Electrostatic–Hydrophobic Competition Induces Re-Entrant Phase Separation in Designer Polypeptides

Abstract

The liquid–liquid phase separation (LLPS) of intrinsically disordered proteins is governed by a delicate balance among electrostatic, hydrophobic, and aromatic interactions, yet how residue identity reshapes this balance remains incompletely understood. Although significant progress has been made toward establishing a molecular grammar linking sequence to phase behavior, the specific roles of individual hydrophobic residues in modulating the electrostatic–hydrophobic competition have not been systematically dissected. Here, we use a residue-level coarse-grained model to examine the phase behavior of artificially designed peptides with controlled charge patterning and amino acid composition. We find that even when the arrangement of charged and neutral residues is kept fixed, the identity of the neutral residue leads to qualitatively distinct phase behaviors. Phenylalanine monotonically enhances dense-phase formation, whereas alanine progressively suppresses it. In contrast, isoleucine produces a striking re-entrant behavior: increasing its fraction first weakens condensate formation by disrupting electrostatically stabilized associations, but beyond an intermediate composition, effective hydrophobic interactions become dominant and drive the re-emergence of a dense yet relatively loose and dynamic phase. These results illustrate that hydrophobic residues do not merely modulate LLPS strength, but can fundamentally alter the competition between interaction modes and thereby change the phase-separation regime. Our findings are consistent with a picture in which sequence composition modulates condensate formation via a balance among electrostatic, hydrophobic, and aromatic-type interactions, as captured by our coarse-grained model. Preliminary tests on LAF-1-RGG mutants further suggest that these trends may be transferable to biologically relevant IDRs. Together, this work provides additional insights into sequence-dependent LLPS and may inform the rational design of condensates with tunable material properties.

1. Introduction

Biological cells are not homogeneous “soups” but are instead intricate systems replete with complex structures and compartments. In recent years, liquid–liquid phase separation (LLPS) has been established as a crucial biophysical mechanism underlying the formation of membrane-less organelles in the nucleus and cytoplasm, as well as in cellular signaling and cytoskeletal assembly. , LLPS occurs when biomolecules, such as proteins or nucleic acids, undergo a phase transition from a homogeneous solution into coexisting dense and dilute phases via multivalent interactions, giving rise to various biomolecular condensates. − These dynamic, liquid-like condensates can selectively enrich molecules, accelerate biochemical reactions, and respond to environmental cues, thereby providing a new physical framework for understanding spatiotemporal complexity within cells. −

The molecular interactions driving LLPS primarily include electrostatic interactions, hydrophobic interactions, and π–cation and π–π interactions mediated by aromatic residues. ,,, Intrinsically disordered proteins (IDPs) and their low-complexity regions are frequently considered “hotspots” for phase separation due to their ability to engage in multivalent interactions. − Considerable progress has been made toward establishing a “molecular grammar” that links amino acid sequences to phase-separation behavior, as exemplified by the systematic work of Wang et al., which combined comprehensive mutagenesis and experimental validation across all amino acid types. − Building on this foundation, several key questions remain: how do electrostatic, hydrophobic, and aromatic interactions cooperate or compete within a given sequence context? What are their relative contributions to phase separation? Does a dominant interaction type govern phase-separation outcomes across diverse sequence backgrounds? In particular, it remains incompletely understood whether the competition between electrostatic attraction (between oppositely charged residues) and hydrophobic interactions differs fundamentally in its efficacy and mechanistic origin in driving phase separation. Answering these questions is crucial for moving from empirical prediction toward the rational design of phase-separating systems.

A powerful strategy to address this challenge is to use highly simplified model peptides. Such systems allow precise control over the type, order, and number of amino acids, enabling the systematic decoupling of the contributions of different interactions. By systematically altering the “molecular letters” in the sequence, it becomes possible to establish causal links from chemical structure to macroscopic properties. Furthermore, to assess whether principles derived from simple models are transferable to more complex contexts, it is essential to test them in a natural, functionally relevant biological macromolecular system. This “simple to complex” approach is the key to advancing the field from phenomenological description toward predictive understanding.

Based on these considerations, we designed and executed a hierarchical research strategy. We first focused on a series of model peptides with an alternating charged backbone of aspartate (D) and arginine (R), namely, DFR, DIR, and DAR, where the central hydrophobic residue (phenylalanine, isoleucine, or alanine) is systematically varied. This design allows us to directly compare the driving efficacy of aromatic (F) versus aliphatic hydrophobic (I, A) residues within an identical charged framework while maintaining a well-defined charge pattern. Additional peptide series (e.g., with E/K backbones) were used for initial screening and are discussed in the Supporting Information.

Second, to probe the transferability of the trends observed in the model peptides, we selected the intrinsically disordered RGG region of the well-known phase-separating protein LAF-1 as a more complex test system. We constructed the wild-type sequence (LAF-1 WT) together with mutants (LAF-1 A2F and LAF-1 A2I), which replaced all alanine residues with either phenylalanine or isoleucine. To probe the transferability of these trends to a more complex system, we designed three LAF-1 RGG mutants (WT, A2I, and A2F) in which the alanine at a key position is substituted with alanine (control), isoleucine, or phenylalanine, respectively. This allows for a direct comparison of the effects of the three residues on LLPS. The ability to promote phase separation follows the order F > I > A, indicating that aromatic phenylalanine exerts the strongest enhancement, whereas alanine, the least hydrophobic among the three, shows the weakest effect.

In summary, by integrating simplified peptide models with a complex protein system, this study aims to quantitatively dissect how specific hydrophobic residues modulate the balance between charge-driven and hydrophobic LLPS. Our findings complement existing molecular grammar frameworks by providing additional mechanistic insights into the residue-specific modulation of the phase behavior. We anticipate that this work will contribute to a more nuanced understanding of the biophysical interactions governing condensate formation and may inform future rational design strategies for synthetic biology applications.

2. Computational Methods

2.1. Studied Systems

To investigate the stability and dynamics of condensates formed by IDPs, we employed a residue-level coarse-grained molecular dynamics simulation approach ,, that has been recently applied to study LLPS in polyampholytes. , The sequences studied here comprised 40 residues, each represented by a single bead. The beads were either charged or neutral. In this context, “neutral” refers to the charge state (zero net charge) rather than hydrophilicity or hydrophobicity. Neutral beads participate in short-range dispersion interactions and, in our design, represent hydrophobic residues. We chose hydrophobic residues as the neutral beads because (i) they allow systematic tuning of hydrophobicity independently of charge, which is central to our mechanistic question, and (ii) their interaction parameters in the HPS model are well-validated for LLPS studies. Hydrophilic uncharged residues (e.g., glycine, serine) were not considered, as they would introduce additional variables, such as backbone flexibility and hydrogen-bonding capacity, that are beyond the scope of this study. It is important to note that the conclusions drawn here are specific to hydrophobic neutral residues and should not be generalized to all neutral residue types without further investigation.

The reference peptide without hydrophobic residues (ϕ = 0) contained 20 positively and 20 negatively charged residues arranged with a charge-patterning parameter κ = 0.55. This mid-range κ value was previously shown to correspond to intermediate liquid-like properties of the condensate. From this fully charged reference, we generated seven additional peptides with hydrophobic residue fractions (ϕ) ranging from 0.05 to 0.80 (i.e., 2 to 32 hydrophobic residues). To probe hydrophobic interactions of varying strengths, we introduced three different hydrophobic residues: alanine (A), isoleucine (I), and phenylalanine (F). The hydrophobic fraction ϕ is defined as the number of hydrophobic residues divided by the total chain length (N = 40 for the model peptides). For the tripeptide systems, ϕ ranges from 0 (no hydrophobic residue) to 1.0 (fully hydrophobic trimer). For these tripeptide systems, we use the shorthand notation X₄₀ (where X = I, F, or A) to denote the fully hydrophobic sequences with ϕ = 1.0 (e.g., III, FFF, or AAA). The exact sequences of the studied peptides are provided in Tables S1–S6.

In addition to these model peptides, we also investigated mutants of the LAF-1-RGG protein, including the wild-type (WT), A2I, and A2F variants, to test the transferability of our findings to a biologically relevant system. The hydrophobic fraction ϕ = 0.732 for the LAF-1-RGG mutants, consistent with the model peptide series.

2.2. Coarse-Grained Model

We employed the hydropathy scale (HPS) coarse-grained model developed by Dignon et al. − to represent IDP sequences as flexible chains with a single bead per residue. This model and its variants (including the Mpipi model) have been extensively validated against experimental data, including radii of gyration and quantitative in vitro phase diagrams for multiple IDP systems such as FUS and LAF-1. − Directly bonded residues interacted via a harmonic potential

$$U_{\mathrm{b}\mathrm{o}\mathrm{n}\mathrm{d}}(r_{\mathit{ij}})=\frac{1}{2}{k}_{\mathrm{b}}{(r_{\mathit{ij}}-r_0)}^2$$ 1

with spring constant k _b = 20 kcal/(mol·Å²) and equilibrium bond length r ₀ = 3.8 Å. Nonbonded interactions were modeled using a modified Lennard–Jones potential, where the attraction between residues i and j is scaled by their average hydropathy λ _ij

$$\lambda =\frac{{\lambda }_i+{\lambda }_j}{2}$$ 2

$$U_{\mathrm{n}\mathrm{o}\mathrm{n}\mathrm{b}\mathrm{o}\mathrm{n}\mathrm{d}}(r_{\mathit{ij}})=U_{\mathrm{L}\mathrm{J}}(r_{\mathit{ij}})+(1-{\lambda }_{\mathit{ij}}){\epsilon }_{\mathit{ij}}$$ 3

$${U_{\mathrm{n}\mathrm{o}\mathrm{n}\mathrm{b}\mathrm{o}\mathrm{n}\mathrm{d}}(r_{\mathit{ij}})={\lambda }_{\mathit{ij}}U}_{\mathrm{L}\mathrm{J}}(r_{\mathit{ij}})$$ 4

The standard Lennard–Jones potential is

$$U_{\mathrm{L}\mathrm{J}}(r_{\mathit{ij}})=4{\epsilon }_{\mathit{ij}}[{\left(\frac{{\sigma }_{\mathit{ij}}}{r_{\mathit{ij}}}\right)}^{12}-{\left(\frac{{\sigma }_{\mathit{ij}}}{r_{\mathit{ij}}}\right)}^6]$$ 5

where ε _ij = 0.2 is the interaction strength and ${\sigma }_{\mathit{ij}}=\frac{{\sigma }_i+{\sigma }_j}{2}$ is the average diameter. The potential was truncated at r _c = 3σ _ij . Charged residues also interacted via a Coulomb potential with Debye–Hückel electrostatic screening

$$U_{\mathrm{e}\mathrm{l}\mathrm{e}}(r_{\mathit{ij}})=\frac{q_i{q}_j}{4\pi D{r}_{\mathit{ij}}}{\mathrm{e}}^{-r_{\mathit{ij}}/K}$$ 6

with relative permittivity D = 80 and Debye screening length K = 10 Å, corresponding to a physiological salt concentration of approximately 100 mM. The electrostatic potential was truncated at 3.5 nm. We acknowledge that systematic variation of ionic strength could further modulate the observed phase behavior, particularly the charge–hydrophobicity balance and the re-entrant transition thresholds. All simulations in this study were performed at this fixed physiological salt concentration to isolate sequence-specific effects, and future work incorporating salt-dependent screening effects will be important to generalize our findings to broader solution conditions.

2.3. Simulation Protocols

2.3.1. Cubic Box Simulations

For bulk phase characterization, systems containing 200 chains were simulated in a cubic box of side length 30 nm under NVT conditions. Simulations were performed for 1 μs, with some trajectories extended to 5 μs for the LAF-1-RGG systems to ensure convergence. To assess convergence, we monitored the time evolution of the density. These quantities reached stable plateaus within the first 500 ns for the tripeptide systems and within approximately 1 μs for the LAF-1-RGG systems (Figure S1); therefore, data from the subsequent production windows, 500–1000 ns for tripeptides and 1–5 μs for LAF-1-RGG, were used for all analyses. For the LAF-1-RGG protein, cubic boxes of both 30 and 50 nm side lengths were used to check system-size effects.

2.3.2. Slab Simulations

To characterize the surface tension and enable phase coexistence, slab geometry simulations were performed. The same number of chains (200) were initially placed as a dense slab within a rectangular box (L _x = L _y = 15 nm, L _z = 280 nm). Simulations were carried out for 1 μs, with the last 0.5 μs used for analysis.

All simulations were performed using HOOMD-Blue (version 2.9.7) with azplugins (version 0.10.1). A Langevin thermostat with a damping time of 1 ns was used to maintain the temperature. The time step was 10 fs, and periodic boundary conditions were applied in all three dimensions.

2.4. Data Analysis

Statistical uncertainties were estimated using a block-averaging procedure. For each trajectory, the production window (500–1000 ns for tripeptide systems; 1000–5000 ns for LAF-1-RGG systems) was divided into five consecutive blocks of equal length. The mean and standard deviation of each observable (e.g., cluster size, dense-phase fraction, contact frequency) were calculated from the block averages. All error bars reported in the figures correspond to the standard deviation across these five blocks.

2.4.1. Structural Analysis

Structural characterization included the calculation of intrachain and interchain contact numbers, where a contact was defined as a distance of less than 0.6 nm between any two beads, and the distribution of the radius of gyration (R _g) for individual chains in both dense and dilute phases.

$$R_{\mathrm{g}}=\sqrt{\frac{\sum _{i=1}^N{m_i(r_i-r_{\mathrm{c}})}^2}{\sum _{i=1}^N{m}_i}}$$ 7

2.4.2. Dynamics Analysis

To quantitatively characterize phase separation, we performed cluster analysis and calculated an order parameter. Clusters were identified using a distance-based criterion: two beads were considered to belong to the same cluster if their pairwise distance was less than 0.6 nm. To test the sensitivity of our results to this choice, we performed a systematic analysis using distance cutoffs ranging from 0.8 to 1.2 nm (corresponding to approximately 0.5–0.7 nm in bead-to-bead distance) and minimum cluster sizes ranging from 2 to 5 chains. The overall trends, monotonic enhancement by phenylalanine, U-shaped re-entry by isoleucine, and suppression by alanine, were robust across all tested parameters. Based on these sensitivity tests, we chose a cutoff of 0.6 nm and a minimum cluster size of 3 chains for all reported analyses. The order parameter was defined as the normalized density difference between the dense and dilute phases

$$O=\frac{{\rho }_{\mathrm{d}\mathrm{e}\mathrm{n}\mathrm{s}\mathrm{e}}-{\rho }_{\mathrm{d}\mathrm{i}\mathrm{l}\mathrm{u}\mathrm{t}\mathrm{e}}}{{\rho }_{\mathrm{d}\mathrm{e}\mathrm{n}\mathrm{s}\mathrm{e}}+{\rho }_{\mathrm{d}\mathrm{i}\mathrm{l}\mathrm{u}\mathrm{t}\mathrm{e}}}$$ 8

where ρ_dense and ρ_dilute are the average number densities of chains in the two coexisting phases. Alternatively, we computed the fraction of chains residing in the largest cluster f _large, as an order parameter; f _large approaches 1 under strong phase separation and remains low in a homogeneous state. By monitoring the distribution of cluster sizes and the evolution of O over simulation time, we were able to determine the onset of phase separation, distinguish between microphase segregation and macroscopic condensation, and assess the stability of the resulting biomolecular condensates.

3. Results and Discussion

In this study, we systematically investigated how the identity and proportion of hydrophobic residues modulate the phase-separation behavior of net-neutral polyampholytes. , Using a residue-level coarse-grained model, we were able to isolate the contributions of different interactions and uncover a broad spectrum of phase behaviors that depend critically on both hydrophobic residue identity and the specific charged residues present. To construct a comprehensive picture, − we systematically computed the phase diagrams for all systems examined (Figures S2–S29), including both the model peptides and the LAF-1-RGG variants, as a function of temperature and hydrophobic composition.

3.1. Re-Entrant Phase Behavior in DIR Sequences

At 300 K, the DIR sequences exhibit a nonmonotonic transition in their phase behavior as the fraction of isoleucine (ϕ) increases (Figure ). When ϕ = 0 (fully charged D/R alternating sequence), a well-defined dense phase forms due to strong electrostatic attractions between oppositely charged residues. As ϕ increases to 0.2 and 0.4, the dense phase progressively dissolves, leading to a more dispersed, loose structure. At ϕ ≈ 0.6, the system appears most disorganized, with minimal dense-phase formation. Surprisingly, upon further increasing ϕ to 0.8 and 1.0 (fully hydrophobic I₄₀), a dense phase re-emerges, although its structure is looser and more dynamic than that of the purely electrostatic condensate. This re-entrant behavior, , where the system transitions from dense → dilute → dense as the hydrophobic content increases, is a key observation of this study.

1.

Phase snapshots of DFR, DIR, and DAR sequences at 300 K as functions of hydrophobic residue fraction ϕ. DFR shows monotonic enhancement of dense-phase formation with increasing ϕ; DIR exhibits a re-entrant transition: dense → dilute → dense; DAR shows monotonic suppression of phase separation.

3.2. Monotonic Enhancement in DFR Sequences

In stark contrast to DIR, DFR sequences exhibit a monotonic increase in dense-phase stability as the fraction of phenylalanine (ϕ) increases (Figure ). At all ϕ values, a dense phase is observed, and its density increases continuously with ϕ. Even at ϕ = 0.1, the condensate is more compact than the pure electrostatic D/R system. This indicates that phenylalanine strongly promotes phase separation, consistent with its higher hydrophobicity and capacity to form effective attractive interactions with neighboring residues, as captured by the parametrized interaction potentials in this CG model.

2.

Dense-phase density as a function of hydrophobic residue fraction ϕ at 300 K. DFR (red circles) shows monotonic increase; DIR (blue triangles) shows re-entrant behavior with a minimum at ϕ ≈ 0.6; DAR (black squares) shows monotonic decrease. Error bars represent the standard deviations calculated from five blocks of 100 ns each, sampled from the last 500 ns of each trajectory (i.e., the production window spanning 500–1000 ns).

3.3. Monotonic Suppression in DAR Sequences

For DAR sequences, alanine (A) progressively disrupts phase separation (Fig. ). At ϕ = 0.1, a dense phase is still visible but less compact than the pure D/R system. As ϕ increases to 0.2 and beyond, the dense phase gradually disappears, and at ϕ = 1.0 (A₄₀), the system remains completely homogeneous. Alanine, being the least hydrophobic among the three residues tested, fails to provide sufficient attractive interactions to overcome the loss of electrostatic attraction upon mutation.

3.4. Molecular Mechanism of Re-Entrant Behavior: Contact Analysis

To elucidate the molecular origin of the re-entrant phase behavior observed in the DIR sequences, we quantified the number of interchain contacts as a function of the hydrophobic residue fraction ϕ (Figure ). Additional contact analysis for DFR (Figure S29), DAR (Figure S30), EFK (Figure S31), EIK (Figure S32), and EAK (Figure S33) can be found in the Supporting Information.

3.

Number of intrachain (square) and interchain (circle) contacts at 300 K for DIR sequences. Total interchain contacts decrease with increasing ϕ, reaching a minimum at ϕ ≈ 0.6, and partially recover at higher ϕ, consistent with the re-entrant phase behavior shown in Figures and . Error bars represent SD from block averaging.

For the DIR system, the total number of interchain contacts initially decreases with increasing ϕ, reaching a minimum at ϕ ≈ 0.6, and then gradually recovers at higher ϕ, albeit without returning to the level of the purely electrostatic system (ϕ = 0). This nonmonotonic trend directly reflects the competition between electrostatic and hydrophobic interactions. At low ϕ, the incorporation of isoleucine (I) disrupts the electrostatic attraction formed between aspartic acid (D) and arginine (R), weakening interchain attractions and leading to a decrease in contacts. When ϕ exceeds ∼0.6, the electrostatic network is largely replaced by hydrophobic interactions, which re-drive chain aggregation and increase the number of contacts. However, the contacts formed at high ϕ are less numerous and more dynamic than those in the purely electrostatic regime, consistent with the looser and more flexible structure observed in snapshots. Thus, the contact analysis demonstrates that the re-entrant phase behavior , originates from a gradual crossover from electrostatic-dominant to hydrophobic-dominant interactions, with a minimal contact density at the intermediate composition where neither driving force is sufficient to sustain a stable condensed phase.

3.5. Dynamic Properties of Condensates for DIR sequence

To quantitatively characterize the progression of phase separation, we computed the order parameter O for each system (Figures S37–S44), defined as the normalized density difference between the dense and dilute phases (see the Methods section). This parameter ranges from 0 for the dilute phase to 1 for the dense phase, and its time evolution provides a robust measure of condensate stability and dynamics. Figure shows the time evolution of the largest cluster size (top panel of each subplot) and the order parameter (bottom panel) at ϕ = 0.0, 0.2, 0.4, 0.6, 0.8, and 1.0 (from left to right). For ϕ = 0.0 (purely electrostatic D/R system), a large, dense cluster rapidly forms, and the order parameter quickly approaches unity, indicating complete phase separation. As ϕ increases to 0.2 and 0.4, both the largest cluster size and the order parameter decrease significantly, reflecting disruption of the electrostatic network by the hydrophobic isoleucine residues. At ϕ = 0.6, the system exhibits the weakest phase separation, with only small transient clusters and an order parameter barely above zero. Interestingly, at ϕ = 0.8 and 1.0, both the cluster size and order parameter recover, albeit to lower levels than the ϕ = 0.0 case, demonstrating a re-entrant behavior driven by hydrophobic interactions. These trends directly correlate with the contact analysis (Figure ) and highlight the competition between electrostatic and hydrophobic forces that governs the phase behavior of DIR sequences.

4.

Dynamics of cluster growth and phase separation for the DIR sequence at varying hydrophobic fractions (ϕ). Each subplot corresponds to a different ϕ value, ranging from 0.0 to 1.0 (left to right). In each panel, the top traces display the corresponding order parameter, which measures the extent of macroscopic phase separation (ranging from 0 to 1), while the top traces show the time evolution of the size of the largest cluster (defined as the number of chains in the largest aggregate). For ϕ = 0.0 (purely electrostatic D/R system, first column), a large cluster forms rapidly within ∼100 ns, and the order parameter quickly approaches unity, indicating fast and complete liquid–liquid phase separation (LLPS). As ϕ increases to 0.2 and 0.4, the dense phase becomes less stable, as reflected by smaller cluster sizes and lower plateau values of the order parameter, consistent with charge-screening effects that suppress condensation. At intermediate ϕ = 0.6, partial re-entry into a denser state is observed, and at ϕ = 0.8 and 1.0, the system again shows robust cluster growth and near-complete phase separation, driven by enhanced hydrophobic interactions.

3.6. Transferability to the LAF-1-RGG Protein

To assess the transferability of these findings to a natural protein system, we simulated the LAF-1-RGG domain and its mutants (Figure ). The wild-type LAF-1-RGG (ϕ ≈ 0.723, with multiple aliphatic residues) formed a moderately dense condensate. The A2F mutant, in which all alanine residues were replaced by phenylalanine, exhibited a markedly denser and more stable condensate. In contrast, the A2I mutant showed a reduced dense-phase density compared to A2F, and the WT control (no substitution) exhibited the lowest dense-phase density. The observed increase in dense-phase density for the LAF-1-A2I mutant at elevated isoleucine content (ϕ ≈ 0.7–0.8) is consistent with the rising branch of the re-entrant phase diagram (Figures S34–S36). At this composition, the system has surpassed the minimum of the U-shaped curve (located at ϕ ≈ 0.6 in the model peptides), and effective hydrophobic interactions now drive chain association. This trend mirrors the behavior seen in the DIR model peptides (Figure ) and reinforces the notion that isoleucine can promote phase separation at sufficiently high fractions, despite its disruptive effect at intermediate compositions. The ability to promote phase separation follows the order F > I > A, consistent with the increasing hydrophobicity of the substituted residue and with the trends observed in the model peptide systems. We acknowledge that testing on only two mutants is insufficient to establish a general rule; however, these results provide a preliminary indication that the residue-specific trends identified in the simple tripeptide models are qualitatively recapitulated in a more complex IDR context.

5.

Representative snapshots of the wild-type (A), A2I (B), and A2F (C) LAF-1-RGG protein at varying temperatures. Each snapshot is taken from the equilibrated production phase of 5 μs simulations.

3.7. Comparison with E/K-Based Model Peptides

To assess whether the trends observed in the D/R-based model peptides are generalizable to other charged backbones, we also performed simulations on a series of E/K-based sequences (alternating glutamic acid and lysine) with the same hydrophobic substitutions (alanine, isoleucine, and phenylalanine) across the same range of hydrophobic fractions (ϕ = 0.0 to 1.0). In contrast to the D/R systems, none of the E/K-based peptides formed stable dense-phase condensates at 300 K across the entire ϕ range examined (Figures S31–S33). The systems remained homogeneous with only transient, small clusters that did not coalesce into a macroscopic dense phase. This striking difference underscores the importance of charged residue identity beyond net charge and charge patterning. While both D/R and E/K backbones share the same charge periodicity (κ = 0.55) and net-neutral composition, the E/K sequences fail to drive LLPS under the same conditions. This behavior likely arises from the longer and more flexible side chains of glutamic acid (E) and lysine (K), which, in the coarse-grained model, led to weaker effective interchain interactions compared to the more compact side chains of aspartic acid (D) and arginine (R). These results indicate that the phase behavior of polyampholytes is governed not only by charge patterning and hydrophobic content but also by the specific chemical nature of both the charged and neutral residues, consistent with the residue-specific hierarchy observed throughout this work.

3.8. Role of Specific Charged Residues: R₅₀ vs K₅₀

To further dissect the contribution of residue identity beyond charge state, we simulated poly-arginine (R₅₀) and poly-lysine (K₅₀) with electrostatic interactions completely suppressed (Figure ). Under these conditions, K₅₀ remained homogeneous across the temperature range examined, with a consistently low density of approximately 33 mg/mL. In contrast, R₅₀ still exhibited clear phase separation even without electrostatics, reaching a dense-phase density of 606 mg/mL at 380 K. This striking difference highlights the unique effective interaction capability of arginine relative to lysine, which is captured by the parametrized LJ interaction strengths in the HPS model. Whereas lysine remains unable to sustain condensation in the absence of electrostatic attraction, arginine can still promote intermolecular association, indicating that charged residue identity contributes to phase separation through mechanisms beyond simple charge effects, specifically, through differences in effective contact strengths that serve as a coarse-grained proxy for more complex molecular features (e.g., hydrogen-bonding capacity and side-chain geometry).

6.

Density profiles and phase behavior of (A) K₅₀ and (B) R₅₀ at various temperatures with electrostatic interactions suppressed. R₅₀ exhibits clear phase separation even in the absence of electrostatics, whereas K₅₀ remains homogeneous across all temperatures.

4. Discussion

Recent theoretical and simulation studies have further dissected the role of short-range interaction types and sequence patterns. For example, they systematically investigated how the balance between hydrophobic and electrostatic interactions dictates phase separation, while exploring the effect of residue-specific contact energies on condensate stability. − Our work complements these studies by focusing on the specific modulation by phenylalanine vs isoleucine within a charge-patterned framework. In this study, we systematically investigated how the identity and proportion of hydrophobic residues modulate the phase-separation behavior of net-neutral polyampholytes. Using a residue-level coarse-grained model, we were able to isolate the contributions of different interactions and uncover a broad spectrum of phase behaviors that depend critically on both the hydrophobic residue identity and the specific charged residues present.

Our results show that phenylalanine (F) is a strong promoter of phase separation. Its effect is monotonic: increasing the fraction of F leads to denser and more stable condensates, with higher surface tension and slower dynamics. This behavior is consistent with the higher effective interaction strength associated with aromatic residues in the HPS parametrization, which serves as a coarse-grained proxy for the combined effects of enhanced hydrophobicity and attractive contacts involving aromatic side chains.

Alanine (A), in contrast, acts as a weak hydrophobic residue that cannot compensate for the loss of electrostatic attraction when charged residues are replaced. As the A fraction increases, phase separation is progressively suppressed, and the system becomes homogeneous at high ϕ. This behavior indicates that not all hydrophobic substitutions contribute equivalently to condensate stabilization.

The behavior of isoleucine (I) is more complex and reveals a re-entrant phase transition. At low ϕ, introducing I initially disrupts the electrostatic network formed by aspartic acid (D) and arginine (R), causing the dense phase to weaken. At intermediate ϕ, around 0.6, the system reaches its most disordered and dynamic state, with minimal contacts and high chain mobility. Beyond this point, as isoleucine becomes the dominant residue type, hydrophobic interactions begin to outweigh the loss of electrostatic attraction, and a new dense phase re-emerges. However, this hydrophobically stabilized condensate remains looser and more dynamic than the original electrostatically stabilized one. The DIR sequences therefore illustrate a competition between two interaction regimes, with the minimum in condensation appearing at the composition where neither driving force is sufficiently dominant.

The comparison between D/R-based and E/K-based systems further highlights the importance of charged residue identity. Despite having the same net charge and similar charge patterning, the E/K sequences fail to phase separate at 300 K under conditions where the D/R sequences readily do so. This difference may arise from the longer and more flexible side chains of glutamic acid (E) and lysine (K), which can weaken effective chain–chain interactions, as well as from the stronger effective contact strength of arginine (R) relative to lysine (K) in the CG parametrization, which serves as a proxy for differences in side-chain chemistry beyond simple electrostatics.

These interaction trends were further supported in a biologically relevant protein system through simulations of LAF-1-RGG and its mutants. Replacing aliphatic residues with phenylalanine enhanced condensate formation, whereas substitution with isoleucine produced a weaker effect, with the promoting ability following the order F > I > A, directly mirroring the trends observed in the DFR and DIR model systems. This consistency suggests that the principles identified from simplified peptide models remain informative for understanding and guiding sequence perturbations in naturally occurring phase-separating proteins. However, we acknowledge that testing on only two mutants is insufficient to establish a general rule, and broader experimental and computational validation will be required to determine the universality of our findings.

It is important to note that the conclusions drawn here are specific to hydrophobic neutral residues within the framework of the HPS coarse-grained model. Hydrophilic uncharged residues may exhibit qualitatively different effects due to distinct solvation and conformational properties. Furthermore, our CG model uses effective, isotropic interaction potentials and does not resolve explicit side-chain geometry or directional noncovalent bonds (e.g., π–π stacking or cation−π). The terms “aromatic interactions” or “arginine contacts” in this work refer to the enhanced effective attraction captured by the parametrized LJ and electrostatic terms, which serve as coarse-grained proxies for these complex molecular forces. All simulations were performed at a fixed physiological salt concentration (∼100–150 mM); systematic variation of ionic strength could further modulate the observed phase behavior, particularly the charge–hydrophobicity balance and re-entrant transition thresholds. Despite these limitations, the qualitative trends observed across both model peptides and the LAF-1-RGG system suggest that the residue-specific hierarchy identified here is robust and may inform future rational design strategies.

5. Conclusions

In this work, we used residue-level coarse-grained simulations to determine how the hydrophobic residue identity shapes the phase behavior of net-neutral polyampholytes with controlled charge patterning. Our results corroborate the established view that phase separation depends not only on overall hydrophobic content but also on the specific interaction modes conferred by different residue types. − Even with the same overall arrangement of charged and neutral residues, different amino acids produce qualitatively distinct phase behavior.

A clear interaction hierarchy emerged. Phenylalanine consistently strengthens condensate formation, alanine progressively suppresses it, and isoleucine produces a re-entrant transition from dense to dilute and back to dense as its fraction increases. This nonmonotonic behavior reflects a crossover between two driving regimes: an electrostatically stabilized condensed state at a low isoleucine content and a hydrophobically stabilized, looser condensed state at a high isoleucine content. The contrast between D/R- and E/K-based systems further shows that the charged residue identity also matters, indicating that LLPS is controlled by residue-specific interaction balance rather than charge patterning alone.

Our findings are consistent with a picture in which sequence composition modulates condensate formation via a balance among electrostatic, hydrophobic, and effective aromatic interactions, as captured in the coarse-grained mode. − The consistent trends observed in LAF-1-RGG mutants further suggest that principles derived from simplified designer peptides can help guide the interpretation and engineering of biologically relevant condensates. More broadly, this work illustrates that changing residue identity can do more than strengthen or weaken LLPS; it can shift the dominant driving force and thereby alter the phase-separation regime itself, providing additional insights that complement existing molecular grammar frameworks.

Glossary

Abbreviations

LLPS

liquid–liquid phase separation

IDPs

intrinsically disordered proteins

HPS

hydropathy scale.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.6c05668.

• Sequences of the studied polypeptides (Tables S1–S6); simulation setup and trajectory information (Table S7); convergence test results (Figure S1); phase diagrams (Figures S2–S28); contacts versus order parameter relationships (Figures S29–S33); density profiles for LAF-1-RGG wild-type, A2I, and A2F mutants at 300 K from 5 μs equilibrated trajectories (Figure S34); temperature dependence of the unbound monomer fraction for the same mutants (Figure S35); dense-phase density of LAF-1-RGG mutants as a function of isoleucine content at 300 K (Figure S36); cluster dynamics for DIR at different order parameters (Figures S37–S43); and order parameter O versus volume fraction ϕ at 300 K for DFR, DIR, DAR, EFK, EIK, and EAK (Figure S44) (PDF)

∥.

W.-B.K., J.G., D.-C.W., and G.Z. contributed equally to this work. W.-B.K. and D.-C.W.: conceptualization, software, investigation, and writing of the original draft. G.Z. and L.B.: methodology, investigation, data curation, and visualization. E.-B.H., M.-C.L., and L.-T.C.: conceptualization, supervision, and review and editing.

This study was supported by the National Natural Science Foundation of China (Grant No. 11947006), the Joint Hubei Provincial Natural Science Foundation (Grant No. 2026AFC0314), the Key Project of the Department of Education of Hubei Province (Grant No. D20242104), and the Guangxi Natural Science Foundation (2026GXNSFHA00640263).

The authors declare no competing financial interest.