Main

Biomolecular condensates perform a wide variety of functions in cells and also have applications in biotechnology1,2,3,4. The phase separation of molecules into distinct phases provides increased control over reactions in space and time3,4. Many condensates in cells have an internal architecture that is strongly linked to their function, with examples including stress granules5,6, nucleoli7, nuclear speckles8 and the mitochondrial nucleoid9. Sub-compartments with different compositions result in further spatiotemporal control over the reactions that occur2,10,11.

Given the close relationship between mesoscale architecture and function, the limited synthetic control over condensate architecture poses a very challenging problem12,13,14,15,16. In vitro, methods to change the architecture of condensates from the thermodynamically stable state are limited10,15,16,17,18,19,20. In vivo, there is limited understanding10,15,21 of what influence the out-of-equilibrium environment of a cell has on the observed complex condensate architectures10,22,23,24,25,26.

In this Article we provide a method to understand out-of-equilibrium architectures and design condensates with desired architectures in vitro. Here, ‘architecture’ refers to the mesoscale organization of the various phases—the number of droplets of each phase and where they are located. By forcing a change in the composition of the condensates, we induce the nucleation of our chosen droplets and create complex architectures in a multiphase condensate model. Condensate architecture thus becomes an independent experimental variable. Access to complex architectures will enable researchers to incorporate increasingly sophisticated compartmentalization and functionality in condensates for biotechnology applications27, artificial cells28,29,30,31,32,33 and origin-of-life research34. As an example, we show how a custom architecture with enlarged interface area increases the uptake rate of cargo molecules.

Results

Complex architectures in multiphase condensates

Throughout this Article we explore how to design condensates with different mesoscale architectures (Fig. 1a). At thermodynamic equilibrium, condensates can be terminally viscous or elastic, and before reaching equilibrium, condensates can have a variety of architectures. A simple architecture will have only one droplet of each phase per condensate, with minimized interfacial tension. A complex architecture will have more droplets of each phase, organized with a higher interfacial area. To study condensate architecture, we use a model complex coacervate system that was previously used for artificial life research35,36,37,38. Coacervates are often used as a model for membraneless organelles39,40. Our model system contains two amylose compounds that have been quaternized or carboxymethylated to have a positive or negative charge, respectively (Fig. 1b and Methods). Negatively charged single-stranded DNA, a stabilizing terpolymer, salt and buffer are also present. The favorable electrostatic interaction between the positively charged Q-amylose and negatively charged Cm-amylose and DNA causes multiphase separation, similar to previously described systems41,42, where multiphase condensates are formed in a dilute phase (Fig. 1c, phase 1). The DNA has a higher density of negative charges than the Cm-amylose, causing the formation of two dense phases within the condensate, a DNA-poor phase (phase 2) and a DNA-rich phase (phase 3). The Q-amylose and DNA have the highest concentration in phase 3, and Cm-amylose is most abundant in phase 2 (Extended Data Fig. 1a).

Fig. 1: The model multiphase condensate system.
figure 1

a, Condensates can have various architectures. This Article shows how condensates with complex architectures can be designed. b, The model condensates are prepared by mixing Cm-amylose, Q-amylose and DNA in a solution with a terpolymer stabilizer, salt and buffer (Methods). c, Electrostatically driven phase separation35,50 yields a dilute phase (phase 1) and condensates containing phase 2 surrounding phase 3 at 40 °C (Extended Data Fig. 1). d, Upon heating the condensates to 65 °C, this architecture is maintained. e, However, cooling the condensates back to 40 °C causes the nucleation of droplets of phase 2 in phase 3 (orange arrowheads) and droplets of phase 3 in phase 2 (yellow arrowheads). Not the temperature itself, but how this temperature is reached determines which (transient) architecture is obtained (Extended Data Fig. 2).

At 40 °C, phase 1 surrounds the condensates. The condensates contain phase 2, which (partly) engulfs phase 3 (Fig. 1c). This architecture is retained when condensates are heated to 65 °C (Fig. 1d). However, rapid cooling of these condensates back to 40 °C changes their architecture (Fig. 1e). Droplets of phase 3 nucleate in phase 2 (Fig. 1e, yellow arrowheads) and droplets of phase 2 nucleate in phase 3 (Fig. 1e, orange arrowheads). After cooling, we thus obtain condensates consisting of phase 2 (partly) surrounding droplets of phase 3, some of which contain droplets of phase 2. By comparing the architecture of condensates prepared at 40 °C (Fig. 1c) and those that are heated to 65 °C and subsequently cooled to 40 °C (Fig. 1e), we observe that the architecture of the condensate is determined not by the temperature itself, but how this temperature is reached (Extended Data Fig. 1). Interestingly, heating condensates with nucleated droplets back to 65 °C removes these droplets (Extended Data Fig. 2). Rapidly cooling to 40 °C then induces nucleation again. The process is reversible for at least five cycles of cooling and heating. Additionally, we observe that the small nucleated droplets are very similar in size directly after nucleation, as expected43,44,45.

The presence of the small nucleated droplets increases the total interfacial area of the condensate. Considering the interfacial tension, this implies that an architecture with small droplets (Fig. 1e) is higher in energy than an architecture where all the liquid of that phase is in the same droplet (Fig. 1c). The small nucleated droplets fuse together upon contact. Thus, the architecture with small nucleated droplets could be considered a kinetic product or transient state (Fig. 1e). Next, we will explore why rapid cooling results in the formation of higher-energy transient architectures.

Diffusion-limited nucleation

The concentration of DNA and amyloses in the three phases was estimated using fluorescence intensity (Methods and Extended Data Fig. 3). The projected phase diagram (Fig. 2a) shows the estimated DNA concentration (x axis) in the three phases at different temperatures (y axis) and a possible location of the binodal (solid black line). When the system is cooled from 65 to 40 °C, the amount of DNA in phase 1 does not change substantially. Cooling decreases the amount of DNA in phase 2 and increases it in phase 3. Decreasing the temperature may increase the relative strength of the hydrophobic interactions of the DNA. If phase 2 is cooled from 65 to 40 °C without changing composition, it will deviate from the binodal and cross into the metastable regime or spinodal regime of the phase diagram (vertical arrow). To get back on the binodal, the solution needs to phase-separate (curved arrows) along the tie line (dashed line). Nucleating droplets of the DNA-rich phase 3 in phase 2 deplete the remaining phase 2 in DNA, allowing it to reach the binodal. Similarly, phase 3 may nucleate droplets of phase 2 to change its composition.

Fig. 2: Diffusion-limited nucleation creates long-lived complex architectures.
figure 2

a, A projection of the estimated phase diagram (Extended Data Fig. 3) shows that phase 2 becomes more depleted in DNA, and phase 3 becomes further enriched in DNA when cooling. During cooling, the composition may deviate from the binodal (solid lines), causing droplet nucleation (arrows). Thus, phase 2 may nucleate DNA-rich droplets of phase 3 to become more DNA-poor. Similarly, phase 3 may nucleate droplets of phase 2 during cooling. b, Using FRAP, we determine the diffusion coefficient of the DNA and amyloses at 50 °C in phases 2 and 3. For each data point, we fit the fluorescence recovery and obtain the characteristic recovery time τ (Extended Data Fig. 4a). For all three molecules in phases 2 and 3, diffusion limits the condensates’ ability to change composition in response to changes in the environment. c, As such, droplets may nucleate when cooling the condensates from 65 to 40 °C. More droplets at a closer distance nucleate on cooling more quickly (Extended Data Fig. 4b). d, Nucleated droplets may fuse over time, creating a ‘simpler’ architecture. e, Notably, even when incubating at 40 °C, this process can take hours depending on the radius of the condensate. This is due to slow diffusion. Additional images and fitting information are provided in Extended Data Fig. 4c,d.

Source data.

Overall, staying on the binodal results in a lower-energy architecture, whereas the high-energy transient architecture (Fig. 1e) can be formed when the composition deviates from the binodal. The phases can change composition by exchanging molecules over the interface between them. For example, during cooling, DNA may move from phase 2 to phase 3 across their interface. Notably, the viscosity of condensates is typically orders of magnitude higher than that of the dilute phase46, leading to potentially slow molecule movement in the condensates. We measured the diffusion coefficient of the DNA and amyloses using fluorescent recovery after photobleaching (FRAP; Fig. 2b and Extended Data Fig. 4a). These experiments were performed at 50 °C, the approximate temperature at which we observe droplet nucleation during cooling from 65 to 40 °C. The recovery time, τ, was measured for different bleaching areas. We fit the area against τ to yield diffusion coefficients (Fig. 2b). For all three biomolecules, the diffusion coefficient in phase 3 is lower than that in phase 2, indicating a higher density in phase 3 and a higher concentration of biomolecules, matching the trend in concentrations in the phase diagram (Extended Data Fig. 3). The molecules experience slow diffusion in the condensate (D = 2–9 μm2 s−1). For reference, this is similar to the diffusion coefficient of green fluorescent protein (GFP) in Escherichia coli cytosol (7.7 ± 2.5 μm2 s−1)47.

Given that diffusion is slow in these condensates, it is difficult for them to change composition in response to environmental changes. Moving molecules and exchanging them over the interface may not change the composition at the required rate. Nucleating the droplets offers a solution to this. Accordingly, we expect more droplets to nucleate, in total and per unit of time, when cooling quickly. Figure 2c shows a condensate before and after cooling from 65 °C to 40 °C at different cooling rates. Indeed, the amount of nucleated droplets observed at 40 °C depends strongly on the cooling rate, with many more droplets observed when cooling at 20 °C min−1 than at 0.5 °C min−1 (droplets of phase 3 in phase 2 shown by yellow arrowheads and droplets of phase 2 in phase 3 shown by orange arrowheads; Extended Data Fig. 2c). Notably, the distance between where droplets of phase 3 begin nucleating in phase 2 and the interface between phase 2 and 3 (white arrows) also depends on the cooling rate (Extended Data Fig. 4b). The solution close to this interface may exchange material over the interface to reach a new composition without the need to nucleate a new phase.

Over time, nucleated droplets fuse with each other and the bulk droplet until the lower-energy architecture is reached (Fig. 2d). Due to the slow movement through phases 2 and 3, it may take an extended amount of time for nucleated droplets to fuse. We have studied this process quantitatively by determining the number of phase 3 droplets in phase 2 over time after nucleation while incubating at 40 °C (Fig. 2e and Extended Data Fig. 4c). We find that the number of nucleated droplets is higher in larger condensates, as expected. Additionally, the number of droplets reduces over time via the fusion of two nucleated droplets (D + D → D) and fusion of a nucleated droplet with a large phase 3 droplet (D + P → P)16 (Extended Data Fig. 4d). Although the complex architectures are transient structures, because of the slow diffusion they are present for hours at elevated temperatures, depending on the size of the condensate. The limited diffusion in these condensates, on the same order of magnitude as observed in cells, explains both why droplets can nucleate and also be long-lived.

Predicting nucleation and architectures

The combination of fast compositional changes with limited diffusion results in droplet nucleation and thus complex condensate architectures. However, before we can use this mechanism effectively to design condensates with chosen architectures, two important questions remain. First, how do we control the volume that is nucleated? Second, how can we design condensates with nucleated droplets of a different phase? Next, we will discuss the method for answering these questions (Fig. 3), apply this to our model system undergoing temperature changes (Fig. 4), design condensates with another complex architecture (Fig. 5) and show an application of this method for drug-delivery systems (Fig. 6).

Consider a solution with composition ϕ (Fig. 3a). This point is found within the phase-separating region, surrounded by the binodal (solid line). The solution ϕ will undergo multiphase separation along the tie plane (blue) to form three phases with compositions ψ1 (circle), ψ2 (triangle) and ψ3 (square). The architecture of the system may be condensates containing phases 2 and 3 surrounded by phase 1. The location of the binodal and the tie plane are determined by the relative interaction strengths of the molecules, which in turn depend on the experimental conditions. We will now change the location of ψ3 (blue square) in different directions and see how this may change the architecture.

Fig. 3: Using phase diagrams to understand complex architectures.
figure 3

a, Consider a solution with total composition ϕ, which phase-separates (PS) along the tie plane (blue) to reach solutions on the binodal (solid lines). Three phases are made, with composition ψ1, ψ2 and ψ3. b, Starting from the solution in a, the tie plane and binodal are shifted by changing the experimental conditions. Phases 1 and 2 have not changed substantially in composition. Phase 3 (blue square) is no longer on the binodal and will phase-separate along the edge of the tie plane into phases 2 and 3 (yellow triangle and square). v, the volume fraction (0 to 1) of phase 2, can be determined from the relative lengths of the tie-line segments using a mass balance. c, From the situation in a we can move the tie plane (red) such that the blue-square solution falls on another tie-plane edge. This solution now phase-separates into phases 1 and 3. v and v′ are the maximum volume fractions nucleated, because exchange over the existing interface also facilitates composition changes, limiting how much nucleation occurs. d, From the situation in a the tie plane is once again moved, in this case such that the blue-square solution is in the new tie plane (green). The blue square achieves the composition of the green square due to a combination of nucleating v′ of phase 1 and v of phase 2 (here shown in two steps). Constructing phase diagrams with tie lines or planes provides important insights and predictive capabilities about condensate architecture (shown below the phase diagrams).

Figure 3b shows a new tie plane (yellow) and a new location for ψ3 (yellow square). The blue-square solution that we have formed no longer lies on the binodal and would like to change composition towards the yellow square. Notably, the blue square lies on an edge of the new tie plane and it can thus phase-separate along this edge (arrows) into the new ψ2 (yellow triangle) and ψ3 (yellow square) compositions. Now, all of the solutions are again found on the binodal. If we consider the architecture of our system, phases 1 and 2 have not changed much. However, phase 3 now contains some droplets of phase 2. v is the volume fraction of phase 2 formed (0 to 1), and (1 − v) is the volume fraction of phase 3 that remains (0 to 1). With a mass balance we can find these fractions, with v being the length of the tie-plane edge from the blue square to the yellow square, divided by the total length of this tie-plane edge.

Now consider changing the conditions of our system from Fig. 3a to Fig. 3c. The blue-square solution now lies on another edge of the new tie plane (red). This solution wants to change composition to end up on the binodal. To accomplish this, it will phase-separate along the edge of the tie plane and nucleate a v′ volume fraction of phase 1 (red circle) and obtain a (1 − v′) fraction of phase 3 at the new composition (red square). In the accompanying scheme, we see droplets of phase 1 in phase 3. Notably, the calculated volume fraction of phase 1 is the maximum amount that may nucleate. Molecules may be exchanged over the interfaces to change composition if the experimental conditions are changed slowly. If the composition is partly changed via molecule exchange at the interface, less nucleation is required. Thus, v and v′ are maximum amounts and (1 − v) and (1 − v′) are minimum amounts.

Finally, we will consider the case where the blue square ends up inside the new tie plane (green), rather than on the edge (Fig. 3d). Similarly to how the three phases initially formed from ϕ, the blue square can phase-separate into ψ1 (green circle), ψ2 (green triangle) and ψ3 (green square). The volume fractions are determined most easily by first splitting the solution into v′ of phase 1 and a mixture of phases 2 and 3 along the additional dashed line. This mixture is then split into v of phase 2 and (1 − v − v′) phase 3. The scheme now shows phase 3 containing droplets of both phases 1 and 2.

Constructing a phase diagram with tie planes allows us to predict the condensate architecture in great detail. We can learn the compositions of the nucleated liquids, in which solutions they will appear and what their maximum volume fractions are. In our example (Fig. 3), the three molecules, α, β and γ, partition into the three phases to different extents. To calculate the volume fractions correctly with small error, all compounds with substantially unequal concentrations in the phases should be incorporated into the phase diagram. Next, we will apply this method to understand the nucleation events in our model system when changing temperature (Fig. 1e).

Understanding the complex architecture

Figure 4a shows the phase diagram for our model system at 65 °C (confocal image in Fig. 1d, condition A). The total composition ϕ separates into the three phases \({\psi }_{1}^{\rm{A}}\), \({\psi }_{2}^{\rm{A}}\) and \({\psi }_{3}^{\rm{A}}\) via the tie plane (blue). The system is then cooled to 40 °C, or condition B (Fig. 4b and Extended Data Fig. 5a–c). Now, the binodal and the tie plane (yellow) have shifted. The blue-square and blue-triangle solutions are on the red edge of the tie plane. \({\psi }_{2}^{\rm{A}}\) will split into a maximum of 0.20 \({\psi }_{3}^{\rm{B}}\) and a minimum of 0.80 \({\psi }_{2}^{\rm{B}}\). Similarly, \({\psi }_{3}^{\rm{A}}\) will split into a maximum of 0.07 \({\psi }_{2}^{\rm{B}}\) and a minimum of 0.93 \({\psi }_{3}^{\rm{B}}\). We predict that giving the system more time to exchange molecules over the interfaces between phases 2 and 3 will result in a smaller nucleated volume. We tested these predictions by quantifying the volume fractions for eight condensates of different sizes (Extended Data Fig. 5d,e). Indeed, we observed that small condensates nucleate no droplets or a small volume fraction, whereas larger condensates approach the predicted maximum/minimum volume fractions. Given that phases 2 and 3 change composition along the red edge of the tie plane, there is no reason to nucleate phase 1 during temperature changes, and no such events are observed. Interestingly, if a composition change in the opposite direction took place due to heating instead of cooling, we would predict that mixing of phases 2 with 3 occurs instead of nucleation. This explains our observations of the nucleated droplets disappearing when heated (Extended Data Fig. 2)—they dissolve or mix with the surrounding liquid.

Fig. 4: Understanding the architecture due to temperature changes.
figure 4

a, We construct diagrams similar to those in Fig. 3a,b to understand how cooling our model system changes the architecture. At 65 °C, our total solution ϕ (black) phase-separates into the three phases on the tie plane (blue). b, Changing the temperature to 40 °C changes the binodal location and tie plane (yellow). Phases 2 and 3 both change composition substantially along the red edge of the tie plane. We thus expect phase 2 to nucleate a maximum of 0.20 volume fraction of phase 3. We would expect phase 3 to nucleate a maximum of 0.07 volume fraction of phase 2. No amount of phase 1 nucleation is necessary to achieve the new compositions. The predicted volume fractions fit well with the observed volume fractions, particularly for larger condensates (Extended Data Fig. 5). c,d, Composition changes can force nucleation or mixing of phases. We can construct a look-up diagram using the mass balance to link the new location of the square phase to the forced nucleation of phase 1 (circle, blue), mixing with phase 1 (circle, yellow), nucleation of phase 3 (c) or phase 2 (d) (triangle, red) or mixing with phase 3 (c) or phase 2 (d) (triangle, green). e, The square phase is phase 2 in c and phase 3 in d, so the legend applies to both.

Source data.

Designing complex architectures using controlled nucleation

New architectures can also be designed using this method. Using a mass balance, we explored these options for phase 2 (Fig. 4c) and phase 3 (Fig. 4d). A legend is provided in Fig. 4e. Note that the phase-changing composition is represented by a square in Fig. 4c,d to ensure that the legend applies to both. When changing the composition of the square phase to another location in the phase diagram, the colors at this location show the maximum amount of triangle phase to be nucleated (red scale) or mixed with (green scale) and the maximum amount of circle phase to be nucleated (blue scale) or mixed with (yellow scale). Along the red tie line, the other dense phase is predicted to be nucleated, and along the blue tie line we expect the nucleation of droplets of the dilute phase. Generally, moving the composition further away from the compositions of the circle or triangle can cause nucleation of the circle or triangle. Moving towards them requires mixing. Using this background, we can now design condensates with our chosen architecture.

As an example, we set out to prepare condensates with droplets of phases 1 and 2 in phase 3 and droplets of phases 1 and 3 in phase 2. This requires phase 2 to move in composition towards a lower concentration of DNA and a higher concentration of amyloses (Fig. 4c,e). Phase 3 should become denser in both DNA and amyloses (Fig. 4d,e). We hypothesize that this change can be achieved by lowering the ionic strength. Ions screen charges and decrease electrostatic interaction strengths. Lowering the ionic strength is thus expected to increase amylose concentrations in phases 2 and 3 in comparison to phase 1 and to increase the DNA concentration in phase 3 in comparison to phase 2, as it interacts more strongly with Q-amylose.

Figure 5a,b shows representative examples of condensates with fluorescently labeled DNA, Cm-amylose and Q-amylose at room temperature in 165 mM KCl. Previous experiments were performed at higher temperatures and 100 mM KCl. At 165 mM KCl, we obtain a multiphase condensate. We diluted this sample to obtain condensates in 55 mM KCl (Fig. 5c,d and Extended Data Fig. 6). Rapidly lowering the salt concentration causes the formation of a complex structured condensate. As most easily seen in the images with all compounds shown, droplets of phase 1 nucleate in phase 2 (gray arrowheads) and in phase 3 (white arrowheads). Additionally, phase 3 droplets nucleate in phase 2 (yellow arrowheads) and phase 2 droplets nucleate in phase 3 (orange arrowheads) (most easily seen in Fig. 5d, DNA channel). Notably, diluting the solution while keeping the salt concentration constant does not cause the formation of this architecture (Extended Data Fig. 7a). Condensates prepared immediately at 55 mM KCl also do not have this complex architecture (Extended Data Fig. 7b). This finding is specifically due to decreasing the salt concentration to 55 mM, such that we have nucleated phases 1 and 2 in phase 3 and droplets of phases 1 and 3 in phase 2, confirming our prediction.

Fig. 5: Designing condensates with complex architectures.
figure 5

We set out to create condensates with droplets of phases 1 and 2 in phase 3 and droplets of phases 1 and 3 in phase 2. Using Fig. 4c–e, we predicted that a decrease in ionic strength may achieve the desired effect. a,b, The model system contains multiphase condensates at room temperature at 165 mM KCl. c,d, Rapidly diluting the system lowers the salt concentration to 55 mM KCl, causes the nucleation of droplets of phase 1 in phase 2 (gray arrowheads) and phase 3 (white arrowheads). Additionally, droplets of phase 3 are nucleated in phase 2 (yellow arrowheads) and droplets of phase 2 are nucleated in phase 3 (orange arrowheads) (Extended Data Fig. 6). This architecture is not formed when diluting everything without changing the salt concentration (Extended Data Fig. 7a) or when preparing the condensates at 55 mM immediately (Extended Data Fig. 7b). e, The theoretical maximum volume fractions that are nucleated are determined for phase 2 to be 0.29 phase 1 and 0.09 phase 3. f, Phase 3 may nucleate a maximum of 0.22 dilute phase and 0.20 phase 2. Detailed predictions are provided in Extended Data Fig. 8.

Source data.

We can find the theoretical maximum volume fractions that we can nucleate in phases 2 (Fig. 5e) and 3 (Fig. 5f) using the method from Fig. 3d. The compositions at 165 mM are shown in blue and the composition and tie plane at 55 mM are shown in yellow. Phase 2 will nucleate a maximum of 0.29 volume fraction of phase 1 and a maximum of 0.09 volume fraction of phase 3. Phase 3 will nucleate a maximum of 0.22 volume fraction of phase 1 and a maximum of 0.20 volume fraction of phase 2. In experiments (Fig. 5c,d) we observe smaller nucleated amounts than the theoretical maximum, probably due to exchange over the interfaces facilitating composition changes. Within a short distance from the interface, we tend to observe no nucleation at all. This is most easily seen for the nucleation of phase 1 in Fig. 5c, which starts at a distance from the interface of the condensate with phase 1. Extended Data Fig. 8 shows more detailed predictions starting at 165 mM KCl.

The presented method explains the observed complex structures (Fig. 4) and enables the design of condensates with chosen complex architectures (Fig. 5). Preparing condensates under initial conditions and then changing these conditions rapidly gives access to impermanent, but often long-lived, architectures. Phase diagrams with tie lines or planes allow one to plan which complex architecture to create. If an architecture with a larger volume fraction of nucleated droplets is preferred, one can increase the rate at which the environmental change is applied (Fig. 2b), create larger condensates (a larger distance from the interface, Fig. 2b), work with denser liquids (nucleation is diffusion-limited; a lower diffusion coefficient reduces the critical distance, Fig. 2c) or force a larger composition change (Fig. 3). If a transient architecture is preferred to remain for longer, one can work with larger condensates (Fig. 2e) or store them at lower temperature (slower diffusion and fusing).

Application of complex architectures

As an example of how complex architectures can be used, we consider the case in which condensates are used as a drug-delivery system48,49. Control over the rate of cargo uptake and release is important for this application. We set out to load our cargo, GFP, into phase 3. First, we tested the ability of condensates to take up GFP, supercharged GFP with net 30 negatively charged residues (−30GFP), and an enhanced GFP coupled to a 20-nucleotide ssDNA strand (DNA–GFP) (Supplementary Methods and Fig. 6a). We found that coupling DNA to our cargo allowed it to be taken up by phase 3 of the condensates (Fig. 6b). Second, 12 condensates with a small interface area (Fig. 6c) and 12 condensates with a larger interface area (Fig. 6d) but similar final conditions were created. Third, we added DNA–GFP to the condensates and studied the uptake from phase 1 to 2 to 3. We found that condensates with a larger interface area take the cargo up into phase 3 substantially more quickly. The average intensity of GFP in phase 3 for the 24 condensates was plotted over time and fitted as a first-order reaction (Fig. 6e). The uptake time is τ = 30 ± 1 min for the simple architecture and τ = 4.1 ± 0.2 min for the complex architecture—over seven times faster. Thus, being able to control the architecture of condensates as an independent variable also yields control over the interface area and exchange rates. When cargo is added in before the complex architecture is created, it is found at equal concentrations in both small and large droplets of phase 3, confirming that size affects the uptake rate, not the final concentration (Extended Data Fig. 9).

Fig. 6: Control over cargo uptake rate via control of the condensate architecture.
figure 6

a, Confocal images of the uptake of 100 nM GFP, −30GFP or DNA–GFP into condensates. b, Although GFP and −30GFP uptake by phase 3 is weak, DNA–GFP has a strong preference for moving into phase 3, independent of the size of the droplets (n > 50 per box plot; data are presented as mean, 25th and 75th percentiles and s.d.; Extended Data Fig. 9). c,d, Lowering the [KCl] concentration in the solution before (c) or after (d) droplet formation creates a simple architecture with a small phase 2–phase 3 interface or a complex architecture with larger interface, respectively. DNA–GFP uptake is substantially faster in small droplets of phase 3, which have a relatively large interface area. e, The uptake of DNA–GFP in the condensate with a complex architecture is over five times faster than that for the simple architecture. For both the simple and complex architectures, 12 condensates of similar size were used.

Source data.

Discussion

Transient multiphase architectures can be obtained using diffusion-limited nucleation in a condensate. Using a phase diagram and tie planes, one can understand how these architectures are formed and design new architectures. One can plan what composition changes lead to which nucleation or mixing events and determine the maximum volume fractions involved. Complex architectures can be long-lived structures. Access to complex architectures will enable researchers to incorporate increasingly sophisticated compartmentalization and functionality into condensates for various applications. Additionally, insight into how complex architectures arise out of equilibrium may contribute to understanding the dynamic behavior and structural diversity of condensates.

Methods

Materials

All chemicals were used as received unless otherwise stated. Poly(ethylene glycol) monomethyl ether (2 kDa) was purchased from RappPolymere, and trimethylene carbonate (1,3-dioxan-2-one) was purchased from Actu-All Chemicals. Amylose (12–16 kDa) was supplied by Carbosynth and (3-chloro-2-hydroxypropyl) trimethylammonium chloride (65 wt% in water) by TCI Europe. DBCO-Cy3 and DBCO-Cy5 were purchased from ThermoFisher. The 38-nucleotide-long ssDNA and HPLC-purified FITC- or Cy5-labeled 38-nucleotide-long ssDNA (5′ TTTTTTTTTTCAGTCAGTCAGTCAGTCAGTCCATAAGG) were obtained from Integrated DNA Technologies (IDT). All other chemicals and reagents were supplied by Sigma-Aldrich.

Formation of multiphase condensates

The modification and synthesis of Q-amylose, Cm-amylose and terpolymer, as well as the expression of the GFP variants and DNA–GFP conjugation are described in the Supplementary Information (Extended Data Fig. 10). Q-amylose and Cm-amylose were dissolved at a concentration of 1 mg ml−1 in buffer (20 mM HEPES, 100 mM KCl, pH 7.5). DNA was dissolved in Milli-Q to reach a final concentration of 250 μM, of which 2.5 μM was either labeled with FITC or Cy5. Cy5-labeled DNA was used for Fig. 6 and Extended Data Fig. 9. In other cases, FITC-labeled DNA was used. To prepare condensates, 50 μl of buffer, 25.3 μl of Cm-amylose and 3.2 μl of ssDNA solutions were mixed in a low-DNA-binding tube, then 24.7 μl of Q-amylose solution was added while shaking at 1,500 r.p.m. at room temperature to obtain an amylose charge ratio of 2.1:1 (Q-amylose:Cm-amylose). After 1 min, 100 nM GFP variants were added for the experiments that used GFP. After 6 min, terpolymer (2.5 μl, 50 mg ml−1 in PEG350) was added to the solution to stabilize the condensates. The samples were placed in an Ibidi slide with a lid to prevent the evaporation of water during heating.

Confocal microscopy

The images in Figs. 1 and 2, as well as Extended Data Figs. 1, 2, 4 and 5, were taken with a Leica Stellaris 5 confocal microscope (white-light laser) equipped with a ×63 oil immersion Leica 1.4 NA objective. The images in Figs. 5 and 6 and Extended Data Figs. 6, 7 and 9 were taken with a Leica TSC SP8 confocal microscope equipped with 488-nm, 552-nm and 638-nm lasers, a hybrid (HyD) and photomultiplier tube (PMT) detector, and an HC PL APO CS2 ×20/0.75 dry objective. Cooling and heating experiments were performed with a TS102SI Instec rapid heating and cooling stage. Concentrations in the phase diagrams were estimated based on fluorescence intensity in comparison to a reference sample at the same pH, salt concentration and temperature. The determined concentrations are estimates, as factors such as the effect of viscosity on dye intensity were not taken into account.

Analysis

Analysis was performed and figures were prepared using Fiji (v. 2.3.051), Origin (2017) and Adobe Illustrator (v. 27.8.1).

Statistics and reproducibility

Figures 1, 2, 5 and 6 show representative images. The cooling experiment shown in Fig. 1 was repeated five times, and other condensates (≥50) were observed to undergo similar changes. Eight different-sized condensates were used for the investigations in Fig. 2d,e. Data in Fig. 4c,d are presented as the average of n ≥ 5 data points. Figure 2b includes n ≥ 10 data points per biomolecule and phase combination. The R2 values for Fig. 2e are reported in Extended Data Fig. 4d.

Reporting summary

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