Compartmentalized cytoplasmic tradewinds direct soluble proteins

Abstract

Proteins essential for signaling, morphogenesis, and migration traverse the complex intracellular landscape via vesicular trafficking, microtubule-based transport, and diffusion. However, the precise mechanisms guiding soluble proteins toward their functional destinations have remained elusive. Here, we demonstrate that soluble proteins are directed toward the cell’s advancing edge through advection—diffusion enhanced by intracellular fluid flow. We reveal that advective transport occurs within a specialized compartment at the cell’s leading edge, separated from the rest of the cytoplasm by an actin-myosin condensate barrier. The barrier limits protein mixing between the compartment and the rest of the cytoplasm, maintaining localized protein concentrations. Contraction at the barrier generates a molecularly non-specific fluid flow that drives the forward movement of treadmilling actin monomers, actin-binding proteins, adhesion molecules, and even inert proteins. Dynamic changes in the local curvature of the barrier steer the fluid flow to direct proteins toward protrusive regions of the leading edge. This advective mechanism synchronizes protein distribution with local changes in cell morphology. Outside this compartment, diffusion dominates as the principal mode of soluble protein transport. Our findings uncover previously unrecognized compartmentalization strategies that regulate soluble protein concentrations and coordinate their efficient distribution for homeostasis, protrusion, and adhesion.

Cells use fluid flow to deliver proteins to their leading edge. An actin barrier creates a compartment where contraction drives flow, steering proteins toward growing regions. This mechanism coordinates protein distribution with cell shape changes.

Introduction

Cells depend on the precise localization of proteins to coordinate essential processes such as signaling, morphogenesis, and migration^1,2. While vesicular trafficking and motor-driven transport along cytoskeletal filaments provide directionality to membrane-bound cargo^3–6, soluble proteins move through the cytoplasm primarily by diffusion⁷. However, because diffusion can disperse molecules in all directions, it is poorly suited for targeted delivery. This raises the question: How do cells guide soluble proteins toward specific regions where they are needed?

Among soluble proteins, actin offers a compelling system to study these mechanisms due to its abundance and essential role in cellular dynamics. Actin filaments treadmill by polymerizing at their barbed ends and depolymerizing at their pointed ends, requiring the continuous recycling of actin monomers to regions of active filament growth. Because de novo protein synthesis is too slow to meet the demands of dynamic cellular processes, cells must reuse existing proteins. Although the molecular steps of actin treadmilling are so well understood that they can be reconstituted in vitro^8,9, how cells organize and direct the transport of soluble actin monomers within the cytoplasm remains unknown.

This knowledge gap is evident in cartoon models of treadmilling, which simplify the process of actin monomers returning toward the advancing cell edge with a simple arrow¹⁰. So little is understood because observing soluble monomer movement in living cells is very challenging. Despite being the most abundant cytoplasmic protein, the actin monomer is one of the most difficult to visualize; it is hard to distinguish the movement of individual monomers from the dense, continuously remodeling network. Early simulations and more recent models suggest that actin monomers might move faster than expected by diffusion alone^11,12, but experimental results have been varied. Some studies were unable to detect biased monomer movement and, instead of fast monomer movement, reported unexpectedly slow movement relative to cell migration rates¹³. These discrepancies highlight the need for insights into how soluble proteins, including actin monomers, are transported to regions where they are required for cellular functions.

Here, we overcame the technical challenges hampering progress to visualize and quantify soluble protein transport. Using a combination of photobleaching, localized photoactivation, single-molecule tracking (SPT), and fluorescence correlation spectroscopy (FCS), we uncovered key mechanisms that govern protein distribution at the leading edge. We found a cytoplasmic compartment at the cell’s front that separates proteins from the rest of the cytoplasm and maintains localized concentrations. Within this compartment, soluble proteins are directed toward the leading edge faster than diffusion by a fluid flow. Dynamic changes in the curvature of the barrier between the compartment and the rest of the cytoplasm precisely target proteins traveling by flow to sites of active protrusion and adhesion assembly. These discoveries reveal a pseudo-organelle, separated from the cell body by an actin-myosin condensate, that actively regulates soluble protein distribution, offering insights into intracellular organization and the dynamic control of cell morphology and motility.

Results

Quantification of anterograde transport

To measure the forward velocity of actin monomers, we used bleach-labeling to mark a subset of fluorescent actin¹¹. We chose NG108 cells, which are non-ruffling and possess a morphologically uniform lamella, to avoid shape changes, and to target a single rather than multiple morphological regions of the cell^11,12. Bleaching the rear of the lamella of these cells ensures that the labeled monomers will travel the maximum possible distance to reach the leading edge, where polymerization is most likely to occur.

We found that a distinct thin black line appeared at the cell front within seconds of bleaching (Fig. 1a and Supplementary Movie 1). The line marked the incorporation of nonfluorescent monomers newly transported to the cell front from the bleach region at the rear of the lamella. The striking thinness of the line is likely due to the very high concentration of barbed filament ends available for polymerization, acting as a strong sink^14,15.

Fig. 1. Transport of depolymerized actin to the polymerizing front is facilitated by myosin contraction.

a NG108 cells expressing egfp-actin were bleached between 0.3 and 1.3 s. By 3.9 s, the bleached monomer has been transported (recycled) to the front of the cell, repolymerized at the leading edge, and traveled rearward (thin dark line indicated by arrow). b Forward actin transport still occurs when blebbistatin is active, and myosin II ATPase is inhibited in cells expressing Apple-actin. However, the dark line appears later (t = 17.8 s) and is more diffuse than in (a). Repeating the experiment in the same cell but using 488 nm light between t = 0 and 2.5 s to inactivate blebbistatin rescues the rapid actin forward transport and the sharpness of the line. c Lines indicated in (b) were used to create kymographs, line intensity as a function of time plots. Kymographs are aligned to the starting time of bleach, and the time until actin appears as a narrow band at the cell front (arrowhead) is longer when blebbistatin is active. d The apparent velocity (measured as indicated in Supplementary Fig. 1a) in blebbistatin-treated and control cells, as well as a subset of matched treated and photonic washout experiments, indicates a significant difference (n = 37 control, 16 blebbistatin (20 μM), 9 photonic washouts). p values are two-tailed t-test comparisons. Box plots show median (center line), interquartile range (25th–75th percentiles; box), whiskers to the most extreme non-outlier values, and notches as 95% CIs. Scale bars, 5 μm. n refer to the number of cells analyzed, all panels represent at least three independently performed experiments. Source data are provided as a Source Data file.

The original bleach and new black line are the two endpoints of the forward-directed arrow depicted in cartoon models¹⁰. Measuring the distance between the endpoints and dividing by the elapsed time yielded an apparent anterograde velocity of 3.6 ± 1.1 μms⁻¹ (Supplementary Fig. 1a, b), nearly fifty times faster than the rearward movement of the original bleach line (0.08 ± 0.02 μms⁻¹)¹⁶ (Supplementary Fig. 1c).

Anterograde transport is dependent on myosin II

There have been conflicting studies on whether myosin II facilitates this rapid forward movement, with some suggesting it plays a critical role^13,17, while others argue it is not involved at all¹⁸. To address this discrepancy, we combined bleach-labeling measurements with a myosin II inhibition and photonic washout assay, enabling us to measure the speed of soluble protein with and without myosin II activity in the same cell.

NG108 cells expressing Apple-actin were treated with the wavelength-dependent version of the myosin II ATPase inhibitor, blebbistatin¹⁹. Following bleaching, a dark line still appeared at the leading edge, but the line appeared later and was more diffuse than in control cells (Fig. 1b, blebbistatin active). The delayed appearance of the line at the edge indicated that blebbistatin treatment reduced the apparent forward speed (Fig. 1c). When blebbistatin was photo-inactivated with 488 nm light within the same cell, the sharpness of the line was restored, and the speed of anterograde movement returned to control levels (Fig. 1b–d, blebbistatin inactive/washout).

Control experiments demonstrate that neither blebbistatin nor the Rho-kinase inhibitor Y-27632 changed the retrograde speed of the actin network. However, both treatments slowed forward transport by 2–4-fold (Supplementary Fig. 1b, c)²⁰. These data suggest that active myosin contraction is required to facilitate the rapid forward movement of soluble actin protein.

Advection: molecularly non-specific anterograde transport mechanism

The exact mechanism of myosin II facilitated transport is unclear. Modeling efforts that proposed a role for myosin II in accelerating transport suggested that myosin II creates a forward fluid flow^11,13, but direct experimental evidence has been lacking. To address this, we devised a cellular analog of the classical engineering experiment to measure flow—the dissipation of material from a continuous emission point source²¹.

We expressed photoactivatable green fluorescent protein (PaGFP)-tagged actin, which remains dark until activated by UV light²². Using CAD cells, which have a uniform circular geometry, we used a focused UV beam to continuously locally activate actin while simultaneously imaging its dispersion. This approach is the inverse of a fluorescence loss in photobleaching (FLIP) experiment²³, and we have named it ‘FLOP’ for Fluorescence Leaving the Original Point. Using FLOP, we found that activating within the lamellipodial region produced an asymmetric plume of fluorescence extending toward the leading edge. This pattern aligned with analytical models of dispersion by diffusion coupled with fluid flow (Fig. 2a, b and Supplementary Fig. 2)²¹. The plume advanced until the fluorescently labeled monomers reached the cell front, where they were incorporated into the actin network (Fig. 2a, d and Supplementary Movie 2).

Fig. 2. Myosin mediates the advective flow of actin.

a Photoactivation of a diffraction-limited spot in a CAD cell expressing PaGFPactin shows asymmetric dispersion toward the front of the cell. b Graphic illustrating how the asymmetry is analyzed from fluorescence intensity perpendicular and parallel to the cell leading edge of control and blebbistatin cells. A transport metric is defined as the ratio of the area under each half of the intensity curve, indicated by different shading levels on either side of the center line. c Photoactivation of a diffraction-limited spot in a blebbistatin-treated CAD cell expressing PaGFPactin shows symmetric movement in all directions away from the activation spot. d Stacked line-intensity plots of data summed for t = 1.5, 3, 7.5 s after the initiation of spot activation indicate that the asymmetry in the control cell in (a) only occurs perpendicular to the direction of the cell's leading edge. e Box and whisker plots of the transport ratio, including all individual data points, indicate fluorescence dispersion toward the leading edge (perpendicular) in control cells. (n = 28 control, 32 blebbistatin (0.2 μM)). f Stacked line-intensity data plots summed for t = 1.5, 3, 7.5 s after the initiation of spot activation indicate the symmetric fluorescence distribution in the blebbistatin-treated cell in (c). g Transport ratio analysis as described in (b) reveals asymmetry for the spread of actin mutants that cannot be incorporated into the network (n = 13, G13R, 14, R62D). h Asymmetric spread was also observed for free dye, mEos3.2 (n = 13), parallel to the edge and perpendicular to the edge. i Sum of 3891 single-molecule frames acquired at 8 ms/frame from spot activation of mEos-R62D, yielding 1451 tracks > 5 steps. p values are two-tailed t-tests in (e, h), and one-way ANOVA with post-hoc Tukey HSD test in (g). Box plots show median (center line), interquartile range (25th–75th percentiles; box), whiskers to the most extreme non-outlier values, and notches as 95% CIs. Scale bars, 5 μm. All panels represent at least three independently performed experiments, n= number of cells analyzed. Source data are provided as a Source Data file.

To quantify this asymmetry, we measured the fluorescent intensity on either side of the center of the activation point along a line perpendicular to the cell edge (Fig. 2b, cyan and magenta) and calculated the ratio of intensity toward and away from the edge. Ratios greater than one indicated a bias toward the leading edge. The bias only occurred in this direction; fluorescence measured along a line parallel to the cell edge showed ratios near one, consistent with symmetric dispersion, which would be expected if diffusion were the sole transport mechanism (Fig. 2b, orange and green). These tracer experiments indicate flow toward the leading edge.

We then inhibited myosin ATPase activity with either the wavelength-independent version of blebbistatin, nitro-blebbistatin²⁴, or the Rho-associated protein kinase (ROCK) inhibitor, Y27632 (Supplementary Movie 3). Myosin inhibition consistently produced symmetric fluorescence profiles both parallel and perpendicular to the cell edge (Fig. 2e and Supplementary Fig. 3a, b), without affecting the rate of parallel fluorescence spread (Supplementary Fig. 3f), demonstrating that myosin II contraction is required for fluid-driven transport. This flow-facilitated diffusion is analogous to convective heat transfer and is referred to as advection in mass transport.

Since monomer incorporation into the network could contribute to the observed asymmetry, we analyzed the dispersion patterns of two actin mutants, G13R and R62D, which cannot incorporate into the network²⁵. Both mutants displayed similar distributions to wild-type actin. Although neither mutant was incorporated into the network, they retained interactions with different actin-binding proteins: G13R binds profilin, which regulates network assembly at the cell edge, while R62D binds cofilin, a disassembly factor not localized to the leading edge²⁵. These findings suggest that neither direct incorporation into the network nor interaction with profilin explains the observed transport asymmetry (Fig. 2g and Supplementary Movie 3)²⁶.

This independence from network binding prompted us to explore whether this transport mechanism extends beyond actin recycling to function as a general, molecularly non-specific system for directing proteins toward the cell front. Since our initial attempts to track the distribution of untagged PaGFP were hindered by its inadequate ON/OFF contrast ratio²⁷, we switched to mEos3.2, a high-contrast photoactivatable probe used in single-molecule microscopy²⁸. Using mEos3.2 and the mutant mEos3.2-R62D, we observed forward-biased transport toward the leading edge (Fig. 2h, i). Additionally, mEos3.2-R62D accumulated at the leading edge, consistent with results predicted by fluid-flow-enhanced transport modeling (Fig. 2i¹³).

Molecular characterization of transport toward the cell front

To further investigate the transport of molecules toward the front of the cell, we performed FLOP experiments in cells expressing mEos3.2-Lifeact. We then used single-particle tracking (SPT) to follow individual molecules and identified the compass angle at which the trajectories crossed a bounding circle surrounding the excitation beam (Fig. 3a–d). Comparing tracks from the lamella and cell body, we found molecules activated in the lamella demonstrate a directional bias toward the cell front, consistent with our tracer experiments, contrasting the symmetric distribution within the cell body (Fig. 3c, d). Similarly, Momentum Scaling Spectrum (MSS) analysis of spot-activated mEos3.2-Lifeact, as well as whole-field illumination JF549 tagged Halo-G13R exhibit a slightly larger population of molecules with S_MSS > 0.5 (indicative of super diffusion or advection) in the lamella compared to the cell body (Supplementary Fig. 4a, b)^29,30. These results provide direct experimental evidence of actin transport directed toward the cell front in the lamella.

Fig. 3. Molecular transport of actin in the lamella and cell body.

a, b Spot activation of mEos3.2 Lifeact in either the lamella or the cell body in CAD cells. Randomly colored trajectories 10 frames or longer were detected during a 50-frame window. Circles indicate a 2 μm boundary around the center of the activation spot. c, d Rose plots illustrate the angle at which trajectories moving away from the activation spot cross the bounding circle in the lamella or the cell body. Rayleigh Tests indicate that (c) is not uniformly distributed (p = 0.0002, while (d) may be uniformly distributed (p = 0.36). Both the Mardia–Watson–Wheeler and Kuiper two-sample tests detect a significant difference between (c) and (d) p = 0.001. e Representative FCS autocorrelation curves (black lines) of Halo Actin labeled with JF549 fit to a model of two diffusion coefficients and flow. The model fits of the data (black trace) in the lamella (D = 11.1 μm²s⁻¹, v = 1.8 μms⁻¹) and cell body (D = 9.64 μm²s⁻¹, v = 1.1 μms⁻¹), and the residual panel indicates the accuracy of fit. f, g Primary diffusion is not significantly different in the cell lamella or body. The flow velocity is much larger in the lamella and statistically different from the cell body (n = 40 cell body measurements in 24 cells, and 20 lamella measurements in 20 cells), over five independent replicates. p value values two-tailed t-tests. Box plots show median (center line), interquartile range (25th–75th percentiles; box), whiskers to the most extreme non-outlier values, and notches as 95% CIs. Source data are provided as a Source Data file.

To further quantitatively characterize the transport in the lamella and the cell body, we employed FCS. FCS tracks fluctuations in fluorescence intensity within a focused beam over time and applies an autocorrelation function to extract molecular movement parameters³¹. We modeled the transport dynamics by incorporating free monomer diffusion, fluid flow-assisted transport, and interactions between monomers and other proteins. This model enabled us to disentangle the contributions of diffusion and flow to soluble molecular transport (Fig. 3e, see “Methods” section, Eq. 2)^32,33. Residual analysis confirmed the quality of the fit of the model to the data.

We found calculated diffusion rates were similar in both compartments and aligned with previously reported values (Fig. 3f)^11,34, but the velocity in the lamella was significantly higher and statistically distinct (Fig. 3g). Repeating these measurements using the non-polymerizable actin mutant G13R yielded similar results (Supplementary Fig. 8a–c). Together, these data support the conclusion that facilitated diffusion (advection) plays a larger role in protein transport within the lamella than in the cell body.

Anterograde transport is compartmentalized to the front of the cell by a leaky condensate barrier

We then sought to characterize why the transport mechanisms in the lamella were distinct from those in the cell body by performing FLOP experiments in both regions. We discovered that the fluorescence of both untagged PaGFP and PaGFP-actin remains predominantly confined to the compartment where it was activated, regardless of whether the activation occurred in the lamella or the cell body (Fig. 4a). Fluorescence spread into both compartments only when the activation spot was positioned at the boundary (Fig. 4c). These results suggest that the actin-myosin boundary between the compartments acts as a barrier to separate proteins in the lamella from the rest of the cytoplasm.

Fig. 4. Actin-myosin condensate barrier compartmentalizes the cell cytoplasm.

a Spot activation of PaGFP actin in the cytoplasm of a CAD cell demonstrates fluorescence spreading throughout the cell body. By the 30 s, some actin has traveled into the lamella compartment and become incorporated into the leading edge. b, d, Representative baseline-subtracted, normalized raw fluorescence traces from volume-normalized intensity of PaGFP activated in either the lamella or the cell body measured in one-micron square regions of interest (ROIs) placed equidistant from the activation centroid. Curves demonstrate that the rise time (time between 90% and 10% signal, t_90–10) in the acceptor compartment is always slower than the rise time in the donor compartment, independently of which compartment is the donor. c–e In contrast, when the activation spot is placed on the lamella-cell body barrier, actin rapidly spreads in both compartments. d with the difference in rise times approximately zero (e) (n = 6 cells, 6 independent experiments). Error bars are median ± 95% confidence limits. f The difference in rise time between compartments in cells activated in each compartment separately, expressed as the average of the difference in rise time across the barrier in both directions, provides a conservative estimate of interface-associated delay in protein transfer (n = 6 cells from 2 independent experiments, yielding 12 bidirectional measurements (2 per cell)). Error bars, median ± 95% confidence limits. g Maximum projection of a 3D SIM image of mEmerald MLC, Alexa 647. Phalloidin-labeled CAD cells, color-coded for height, indicate that actin-myosin is arranged as a vertical barrier between the lamella and the cell body. h iPALM image of f-actin in CAD cell rendered at 15 nm isotropic resolution, gamma-adjusted color-coded for height with Peak Selector software³⁷. Blue and magenta colors in the middle of the inset box indicate actin arcs that form the barrier at the rear of the lamella. i y–z and x–z inset panels from (h) indicate a separate compartment exists between the leading edge and the actin arcs at the base of the lamella. The color code is the same as in (g). Scale bars, 5 μm. Source data are provided as a Source Data file.

To test whether the interface limits protein exchange beyond geometric differences, we quantified volume-normalized PaGFP TIRF signals at ROIs equidistant from the activation centroid: within the same compartment (donor) and across the interface in the acceptor compartment (Fig. 4b, f). Under the continuous activation used here, these traces show the monotonic rise to a plateau expected for a constant source in standard heat/mass-transfer problems^35,36 (Fig. 4b, d). We found that the acceptor always showed a slower 10–90% rise time within the same cell (Fig. 4b). When the activation was instead centered on the interface, the two sides rose together. No directional bias was detected: the 10–90% rise time difference was not significantly different from zero using the labeling shown (arbitrary assignment, one-sample t-test, p = 0.2, n = 6), and this conclusion was unchanged using a fixed anatomical convention (lamella-cell body; p = 0.9, n = 6), indicating that there is no systematic geometric or optical asymmetry between compartments (Fig. 4c–e). We then alternated the donor and acceptor within the same cell. A geometry-only mechanism would reverse direction and reverse sign. Instead, the direction-average of the across-within compartment differences remains positive, with no significant difference between the two activation directions (paired t-test, p = 0.2, n = 6), suggesting geometry alone is insufficient and providing a conservative measure of interface-associated transfer delay with the actin-myosin condensate (Fig. 4f).

We next investigated the structure of the condensate barrier using 3D structured illumination microscopy (3D-SIM). 3D-SIM revealed a vertical concentration of actin and myosin at the lamella-body boundary (Fig. 3g and Supplementary Figs. 5 and 6), with myosin primarily confined to the barrier. Only a few fibers extended into the lamella, and no myosin was detected at the cell leading edge (Supplementary Fig. 5b–e). This finding suggests that myosin’s functional role is primarily confined to the barrier region.

To gain further insight, we used interferometric photoactivation localization microscopy (iPALM)³⁷, which reconstructed the actin nanoarchitecture with a 15 nm isotropic resolution and allowed us to create 15 nm virtual optical sections to explore compartmentalization. iPALM confirmed that the barrier effectively extends through the entire thickness of the cell and segregates the lamella from the cell body in both NG108 and CAD cells, creating a separate compartment within the cytoplasm (Fig. 4h, i and Supplementary Fig 7).

Anterograde transport directionally targets necessary proteins to locally advancing edge

The localization of advective transport to a compartment at the front of the cell suggests that this flow plays a role in specializing the function of the cell front, regulating actin-mediated protrusion and the formation of adhesion complexes to bind to the extracellular matrix. To test this hypothesis, we used the FLOP technique to evaluate the movement of components crucial for edge extension and adhesion formation. Analyzing the dispersion of the Arp3 component of the Arp2/3 complex, vinculin, and paxillin demonstrated that these proteins also exhibited advective transport toward the leading edge (Fig. 5a). The advective transport shows a small decrease over this range of particle size (Supplementary Fig. 9), and the decrement agrees well with the theoretical velocity hindrance of similar sized particles in cylindrical pores³⁸. In contrast to diffusion, which experimentally and theoretically has a strong dependence on protein size^38,39, advective transport should continue until the particle physically occludes the actin meshwork^38,39. Thus, this facilitated transport mechanism enables a wide range of proteins required for cellular migration and adhesion to reach the leading edge faster than diffusion.

Fig. 5. The actin-myosin condensate barrier targets polymerizable proteins to advancing regions of the leading edge.

a Asymmetry analysis of spot activation of mEos3.2 Arp3, vinculin, and paxillin in CAD cells illustrates that they are targeted toward the leading edge, where polymerization and adhesions initiate (n = 14, Arp3, 9 vinculin, 10 paxillin cells). b Temporal color-coded spot activation of mEos3.2 actin in an NG108 cell advancing preferentially on the right illustrates that this is the region where recently activated actin is targeted (arrows). c Maximum projection of 3D SIM images of Alexa 647-phalloidin and mEmerald MLC in control cells and cells treated with 0.2 μM nitro-blebbistatin before fixation illustrates the broadening and flattening of the actin-myosin arcs at the border between the lamella and cell body when myosin is inhibited. d These observations were quantified by fitting the arcs to ellipses and calculating ellipticity, which reveals flatter arcs in the blebbistatin-treated cells (n = 19 control cells, 87 measurements, 17 blebbistatin cells, 100 measurements). Ellipticity =√(1-(minor)²major⁻²)) e An overlay of two different mNeon MLC image time points (t = 0, and t = 45 s) with arrows indicating where the edge retracts and extends during the 45 s. f Contour of the same two-time points in (e) with arrows indicating changes in the position and curvature of the arcs corresponding to the retraction and advancement of the leading edge. g CAD cell expressing mEmerald MLC were treated with a high-power localized beam of 405 light positioned over one arc, causing the arc to be disrupted and only the leading edge in front of that single arc to collapse. The yellow line indicates the position of the edge prior to MLC disruption. p values two-tailed t-tests. Box plots show median (center line), interquartile range (25th–75th percentiles; box), whiskers to the most extreme non-outlier values, and notches as 95% CIs. Scale bars, 5 μm. Source data are provided as a Source Data file.

To further investigate the connection between advective transport and protrusion, we examined how transport is connected to the uneven advancement of different regions of the leading edge. In NG108 cells, we spot-activated mEos3.2-actin and observed that newly activated actin preferentially moved towards and became incorporated into the actively advancing region. Since actin was not uniformly incorporated along the front of the cell, this behavior suggests a directional bias in the flow (Fig. 5b). This bias is easier to visualize in 3T3 fibroblasts, which rapidly switch the local region of the edge that is advancing. In these cells, we used a narrow strip of UV light oriented perpendicular to the leading edge to activate PaGFP-actin in the center of the cell. The activated actin remained largely confined to the lamella, with minimal spreading to the cell body, even after extensive in-compartment mixing. This additional indicator of compartmentalization also shows the spread of activation was asymmetric: it rapidly extended toward the advancing side of the lamella. As the advancing region shifted, the fluorescence spread followed, switching to the opposite side of the activation line (Supplementary Fig. 10).

Finally, we examined the role of the actin-myosin condensate at the rear of the lamella in directional targeting protein delivery. In CAD cells, this barrier forms well-defined arcs. By quantifying the curvature of these arcs using ellipticity measurements, we found that inhibiting myosin activity to block advection caused the arcs to flatten and elongate (Fig. 5c, d). Additionally, the curvature, orientation, and position of the arcs were dynamically adjusted in control cells to align with the direction of local edge advancement or retraction (Fig. 5e, f). Disrupting a single arc through laser ablation resulted in the collapse of the leading edge only at the region directly in front of the arc, indicating that these arcs are necessary to maintain actin monomer transport for effective cellular protrusion (Fig. 5g). These findings suggest that the curvature and the orientation of the actin-myosin arcs direct cytoplasmic flow to regulate both the speed and direction of soluble protein transport toward advancing regions of the cell’s leading edge.

Discussion

Our results reveal that an actin-myosin barrier limits protein transfer throughout the cytoplasm by creating a separate cytoplasmic compartment at the front of the cell. The barrier retards the movement of soluble protein across the cytoplasm, similar to other actin-myosin condensates^40–42. Contraction of the barrier creates a forward-directed fluid flow that continuously and non-specifically transports protein to the cell edge. The local curvature of the barrier targets the delivery of soluble proteins to advancing regions of the edge, where fluid likely leaves the cell through aquaporin channels¹³. This compartment, equipped with a directionally tunable, molecularly non-specific transport mechanism, enables rapid and localized changes in the supply of polymerizable proteins, facilitating cellular protrusion and adhesion formation.

Although the concept of myosin-dependent forward fluid flow has been postulated^11,13,17,43, direct experimental evidence of the mechanism underlying the forward cartoon arrow has been lacking until now. We overcame the challenges associated with visualizing this process by combining photobleaching, FLOP, FCS, SIM, iPALM, and SPT. Our data offers insight into earlier studies, such as those suggesting dual sources of actin monomers for leading-edge incorporation¹⁶ or the presence of multiple actin networks at the cell front⁴⁴. Our findings indicate that monomers from the cell body must cross an actin-myosin transport barrier to contribute to leading-edge dynamics, underscoring the need to reinterpret previous data in light of our results. We propose that transport within each compartment is likely optimized to fulfill the specific functional requirements of that compartment.

The existence of a distinct compartment at the front of the cell allows proteins to travel only short distances, enabling rapid and precise control of protein localization. This concentration of proteins may further enhance biochemical reactions, allowing the cell to efficiently probe and interact with its environment. Rapid deployment of proteins such as actin is essential, as de novo synthesis, regulated by localized mRNA, is too slow to meet the immediate demands of protrusion and occurs primarily to stabilize structures over longer periods⁴⁵. Our observation that actin-binding proteins are also transported via advection suggests that this transport mechanism targets sites of active actin polymerization, which we have shown influences polymerization-dependent integrin conformations at the protruding edge⁴⁶. Thus, the dynamic tuning of the direction and enhanced protein delivery along the leading edge demonstrates the involvement of the actin cytoskeletal architecture in facilitating cell shape changes and migration.

In summary, our findings redefine the front of the cell as a dynamic, hybrid structure—part membrane-bound and part ‘membrane-less’ condensate-bound—a pseudo-organelle that continuously directs polymerizable proteins to where they are most needed. This targeted delivery enhances the cell’s ability to adapt rapidly to extracellular cues, highlighting the sophisticated coordination of cellular structures and functions required for motility and environmental interaction.

Methods

Coverslip cleaning and preparation

For single-molecule, structured illumination, fluorescence correlation microscopy (FCS), and photoactivation experiments, # 1.5 coverslips were purchased from Warner Scientific (25 mm) and sonicated for 30 min in Luminox (Alconox). After five washes in approximately 300 mL of Milli-Q water, the coverslips were rinsed in acetone, air dried, and baked in a 110 °C oven for 45 min. Coverslips were cooled to room temperature and then plasma cleaned with a Harrick plasma sterilizer. Coverslips were baked again at 110 °C for 45 min in a covered crystallizing dish. 0.75 mL of HMDS (Sigma), which was stored under nitrogen gas, was introduced into the center of the dish to vaporize and silanize the coverslips. After an additional hour of baking, coverslips were cooled and stored in airtight containers until use. Coverslips were mounted in AttoFluor Chambers (Invitrogen) or CAMPO Chambers (LiveCell Instruments) for imaging. MatTek dishes (MatTek Life Sciences) were used for Duo Scan experiments. For iPALM experiments, # 1.5 coverslips were processed to embed gold nanorods to act as fiducials for drift correction according to previously published protocols³⁷. These coverslips were then cleaned and silanized as described above. All coverslips or dishes were coated with 40 μg/mL mouse laminin (MP Research) or 5 μg/mL human plasma fibronectin (MP Research) at 4 °C overnight before use.

Microscopy

All live cell microscopy was performed at 37 °C using a WPI Air Therm PID temperature controller, except for the FCS experiments, which used a Tokai Hit stage top incubator with 5% CO₂. All experiments were performed in phenol red-free growth media specific for the cell line, supplemented with 15 mM HEPES.

TIRF and SMLM

A TIRF microscope was constructed around an Olympus IX-71 stand using a 60 × 1.49 NA objective. Coherent 405, 488, 561, and 637 nm lasers were coupled to the microscope through free space on a separate laser sled⁴⁷. An ANDOR iXion 897 EMCCD camera, under the control of Solis software, was used for image acquisition. The TIRF angle was controlled by a programmable Thor Labs motor (Z812B) and custom LabVIEW software. The AOTF (AA Optoelectronics) and the lasers were controlled with AA MDS and Obis software, respectively. The filter set was a Semrock Penta set, Di01-FF409/49/57/652/759, 432/514/595/681/809.

Multi-color TIRF

A custom-built electronic circuit accomplished the frame switching between 561 and 637 nm lasers on the TIRF microscope⁴⁷. For two-color live cell imaging, results were denoised using the Fiji CSBDeep Noise2Void plugin⁴⁸, with training consisting of 30 epochs, patch size = 64, and a neighborhood radius of 5.

Spot-activation for FLOP

A second Coherent 405 nm laser, equipped with a fiber optic, was introduced into the fluorescence light path using a Semrock FF458-Di02 dichroic mirror, creating a diffraction-limited spot in the TIRF microscope’s image field.

SIM

Structured illumination experiments were performed using a Zeiss Elyra 7 with 488 and 642 nm lasers and a Plan-Apo 63x 1.4 NA Oil DIC objective. Filterset#1 BP 57- = 620 + LP 655, Filterset#2 BP 420–480 + BP 495–550.

Zeiss Duo scan

Images for NG108 stripe bleaching were collected at 10 frames per second (Figs. 1 or 2) frames per second for box activation (Supplementary Fig. 6). Box activation images collected with the Zeiss Duo Scan were filtered with a Gaussian blur (σ = 0.6) followed by an unsharp mask (σ = 0.75). Photobleaching of time-lapse movies was corrected using the exponential fit option in Fiji^49,50. Images were collected with a 63x Plan Apo 1.4 NA objective and DPS532-75 and DPSS 561-10 filters.

iPALM

The iPALM microscope combines single-molecule localization with multi-phase interferometry, and its usage has been previously described in ref. ³⁷. In summary, cells were cultured on silanized coverslips embedded with gold fiducials (Nanopartz, Inc.), which were passivated with an approximately 50 nm layer of SiO₂ for calibration and drift correction. Following fixation and labeling as outlined below, the cells were covered with STORM buffer containing TRIS-buffered glucose, glucose oxidase, catalase, and mercaptoethanol amine⁵¹, and sealed with epoxy. Fluorescence was captured using a pair of 60x 1.49 NA Apo TIRF objectives (Nikon) along with two emission filters (LP02-647RU and FF01-720/SP from Semrock). Interference images were digitized using three iXon3-DU897E EMCCD cameras (Andor Technologies). The iPALM data were analyzed and images rendered using the PeakSelector software (Janelia Research Campus), which is available at https://github.com/gleb-shtengel/PeakSelector)^37,52,53. Images were rendered using Group Peaks in Peak Selector. x–z and y–z views were created using the Volume Viewer plug-in in Fiji with a 10 x scale in z⁵⁰.

FCS

FCS curves were collected using a Leica TCS SP8 Falcon FLIM/FCS microscope controlled using Leica LAS X (3.5.7.23225) and LAS X FLIM/FCS (3.5.6) software. Imaging was performed with Leica APO 86×/1.20 NA water-immersion objective with a motorized correction collar (mottCorr, Leica). Prior to FCS measurements, the mottCorr was adjusted using x–z-scans in reflection mode to yield the sharpest cellular image.

Cell culture and transfection

Cell lines and culture

CAD (Cat.-a-differentiated) cells (Sigma-Millipore, 08100895-DNA-5ug) were grown in DMEM/F12 media containing 2 mM glutamine (and supplemented with 8% fetal bovine serum). NG018-15 cells (ATCC, HB-12317) were grown in DMEM containing 1× HAT supplement and 10% fetal bovine serum. NIH 3T3 cells (Duke University Medical Center Cell Culture Core, ATCC, CRL-1658) were grown in DMEM High glucose supplemented with 10% fetal calf serum. All media and supplements were obtained from Gibco; serum was obtained from Cytivia-Hyclone.

Labeling

Cells expressing Halo-tagged proteins were labeled for FCS or two-color live-cell imaging by incubating with JF549⁵⁴ or JFX650⁵⁵, respectively, at a concentration of 200 nM for 30 min. Cells were then rinsed twice in warmed complete media before imaging.

Fixation

All cells were fixed at 37 °C with 2% paraformaldehyde in PHEM (pH 6.8)⁵⁶. After 5 min, the fixation media was replaced, and the fixation process continued for another 25 min. Cells were rinsed 3× for 5 min each in PHEM before labeling with Alexa 647plus phalloidin (Invitrogen).

Transfection

NG108, CAD, and NIH 3T3 cells were transfected using Invitrogen LTX Plus according to the manufacturer’s protocol. After 3 h, the transfection media were exchanged for complete growth media, and the cells were imaged the following day.

Constructs

Egfp-Actin was purchased from Clontech (cat #6116-1). Apple, mEos3.2, and Emerald actin were gifts from Michael Davidson (Addgene nos. 54862, 57446, and 53978, respectively). Halo-actin was a gift from Kai Johnsson⁵⁷.

PaGFP-Actin was constructed by exchanging the color from PaGFP-C1 (gift from George Patterson, Addgene no. 11910) and egfp-Actin using BsPE1 and BamH1^22,46.

mNeon LC-Myosin-N7 was constructed by exchanging color with mEmerald-LC-Myosin-N7 (gift from Michael Davidson, Addgene no. 54146) and mNeon Lifeact (Addgene no. 9887) using BsrGI and AgeI.

mEos3.2 ARC3 was constructed by exchanging color with egfp-ARC3, a gift from Matt Welch (Addgene no. 8462)⁵⁸ and mEos3.2-N1 (Addgene no. 54525) using SmaI and Not I.

mEos2 Paxillin-22 (Addgene no. 57409) and mEos3.2 Lifeact-7 (Addgene no. 54696) were gifts from Michael Davidson.

mEos3.2 Vinculin-N-21 was constructed by exchanging color with mEos2-vinculin-14 (Addgene 57438) and mEos3.2-N1 (Addgene no. 54525) and is available as (Addgene no. 6692).

pENTR-NLS-actin-R62D (Addgene no. 11831) was used to construct the mEos3.2 actin-R62 plasmids using SalI-HF and BmbG1 to combine the mutated region of actin with existing mEos3.2 actin plasmids. The insert regions of the resultant plasmids were verified by sequencing.

YFP NLS Beta-Actin G13R (Addgene no. 60615) was used to construct the mEos3.2 actin-G13R plasmid using SalI-HF and BamH1 to combine the mutated region of actin with existing mEos3.2 actin plasmids. The insert regions of the resultant plasmids were verified by sequencing.

Chemical reagents and dosage

Reagents were obtained from the following sources: ±Blebbistatin (Calbiochem, #203390), (s) Nitro-Blebbistatin (Cayman Biochemicals, #13891)), Y-27632 (Cayman Biochemical, #10005583, CAD) or Y-27632 (Calbiochem, #688000, NG108), Alexa Fluor Plus 647 Phalloidin (Invitrogen, #A30107). Digestion Enzymes were obtained from NEB Biolabs (HF versions used when available). Titration experiments for all pharmacologicals spanned literature values to the lowest dose that yielded quantifiable results over the time course of the live cell experiment, with cells treated with these doses, fixed and stained with phalloidin to document morphological integrity (Supplementary Fig. 3).

Over-expression controls

Morphology

All three cell lines were transfected with egfp-actin, fixed, and labeled with Alexa 647. TIRF images were obtained of fields containing transfected and untransfected cells to compare the effects of transfection on actin network morphology (Supplementary Fig. 11).

Western blotting

Cells were plated at the same density and with the same transfection protocol used for live-cell experiments. Negative controls run in parallel were transfections with the same concentration of the fluorescent backbone vector. Approximately 24 h after transection, fluorescence expression was verified using an EVOS FL fluorescent microscope. Cells were trypsinized, centrifuged in complete media, resuspended in media, and counted 3×. The 100,000 cells from each transfection were centrifuged, and the pellets were frozen at −80 °C.

Samples were lysed directly in 100 μl of 1× SDS (40 mM Tris-CL, pH 6.8, 0.5% SDS, 5% glycerol, bromophenol blue to desired color). Samples were then boiled for 8 min and loaded onto the gel (35 μl on gel), except for vinculin, which was run at 25 μl.

Total protein lysates were separated into 4–12% Bis-Tris CriterionXT Precast Protein Gels (BioRad, 3450123), transferred on a PVDF membrane (Millipore, Immobilon-FL), and transferred using a semi-dry apparatus (BioRad).

Antibodies were diluted in [1:1 (1× PBST, Phosphate Buffered Saline-0.1% Tween-20, pH 7.4): (Aquablock, Arlington Scientific)]. Antibodies used were: anti-actin (Sigma, A5441), anti-paxillin (Sigma, P1093), GFP (Abcam, AB38689), GAPDH (Applied Bioscience, AM4300), Eos (Novus, NP3-05557), Arp3 (Proteintech, 13822), and Vinculin (Sigma, V9131).

Secondary antibodies: Licor 1:10000 (Goat anti Rabbit 680 (926-68071), Goat anti Mouse 800 (926-32210), Donkey anti Mouse 680 (926-68072) in 1:1 PBST: Aquablock.

Gels were scanned with a LI-COR Odyssey Imaging System (9120, using Odyssey Software V3.0).

Analysis

Transport ratio

Baseline images were collected before initiating activation (typically 50–75 frames). These images were averaged, and the result was subtracted from each image frame in the experimental time series. After activation, the first 500 frames (1000 frames for Arp3, vinculin, and paxillin) were summed to create an image. This image was contrast-stretched, and the intensity was recorded through a 5-pixel-wide line drawn orthogonal to the leading edge and through the center of the activation spot or a 5-pixel-wide line drawn parallel to the edge and through the spot center.

Software (TransportRatio.m) written in Matlab normalized each of these intensity curves, smoothed them with a spline fit, detected the peak (center of the activation spot), and integrated the area under the curve on both sides of the center. The software is available at github.com/galbraithlab/advectus. These two area calculations were then compared. The ratio of these areas (leading edge/cell body or parallel side1/parallel side2) was defined as the transport ratio. Since there was no morphological rationale for the side-to-side calculation, the choice of which side was “leading” or “body” was randomized.

In the presence of fluid flow, the areas of the halves of the curve drawn orthogonal to the leading edge will not be equal. When the area of the curve closer to the leading edge is divided by the area closer to the cell body, the ratio will be greater than 1. However, the forward-directed fluid flow would not be expected to affect the lateral distribution of mass in the direction parallel to the leading edge. In this direction, the ratio is expected to be 1.

Slope of parallel spread

Parallel profiles of FLOP data were spline fit, normalized, and their full width ¼ max (FWQM) were determined in Matlab. FWQM minimizes emphasis on the activation beam profile. To linearize the spread over time, Theil–Sen slopes, were calculated for the FWQM² over the time interval of 1–10 s. Regression coefficients less than 0.5 were discarded, as were experiments with off-center activation beams. The activation beams of the analyzed control and blebbistatin-treated cohorts were located at the center of the lamella width with a median absolute deviation (MAD) of 0.05.

Advective transport simulation

Plume profiles in Supplementary Fig. 2 were generated in Matlab by numerically integrating the solution for a continuous injection at the origin into a steady plane flow in the x–y plane²¹:

$$C\left(x,y,t\right)=\frac{C_{o}Q}{4\pi D}{\int }_{\theta =0}^{\theta =t}\frac{1}{\left(t-\theta \right)}\exp \left\{-\left(\frac{x^2+y^2}{4D}\right)\cdot \frac{1}{\left(t-\theta \right)}+\frac{2vx}{4D}-\frac{v^2}{4D}\left(t-\theta \right)\right\}d\theta$$ 1

Where C_oQ is a constant injection rate of material, D is the diffusion coefficient, and v is the velocity of the flow in the x direction.

Localization and tracking

All SPT was performed with TrackIt⁵⁹, using the nearest neighbor tracking algorithm with a threshold of 2.5, tracking radius of 5 pixels, maximum gap of 1 frame, and minimum track length of 5 steps.

Moment scaling spectrum (MSS)

The moment scaling spectrum (MSS)^29,30,60 was calculated from single-molecule track data obtained using TrackIt⁵⁹. The Matlab analysis script, MSScalc.m, is available from github.com/galbraithlab/advectus. The MSS calculates scaling exponents for the higher orders of the mean displacement curve, with the commonly used mean squared displacement (MSD) being the second-order moment used to obtain the diffusion coefficient. The slope of the exponent curve versus the moment order characterizes the type of motion, with a slope > 0.5 being enhanced diffusion, 0.5 being diffusion, and <0.5 being subdiffusive. Tracks shorter than 9 steps and exponent-moment plots with a regression coefficient of determination (r²) less than 0.75 were excluded.

Fluorescent correlation spectroscopy (FCS)

The FCS curves were exported to Matlab, where the autocorrelation curve was fitted using a two-component model that accounted for both diffusion and flow, with the flow affecting both species equally^32,33. A two-component model reduced the fit residuals and was biologically reasonable based on FRAP measurements in the literature, indicating that diffusing actin exists as either a diffusing oligomer (or other complex) or monomeric actin transiently bound to fixed targets⁶¹. The analysis script, FCSfit.m, is available at github.com/galbraithlab/advectus.

$$G\left(\tau \right)=\left[1-T+T{e}^{-\tau /{\tau }_{Triplet}}\right]\cdot \left[\begin{array}{c}A\cdot \frac{1}{1+\frac{\tau }{{\tau }_{D1}}}\cdot \sqrt{\frac{1}{1+\frac{\tau }{{\kappa }^2{\tau }_{D1}}}}\cdot e^{-\left({\left(\frac{\tau }{{\tau }_{flow}}\right)}^2\cdot \frac{1}{1+\frac{\tau }{{\tau }_{D1}}}\right)}+\hfill \\ \left(1-A\right)\cdot \frac{1}{1+\frac{\tau }{{\tau }_{D1}}}\cdot \sqrt{\frac{1}{1+\frac{\tau }{{\kappa }^2{\tau }_{D2}}}}\cdot e^{-\left({\left(\frac{\tau }{{\tau }_{flow}}\right)}^2\cdot \frac{1}{1+\frac{\tau }{{\tau }_{D2}}}\right)}\hfill \end{array}\right]$$ 2

Where T is the triplet state fraction, τ_triplet is the triplet state lifetime in ms, τ_D1, τ_D2 is the diffusion times of species 1 and 2 in ms, A is the fraction of molecules with diffusion time τ₁, τ_flow is the flow time in the volume in ms, and κ is the structure parameter of the focal volume (κ = 5). The diffusion coefficient was calculated by: $D=\frac{r_o^2}{4{\tau }_D}$ and the velocity by: $v=\frac{r_o}{{\tau }_f}$ (r_o = 0.2543 µm). All traces were individually analyzed, provided they contained at least 10 counts and exhibited no obvious shift in count baseline.

Barrier analyses

A diffraction-limited spot of 405 nm laser light was used to photoactivate either in the lamella or the cell body of CAD cells expressing either PaGFP actin or untagged PaGFP. Images were collected every 15 ms by TIRF microscopy. Quantification was performed in Fiji⁵⁰ on one square-micrometer ROIs as follows.

For each ROI, the baseline was the median intensity from pre-activation frames. Signals were baseline-subtracted and normalized by dividing by the baseline (accounts for effective TIRF volume). To make the rise-time measurement consistent and reproducible, we used MATLAB’s built-in movmax function with a look-ahead window equal to 20% of the trace length. The sliding max provides each time point a local reference, allowing automatic location of the first 10% and 90% crossings (t₁₀ and t₉₀). This step is only used to mark timing, not the plotted amplitudes.

For boundary-centered activation, two ROIs were positioned equidistant from the activation centroid and on opposite sides of the lamella-cell body interface (one in the lamella, one in the cell body). For within-compartment activation, a donor ROI was placed in the activation compartment, and an acceptor ROI was placed in the other compartment at the same distance from the activation centroid. Donor and acceptor labels were randomly assigned for boundary activations.

For within-compartment activations, the within rise time is the (t₉₀–t₁₀)_donor, and the across-interface rise time is (t₉₀–t₁₀)_acceptor. The across-within difference, (t₉₀–t₁₀)_acceptor -(t₉₀–t₁₀)_donor measures the delay encountered crossing the interface (interface delay + any residual geometric effects). In matched, same-cell experiments, we repeated the measurement with the direction reversed (donor becomes acceptor and acceptor becomes donor) and report the direction average. Because geometric height/taper differences bias the two directions with opposite sign, where a true interfacial delay contributed the same sign, the direction average provides a conservative estimate of interface (actomyosin condensate) associated delay, with no significant difference between the directions (paired t-test, p = 0.2, n = 6). Directional effects were assessed with a paired t-test across cells, and boundary-centered symmetry was assessed with a one-sample t-test of the lamella–body difference (using either random labeling or a fixed anatomical convention, single-sample p-test, p = 0.2 or 0.9, n = 6, respectively).

Western blots

The gel analysis Fiji⁵⁰ plug-in was used to quantify protein density.

Statistics

The t-tests and Z-tests were performed in Excel. ANOVA with Tukey post-ANOVA tests were performed using online software available at astatsa.com. Circular statistics were performed using the Circular Statistics Toolbox in Matlab.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

C.G.G. and J.A.G. conceived of the project, performed experiments and analyzed. B.P.E. made substantive contributions to paper organization and performed FCS experiments and analysis, and U.B. performed iPALM experiments and analysis. C.G.G. and J.A.G. drafted the manuscript with contributions from all authors.

Peer review

Peer review information

Nature Communications thanks Xuebiao Yao and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The fluorescence intensity traces, SPT movies, virtual iPALM sections, processed SIM images, and representative raw imaging sequences generated in this study have been deposited in Zenodo https://zenodo.org/records/18462015. Source Data generated in this study are provided in the Source Data file. Complete unprocessed raw micrograph datasets (>150 GB) are not deposited in a public repository due to file size constraints that exceed practical repository limits; these data are available from the corresponding author(s), with a response within 30 days. All plasmids are either previously published and referenced or, as described in the methods, created by simple enzymatic digestion and ligation of constructs available through Addgene. These resultant plasmids are available upon request from the corresponding author. Source data are provided with this paper.

Code availability

Matlab code used to process transport ratios, calculate MSS from Track-it trajectories, and fit FCS data is available at http://github.com/galbraithlab/advectus and is linked to https://zenodo.org/records/18462015⁶².

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-70688-6.