Spatial confinement of active microtubule networks induces large-scale rotational cytoplasmic flow

Significance

At the microscopic scale, collective behaviors of motile units can induce directed fluid flow on a larger length scale than individual units based on their hydrodynamic interactions. Here, we found that the motor-driven extensile behaviors of microtubule bundles in the cytoplasm induce rotational flow in a cell-sized confined space on length scale and timescale that were 10- to 100-fold longer than the vortex flows emerging in the bulk space. These scale differences were derived from mechanical force generation by microtubule bundle elongation near the physical boundary and the transmission of this force over the microtubule network. These findings suggest that the microtubule cytoskeleton utilizes not only hydrodynamic interactions but also mechanical interactions to induce large-scale cytoplasmic flow.

Abstract

Collective behaviors of motile units through hydrodynamic interactions induce directed fluid flow on a larger length scale than individual units. In cells, active cytoskeletal systems composed of polar filaments and molecular motors drive fluid flow, a process known as cytoplasmic streaming. The motor-driven elongation of microtubule bundles generates turbulent-like flow in purified systems; however, it remains unclear whether and how microtubule bundles induce large-scale directed flow like the cytoplasmic streaming observed in cells. Here, we adopted Xenopus egg extracts as a model system of the cytoplasm and found that microtubule bundle elongation induces directed flow for which the length scale and timescale depend on the existence of geometrical constraints. At the lower activity of dynein, kinesins bundle and slide microtubules, organizing extensile microtubule bundles. In bulk extracts, the extensile bundles connected with each other and formed a random network, and vortex flows with a length scale comparable to the bundle length continually emerged and persisted for 1 min at multiple places. When the extracts were encapsulated in droplets, the extensile bundles pushed the droplet boundary. This pushing force initiated symmetry breaking of the randomly oriented bundle network, leading to bundles aligning into a rotating vortex structure. This vortex induced rotational cytoplasmic flows on the length scale and timescale that were 10- to 100-fold longer than the vortex flows emerging in bulk extracts. Our results suggest that microtubule systems use not only hydrodynamic interactions but also mechanical interactions to induce large-scale temporally stable cytoplasmic flow.

Many biological systems exhibit directed fluid flow on various length scales from the order of 10 m, such as a school of fish (1), down to the order of 10 µm, such as a bacterial suspension (2–5). Swimming bacteria generate dipole fluid flow around themselves (6). At high concentrations of bacteria, the dipole fluid flow induces orientation of neighboring bacteria, leading to large-scale spiral-pattern organization generating the vortex flow (2–5).

Inside cells, active cytoskeletal networks mainly composed of polar filaments and molecular motors generate fluid flow in the cytoplasm, a process called cytoplasmic streaming (7–14). Although still little is known about the biological roles of cytoplasmic streaming, streaming is thought to be necessary for long-distance transport of micrometer-sized components, such as organelles (9). The active cytoskeletal system is classified into two categories, that is, actin and microtubule systems, according to filament type. In both systems, the polar filaments (actin filaments or microtubules) are polymerized and depolymerized using chemical energy obtained from ATP or GTP hydrolysis, and the molecular motors (myosin for actin filaments, kinesin and dynein for microtubules) convert chemical energy of ATP hydrolysis into mechanical work, resulting in the sliding of overlapping filaments. Thus, actin and microtubule systems in the cytoplasm are highly dynamic.

Actin-based cytoplasmic streaming has been extensively studied, and the following mechanism has been generally accepted: organelle-bound myosins moving on actin filaments produce hydrodynamic flows, which align neighboring actin filaments to induce large-scale streaming (10–12, 14). Similar to actin-based streaming, organelle-bound kinesins moving on microtubules are thought to generate large-scale streaming (9). Meanwhile, recent in vitro experiments have demonstrated that purified microtubules are organized in extensile bundles by artificially clustered kinesins, and the network of these bundles generates fluid flows (15). However, it is still unclear whether and how microtubule bundle elongation can drive cytoplasmic flow in cells. As revealed in bacteria-driven vortex flow (2–5), the hydrodynamic interaction between microtubule bundles is probably essential. Bundle elongation generates hydrodynamic flows around itself, which might be able to align neighboring filaments to induce large-scale flow. In addition to these hydrodynamic interactions, the physical boundaries of the cells potentially contribute to initiating the organization of large-scale cytoplasmic flow. This is because the persistence length of microtubules (1‒2 mm: ref. 16) is nearly equal to or longer than the cell size (10‒1,000 μm). In fact, a recent theoretical study of actin-based streaming indicated that, in a cylindrical cell, actin filaments located on the side preferentially reorient from circumferential to longitudinal so that the filaments minimize their physical curvature, thereby directing cytoplasmic streaming (12). This work allows us to speculate that the mechanical interactions between microtubules and the physical boundary more strongly affect streaming, because the flexural rigidity of microtubules is much higher than that of actin filaments (17, 18).

Here, we adopted metaphase Xenopus egg extracts as a model system of the cytoplasm (19) and examined whether and how the microtubule system in the cytoplasm induced directed fluid flow. We found that the confinement of cytoplasmic microtubules in droplets induces rotational flow on length scale and timescale that were 10- to 100-fold longer than the vortex flows emerging in bulk extracts. Quantitative analysis and comparison with the bacteria-driven flows (2–5) suggested that, in addition to hydrodynamic interactions, mechanical interactions of microtubules play important roles in the self-organization of large-scale cytoplasmic flow.

Results

Vortex Flows Emerged in Bulk Cytoplasmic Extracts at the Lower Activity of Dynein.

We first observed self-organization of microtubule networks in bulk metaphase extracts. To promote the assembly of dense microtubule networks, extracts stored on ice were mixed with fluorescently labeled Taxol, a microtubule-stabilizing agent (100 nM, unless otherwise stated, at which the depolymerization rate is decreased, but the frequency of microtubule catastrophe is not affected: ref. 20). To eliminate the effects of actin dynamics (21, 22), 2 μM cytochalasin D, an actin-depolymerizing agent, was added to the extracts. The extracts were then perfused into a large flow chamber (20 × 20 mm² with a height of ∼100 μm), sealed to prevent unexpected fluid flow, and warmed to 20 °C to initiate microtubule polymerization. As the polymerization proceeded, multiple microtubule asters assembled over the entire region and connected with each other, organizing a lattice network (Fig. 1A). These results are consistent with a previous report (23).

Fig. 1.

Vortex flows continually emerge and disappear at multiple places in bulk extracts. (A and B) Epifluorescence images of microtubule networks self-organized in bulk Xenopus egg extracts without (A) and with (B) the dynein inhibitor p150-CC1. (Scale bars, 100 μm.) (C) Flow field of the bulk extracts measured by PIV (grid size, 9.5 μm). The vectors (Left) or flow streamlines (Right) were merged with the bright-field image. The blue boxes show places where typical vortex flows emerged. Not all vectors are shown for clarity. (Scale bar, 100 μm.) (D) Flow velocity distribution shown in C. (E) The normalized spatial correlation function C(r, t) as a function of lateral separation r. Black, orange, green, and cyan lines indicate the spatial correlation functions at 155, 161, 167, and 171 min, respectively. In all microscopic images, 0 min indicates the timing of elevating the temperature from 0 to 20 °C.

Because dynein participates in aster formation (23), we assumed that dynein played a dominant role in this lattice formation. As expected, adding a dynein inhibitor (24, 25) prevented lattice formation, and the network became disordered (Fig. 1B and Fig. S1). In the presence of a dynein inhibitor, cytoplasmic granules appeared to flow. Particle image velocimetry (PIV) measurements of the granules clarified that turbulent-like complex flow emerged in the entire region of bulk extracts (Fig. 1C). The complex flow continued for at least several hours, and the steady-state flow velocity was 0.36 ± 0.26 μm⋅min⁻¹ (mean ± SD; Fig. 1D). During this period, local vortex flows were continually generated at multiple places (Fig. 1C, blue boxes) and disappeared within a minute after generation. Consistent with this, the time correlation of the flow evaluated over the entire region of the complex flow was decayed within 1 min. To examine the spatial correlation of the complex flow, we calculated the equal-time velocity–velocity correlation function C(r, t), defined as follows:

$$\text{C}\left(r,t\right)={\langle \frac{\overrightarrow{v}\left(\overrightarrow{x}+\overrightarrow{r},t\right)\mathrm{\cdot }\overrightarrow{v}\left(\overrightarrow{x},t\right)}{\left|\overrightarrow{v}\left(\overrightarrow{x}+\overrightarrow{r},t\right)\right|\left|\overrightarrow{v}\left(\overrightarrow{x},t\right)\right|}\rangle }_{x,\theta },$$

where ${\langle \hspace{0.17em}\rangle }_{x,\theta }$ indicates the average over space coordinates $\overrightarrow{x}=\left(x,y\right)$ and all angles θ of $\overrightarrow{r}$. The $\text{C}\left(r,t\right)$ shows the orientation correlation of the flow between two points with the distance r at time t. The $\text{C}\left(r\right)$ exponentially decayed and reached zero at ∼50 μm (defined as the correlation length; Fig. 1E), which was comparable to the size of the vortex flow (Fig. 1C, blue boxes). The correlation length was also comparable to the length of single microtubule bundles (34.4 ± 10.1 μm, mean ± SD; n = 63 bundles; Fig. S2), rather than the length of single microtubules (14.5 ± 6.4 μm, mean ± SD; n = 122 microtubules). To investigate how microtubule bundles induce complex flows, we observed bundle behaviors by confocal microscopy and found that the bundles elongated (Movie S1). These results suggested that the complex flows emerged through hydrodynamic interactions between extensile microtubule bundles based on a mechanism similar to that of bacteria-driven vortex flows (2–5). The complex flow driven by microtubule bundle elongation was also observed in vitro (15), but the correlation length was fourfold longer than that in our work. This fact suggests that the correlation length depends on the bundle density, because the microtubule bundle density in the previous work was much higher than that in our work.

Fig. S1.

The behavior of the microtubule network in the presence of a dynein inhibitor and high concentration of Taxol. Time-lapse images of the microtubule network in the presence of 2.5 μM Taxol. The microtubules were organized in a random network in bulk extracts, similar to those in the presence of 100 nM Taxol (Fig. 1B). However, in contrast to the result in the presence of 100 nM Taxol and 800 nM p150-CC1 (Fig. 1B), the network showed contractile behaviors in the presence of 2.5 μM Taxol and 800 nM p150-CC1, as reported previously (43). Because microtubule polymerization and depolymerization were almost completely suppressed at 2.5 μM Taxol, these results suggest that microtubule stability is required for the network contraction. Although the contractile behavior is consistent with the previous report (43), it should be noted that there may be differences between our results and the previous results because we used freshly prepared cytoplasmic extracts, whereas the previous work used extracts that had been frozen. (Scale bar, 100 μm.) [OG-Taxol] = 500 nM, [Unlabeled Taxol] = 2,000 nM, and [p150-CC1] = 800 nM.

Fig. S2.

The distribution of microtubule bundle length. The bundles were scanned along the z axis using a confocal microscope. We then analyzed only the bundles for which the entire structure was captured in a single-slice image, with both ends clearly distinguished. The length of single microtubule bundles was 34.4 ± 10.1 μm (mean ± SD; n = 63 bundles). [OG-Taxol] = 100 nM; [p150-CC1] = 800 nM.

Encapsulation of the Extracts in a Droplet Induces Rotational Cytoplasmic Flow at Length Scale and Timescale That Are Larger than the Vortex Flows in Bulk Extracts.

We next investigated the effects of spatial confinement by encapsulating the extracts in various-sized droplets (Fig. 2A). Due to the high mass density of the extracts, the droplets were sedimented on the coverslip and deformed in the z axis. [The ratio of the droplet height to its diameter was 0.465 ± 0.050 (mean ± SD), almost independently of the droplet size (n = 25 droplets).] After the encapsulation, the droplets were warmed to initiate microtubule polymerization. We found that, in the presence of dynein inhibitor, rotational cytoplasmic flow emerged inside the droplets, visualized by passivated tracer particles (Fig. 2B, Fig. S3A, and Movie S2) and the organelles (Fig. 2C and Movie S3). The flow resembles microtubule-based rotational cytoplasmic streaming observed in embryos, oocytes, and eggs of several species (7–9). Probably due to the droplet deformation in the z axis, rotational flow along the z axis was rarely observed (only a few percentage of all droplets; Movie S4), suggesting that rotational dynamics could be discussed in the context of a 2D plane. We did not include the results of the droplets showing the rotational flow along the z axis. Droplets containing the extracts without Taxol or with nocodazole (a microtubule-depolymerizing agent) did not show rotational flow (Fig. S3 B and C), indicating that a sufficient density of microtubules was required to induce flow. Compared with the vortex flows observed in bulk extracts, which persisted for 1 min, the rotational flow in the droplets was sustained for over 10- to 100-fold longer. The angular velocity of the flow ${\langle \omega \rangle }_x$ gradually increased after the increase in temperature and then remained constant for several tens of minutes (Fig. 2D). Collectively, the spatial confinement induces temporally stable microtubule-driving rotational flow.

Fig. 2.

Encapsulation of the extracts induces rotational cytoplasmic flow. (A) Schematic illustration and confocal images of the extracts encapsulated in droplets. (B and C) Time projection of PEG-coated 0.25-μm beads (B) and organelles (C) in droplets. Arrows represent the rotational direction. 0 min indicates the timing of elevating temperature. For the experiment in C, [OG-Taxol] = 200 nM. (D) Time course of the angular velocity of the beads ${\langle \omega \rangle }_x$ in three droplets. The lines and their surrounding areas represent means ± SDs, respectively. (E) Relationship between the droplet radius R and ${\langle \omega \rangle }_{x,t}$ without distinction of [p150-CC1] = 400, 800, or 2,000 nM. ${\langle \omega \rangle }_{x,t}$ indicates the absolute value of the average angular velocity of beads in the overall region of a droplet while rotational flow continues at a constant velocity. We confirmed that ${\langle \omega \rangle }_{x,t}$ and the frequency of rotational flow did not differ between the three concentrations of p150-CC1. (F) Flow field for a droplet showing rotational flow measured by PIV. Not all vectors are shown for clarity. (G) Flow velocity profiles of various sized droplets calculated from PIV measurements (d, distance from the droplet center). (H) The relationship between R and the flow velocity near the boundary, defined as the flow velocity averaged over the region of $d/R>0.8$. In all microscopic images, dashed lines indicate the droplet boundaries. (Scale bars, 30 μm.)

Fig. S3.

Dynein inhibition and sufficient microtubule density are required for rotational cytoplasmic flow. (A–C) Images of PEG-coated beads in a droplet (Left), time projection images of the motion of the beads (Middle), and time courses of the angular velocity ${\langle \omega \rangle }_x$ of the beads in the extracts (Right) under various conditions. The lines and their surrounding areas represent means ± SDs, respectively. Note that all droplets observed in these experiments ranged from 100 to 400 μm in diameter, corresponding to the diameter of the droplet showing rotational flow in the presence of p150-CC1 and Taxol. (A) In the presence of 0.5 mg⋅mL⁻¹ anti-dynein antibody 70.1 (instead of p150-CC1). This antibody blocked the formation of a dynein/dynactin complex, similar to p150-CC1. Bead diameter: 0.25 μm. Similar to p150-CC1–treated extracts, a droplet containing the antibody-treated extracts showed rotational flow. (B) Extracts containing no Taxol. All droplets containing these extracts did not show unidirectional flow (n = 12 droplets), suggesting that a sufficient density of microtubules is required to induce rotational cytoplasmic flow. The extracts contained 1,000 nM p150-CC1. We confirmed that extracts containing 800 or 2,000 nM p150-CC1 also showed no cytoplasmic flow. Bead diameter: 2.0 μm. (C) In the presence of 50 μM nocodazole. All droplets containing the nocodazole-treated extracts did not show rotational flow (n = 7 droplets), demonstrating that microtubules are essential for inducing flow. The extracts contained 800 nM p150-CC1. Bead diameter: 0.25 μm. For all microscopic images, dashed lines indicate the droplet boundaries. (Scale bars, 30 μm.) 0 min, timing of the encapsulation into droplets. For all experiments, [OG-Taxol] = 100 nM.

Rotational cytoplasmic flow was observed in droplets with a diameter of 100‒700 μm (Fig. 2E), that is, the length scale of the rotational flow in droplets was 10-fold larger than the vortex flows in bulk extracts. The mean angular velocities ${\langle \omega \rangle }_{x,t}$ in single droplets exhibited a linear correlation with R⁻¹ (R, radius of the droplet; Fig. 2E), implying that the flow velocity near the boundary was nearly constant among various-sized droplets. In fact, PIV analysis showed that, although there were regional variations in the flow velocities in droplets (Fig. 2 F and G), the flow velocity near the boundary was almost independent of the droplet size (Fig. 2H). In droplets smaller than 100 μm in diameter, the microtubule bundles formed a cortex-like structure (Fig. S4A), similar to the structures organized by purified microtubules (15, 26) and actin filaments (27). The intersections of the fitting lines and the horizontal line in Fig. 2E predicted that the maximum diameter of a droplet showing rotational flow was ∼800 μm, which is nearly equal to the maximum diameter of that of droplets showing unidirectional flow. In droplets of over 700 μm in diameter, multidirectional flow was observed (Fig. S4B and Movie S5). Thus, the confinement induced rotational cytoplasmic flow not only at the larger timescale but also at the larger length scale than the vortex flows emerging in bulk extracts.

Fig. S4.

A cortex-like structure in a small droplet and multidirectional cytoplasmic flow in a huge droplet. (A) Confocal images of microtubules in a small droplet of 60 μm in diameter. The images show the bottom (Left) and the equatorial plane (Right: 30 μm away from the bottom coverslip) of the droplet. In droplets smaller than 100 μm in diameter, microtubule bundles localized beneath the droplet boundary, forming a cortex-like structure, as observed in purified microtubules cross-linked by artificially clustered kinesin motors (15, 26) and in purified actin filaments in droplets with a diameter smaller than 13 μm (27). This size difference could be caused by differences in the bending stiffness between microtubules and actin filaments. (Scale bar, 15 μm.) (B) A time projection image of 0.25-μm PEG-coated beads in a droplet of over 700 μm in diameter (Movie S5). Multidirectional flow was observed (three arrows in the image show the flow directions). (Scale bar, 60 μm.) 0 min, timing of the encapsulation into droplets. For all experiments, [OG-Taxol] = 100 nM; [p150-CC1] = 800 nM.

Microtubule Bundles Are Arranged in a Rotating Vortex.

To clarify the mechanism for rotational flow, we observed microtubule networks by confocal microscopy. When the extracts encapsulated in droplets were warmed to initiate microtubule polymerization, microtubule bundles were assembled in random orientations, forming a random bundle network (Fig. 3A, 11 min). The neighboring bundles located near the droplet boundary were then oriented in the same direction, and the entire network began to rotate (Fig. 3A, 15 min). Eventually, most of the bundles oriented in the rotational direction, resulting in the formation of a vortex structure (Fig. 3A, 27 min; Movie S6). This structure resembles the spiral-arrayed microtubules observed in embryos, oocytes, and eggs of several species (7–9, 28). However, unlike the microtubule array observed in animal cells, most of the bundles in droplets seemed not to be anchored to the lipid membrane (29) and the bundles did not exhibit wave-like motion (9, 29). Meanwhile, purified microtubules and artificial kinesin clusters also formed a vortex structure in confined space, but this structure did not rotate (30). The difference suggests the role of nonmotor cross-linkers [e.g., microtubule-associated proteins (MAPs)] on the vortex rotation, because nonmotor cross-linkers bridge between microtubules and between microtubule bundles, modifying the mechanical properties of the microtubule bundle network. Although time-lapse observations (Fig. 2C and Movie S3) were not sufficient to conclude whether the organelles were transported along microtubules by kinesins, the organelles could enhance the fluid flow even when just bound on the rotating bundles.

Fig. 3.

Spatial confinement arranges microtubules in the vortex array. (A, Left) Time-lapse series of confocal images of microtubule bundles showing the vortex formation process. Every image is the maximum projection image along the z axis (range, the equatorial plane of the droplet ± 20 μm). (Right) Each histogram shows angles of microtubule bundles (θ) measured in the left image. θ is defined as the angle of the bundles to the line connecting the droplet center with the bundle midpoint, as shown in the image at 11 min (in this image, θ = −63°). (Scale bar, 30 μm.) (B) Time-lapse images of an elongating bundle near the boundary shown in the yellow box in A. Arrows, ends of bundles. (Scale bar, 20 μm.) In all microscopic images, dashed lines indicate droplet boundaries, and 0 min indicates the timing of elevating the temperature. [OG-Taxol] = 500 nM.

The microtubule bundles in the rotating vortex array elongated and buckled or bent after they reached the droplet boundary (Fig. 3B). During these processes, the end of the bundle contacting the boundary did not slip along the boundary. Although there was no other experimental evidence, this fact implies the existence of friction between the end of a bundle and the boundary. Moreover, as mentioned above, the flow velocity near the boundary was nearly constant among droplets of various sizes (Fig. 2H). These results suggest that the driving force for the rotational flow was produced on the boundary. Taken together, these data led us to speculate that one end of the bundle pushed the boundary, and this pushing force then rotated the entire bundle network to drive the rotational flow.

Microtubule–Microtubule Sliding Propelled by Molecular Motors Drives Rotational Flow.

To test our hypothesis, we examined the dynamics of individual bundles in detail by compressing the droplets with two coverslips (Fig. 4). In compressed droplets, rotational flow did not occur, but the bundle dynamics were clearly observed because the motion was restricted in a 2D plane. The bundles elongated and then buckled (or bent) when they reached the droplet boundary (Fig. 4 A and B, and Movie S7). The elongation rate was 3.58 ± 1.61 μm⋅min⁻¹ (mean ± SD; n = 19 bundles; Fig. 4C), almost equal to the velocity of rotational flow near the boundary (3.19 ± 0.63 μm⋅min⁻¹, mean ± SD; n = 10 droplets; Fig. 2H). Although hydrodynamic interactions in the droplets compressed by two coverslips (height, 5‒10 μm) may be different from those in droplets showing rotational flow (height, >50 μm), there was good agreement between the elongation rate and the flow velocity near the boundary, strongly suggesting that microtubule bundle elongation rotated the bundle network, generating rotational cytoplasmic flow.

Fig. 4.

Microtubule bundles exhibit extensile behavior. (A) Time-lapse series of confocal images of microtubule bundles in a compressed droplet. Every image is the maximum projection image along the z axis. To examine the behaviors of individual microtubule bundles in the steady state, we observed them after 2–4 h from the droplet formation, which corresponded to the time period when most droplets showed rotational flow. Arrows, both ends of the bundle. (Scale bar, 30 μm.) 0 min, timing of elevating the temperature. (B) Typical time courses of the microtubule bundle length in the compressed droplet. Asterisks and double daggers indicate the timing of bundle disassembly and fusion to other bundles, respectively. (C and D) Distributions of the bundle elongation rate (C) and the bundle lifetime (D) (period between the assembly time and disassembly time) in compressed droplets. Solid lines indicate the mean rate and lifetime.

Two possibilities exist for microtubule bundle elongation: (i) microtubule polymerization and (ii) microtubule–microtubule sliding propelled by molecular motors, presumably kinesins (15, 26). First, we investigated whether the microtubule polymerization reaction was essential to drive rotational flow by varying Taxol concentrations from 50 to 500 nM. Because Taxol reduces the rates of both growing and shortening of microtubules (20), we expected that an increase in the Taxol concentration may decrease the angular velocity of the flow. However, the angular velocities were not decreased by increasing Taxol concentrations (Fig. 2E), even though the microtubule-growing rates at 100 and 500 nM Taxol are about 50% slower than that at 50 nM Taxol in vitro (20). Meanwhile, ATP depletion with apyrase (an ATP-hydrolyzing enzyme) or inhibition of ATP hydrolysis of kinesins with AMP-PNP (a nonhydrolyzable analog of ATP) blocked the rotational flow, even though microtubule bundles were assembled under both conditions (Fig. S5). Taken together, these results show that the driving force for rotational cytoplasmic flow is derived from microtubule–microtubule sliding propelled by kinesins, rather than from microtubule polymerization.

Fig. S5.

Rotational cytoplasmic flow requires ATP. (A) A time projection image (Left) and time course of the angular velocity ${\langle \omega \rangle }_x$ of 0.25-μm PEG-coated beads (Right) in a droplet containing 10 U⋅mL⁻¹ apyrase. The lines and their surrounding areas represent the mean ± SD, respectively. (B) A time-lapse series of the maximum projection confocal images of microtubules in a droplet containing apyrase-treated extracts. The network structure did not change for over 2 h in all droplets (n = 10 droplets). (C) A time projection image (Left) and time course of the mean angular velocity of 0.25-μm PEG-coated beads (Right) in a droplet containing 3 mM AMP-PNP. The lines and their surrounding areas represent means ± SDs, respectively. These extracts did not exhibit rotational flow, similar to those of the apyrase-treated extracts in all droplets (n = 12 droplets). (D) A time-lapse series of the maximum confocal projection images of microtubules in the droplet containing AMP-PNP–treated extracts. Microtubules were organized into a highly stable structure in all droplets (n = 9 droplets). The lengths of individual microtubule bundles did not change for 2 h (see the representative examples shown in the images; each end of the microtubule bundles is indicated with an arrow), supporting our model that the elongation of microtubule bundles is mainly due to microtubule–microtubule sliding driven by molecular motors. For all microscopic images, dashed lines indicate the droplet boundaries, and the extracts contained 500 nM OG-Taxol to clearly observe microtubules. [p150-CC1] = 800 nM. (Scale bars, 30 μm.) 0 min, timing of the encapsulation into droplets.

Model for the Organization of a Vortex Structure Inducing Rotational Cytoplasmic Flow.

Based on these results, we propose a model for rotational cytoplasmic flow in a confined space under suppression of dynein activity (Fig. 5A). As the microtubule polymerization proceeds, microtubules are bundled in parallel or antiparallel arrangements by kinesins and MAPs (phase 1). These bundles are randomly oriented and connected by cross-linking proteins (kinesins and MAPs), forming a random microtubule bundle network (phase 2). Each microtubule bundle is elongated by kinesins. Once the extensile bundles reach the droplet boundary, they start to generate the moment of force against the microtubule bundle network (phase 3). Because both the magnitude and direction of the moment of force generated by each microtubule bundle are varied, and because the bundle density is low (∼0.0002 bundles⋅μm⁻³ at 500 nM Taxol; estimated from Fig. 3A, 27 min), the balance between clockwise and anticlockwise rotational forces can be easily disrupted. This spontaneous symmetry breaking induces rotation of the entire microtubule bundle network, which initiates rotational cytoplasmic flow (phase 4). The flow promotes the arrangement of microtubule bundles into a vortex pattern, and the bundle alignment accelerates the flow. This positive-feedback loop maintains the flow unidirectionally over several tens of minutes (phase 5). In this context, the rotational direction is not determined in one direction, and this is consistent with our results (clockwise, 51%; anticlockwise, 49%; n = 47 droplets). This is in contrast to the observation that contractile actin bundles organize the vortex structure with a certain direction of chirality in a circular cell (31). In addition, the actin vortex did not show rotational flow. However, these results suggest that vortex formation could be one of the general features among extensile and contractile systems.

Fig. 5.

Directional stability and mechanism of rotational cytoplasmic flow. (A) Schematic illustration showing the mechanism of the rotational flow. (B) Time projection of the beads (Left) and ${\langle \omega \rangle }_x$ (Right) with 50 nM Taxol. The lines and their surrounding areas represent means ± SDs, respectively. (Scale bar, 30 μm.) (C) Distributions of stable rotational flow (blue), unstable rotational flow (the flow changing its direction oppositely at least once per 30 min; red), and no rotational flow (${\langle \omega \rangle }_{x,t}$ < 0.1 deg⋅min⁻¹; gray; Movie S9) at various droplet diameters. Each bar includes 10–44 droplets.

The flow direction occasionally changed (Figs. 2D and 5B, and Movie S8). The mechanism of this directional change can be speculated based on the observation that the assembly and disassembly of microtubule bundles continually occurred (Fig. 4B). The bundle density was so low that a few newly assembled bundles reaching the boundary may easily overcome the force generated by preexisting bundles to change the rotational direction. In fact, with increasing Taxol concentration, the frequency of the directional change decreased, that is, the rotational flow became more stable (Fig. 5C). Because Taxol stabilizes microtubules (i.e., the frequency of complete disassembly of microtubules triggered by catastrophe decreases with increasing Taxol concentration; ref. 20), these results are consistent with our model.

Discussion

We have demonstrated that, under suppression of dynein activity, cytoplasmic microtubules were organized in extensile bundles, and these bundles induced directed cytoplasmic flow, the length scale and timescale of which depended on the existence of spatial confinement.

Confinement of extracts in droplets induced submillimeter length scale rotational flow, which was 10-fold larger than the vortex flows emerging in bulk extracts. The length scale difference could be caused by mechanical interactions between the bundle and the droplet boundary. First, the microtubule bundle network in the cytoplasm could be elastic enough to transmit mechanical force over the longer distance than the hydrodynamic interaction because microtubules in the cytoplasm are so short (14.5 ± 6.4 μm, mean ± SD; n = 122 microtubules) that they can act as rigid rods (16) and are connected with nonmotor cross-linkers, such as MAPs, which enhance the elasticity of the overall network. Accordingly, microtubule bundle elongation can generate a pushing force against the overall microtubule bundle network once the elongating bundles reach the droplet boundary. Although hydrodynamic interactions of microtubule bundles may also contribute to the spontaneous alignment of bundles into the vortex pattern because extensile microtubule bundles can be regarded as pusher-type motile units, the alignment cannot be explained solely by the hydrodynamic interactions. To orient the bundle with a length of 21.9 μm (the mean length of the bundles in droplets) by the hydrodynamic interactions from the angle θ = 0° to 60° (θ = 60° is the typical bundle angle in the vortex; Fig. 3A) before the bundle disassembly (bundle lifetime, 20.9 ± 6.74 min; mean ± SD; Fig. 4D), the flow velocity gradient per micrometer should be larger than 0.05 min⁻¹, which was calculated under the assumption that the bundle is advected at the same velocity of fluid flow. Therefore, the flow velocity in the bulk extracts (0.36 ± 0.24 μm⋅min⁻¹) is too slow to arrange the bundles into the vortex pattern because the maximum flow velocity gradient generated by the bulk flow was estimated as only 0.033 min⁻¹ (0.36 μm⋅min⁻¹ × 2/21.9 μm), and it would take more than 31.9 min to orient the bundle from θ = 0° to 60°. Thus, the mechanical interactions of microtubule bundles could be essential for the arrangement of the bundles into the large length scale vortex pattern.

Not only the length scale, but also the timescale of the rotational flow is 10- to 100-fold longer than those of the vortex flows emerging in bulk extracts. This timescale difference could be explained by the positive-feedback loop (Fig. 5A); the fluid flow promotes the alignment of the bundles, and the bundle alignment accelerates the fluid flow. Moreover, once the microtubule bundles are arranged in the vortex array, the vortex structure would be more easily maintained than the random network because of the presence of mechanical interactions between microtubule bundles. The vortex structure is primarily composed of parallel-arrayed bundles, and many cross-linkers can bridge the overlapping regions simultaneously, stabilizing the vortex array. In contrast, in the case of randomly oriented arrays, cross-linkers can bridge only the crossing points of bundles, and thus, the random bundle network is easily remodeled by fluid flow and the mechanical stress generated by bundle elongation. Accordingly, we concluded that not only the length scale, but also the timescale difference is caused by the mechanical interaction.

Both the rotational flow emerged in droplets and the vortex flows emerged in bulk space resemble the flow patterns of bacterial suspensions with and without spatial confinement (4, 5). However, the effects of the spatial confinement are essentially different. Although confinement increases the temporal stability of the bacteria vortex similar to our results, it does not change the length scale of the vortex, that is, the vortex size of bacteria in droplets is comparable to the vortex size in bulk space (4, 5). The difference may be attributed to a lack of specific connections between bacteria, presumed by the bacterial motions during the circulation in a droplet. Whereas the bacteria located near the boundary did not move, the others were smoothly swimming even while contacting with those immotile bacteria (4). Compared with bacteria, the vortex-arrayed microtubule bundles rotated more like an elastic structure. In summary, although further quantitative studies with a combination of numerical simulations should be performed to reveal the respective roles of hydrodynamic and mechanical interactions, our results suggest that microtubule networks use, in addition to hydrodynamic interaction, mechanical interaction derived from microtubule cross-linkers (motor and nonmotor proteins) and the physical boundary to induce cytoplasmic flow over larger length scale and timescale.

Materials and Methods

Xenopus egg extracts were prepared as described previously (19). All chemical reagents, proteins, and polyethylene glycol (PEG)-coated beads (32, 33) were mixed with the extracts on ice before droplet formation. Oregon Green-labeled Taxol (100 nM, unless otherwise stated) and 2 μM cytochalasin D (for all experiments) were added to the extracts. To suppress dynein activity, p150-CC1 was added (800 nM, unless otherwise stated). The lipid–oil mixture was prepared as previously reported (34–36). The egg extracts were added to this mixture, and the droplets were then immediately formed by tapping on ice or at 16 °C. The extract-in-oil droplets were perfused into the chamber assembled with siliconized coverslips, which were spaced with double-faced tape pieces, and the chamber was sealed with Valap. All images were acquired at 20 ± 1 °C and analyzed with LabVIEW and ImageJ.

SI Materials and Methods

Xenopus Egg Extracts and Proteins.

Metaphase Xenopus egg extracts were prepared as described previously (19). The extracts were stored on ice and used within 24 h. We did not use extracts that had been frozen because we confirmed that freeze–thawing of the extracts impaired normal microtubule dynamics. To eliminate the effects of actin dynamics (21, 22), all experiments were performed in the presence of 2 μM cytochalasin D, an actin-depolymerizing agent. For all experiments except ATP depletion experiments, the energy mix (7.5 mM creatine phosphate, 1 mM ATP, and 1 mM MgCl₂) was added to the extracts. The dynein inhibitor p150-CC1 was expressed and purified as described previously (37) and stored in PBS [PBS(–)]. Anti-dynein intermediate chain antibody 70.1 was purchased from Sigma-Aldrich (D5167), concentrated using a centrifugal filter (4304; Millipore), and stored in XB buffer (10 mM HEPES, pH 7.7, 100 mM KCl, 1 mM MgCl₂, 0.1 mM CaCl₂, and 50 mM sucrose).

Preparation of the Lipid–Oil Mixture.

The lipid–oil mixture was prepared as reported in our previous studies (34‒36). First, l-α-phosphatidylcholine from chicken egg yolk (egg PC; 840051P; Avanti Polar Lipids) was dissolved in chloroform (034-02603; Wako; dehydrated with molecular sieves 4A from Nacalai Tesque) to a final concentration of 20 mM. Next, 50 μL of this lipid solution was dispensed into 1.5-mL glass test tubes. The chloroform was completely removed by evacuation overnight inside a vacuum desiccator, forming a dry lipid film on the bottom surface of the tubes. The dry lipid film was mixed with 1 mL of mineral oil (23306-84; Nacalai Tesque), heated to 80 °C, and dissolved by vortexing. Then, the lipid–oil mixture was sonicated for 90 min in a bath sonicator at 60 °C and 60-W power (AS12GTU; As One). Immediately after sonication, the lipid–oil mixture was vortexed and cooled to room temperature. The lipid–oil mixture was stored at room temperature under dark conditions and used within 1 wk. The lipid concentration in oil was 1 mM.

Preparation of PEG-Coated Beads.

PEG-coated beads were prepared as described previously (33). First, 50 μL of carboxylate-modified fluorescent beads (0.25 μm; F-8810, Life Technologies) was dialyzed against 1 L of PBS(−) for over 1 h using a cellulose membrane (69570; Thermo Scientific). The beads were mixed with n-hydroxysulfosuccinimide (Sulfo-NHS; 24510; Thermo Scientific), 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide HCl (EDC; 22980; Thermo Scientific), and methoxypolyethylene glycol amine (5,000-Da PEG; 06679; Sigma-Aldrich) dissolved in 500 μL of PBS(−). The final concentrations of all three reagents were 8–16 mg⋅mL⁻¹. To avoid aggregation, the bead solution was gently mixed using a rotator (MTR103; Matsuura) at room temperature for 2 h. PEG molecules and carboxyl groups on the bead surface formed amide bonds mediated by EDC and Sulfo-NHS. EDC was used immediately after dissolving in PBS(−) because EDC is easily degraded in aqueous solution (33). After incubation, the bead solution was concentrated using a centrifugal filter (4304; Millipore) until the volume decreased to ∼50 μL. Finally, the concentrated solution was dialyzed against PBS(−) using a cellulose membrane overnight at 4 °C to remove unreacted EDC, Sulfo-NHS, and PEG. The bead solution was stored at 4 °C until use.

Microscopy.

Epifluorescence and bright-field images of the droplets were acquired using an upright microscope (Axio Imager; Carl Zeiss) equipped with a 40× objective (0.75 N.A.; Carl Zeiss) and a charge-coupled device (CCD) camera (AxioCam MRm; Carl Zeiss), or a custom-built inverted microscope equipped with a 10, 20, or 40× objective (Uplan FL N 10×/0.30, UplanApo 20×/0.70, or UPlanFI 40×/0.75 Ph2; Olympus), an electron-multiplying charge-coupled device (EM-CCD) camera (iXon3 DU-897E-CS0-#BV; Andor Technology), a stable solid-state lamp as an epifluorescence light source (SOLA light engine; Lumencor), a 150-W halogen lamp as a bright-field light source (LS-LHA; Sumita Optical Glass), a XYZ-motorized sample stage driven by stepping motors (SGSP-13ACT-BO; Sigma-Koki), a motorized filter wheel (FW103H; Thorlabs), and two motorized shutters for epifluorescence and bright-field light sources (SSH-R; Sigma-Koki). The motorized stage, shutters, and camera were controlled by LabVIEW software (National Instruments). The custom-built heat blocks, which connected to the water bath (NESLAB RTE-7; Thermo Scientific), were mounted on the objective lens and the sample holder to maintain the sample temperature at 20 ± 1 °C. For time-lapse observation, images were acquired every 1 min for over 3 h to observe fluid flow.

Confocal fluorescence images of the droplets were acquired using an inverted microscope (IX71; Olympus) equipped with a 40× objective (UPlanFL N 40×/1.30 Oil, PlanApo 40×/0.90W LSW, or UPlanFI 40×/0.75 Ph2; Olympus), a confocal scanner unit (CSU10; Yokogawa), an EM-CCD camera (iXon+ DU-888E-C00-#BV or iXon3 DU-897E-CS0-#BV; Andor Technology), excitation lasers (488 and 568 nm; 643-YB-A01; Melles Griot), a motorized filter wheel, and shutters (MAC5000; Ludl), and a motorized sample stage (KS-O; Chuukousha Seisakujo). The motorized stage, shutters, and camera were controlled by LabVIEW software (National Instruments). All experiments were conducted at 20 ± 1 °C. For time-lapse observation of microtubule bundles in compressed droplets, confocal images were acquired every 2 min for 2 h. In AMP-PNP and apyrase treatment experiments, confocal images were acquired every 10 or 20 min, respectively, for 2 h.

For all experiments, LabVIEW (National Instruments) and ImageJ (https://imagej.nih.gov/ij/) were used to analyze microscope images.

Bulk Experiments and PIV Measurements.

Oregon Green-labeled Taxol (OG-Taxol; P22310; Life Technologies) and p150-CC1 (800 nM) (or anti-dynein intermediate chain antibody (0.5 mg⋅mL⁻¹)) were added to extracts stored on ice. For observation of behaviors of microtubule networks in bulk extracts, we dropped 40 μL of the extracts on a siliconized coverslip (custom order; Matsunami) coated with Pluronic F-127 (P2443; Sigma-Aldrich), placed a 20 × 20-mm² F-127–coated siliconized coverslip on the drop, and gently compressed with the two coverslips by hand. The chamber was immediately sealed with Valap (a mixture of Vaseline, lanolin, and paraffin at equal weight ratios). The height of the chamber was ∼100 μm, as estimated by the volume of the extracts and the size of the coverslip. We consider that this chamber can be regarded as an infinitely large 2D system for microtubules.

To quantify fluid flow in the bulk extracts, PIV measurements using PIV software (PIVLab; cf. ref. 38) were performed for bright-field images. The grid size of PIV measurements was 9.5 μm. To measure the fluid flow at steady state and to avoid the effects of artifactual flow caused by the chamber preparation, the PIV measurements were performed 2 h after chamber preparation. Fluorescence images of microtubules in bulk extracts, as shown in Fig. 1 A and B, were acquired 30 min after preparation. We confirmed that artifactual flow did not occur during image acquisition.

We observed the bundle behaviors by confocal microscopy. The bundles were scanned along the z axis. Because the objective lens was continually moved at a constant speed of 2 μm per frame, single-slice images were projection images with a height of 2 μm. We analyzed only bundles for which the entire structure was captured in single-slice images, with both ends clearly distinguished. To minimize artifactual flow caused by the chamber preparation, a cylindrical hall (height, 25 μm; diameter, 10 mm) was constructed on a coverslip by using photoresist (SU-8 3025; Nippon Kayaku), and then the extracts were placed in the hall, covered with a siliconized coverslip, and sealed with Valap. Both coverslips had been coated with Pluronic F-127 before use.

Measurement of the Length of Single Microtubules.

To measure the length of single microtubules, microtubule bundles were dissociated into single microtubules, as described previously (39). To organize microtubule bundles, the extracts with OG-Taxol (100 nM) and p150-CC1 (800 nM) had been incubated for 30 min at 20 °C before being subjected to dissociation. Single microtubules were observed by confocal microscopy. As noted previously, a few bundles remained, even after the dissociation process (39). We avoided measuring the filaments for which the fluorescence intensity was obviously higher than that of other filaments.

Formation and Observation of Extract-in-Oil Droplets.

OG-Taxol, PEG-coated beads [1% (vol/vol)], and p150-CC1 [or anti-dynein intermediate chain antibody (0.5 mg mL⁻¹)] were added to extracts stored on ice. The 0.25-μm beads were added to characterize the rotational motion of the cytoplasm. We were unable to determine whether the size of the beads was small enough not to be captured by the microtubule network (i.e., the rotational motion of the beads may not reflect fluid flow, but instead may reflect the rotational motion of the microtubule network if small beads were captured by the network). However, the bead motion clearly characterized the rotational behavior induced in the droplets, implying that our analyses using the beads would satisfy our requirements. In addition, the lifetime of single microtubule bundles was 20.9 ± 6.74 min (mean ± SD; Fig. 4D), suggesting that the beads were not continuously captured during the observation (which lasted several hours). We usually used 800 nM p150-CC1; however, Fig. 2E includes the results obtained in the presence of 400 and 2,000 nM p150-CC1 as well. The differences between these concentrations did not affect the frequency or angular velocity of rotational flow. For observation of the bead motion under suppression of dynein activity, the lipid–oil mixture was preincubated at 16 °C before droplet formation. Meanwhile, for observation of the vortex formation from the start of microtubule polymerization under suppression of dynein activity, the chamber containing the droplets was incubated on ice for 10 min before observation. The frequency and angular velocity of the rotational flow showed no significant differences between the two sample preparation procedures.

The extract-in-oil droplets were perfused into the chamber assembled by two siliconized coverslips spaced with two double-faced tape pieces (thickness of the tape, ∼300 μm). For observation of large droplets, to avoid attachment of the droplets on the top coverslip, we used a chamber with a thickness of ∼600 μm, which was prepared by overlapping two pieces of the double-faced tape. The thickness of 600 μm was sufficiently thick because the ratio of the droplet height h to its diameter D was 0.465 ± 0.050 (mean ± SD; n = 25 droplets; e.g., Fig. 2A) and because the diameters of all droplets we observed were less than 1,000 μm. After perfusion, the chamber was immediately sealed with Valap.

To observe the 3D shapes of the droplets, 1 μM tetramethylrhodamine (TMR)-labeled dextran (10 kDa; D-1816; Life Technologies) was added to the extracts. Organelles were stained with 100 nM octadecyl rhodamine B (R18; O-246; Molecular Probes), a membrane-binding fluorescent compound.

To hydrolyze endogenous ATP in the extracts, apyrase (A6132; Sigma-Aldrich) was added to the extracts at a final concentration of 10 U⋅mL⁻¹, and the extracts were incubated at 16 °C for over 30 min. The extracts were then incubated on ice for over 10 min to depolymerize microtubules and subsequently encapsulated in droplets. Alternatively, the extracts were incubated at 16 °C for over 15 min to promote microtubule polymerization. Then, immediately after the addition of apyrase, the extracts were encapsulated in droplets. In the former protocol, no rotational flow was observed in all droplets (n = 23 droplets). Meanwhile, in the latter protocol, rotational flow was observed in some droplets (20%; n = 10 droplets), but stopped within 30 min from the encapsulation. This was consistent with previous work reporting that it took 20–30 min to deplete endogenous ATP from Xenopus egg extracts using 10 U⋅mL⁻¹ apyrase (40). The droplet in Fig. S5B was prepared by the latter protocol, showing no flow from the beginning.

To inhibit the activities of molecular motors, we added AMP-PNP (A2647; Sigma-Aldrich) to the extracts at a final concentration of 3 mM. Note that the bundles assembled in AMP-PNP–treated extracts were thicker than those assembled in apyrase-treated extracts. This could reflect that the K_a (association constant) of kinesins for binding with microtubules in an AMP-PNP state is the same or higher (depending on the kinesin family members) than that of kinesins for binding with microtubules in a nucleotide-free state (41, 42).

Analysis of the Rotational Flow and Bundle Behaviors in Extract-in-Oil Droplets.

In all experiments, the center of the droplet was determined as the center of the circle that was fitted best to the droplet boundary in the bright-field image. When the beads (the tracer particles of flow) that were positioned in the region of d/R > 0.5 (d, distance from the droplet center; R, radius of the droplet) moved in the same cross-sectional image for over 30 min, we judged that the cytoplasm flowed parallel to the focal plane. We analyzed only those droplets. Based on this definition, the maximum tilting angle of the rotational flow among all analyzed droplets was ∼11°. This is because the droplet was scanned along the z axis every 5 μm, and the minimum distance between the droplet center and the beads that showed rotational motion in the outer region was ∼25 μm (the diameter of the smallest droplet showing rotational flow was ∼100 μm), that is, 11° was the result of 180 × π⁻¹ × tan⁻¹(5/25). The positions of the beads were tracked using an ImageJ plugin (Particle Track and Analysis). The angular velocity of the beads around the droplet center $\omega$ was calculated by the following equation:

$$\omega \left(t\right)=\left\{{\sin }^{-1}\left(\frac{\overrightarrow{X\left(t\right)}\times \overrightarrow{X\left(t+\mathrm{\Delta }t\right)}}{\sqrt{{\left|\overrightarrow{X\left(t\right)}\right|}^2{\left|\overrightarrow{X\left(t+\mathrm{\Delta }t\right)}\right|}^2}}\right)\right\}/\mathrm{\Delta }t,$$

where $\overrightarrow{X\left(t\right)}$ is the planar vector from the center of the droplet to the bead position at time t on the X–Y plane. The positive and negative values of $\omega \left(t\right)$ indicate anticlockwise and clockwise motion of the bead, respectively. In Fig. 2E, ${\langle \omega \rangle }_{x,t}$ is the absolute value of the angular velocity of all beads averaged over time and the region inside the droplet during the period when the flow continues at a constant velocity. Whether or not the flow continues at a constant velocity was judged from the time course of the angular velocity (e.g., Fig. 2D).

Flow velocity profile during rotational flow in droplets (Fig. 2G) was measured by PIV of bright-field images (PIVLab; cf. ref. 38; grid size, 11 μm). The plots in Fig. 2G indicate the flow velocity averaged over the regions where d ranges in 0‒15, 15‒30, 30‒45 μm, and so on, then the mean values were plotted against d/R. The flow velocity near the boundary (Fig. 2H) was defined as the flow velocity averaged over the region of d/R > 0.8.

We observed vortex formation by confocal microscopy. To observe the behaviors of individual bundles, the droplets were scanned along the z axis (range, the equatorial plane of the droplet ± 20 μm). Because the objective lens was continually moved at the constant speed of 2 μm per frame, single-slice images were projection images with a height of 2 μm. We analyzed only the bundles for which the entire structure was captured in a single-slice image. The angle of individual microtubule bundles θ was defined as the angle (in degrees) between a microtubule bundle and the line connecting the droplet center with the bundle midpoint, as shown in Fig. 3A (11 min).

To compress the droplets, we dropped 2–4 μL of the extract-in-oil droplets, which had been prepared on ice, on a siliconized coverslip, gently placed a 20 × 20-mm² siliconized coverslip on this mixture, and compressed with the two coverslips by hand. This chamber was sealed with Valap. The height between the two coverslips was 5–10 μm. To measure the elongation rates of individual bundles, the bundles were scanned along the z axis using a confocal microscope. Because the objective lens was continually moved at a constant speed of 2 μm per frame, single-slice images were projection images with a height of 2 μm. We analyzed only the bundles for which the entire structure was captured in a single-slice image. To investigate the steady state of bundle behaviors under suppression of dynein activity in the compressed droplets, the bundles were analyzed after a few hours from the encapsulation, which correspond to the time period when most droplets showed cytoplasmic flow. We stopped tracking microtubule bundles when they fused to the others, and the lifetimes of these fused-bundles were not included in Fig. 4D.