap junctions [39]. Our model may only represent the cell behaviors under specific conditions and do not rule out the possibility of other calcium signaling mechanisms.
In general, calcium dynamics can be influenced by a large number of factors and mechanisms. Computational models of calcium dynamics and signaling in vascular regulation have been reviewed previously [40]. The relationships between the cytosolic calcium, membrane potential, coupling current and voltage gated channels can be context-dependent (e.g., agonist, concentration and duration) and each factor may influence the calcium dynamics through multiple feedback loops direct or indirectly. As an example, the inhibitory effects of verapamil (an L-type voltage gated calcium channels inhibitor) on histamine-induced calcium signaling in human umbilical vein endothelial cells were observed only at submaximal histamine concentrations, suggesting that voltage gated calcium channels might play a role at a specific range of agonist concentration in HUVECs [41]. Non-selective cation channels were also reported to be responsible for histamine-activated Ca 2+ influx [42]. For histamine-activated calcium signaling in endothelial cells, it has been demonstrated that the membrane potential plays a key role for fine-tuning the calcium influx [43]. In particular, the membrane potential can effectively regulate calcium influx which is responsible for the plateau phase of agonist-mediated calcium dynamics. In addition, activation of the calcium-activated potassium channel triggers membrane hyperpolarization and, thus, provides a driving force for cation influx (Na + , K + and Ca 2+ ) [44]. In the future, additional studies should be performed to determine the contribution of voltage-gated calcium channels, membrane potential and gap junction mediated electrical coupling with the regulation of cytosolic calcium concentration and connexin phosphorylation.
Our results underscore the possibility that cellular architecture modulates histamine-mediated cytoskeletal reorganization and cell contractility via GJIC-mediated collective calcium signaling. Unlike monolayers and individual cells that exhibit cytoskeletal reorganization, histamine has minimal effects on the actin cytoskeleton in capillary-like networks and hexagonal cell networks. To study the roles of calcium oscillation on cytoskeletal reorganization, histamine pulses that impose calcium oscillations are applied to individual cells and monolayers. The histamine pulses have a weak effect on the actin cytoskeleton compared to continuous histamine treatment. On the other hand, suppressing calcium oscillations in self-assembled networks by GJIC inhibition induces cytoskeletal reorganization and cell contraction. Therefore, the effects of histamine can be modulated by the calcium dynamics and are architecture-dependent. Physiologically, the architecture dependence may represent one of the ways of endothelial cells to respond to the stimuli differentially in various vascular structures. The architecturedependent calcium oscillations might contribute to the specificity of histamine-induced permeability in venules, but not in capillaries [15]. Our results, however, do not rule out other potential mechanisms in generating the specificity such as the differences in calcium signaling mechanisms between venules and capillaries. Further investigation is required to test this hypothesis and study the architecture dependence in other models (S9 Fig). As a universal second messenger, collective calcium signaling may serve as a broad mechanism in regulating various cell functions. Other factors may also be considered in the computational and experimental models to decipher the regulation and functional implication of collective calcium signaling in diverse cellular architectures.

Materials and Methods
Cell culture HUVEC were purchased from BD Bioscience (Bedford, MA). The cells were cultured in Medium 200 with low serum growth supplement (Life Technologies, Carlsbad, CA) and 5% CO 2 at 37°C. HUVEC from passages three to six were used in all experiments.

Capillary-like network formation
Matrigel (growth factor reduced, BD Bioscience) was thawed with ice at 4°C refrigerator overnight, added into 48-well plates (130 μL per well) and gelled at 37°C for 1 hour. HUVEC were harvested from tissue culture dishes at a confluency of~90% using 0.25% Trypsin-EDTA (Life Technologies) for 1−2 minutes). The cells were then seeded on top of the gelled matrigel at a density of 300−400 cells/mm 2 . At 8 hours after cell seeding, images of capillary-like networks were taken by a CCD camera (Cooke SensiCam) with an epi-fluorescence microscope (Nikon TE2000-U).

Cell patterning
Plasma lithography was employed to pattern cells into hexagonal cell networks that mimic the capillary, and into linear cell networks for studying the effects of the number of neighboring cells on collective calcium signaling. Briefly, a positive master (50 μm in height) on silicon wafer was fabricated with SU-8 (MicroChem) by photolithography. The polydimethylsiloxane (PDMS) pre-polymers (Sylgard 184, Dow Corning) were casted on the master and cured at 65°C for 4 hours. The widths of linear patterns were 20 μm, 40 μm and 100 μm. The width and segment length of hexagonal channels were 20 μm and 250 μm respectively. The mold was placed in conformal contact on a native polystyrene substrate (24-well plate from BD Bioscience), exposed to atmospheric plasma for 20 minutes and removed. The HUVEC were then seeded (800 cells/mm 2 ) onto the substrate. After overnight culture, cells selectively adhered to the plasma treated area and formed microengineered cell networks.

Calcium imaging and analysis
Cell networks were washed gently with HBSS for three times, incubated with 10 μM Fluo-3/ AM (Biotium) in HBSS at 37°C for 45 minutes in the dark, and washed with HBSS for another three times to remove the residual Fluo-3/AM. Fluorescence images were captured at 100 ms intervals with an exposure time of 400 ms for 5 minutes once the histamine (Sigma) solution was applied to the cells with a pipette at a final concentration of 5 μM. Since the largest frequency of calcium oscillations found in non-excited cells is about 0.1 Hz, 400 ms exposure time should capture all the calcium spikes [45]. The images were analyzed by ImageJ (NIH). To investigate the effect of gap junctions on collective calcium signaling, cells were pre-incubated with 40 μM 18β-GA (Sigma) for 30 min and stimulated with histamine in the presence of 18β-GA. The relative change in calcium was calculated as previously mentioned [46]. The frequency of calcium oscillations was estimated by utilizing the time interval between the first two calcium spikes. To quantitatively differentiate calcium oscillations that have two or more calcium spikes from small and short repetitive fluctuations in cytosolic calcium, amplitude of the calcium signal relative to the baseline noise was used. A calcium fluctuation with peak-to-baseline ratio larger than 3 was counted as a calcium spike.

Immunostaining and imaging
Cells were fixed with 4% paraformaldehyde for 15 minutes, washed with PBS twice, permeabilized with 0.2% (w/v) Triton X-100 (Sigma) for at least one hour, washed with PBS twice and blocked with 1% BSA (sigma) for 5 hours. Alexa Fluor 555 Phalloidin (Life Technologies) was diluted at 1:30 ratio with 1% BSA solutions and applied to the fixed cells at 4°C overnight. Cells were washed three times with PBS (5 minutes each time) before imaging. For histamine H1 receptor immunofluorescence, cells were fixed and blocked as described above, and incubated overnight at 4°C with rabbit histamine H1 receptor antibody (Novus Biologicals, rabbit polyclonal IgG, 1:150 diluted). After washing three times with PBS, cells were further incubated at room temperature for 1 hour with Alexa Fluor1 555 donkey anti-rabbit IgG antibody (Life Technologies). To study the roles of calcium signaling in cytoskeletal reorganization, cells were incubated with 15 μM BAPTA-AM (EMD Millipore) for 30 minutes at 37°C in the dark and washed three times with HBSS before histamine applications. To further resolve the role of calcium oscillations, three pulses (30 seconds) of histamine solution were applied to the monolayer and single cells with an interval of 1 minute HBSS wash between pulses to mimic natural calcium oscillations.

Capillary-like network shrinkage and particle image velocimetry (PIV) analysis
For estimating the shrinkage ratio of capillary-like networks, ImageJ was used for measuring the areas of cell networks before drug application and 5 minutes after drug application. For PIV analysis, cell nuclei were stained with Hoechst 33342 (Life Technologies) for 20 minutes at 37°C, and fluorescence images (blue) of capillary-like networks before and after drug applications were analyzed with ImageJ and PIV plugins.

Statistical analysis
Student's unpaired t-tests (two-tailed) were performed to compare two experimental groups. A one-way ANOVA with Bonferroni post-tests was used to compare the means of three or more experimental groups. All data are presented as mean ± standard error of the mean (SEM).