Introduction
Membrane remodeling and transient pore formation are central to membrane fusion, pore-forming peptides and toxins, immune-cell cytotoxicity, and field-assisted membrane permeabilization.− Electroporation transiently increases membrane permeability and is widely used in molecular delivery, ablation, and bioelectrochemical applications.− Recent studies on electrochemical tissue treatment, pulse-waveform effects, protein-containing membranes, and electrofusion have broadened the mechanistic context of electrically driven membrane remodeling.− Nevertheless, how cholesterol concentration reshapes the pore-formation free-energy landscape remains insufficiently quantified. At the molecular scale, electroporation involves water-defect nucleation, hydrophilic pore formation, pore expansion, and lipid rearrangement.,, Quantifying the free-energy cost of these steps is therefore essential for connecting membrane composition to field-assisted permeabilization.
The transmembrane potential (TMV) is a key electrostatic variable that couples applied fields to water-defect growth, hydrophilic pore stabilization, and pore expansion. Continuum electroporation models describe TMV as lowering pore-formation free energy through a voltage-dependent electrostatic term, whereas atomistic simulations resolve the accompanying water orientation, lipid-headgroup rearrangement, and pore-edge restructuring.,, Because TMVs imposed by external electric fields or charge imbalance can yield consistent electroporation PMFs when treated appropriately, voltage-resolved PMFs are needed to separate cholesterol-dependent changes in poration energetics from direct electrostatic barrier lowering, particularly in cholesterol-containing bilayers with altered pore onset and field sensitivity.,
Earlier atomistic studies established that cholesterol alters field-driven pore onset, but their primary end points were finite-time operational fields and kinetics rather than a continuous poration free-energy profile. Casciola et al. compared POPC/Chol bilayers using protocols mimicking low-intensity millisecond and high-intensity nanosecond pulses. In threshold-bracketing simulations of up to 100 ns, the nanosecond-protocol threshold increased steadily and was approximately doubled at 50 mol % Chol, whereas the millisecond-protocol threshold increased by approximately 50% at 30 mol % Chol and then plateaued; the two protocols also produced distinct pore morphologies.
Fernández et al. defined the minimum porating field E0 from triplicate 25 ns DOPC/Chol simulations using 25 mV nm–1 field increments. E0 was 325 mV nm–1 at 0 and 10 mol % Chol and increased to 475, 600, and 750 mV nm–1 at 20, 30, and 40 mol % Chol, respectively, while individual poration times remained stochastic. These studies demonstrated that cholesterol content and electrical protocol jointly influence finite-time pore onset, but they did not resolve how the free-energy cost is distributed among polar-defect nucleation, metastable pore states, and subsequent pore expansion.
Experimental measurements likewise show that cholesterol effects depend on lipid composition and on the operational end point. In planar phosphatidylcholine bilayers, the minimum breakdown voltage increased with cholesterol loading up to 50 mol % and then decreased at 80 mol %. 1,2-diphytanoyl-sn-glycero-3-phosphocholine (DPhPC) bilayers displayed a nonmonotonic threshold response, with reduced electrical stability at higher cholesterol loading.
Chronopotentiometric measurements on egg-yolk phosphatidylcholine bilayers also reported an increased breakdown potential in the presence of cholesterol, whereas experiments on DOPG/DOPC giant unilamellar vesicles found that increasing cholesterol reduced the probability and rate constant of pore formation while increasing the inferred line tension and prepore energy barrier. Together, these measurements provide method-specific benchmarks for cholesterol-dependent pore onset and mechanics that complement the atomistic pore-formation PMFs examined here.
Molecular dynamics simulations combined with enhanced sampling provide a route to pore-formation free-energy landscapes. However, umbrella-sampling calculations require reaction coordinates that capture both polar-defect formation and pore expansion. Several earlier reaction coordinates showed hysteresis, incomplete sampling, or limited continuity between the nucleation and expansion regimes.,− The joint reaction coordinate proposed by Hub integrates acyl-chain density and pore-radius-like expansion, enabling continuous PMF calculations and estimation of pore-edge line tension., Related atomistic free-energy studies have extended poration simulations to peptide-, drug-, and polycation-perturbed bilayers.− Recent all-atom calculations across complex biological and model membranes further separated pore-nucleation free energy from pore-rim line tension: the former was associated primarily with lipid-tilt energetics, whereas the latter was linked more strongly to bending energy and intrinsic curvature; the same work also showed that lateral lipid sorting can make the local pore-rim composition differ from the nominal bulk composition.
Cholesterol is an endogenous membrane regulator that changes bilayer thickness, packing, lateral diffusion, acyl-chain order, spontaneous curvature, and line tension.− These effects are expected to alter electroporation, but the connection between cholesterol concentration, applied transmembrane potential, pore-nucleation barriers, metastable pore states, and pore expansion remains insufficiently resolved. A free-energy description is therefore needed to distinguish thermotropic phase behavior from the free-energy cost of water-defect nucleation and pore expansion, and from operational pore-onset behavior in direct-field simulations.
To address this gap, we examine how cholesterol concentration and nominal TMV reshape the poration landscape in a controlled binary 1,2-dimyristoyl-sn-glycero-3-phosphocholine/cholesterol (DMPC/Chol) membrane model. The 300 K DMPC/Chol series provides a common phospholipid background and literature-based phase context across 0–66 mol % Chol, while matched DMPC/Chol, POPC/Chol, and DOPC/Chol systems containing 20 mol % Chol resolve lipid-dependent responses at fixed sterol loading. By combining a continuous joint reaction coordinate, umbrella-sampling PMFs, and independent direct-field trajectories, we quantify nucleation-region energetics, metastable states, pore-expansion energetics, and pore-edge line tension while distinguishing composition-dependent membrane responses from direct electrostatic barrier lowering.
This design resolves how membrane composition and applied field jointly remodel the pore-formation landscape within a controlled binary-membrane framework.
Results and Discussion
Figure A provides the phase context for the 300 K DMPC/Chol concentration series. Near this temperature, literature phase diagrams place the liquid-disordered (ld) regime below approximately 6 mol % Chol, coexistence of ld and liquid-ordered (lo) states at approximately 6–27 mol % Chol, the lo regime at approximately 27–50 mol % Chol, and a structured lo region containing cholesterol bilayer domains (CBDs) at approximately 50–66 mol % Chol., Accordingly, the sampled 0 mol % system lies in the ld regime; 10 and 20 mol % lie in the coexistence region; 33 mol % lies in the lo regime; 50 mol % is near the lo/CBD boundary; and 66 mol % is retained as the high-Chol boundary case. These literature-based ranges provide the phase context for interpreting the sampled compositions.
Figure B depicts planar, splay, twist, and bend deformation modes through which cholesterol-dependent packing can modify bilayer mechanics, whereas Figure D distinguishes the direct electrostatic contribution of the applied field from the composition-dependent membrane response. For the 300 K DMPC/Chol concentration series, DMPC was retained as the phospholipid background, and all systems were compared within a common CHARMM36/TIP3P and joint-reaction-coordinate framework. The subsequent comparison of DMPC/Chol, POPC/Chol, and DOPC/Chol at a fixed 20 mol % Chol was performed separately to assess lipid-dependent responses.
Unbiased MD simulations were performed to characterize lipid bilayer properties at varied Chol concentrations, followed by enhanced sampling to construct free-energy landscapes of pore formation. Under simulated TMVs, pore-formation free-energy profiles were analyzed across defect nucleation and pore expansion, together with state-resolved hydration and water connectivity. Baseline-corrected relative electrostatic-potential profiles and pore-centered density and diameter analyses were then used as supporting electrostatic and structural evidence for the PMF interpretation.
Cholesterol Modifies DMPC Membrane Packing and Ordering
Figure summarizes the structural and dynamical properties of the unbiased 512-lipid DMPC/Chol bilayers simulated with CHARMM36/TIP3P at 300 K without an applied electric field. The mean phosphate-to-phosphate thickness increased from approximately 3.82 nm for pure DMPC to 4.47 nm at 33 mol % Chol, followed by a modest decrease to approximately 4.31 nm at 66 mol % Chol (Figure A). The initial thickening observed at 300 K is consistent with X-ray measurements of saturated DMPC/Chol membranes at 303 K (30 °C), which reported cholesterol-dependent increases in bilayer thickness and acyl-chain order.
To quantify molecular area, the composition-weighted mean was calculated as Alipid(xc) = [NDMPCADMPC(xc) + NCholAChol(xc)]/(NDMPC + NChol), where xc denotes the Chol mole fraction and NDMPC and NChol are the numbers of DMPC and Chol molecules, respectively; this formulation avoids an equal-area assumption. Here, AChol(xc) = a0/cos θ(xc), with a0 = 0.36 nm2.Figure B shows that Alipid decreased from approximately 0.56 nm2 in pure DMPC to approximately 0.39 nm2 at 50–66 mol % Chol, whereas AChol remained near 0.36–0.37 nm2 over the Chol-containing systems and ADMPC increased again at the highest loading. The simulated Chol area is close to the approximately 0.37 nm2 experimental partial area reported for Chol in DMPC membranes at 50 mol %.
Membrane-area equilibration and cholesterol tilt-angle distributions supporting this analysis are provided in Figures S3 and S5.
Chol molecules reside among the DMPC acyl chains, while phosphocholine headgroups help shield the sterol hydrophobic surface from water, as described by the cholesterol umbrella/condensing framework., Consistent with this organization, the solvent-accessible surface area of the lipid hydrocarbon tails decreased from approximately 50 nm2 in pure DMPC to approximately 33 nm2 at 66 mol % Chol (Figure C). The decrease indicates reduced hydrocarbon exposure as packing changes with composition. The head-to-tail volume-fraction ratio, fhead/tail, also decreased with xc (Figure D); its relevance to pore-edge curvature is discussed in Section .
The in-plane lateral diffusion coefficient, Dlat, was calculated from Einstein’s relation, ⟨dxy2⟩ = 4DlatΔt, using the linear region between 10 and 50 ns (Figure S8). The value increased slightly from approximately 0.80 × 10–12 m2 s–1 for pure DMPC to approximately 0.93 × 10–12 m2 s–1 at 10 mol % Chol, and then decreased to approximately 0.55 × 10–12 m2 s–1 at 33 mol % Chol; the 50 and 66 mol % values also remained below that of pure DMPC (Figure D). Thus, lateral mobility increased slightly at 10 mol % Chol before declining across the higher-Chol systems.
This direction is consistent with pulsed-field-gradient NMR measurements showing cholesterol-dependent reductions in phospholipid lateral diffusion in oriented phosphatidylcholine bilayers. The segmental order parameter was calculated as SZ = (3⟨cos2 θZ⟩ – 1)/2, where θZ is the angle between the local carbon-chain axis and the membrane normal. SZ increased over most sn-1 and sn-2 chain positions with Chol enrichment (Figure E,F), consistent with the experimentally observed ordering effect in saturated DMPC/Chol membranes.
Localized decreases at the highest concentrations further distinguished the high-Chol response.
Together, Figure shows concentration-dependent changes in DMPC bilayer thickness, molecular area, hydrocarbon exposure, lateral mobility, and chain order. The low-to-medium Chol range is characterized mainly by bilayer thickening, molecular-area condensation, reduced tail exposure, and increased chain order, whereas the 66 mol % system departs from several of these trends. These coupled structural and dynamical changes provide a basis for interpreting the Chol-dependent poration free-energy landscapes examined below.
To assess whether the structural properties observed for DMPC were lipid dependent, matched DMPC/Chol, POPC/Chol, and DOPC/Chol bilayers containing 20 mol % Chol were compared at 310 K (Figure ). DMPC/Chol and POPC/Chol showed similar membrane thicknesses, whereas DOPC/Chol was thinner. Relative to DMPC/Chol, the POPC/Chol and DOPC/Chol systems had larger Alipid, higher Dlat, and lower SZ, with the largest differences generally observed for DOPC/Chol.
Thus, at fixed Chol loading, phospholipid identity and acyl-chain architecture were major sources of variation in membrane structure and dynamics.
Cholesterol Reshapes the Pore-Formation PMF
Before comparing the PMFs, the joint poration coordinate ξp is briefly summarized using Figure C and eq . During nucleation, ξp follows the polar-connectivity coordinate ξch, which describes the development of a polar defect across the membrane core. As ξch enters the switching region, the smoothed Heaviside function Hε activates the pore-radius-like term R(r), allowing the same coordinate to follow hydrophilic pore expansion.
Thus, values below the switching region primarily describe local thinning and polar-defect formation, whereas larger ξp values primarily describe pore expansion. Here, poration resistance at 300 K is characterized by the free-energy cost in the nucleation region and the pore-edge line tension associated with pore expansion. For the concentration series in Figure , the former is reported as a characteristic nucleation-region PMF energy; field-series nucleation barriers are extracted separately from resolved local PMF maxima.
Figure A illustrates pore nucleation/expansion PMFs in DMPC membranes across the cholesterol-concentration series (0–66 mol % Chol). Within this concentration-series PMF reconstruction, the pure-DMPC zero-field characteristic nucleation-region PMF energy was 61.26 ± 3.12 kJ/mol, comparable in magnitude to Hub’s computational value for related pore-formation PMFs (53.98 ± 2.88 kJ/mol). This value serves as the internal reference for analyzing concentration-dependent PMF trends.
Progression along ξp spans local thinning, transition/metastable states, and an approximately linear pore-expansion regime, consistent with prior studies., Across the low-to-medium Chol range, the characteristic nucleation-region free-energy cost increased (Figure C), consistent with composition-dependent packing changes that raise the free-energy cost of forming a connected polar defect.
At 66 mol % Chol, the characteristic nucleation-region PMF energy fell below the low-to-medium-concentration trend. Leaflet-resolved local Chol mole-fraction maps were calculated from the unbiased 512-lipid DMPC/Chol trajectories at 300 K and 0 mV for the 33, 50, and 66 mol % systems (Figures S33 and S34). The heterogeneity index, H, remained at or below the randomized-mixing reference (H = 1) for every composition and tested grid spacing.
At the finest 1.25 nm grid, the 66 mol % estimate was modestly higher than the 33 and 50 mol % estimates; at 1.50 and 2.00 nm, the values converged and remained within the randomized-mixing reference. Together, these results define the 66 mol % system as a high-Chol boundary response characterized by nonmonotonic packing and ordering, with lateral fluctuations remaining within the randomized-mixing reference over the sampled scales.
Figure A showed a progressive reduction of metastable states with increasing Chol content, consistent with Akimov’s elastic-continuum framework as discussed in Section . This trend can be interpreted through Chol-induced modulation of membrane elasticity parameters (B, KA, h, J; Table S3). Chol can shift monolayer spontaneous curvature toward more negative values, and elastic-continuum calculations predict that such curvature changes can reduce the depth of local pore free-energy minima through curvature competition., The decreasing fhead/tail in Figure D further supports this trend, because metastable stability correlates with larger fhead/tail values that better accommodate positive pore-edge curvature.
Pore-edge line tension (Figure B) was estimated from PMF slopes under linear conditions (ξp = 3–5). Under constant lipids, isobaric conditions, and no external surface tension, membrane-water interfacial area remained nearly constant during pore growth, making surface-tension release negligible in these simulations. For pure DMPC, γ = 14.20 ± 0.13 pN, higher than the 7.2 pN large-pore continuum estimate reported by Akimov et al. and the 6.2 pN dynamic-tension estimate for diC13:0 PC summarized by Ahmed et al., but below Hub’s previously reported value under a different atomistic setup (21.83 ± 0.04 pN).,, The diC13:0 PC value provides a nearby short-chain saturated-phosphatidylcholine benchmark; the same review summarizes DOPC estimates of 10.5–11.6 pN, illustrating the dependence on lipid composition, pore scale, and experimental analysis.
Within the present CHARMM36/TIP3P series, line tension increased with Chol, supporting concentration-dependent resistance to pore expansion. At 20 mol % Chol, γ = 18.88 ± 0.08 pN (1.33× pure DMPC); at 50 mol % Chol, γ = 27.08 ± 0.05 pN (1.91× pure DMPC). Transient-pore experiments on DOPC giant unilamellar vesicles likewise showed increasing line tension over the 0–30 mol % Chol range.
Chol-membrane snapshots showed gradual structural transitions from planar to defective configurations, characterized by local thinning and macropore formation (Figure G–J). Chol was relatively depleted from the hydrated pore interface while remaining present at the pore rim. This behavior is compatible with Chol’s molecular asymmetry, namely a small polar hydroxyl group and a bulky hydrophobic sterol ring, which may make water-exposed pore-edge environments unfavorable.
Chol’s negative spontaneous curvature may further disfavor the positive curvature of pore edges. Together, these structural features support a curvature-based interpretation of relative Chol depletion at the hydrated pore interface.
To characterize sterol redistribution relative to channel water, radial distribution functions (RDFs) of Chol groups relative to channel water (biased-field cylindrical water) were calculated (Figure E). The first RDF feature position (r*) and the intrachannel water-atom count along pore formation (ξp, step 0.8) are shown in Figure F, based on final umbrella-sampling frames. As the pore expanded, the Chol distribution showed reduced association with the hydrated pore region, and r* shifted from 1.905 nm at ξp = 0.8 to 2.37 nm at ξp = 3.2.
Beyond ξp = 3.2, the low-r feature was strongly attenuated and the dominant distribution shifted to larger r. The pore-radius-like estimate R(r) = R0ξp (R0 = 0.40433) gives approximately 1.29 nm at ξp = 3.2, which differs from r* = 2.37 nm. Accordingly, r* marks the radial feature position of the Chol–channel-water distribution and tracks sterol redistribution during pore formation, whereas the joint-coordinate radius term independently describes pore geometry.
Figure S9 provides atom-specific short-range RDF controls for CHL1 O3–channel-water O and DMPC P–channel-water O in 20 and 50 mol % Chol membranes at the nearest available representative early defect, connected-pore, and expanded-pore windows. In both compositions, the first CHL1 O3–water peak was strongly attenuated in the expanded-pore state, whereas the DMPC P–water RDF retained a clear first-shell peak. These short-range peaks characterize local association of channel water with the sterol hydroxyl and phosphatidylcholine phosphate groups.
Together with Figure E,F, the results show relative sterol depletion from the hydrated pore interface while phosphatidylcholine headgroups continue to define the pore-water boundary. Similar attenuation in the lo-rich 50 mol % Chol membrane near the lo/CBD boundary extends this response from the 20 mol % coexistence regime to the higher-Chol membrane. Residual Chol at the pore rim and Chol-induced changes in the surrounding bilayer remain compatible with the higher pore-edge line tension.
An independent stretch-compression MD study of DPPC/Chol bilayers similarly reported increasing line tension with cholesterol while several cholesterol molecules remained at the pore rim, further linking reduced Chol-water association with the pore-edge response.
Poration in Chol-containing DMPC membranes showed concentration-dependent remodeling of the free-energy landscape, with higher characteristic nucleation-region PMF energies and shallower metastable states over the low-to-moderate Chol range. At 66 mol % Chol, the lower characteristic nucleation-region PMF energy coincided with nonmonotonic thickness, DMPC molecular area, chain ordering, and sterol association, defining a distinct high-Chol boundary response. This composition approaches the literature-reported high-Chol solubility/domain boundary of DMPC/Chol bilayers, where cholesterol bilayer domains may occur., Leaflet-resolved analysis in Figures S33 and S34 placed lateral-composition variance within the randomized-mixing reference over grid spacings of 1.25–2.00 nm, with a modest increase at 66 mol % only at the finest grid.
Meanwhile, the monotonic increase in γ shows that Chol continues to increase resistance to pore expansion even as the characteristic nucleation-region free-energy cost decreases in the 66 mol % boundary regime. Chol’s molecular structure is compatible with its relative depletion from hydrated pore regions and may contribute to the energetic cost of pore growth. For electrically triggered liposomal delivery, these results suggest that Chol content should be tuned to balance poration resistance, fusion, and long-term carrier stability.
Continuum-Model Comparison Supports the PMF Interpretation
Classical pore theory describes the free-energy change for creating a circular hydrophilic pore as a competition between the positive pore-edge cost and the negative work associated with removing membrane area under lateral tension. In the tension-only limit, W(r) = 2πrγ – πr2σ, where r is pore radius, γ is pore-edge line tension, and σ is membrane surface tension. Setting dW/dr = 0 gives the critical radius rc = γ/σ and the corresponding barrier W(rc) = πγ2/σ.,
To include transmembrane voltage (TMV), the model separates an early hydrophobic defect from a hydrophilic pore and adds a voltage-dependent electrostatic contribution to each state., At a nominal TMV U, the free-energy changes are1ΔWO(r,U)=2πdrΓ0I1(r/λ)/I0(r/λ)−[ε0(εer−εmr)πr2/(2d)]U2;r<rmin2ΔWi(r,U)=2πrγ(r)−Γiπr2−[ε0(εer−εmr)πr2/(2d)]U2;r≥rmin
For r < rmin, ΔWO describes a hydrophobic defect. Here, d is the membrane hydrophobic thickness, Γ0 is the hydrocarbon/water interfacial tension, λ is the characteristic length of hydrophobic interactions, and I0 and I1 are modified Bessel functions of the first kind of orders 0 and 1. The term 2πdrΓ0I1(r/λ)/I0(r/λ) is the interfacial cost of the hydrophobic defect.
For r ≥ rmin, ΔWi describes a hydrophilic pore lined by reoriented lipid headgroups; 2πrγ(r) is the pore-edge cost, −Γiπr2 is the area-dependent contribution associated with the hydrated pore, and γ(r) = γ[1 + C/r5 accounts empirically for the higher edge cost of small pores.,C is the length-fifth correction coefficient, and Γi is the lipid-head/water interfacial tension. In both equations, ε0 is the vacuum permittivity, εer and εmr are the relative dielectric constants of water and membrane, respectively, and −[ε0(εer – εmr)πr2/(2d)]U2 is the favorable electrostatic term produced when water replaces low-permittivity membrane within the pore. Its r2U2 dependence explains why higher TMV preferentially lowers the free energy of larger hydrated defects.
The physically accessible branch is the lower of ΔWO and ΔWi, with rmin marking their crossover. Equations and correspond to the classical-continuum calculation in Figure C.
The classical model represents line tension as a system-wide constant or an empirical function of pore radius. Experimental edge-tension values vary with membrane composition and measurement protocol,, motivating the composition-dependent elastic treatment introduced below.
The elastic-continuum model partly addresses these limitations by introducing composition-dependent elastic parameters, monolayer curvature, and pore-edge deformation terms. It derives a pore-formation free energy that reflects membrane composition while allowing line tension to vary with pore radius and applied lateral tension. The model employs a director-vector field to characterize the average lipid orientation and computes pore-edge free energy from splay, tilt, and lateral-stretch deformations,, as summarized schematically in Figure S2:3W(r)=Wb+Wm−2πr2σ04W(r,L)=Wb+Wm+Wh−2πr2σ0
For hydrophilic pores, the pore boundary of each monolayer was subdivided into horizontal and vertical segments, enabling edge-energy calculation as the sum of elastic-deformation energies in the horizontal (Wb) and vertical (Wm) portions of two symmetric monolayers. Experimental elastic parameters, monolayer thickness, and spontaneous curvature were incorporated to account for membrane composition, with σ0 denoting the lateral tension per monolayer. For hydrophobic defects, the pore structure comprised a cylindrical hydrophobic strip of radius r and variable height 2L, sandwiched between vertical hydrophilic regions of the two monolayers, with free energy Wh.
At r = r*, the optimal hydrophobic-strip height approaches zero (2L* → 0), and the hydrophobic-defect and hydrophilic-pore states have equal free energy. The two states nevertheless remain separated by a small barrier (∼2kBT).Equations and represent the elastic-continuum model shown in Figure B.
To compare the MD trends with elastic-continuum theory, pore-formation PMFs were analyzed for pure DMPC and DMPC containing 10 mol % cholesterol at nominal TMVs of 0 and 400 mV (Figure A). These 200-lipid CHARMM36/TIP3P DMPC systems were analyzed at 300 K, with system details listed in Table S2. The continuum calculations used the parameters below, prioritizing experimental values where available.
The classic continuum model relied on limited parameters. DMPC: d = 3.53 ± 0.03 nm, γ = 2 pN; 10 mol % Chol: d = 3.83 ± 0.05 nm, γ = 3 pN. C = 1.39 × 10–46 m5 is the empirical length-fifth correction coefficient for γ(r) convergence.
Here, γ was an assumed value used for this illustrative classical-model comparison. Notably, adopting an experimentally based γ (e.g., ∼7.2 pN49) reduced agreement with the displayed MD profiles (Figure S1), illustrating sensitivity to the assumed line tension. For the elastic-continuum model, membrane-specific parameters were adopted to reflect composition-dependent differences.
DMPC: Splay modulus (per monolayer) B = 6.8 kBT, lateral stretch/compression modulus (per monolayer) KA = 117 mN/m, hydrophobic thickness h = 1.37 nm, spontaneous curvature JDMPC = +0.075 nm–1., 10 mol % Chol: B = 7.97 ± 0.96 kBT,KA = 198.45 mN/m,h = 1.52 ± 0.03 nm. Thermomechanical measurements of DMPC/Chol bilayers also support the use of composition-specific elastic parameters. A composition-specific value of J for DMPC containing 10 mol % Chol was not available in the literature used here.
For the illustrative calculation, J10%Chol was set to −0.05 nm–1 as an ideal-mixing approximation based on the opposing spontaneous-curvature tendencies of Chol and DMPC; possible lo/ld heterogeneity was not represented. A tilt modulus Kt = 40 mN/m (per monolayer) was assigned to all lipids, approximating oil–water interfacial surface tension. Continuum-model parameters are detailed in Table S1, and the supporting mechanical descriptors are summarized in Table S3.
Both continuum formulations assume a laterally homogeneous membrane described by bulk composition-dependent parameters. The pure-DMPC and 10 mol % Chol systems in Figure provide the most direct composition-parametrized comparison. Accordingly, the continuum curves assess the relative contributions of edge energy, curvature, elasticity, and voltage in these two systems.
The split axis in Figure A labels the two successive stages of the joint coordinate, ξch and R. The left segment reports water-defect connectivity before a continuous hydrophilic pore forms (0 ≤ ξch ≤ 1). After headgroup reorientation produces a connected hydrophilic pore, ξch ≈ 1 and the expansion term dominates, so the right segment is expressed as R(r) = R0ξp.
The axis break marks this descriptor change while retaining the continuous physical pathway from defect nucleation to pore expansion.
At 0 mV and in the absence of applied lateral tension, the rising postbarrier branch in Figure A identifies large-pore expansion as energetically uphill. In an unrestrained membrane, a defect that has crossed the local nucleation barrier but remains on a positive expansion slope is a subcritical pore (or prepore) and tends to close. Umbrella sampling holds such states long enough to map their free energy.
Persistent pore expansion requires additional work from surface tension or TMV: in the tension-only classical model, the larger-scale barrier is πγ2/σ, whereas the negative U2r2 term can reduce or reverse the expansion slope at sufficiently high TMV., Thus, the dashed 400 mV profiles quantify electrostatic lowering of pore-growth free energy relative to the restrained zero-field profiles. A recent review of tension-controlled GUV experiments likewise describes pore formation as a stochastic rate process in which increasing membrane tension lowers the activation barrier and accelerates pore formation.
All three models reproduced generic nucleation-region, metastability, and pore-expansion features. MD additionally captured local membrane thinning through atomistic conformational sampling (quadratic term at ξch < 1). MD and the elastic-continuum model reproduced the direction of the cholesterol effect, including higher nucleation-region free-energy costs and shallower metastable states, through composition-sensitive properties such as negative spontaneous curvature.
Under electric fields, both models showed cholesterol-mediated resistance to pore formation. Differences in absolute barriers, slopes, and pure-versus-Chol responses reflect the force field, water model, reaction coordinate, and sampling protocol used by each model. MD-derived γ remained higher than the elastic-model estimate and showed force-field sensitivity.
Overall, this multiscale comparison provides a physically grounded interpretation of the Chol-dependent trends. The classical model captures generic pore-nucleation and pore-expansion features, while the elastic-continuum model adds composition-sensitive elastic parameters and curvature terms. MD simulations resolve local thinning, hydrated-defect formation, sterol depletion near pore regions, and field-dependent structural responses.
Together, the continuum calculations connect the atomistic profiles to the edge-energy, curvature, elasticity, and voltage terms that govern pore formation.
Cholesterol Content, Lipid Composition, and Temperature Modulate Field-Assisted Pore Formation
Figures A and B show representative top-view configurations from unbiased CHARMM36/TIP3P direct-field simulations at a nominal TMV of 2.0 V and 300 K. A continuous hydrophilic pore formed in pure DMPC, whereas the DMPC bilayer containing 20 mol % cholesterol retained a localized hydrophobic defect over the displayed trajectory. The 2.0 V direct-field trajectories provide representative finite-time structural examples of pore formation, whereas Figure C and D quantify restrained free-energy responses over 0–1.4 V.
Figures C and D show field-dependent PMFs obtained with CHARMM36/TIP3P at 300 K for pure DMPC and DMPC containing 20 mol % cholesterol, respectively. Panel C resolves water-defect nucleation along ξch, whereas panel D follows the joint coordinate ξp through nucleation and pore expansion. Increasing nominal TMV lowered the nucleation-region maximum and progressively reduced the free-energy cost of advancing toward larger hydrated defects.
The two composition series were reconstructed separately and span different coordinate ranges; the nucleation barriers extracted from the local PMF maxima are compared directly in Figure E.
At the molecular level, TMV lowers the PMF by electrostatically stabilizing polar, hydrated defects: the field aligns water dipoles, promotes water penetration and local thinning, and facilitates phosphocholine-headgroup reorientation to form a water-lined defect. This is consistent with the favorable −U2r2 term. Cholesterol modifies baseline packing and pore-edge energetics, whereas TMV supplies the direct electrostatic bias.
At 300 K, the projected area per membrane molecule in the nearest stored intact-like umbrella window varied only slightly and nonmonotonically with TMV for pure DMPC and DMPC containing 20 mol % Chol (Figure S35). This umbrella-conditioned control does not support a large membrane-wide area increase as the main origin of PMF lowering and is not a direct membrane-tension measurement.
Across the simulated TMVs, the 20 mol % cholesterol membrane retained a higher nucleation barrier than pure DMPC, while the barriers in both systems decreased progressively as TMV increased (Figure E). The dashed lines summarize the field-dependent barrier trends over the simulated range.
The unbiased direct-field trajectories in Figure F provide a complementary finite-time kinetic readout at the composition-specific operational pore-onset TMVs listed in the caption and Section . Channel-formation and closure-or-rupture times varied nonmonotonically across the DMPC/Chol concentration series, and postformation pore lifetimes also varied under these composition-specific conditions. These event-based quantities complement the equilibrium PMFs and are interpreted under the stated direct-field conditions.
A separate pure-DMPC force-field comparison is provided in Figure S10. At 300 K, the Berger and CHARMM36/TIP3P PMF series both show progressive lowering of the nucleation-region maximum and movement of the maximum toward lower water-defect connectivity as nominal TMV increases, while the absolute PMF magnitudes and maximum locations differ between parametrizations. The main-text PMFs and composition comparisons in Figure C–H use CHARMM36/TIP3P; the Berger series is shown only in the supplementary comparison.
Figures C and D resolve progressive field stabilization across the sampled potentials. Panel C covers the nucleation regime, whereas panel D shows that postbarrier profiles remain uphill at low field, flatten or develop a finite downhill segment at intermediate fields, and become downhill over the sampled coordinate at the highest field. Together, these profiles trace the progressive electrostatic stabilization of hydrated defects.
Representative water-channel breakage and reclosure events in selected 20 mol % cholesterol umbrella windows are shown in Figure S6, illustrating local window-level variability near the pore-transition region.
The matched 20 mol % cholesterol series extended the field analysis across lipid identities and temperatures. At zero applied field, the temperature-dependent PMFs differed among DMPC, POPC, and DOPC (Figure G). At 330 K, increasing nominal TMV lowered the nucleation-region PMFs in all three lipid systems, with lipid-dependent profile shapes and field sensitivities (Figure H).
These panels use the nucleation-region portion of ξp. Together, Figure G and H resolve lipid-dependent nucleation-region responses at a matched 20 mol % cholesterol loading.
Figure S36A, B directly compares the 20 mol % Chol DMPC, POPC, and DOPC series at 330 K. In all three membranes, the nucleation barrier decreased with TMV and the local PMF maximum generally shifted to lower ξch, with lipid-dependent differences.
A method-resolved comparison with prior simulations and planar-bilayer experiments is provided in Figure S11. The operational definitions in the two earlier MD studies differed: ref bracketed each nsEP and msEP threshold between 100 ns nonporating and porating trajectories, whereas ref . defined E0 as the lowest field at which at least one of three 25 ns replicas porated, using 25 mV nm–1 increments. Figure S11 presents the composition-normalized operational pore-onset TMV used here, defined as the lowest sampled nominal TMV at which a continuous transmembrane water channel was observed across three independently initialized simulations at that sampled condition, alongside selected minimum-porating-field values for DOPC/Chol from ref . and electrically detected breakdown or threshold voltages for planar POPC/Chol and DPhPC/Chol membranes from refs and .
The protocol-specific threshold intervals from ref . are retained as separate method-specific results. Across the low-to-moderate concentration ranges examined, cholesterol generally shifted poration-related field or voltage metrics, with the magnitude and high-concentration response varying with lipid composition, trajectory duration, and detection criterion.
Together, Figure shows within a common CHARMM36/TIP3P framework that applied potential facilitates water-defect formation, while 20 mol % cholesterol retains a higher nucleation barrier than pure DMPC across the simulated TMV range. Temperature and lipid identity further modify the nucleation-region response. The supplementary force-field comparison and literature comparison place these trends in complementary methodological context.
Membrane-Core Hydration, Water Connectivity, and Crossing Kinetics During Restrained Pore Formation
Figure characterizes restrained pore-forming states with three trajectory-resolved observables: membrane-core hydration Hcore, continuous-water probability Pconn, and complete-water-crossing rate. Together, these observables resolve hydration, connectivity, and crossing kinetics along the restrained poration pathway. Full-coordinate and transition-region analyses are provided in Figures S12–S15.
Recent joint-coordinate simulations of asymmetric POPS/POPC bilayers resolved leaflet-localized water entry before formation of a continuous transmembrane polar defect and distinguished the associated nucleation barrier from large-pore line-tension energetics. This provides direct precedent for using hydration and connectivity as descriptors of nucleation-state progression.
ξch is the continuous umbrella coordinate obtained from the smooth slice occupancy of polar atoms (water and phospholipid phosphate oxygens); it is not water-specific. Hcore measures local core-water density relative to bulk water, Pconn is the fraction of frames containing a water-only spanning cluster, and the complete-water-crossing rate counts individual side-to-side water events per unit time. Thus, the four quantities describe biased polar connectivity, hydration, frame-wise water connectivity, and transport dynamics, respectively.
For DMPC containing 20 mol % cholesterol at 300 K and 0 mV, the four profiles in Figure A represent an intact-like or weakly perturbed state (ξp = 0.08), an early hydrated defect (ξp = 0.69), a connected pore near the nucleation-to-expansion transition (ξp = 1.01), and an expanded pore (ξp = 4.93). The progressive replacement of the central water-density minimum by a hydrated core shows that Hcore is a structural descriptor of pore development. The corresponding full-coordinate DMPC concentration and applied-field maps are shown in Figure S12.
Across the DMPC/cholesterol concentration series at 300 K and 0 mV, Hcore increased sharply as the umbrella target approached and passed through the transition region, while its magnitude and the ordering among cholesterol contents varied across windows (Figure B). In the matched 20 mol % cholesterol systems at zero field, DMPC, POPC, and DOPC likewise showed increasing core hydration as ξp approached 1 over 310–330 K, with lipid- and temperature-dependent onset and magnitude (Figure C). The expanded no-field concentration and lipid/temperature matrices are provided in Figure S13.
At 330 K, Pconn generally increased toward the transition region and with nominal TMV, with lipid-dependent response magnitudes (Figure D). At ξp = 0.98, complete water-crossing rates likewise varied with lipid identity and field strength (Figure E). Figures S14 and S15 provide the full field maps and transition-region uncertainty.
Figure S36C–E evaluates these descriptors in the stored windows nearest the local PMF maxima of the pure-DMPC 300 K series. As the maximum shifted to lower ξch, the 600–800 mV windows showed lower Hcore, Pconn, and complete-water-crossing rates than the 200 mV window. This is qualitatively consistent with ref , which found that at higher TMV the nucleation transition can be crossed before a continuous water defect forms.
Because nearest stored windows rather than committor-defined transition states were analyzed, the comparison is qualitative.
For DMPC at 330 K, complete-water-crossing rates were jointly modulated by cholesterol content and nominal TMV (Figure F), producing condition-dependent composition–field response patterns. Together, Figure and Figures S12–S15 show that reaction-coordinate position, membrane composition, temperature, and applied potential jointly shape hydration, connectivity, and water-crossing kinetics in restrained pore states. These state-resolved observables connect the PMF landscape to the emergence of hydrated, water-connected pore states.
Relative Electrostatic-Potential and Pore-Structure Descriptors During Restrained Poration
To relate the hydration results to local electrostatics, we analyzed baseline-corrected, charge-density-derived relative electrostatic-potential profiles across selected umbrella windows (Figure ; complete profile sets are provided in Figures S16–S23). ΔV(z) is obtained from the charge-density distribution along the bilayer normal and reports the potential difference between position z and the bulk-water reference after centering and baseline correction. It integrates the charge and dipole organization of phosphocholine headgroups, cholesterol hydroxyl groups, and interfacial water.
The several-hundred-millivolt central maximum therefore represents a membrane dipole/core electrostatic feature. At 20 mol % cholesterol, 330 K, 0 mV, and ξp = 0.74, all three lipid systems showed a central maximum in ΔVeven(z) (Figure A), and the corresponding ΔVcore values followed DMPC > POPC > DOPC (Figure B). ΔVcore generally decreased as ξp approached 1, with substantial repeat-to-repeat variation.
Figure B summarizes three independent simulations. These interpretations follow established charge-density-based calculations of lipid-bilayer electrostatic potentials and prior experimental and molecular-simulation studies of cholesterol-dependent membrane dipole potentials.−
The central ΔVcore maximum is an internal membrane dipole/core potential relative to bulk water, not an additional applied TMV. During defect growth, water and headgroup reorientation and core hydration screen or remodel this internal potential while the imposed field stabilizes the hydrated defect. Thus, ΔVcore accompanies the electrostatic relaxation of poration but is not itself the PMF barrier.
At 300 K and 0 mV, ΔVcore decreased across much of the reaction coordinate for all DMPC/cholesterol systems (Figure C). This decrease coincided with increasing core hydration and is consistent with screening and reorientation of interfacial dipoles as water occupies the low-dielectric membrane core. Cholesterol-dependent ordering varied among windows, with the 66 mol % system showing pronounced window dependence.
At 20 mol % cholesterol and 300 K, the TMV-associated response also varied with ξp (Figure D), with the largest reductions in expanded-pore windows. Linear fits over 310–330 K yielded predominantly negative temperature coefficients with local deviations (Figure E). At 330 K, the even-component TMV-response coefficients became more negative at larger targets, whereas relatively few odd-component coefficients met the display criterion (Figure F).
Together, ΔVcore and the even and odd components of ΔV(z) map charge-density-derived changes in local electrostatic organization across pore development, temperature, and prescribed TMV.
Figure provides complementary structural descriptors from pore-centered density, core-hydration, and apparent-pore-diameter analyses, with expanded maps and response profiles in Figures S24–S32. In each r–z map, the horizontal coordinate r is the radial distance from the pore center and the vertical coordinate z is the position relative to the membrane midplane. Warmer colors indicate greater relative lipid mass density, whereas dark central regions indicate lipid depletion associated with a hydrated defect; white contours indicate the plotted relative cholesterol-density levels.
The difference map in Figure A visualizes the pointwise contrast between the normalized relative-density patterns at 1400 and 0 mV. These pore-centered maps of restrained configurations resolve pore morphology and local lipid and cholesterol occupancy along the reaction coordinate.
At ξp = 0.74, 330 K, and 20 mol % cholesterol, the relative DMPC-density maps differed between 0 and 1400 mV (Figure A). The threshold-based apparent pore diameter increased from 0.66 to 5.72 nm, with the latter reported as a lower bound because it approached the available radial range. The field-dependent progression is shown in Figure S24, and corresponding temperature- and lipid-resolved density maps are provided in Figures S25 and S26.
Across the full coordinate at 300 K and 0 mV, apparent diameter generally increased with ξp (Figure B and Figure S32A). At matched windows, apparent diameter varied with cholesterol content, with smaller values more frequently observed at higher cholesterol contents; supporting extended-coordinate density maps are shown in Figure S27.
At 20 mol % cholesterol and 300 K, increasing nominal TMV generally increased apparent diameter, particularly at larger ξp (Figure C; Figures S28 and S32B). Core hydration increased toward the transition region in DMPC, POPC, and DOPC at 310–330 K and 0 mV (Figure D and Figure S29). At 330 K and 1000 mV, all three lipid systems showed a marked diameter increase as ξp approached 1, with lipid-dependent magnitudes (Figure E and Figure S31).
At the matched 20 mol % cholesterol loading, these results resolve lipid-dependent differences in hydration and field-assisted pore expansion.
At ξp = 0.98 and 330 K, apparent diameter was generally larger at higher nominal TMVs and lower cholesterol contents (Figure F). The complete DMPC concentration–potential trajectories are shown in Figure S30, while POPC and DOPC field responses are shown in Figure S31. Several high-field estimates reached the radial measurement limit and are reported as lower bounds.
Across the complete matrix, apparent diameter generally increased with field strength, was modulated by cholesterol content, and included local departures from monotonic ordering.
Conclusions
This study combined a continuous joint poration coordinate, umbrella-sampling PMFs, and independent direct-field trajectories to resolve how cholesterol and TMV jointly regulate pore nucleation and expansion in DMPC/Chol membranes. Across low-to-medium cholesterol contents, cholesterol increased the characteristic nucleation-region free-energy cost and pore-edge line tension while making metastable pore states shallower, whereas higher nominal TMV promoted hydrated-defect growth and pore expansion. Direct-field trajectories further revealed composition-dependent channel formation and evolution, together with pore lifetimes defined by continuous-water connectivity, under the respective operational pore-onset conditions.
Together, these findings show that membrane composition modulates the energetics of pore nucleation and expansion, while the electric field drives hydrated defects along this free-energy landscape; their interplay shapes both the poration pathway and its kinetics.
At a matched loading of 20 mol % cholesterol, the DMPC/Chol, POPC/Chol, and DOPC/Chol systems exhibited lipid-dependent nucleation-region free energies, core hydration, electrostatic features, and apparent pore-diameter responses, showing that phospholipid identity and acyl-chain architecture further modulate poration at fixed cholesterol content. The 66 mol % DMPC system showed a distinct high-cholesterol boundary response, in which nonmonotonic changes in packing and ordering coincided with a lower characteristic nucleation-region PMF energy. Its lateral composition fluctuations remained within the randomized-mixing reference over the tested grid scales, while the observed response comprised changes in packing, order, and sterol redistribution.
Within this controlled binary-membrane framework, the characteristic nucleation-region PMF energy and the pore-edge line tension associated with pore expansion quantify composition-dependent poration resistance, while resolved local PMF maxima define the field-series nucleation barriers. By integrating restrained free-energy profiles with finite-time channel events from direct-field trajectories, this study connects membrane composition and nominal TMV to pore energetics, hydration and connectivity, pore structure, and pore-evolution kinetics, providing a unified molecular-scale framework for interpreting composition-dependent electroporation in model membranes.
Methods
Definition of the Joint Reaction Coordinate
The PMF for pore nucleation and expansion was calculated along the joint reaction coordinate ξp introduced by Hub, which combines coordinates for polar-defect nucleation and hydrophilic-pore expansion (Figure C). During nucleati
on, the chain coordinate ξch(r) quantifies the connectivity of a polar transmembrane defect. A membrane-spanning cylinder is divided into Ns slices, and ξch is approximately the fraction of slices occupied by polar heavy atoms; water oxygen atoms and lipid phosphate-group oxygen atoms were included in this polar selection., After a continuous defect forms and ξch approaches 1, expansion is described by the approximate pore radius R(r), obtained from the number of polar heavy atoms in a central slab of thickness D. The coordinate vector r denotes all atomic positions.
The combined coordinate is5ξp(r)=ξch(r)+Hε[ξch(r)−ξchs](R(r)−R0)/R0
R0 normalizes the expansion term and is chosen so that [R(r) – R0]/R0 ≈ 0 in the switching region. The function Hε(x) is a smooth Heaviside function that changes from 0 to 1 over [−ε, ε]. Accordingly, ξp = ξch when ξch < ξchs – ε; when ξch > ξchs + ε and the polar defect is continuous, ξp follows the radius-like expansion term, with R ≈ R0ξp.
For the present systems, R0 was fixed before production sampling from Ns, D, ds, and ξchs.
To maintain comparable intact-state scaling across compositions with different membrane thicknesses, Ns = 26–36 and ds = 0.1 nm were selected so that the flat bilayers gave ξch ≈ 0.2. The cylindrical localization radius Rcyl = 0.8 nm confined the biased defect laterally while allowing overlap between upper and lower water defects. Rcyl set the lateral localization constraint, while atomistic configurations under each composition and thermodynamic condition determined the defect geometry.
The switching parameters were ξchs = 0.925, ε = 0.05, and D = 1.2 nm. The volume assigned per polar atom in the central slab was 0.02996 nm3, corresponding to the molecular volume of water. The joint coordinate was selected to provide a continuous path from pore nucleation to pore expansion, and the present forward/reverse pulling check is shown in Figure S4.,,,
Membrane-Core Hydration, Water Connectivity, and Complete Water-Crossing Analyses
For Figure and Figures S12–S15, bilayer-normal water-oxygen density profiles were centered on the membrane midplane and normalized by the bulk-water density. Core hydration Hcore was the mean normalized water-oxygen density over −0.50 ≤ z ≤ 0.50 nm, using the same definition across all compared conditions. Pconn was calculated as the fraction of analyzed frames containing a continuous water path between the two bulk-water compartments.
A complete crossing event was counted when an individual water molecule moved from one bulk-water side to the other, and the crossing rate was the number of completed events divided by the analyzed trajectory time. Analyses were performed at the specified umbrella target, composition, temperature, and applied TMV. Gray cells denote unavailable data, and Figure S15 reports the transition-region variability associated with Figure E.
Figure S7 reports the water-velocity distribution from an unbiased flat-membrane control as a descriptive dynamical reference.
For Figure S36C–E, Hcore was the annular-volume-weighted water-oxygen density within r ≤ 0.40 nm and |z| ≤ 0.50 nm, divided by the trajectory-specific bulk-water density. Radial bins were 0.08 nm and were Gaussian-smoothed with σ = 0.08 nm. The bulk reference used r ≤ rmax and zbulk,min ≤ |z| ≤ zmax, where rmax = min[4.5, 0.46 min(Lx, Ly)], zmax = min(3.2, 0.48Lz), and zbulk,min = min[2.6, max(1.3, zmax – 0.6)] nm.
Pconn was the fraction of saved frames in which water oxygens within r ≤ 1.00 nm and |z| ≤ 1.85 nm formed an O–O graph with an edge cutoff of 0.35 nm and a connected component spanning z ≤ −1.20 to z ≥ +1.20 nm.
Complete crossings were counted by tracking each selected water-oxygen index from one reservoir (z ≤ −1.50 or z ≥ +1.50 nm), through the central region (r ≤ 1.00 nm and |z| < 1.20 nm), to the opposite reservoir. Return to the original side was not counted, tracking was reset at r > 1.40 nm, and the rate was the total event count divided by the elapsed analyzed time.
For Figure S36C–E, the stored target nearest the local PMF maximum in each 300 K pure-DMPC profile was analyzed. Only nonend point cases with |⟨ξactual⟩ – ξ‡| ≤ 0.05 were retained (200, 600, and 800 mV). The final approximately 5 ns from each of three independent simulations initialized with reassigned velocities were analyzed; Hcore used every fourth saved frame, whereas Pconn and crossing used every saved frame.
Each trajectory contributed one replicate value.
Simulation Settings and Parameters
All membrane systems were constructed using CHARMM-GUI and atomistic simulations were performed with GROMACS 2021., All main-text atomistic MD data sets used CHARMM36 with TIP3P water., The pure-DMPC Berger series in Figure S10 was retained for the supplementary force-field comparison. For the primary 300 K DMPC/Chol series, the verified setup used the Verlet cutoff scheme; particle-mesh Ewald electrostatics with tinfoil boundary conditions, a 1.2 nm real-space Coulomb cutoff, 0.12 nm Fourier spacing, and fourth-order interpolation; and Lennard-Jones interactions switched from 1.0 to 1.2 nm. Temperature was controlled by velocity rescaling with separate lipid and solvent coupling groups, and pressure was controlled semi-isotropically at 1 bar with the Parrinello–Rahman barostat (τP = 5 ps; compressibility 4.5 × 10–5 bar–1).
The primary-series temperature-coupling constants were τT = 1.0 ps for unbiased production, 0.5 ps for pulling, and 0.1 ps for umbrella windows. Bonds involving hydrogen were constrained with LINCS, and production and umbrella-window simulations used a 2 fs integration step. Primary-series system sizes, box dimensions, and 200–300 ns equilibration durations are listed in Table S2, with membrane-area convergence shown in Figure S3.
When a uniform external field Ez was prescribed, the corresponding field-derived potential was defined as Unom = EzLz, where Lz is the box length along the membrane normal. Below, this prescribed value is termed the applied TMV; Section separately defines the baseline-corrected relative electrostatic-potential profiles.
The simulations were organized into distinct data families, with matched comparisons performed within each family. The primary full-coordinate 512-lipid DMPC/Chol concentration series at 300 K comprised 0, 10, 20, 33, 50, and 66 mol % Chol at 0 mV; the corresponding 20 mol % Chol field series used applied TMVs of 0, 200, 600, 800, 1000, 1200, and 1400 mV. The transition-region analyses in Figure C–F and Figures S13–S15 used separately prepared CHARMM36/TIP3P data sets: DMPC at 330 K across 0, 10, 20, 30, 50, and 60 mol % Chol, and matched DMPC, POPC, and DOPC systems containing 20 mol % Chol at 310, 320, and 330 K under zero field and at 330 K under the indicated applied TMVs.
The 200-lipid DMPC/Chol systems supported the continuum-model comparison and related controls listed in Table S2. Within the supplementary force-field comparison shown in Figure S10, the Berger series was analyzed at 300 K and uniquely included a 400 mV profile. Matched zero-field structural and dynamical properties at 310 K are summarized in Figure .
In Figure G,H, ξp is displayed over the nucleation-region interval. Except for box dimensions, the matched POPC/Chol and DOPC/Chol data families used the same atomistic settings, equilibration procedure, pulling protocol, umbrella-window centers and force constants, per-window sampling, discarded equilibration interval, and WHAM reconstruction protocol as the corresponding DMPC umbrella-sampling family described in Sections and .
Trajectory-derived means, error bars, and shaded variability bands were obtained from three independent simulations initialized with reassigned velocities; the corresponding replicate or uncertainty convention is stated in each figure legend. PMF uncertainty bands were estimated separately by Bayesian bootstrap during WHAM reconstruction.
For Figures S35 and S36C–E, pale symbols show individual values from three independent simulations initialized with reassigned velocities, filled symbols show the mean, and error bars show the corresponding sample standard deviation. For Figure S36A,B, symbols and error bars likewise show the mean and sample standard deviation across values extracted from three independently reconstructed replicate PMFs.
For the unbiased direct-field simulations in Figure F, three independently initialized trajectories were analyzed for each Chol composition at the corresponding operational pore-onset nominal TMV: 2.037, 2.034, 2.115, 3.653, 3.860141, and 3.785149 V for 0, 10, 20, 33, 50, and 66 mol % Chol, respectively. Times were measured in ns from the start of each direct-field production trajectory. A continuous transmembrane water channel was defined operationally as a connected water path spanning the bilayer between the two bulk-water compartments.
Channel-formation time was the first saved frame satisfying this criterion. Channel closure was the first subsequent loss of transmembrane connectivity, whereas membrane rupture denoted continued expansion without reclosure during the remaining analyzed trajectory. Pore lifetime was the interval from the first continuous transmembrane water connection to subsequent closure or membrane rupture.
Figure F (a) reports the mean formation and closure-or-rupture times, and Figure F(b) reports the mean pore lifetime; in both panels, error bars represent the variability among the three independent simulations.
Membrane thickness and molecular areas were calculated with FATSLiM from the unbiased trajectories used in Figures and . Lipid-tail solvent-accessible surface area and the head-to-tail volume-fraction ratio fhead/tail were calculated with gmx sasa, with the aliphatic chains assigned to the tail group. Acyl-chain order parameters SZ were obtained with gmx order, and the in-plane lateral diffusion coefficient Dlat was obtained with gmx msd by fitting ⟨dxy2⟩ = 4DlatΔt over Δt = 10–50 ns.
The corresponding MSD curves and fits are provided in Figure S8. The values in Figures and are means from three independent simulations, and the error bars show the corresponding intersimulation sample standard deviation. The same analysis definitions were applied to the matched DMPC/Chol, POPC/Chol, and DOPC/Chol systems.
For Figures S33 and S34, unbiased 512-lipid DMPC/Chol trajectories at 300 K and 0 mV for 33, 50, and 66 mol % Chol were analyzed separately for the two leaflets. The membrane plane was divided into fixed x–y grids, and the local cholesterol mole fraction was xChol = NChol/(NChol + NDMPC). The heterogeneity index was H = Var(xChol,obs)/⟨Var(xChol,rand)⟩.
The randomized reference was generated by relabeling lipid identities while retaining molecular positions and the global composition. H = 1 denotes the randomized-mixing reference, whereas H > 1 indicates excess lateral composition variance relative to that reference. Grid-spacing sensitivity was evaluated at 1.25, 1.50, and 2.00 nm.
Thus, H quantifies local lateral-composition variance relative to randomized mixing over the sampled trajectories and specified grid scales.
Umbrella Sampling and PMF Reconstruction
For the primary full-coordinate DMPC/Chol series, constant-velocity pulling along ξp over at least 150 ns generated configurations spanning pore nucleation and expansion. Nonuniform umbrella windows were centered at ξp = 0.08–0.64 (spacing 0.08; force constant 3000 kJ mol–1), 0.69–1.17 (spacing 0.02; 5000 kJ mol–1), and 1.18–4.93 (spacing 0.15; 400 kJ mol–1) to resolve the nucleation, transition, and expansion regions, respectively. Each window was sampled for at least 60 ns, and the first 30 ns were excluded from PMF reconstruction.
PMFs were reconstructed with gmx wham, with uncertainty estimated by 50 Bayesian-bootstrap iterations. Pore-edge line tension was obtained from the linear expansion region ξp = 3–5 using γ = (1/2π)dG/dR, with R ≈ R0ξp. A forward/reverse pulling comparison for the 20 mol % Chol membrane is provided in Figure S4 as a qualitative hysteresis check.
The matched POPC/Chol and DOPC/Chol umbrella-sampling families used the same pulling duration, window ranges, restraint force constants, per-window sampling duration, equilibration exclusion, and WHAM/bootstrap settings as the corresponding DMPC calculations; only lipid identity, temperature, and prescribed TMV differed.
The concentration-dependent and field-dependent PMF families were reconstructed separately and referenced to their respective intact-bilayer minima. For the concentration-dependent profiles in Figure , the characteristic nucleation-region PMF energy was identified from the marked local maximum or trend inflection point preceding pore expansion. For the field-dependent profiles in Figure C–E, the nucleation barrier was extracted from the resolved local PMF maximum.
The pure-DMPC and 20 mol % Chol field series in Figure C–E both used CHARMM36/TIP3P, enabling direct comparison of the field-dependent barrier trends. For the supplementary comparison in Figure S10, the Berger and CHARMM36/TIP3P pure-DMPC PMFs were treated as separate 300 K force-field series, and each profile was referenced to its own intact-bilayer minimum. The 400 mV profile was available only for the Berger series.
The nucleation-region PMF maxima were identified consistently within each series. The matched DMPC/Chol, POPC/Chol, and DOPC/Chol umbrella families used the same sampling protocol, and Figure G,H display the nucleation-region portion of ξp for the lipid-dependent comparison.
For Figure S36A,B, the barrier and ξch position of the resolved local PMF maximum were extracted separately from PMF profiles reconstructed for three independent simulations initialized with reassigned velocities for the 330 K DMPC, POPC, and DOPC systems over the common 600–1400 mV range; barriers were referenced to the corresponding intact-bilayer minima. The displayed values are the mean and sample standard deviation across the three replicate profiles.
For Figure S35, the projected area per membrane molecule was calculated from each saved periodic box as Aproj = LxLy sin γbox/100, where 100 is the number of membrane molecules per leaflet. The 300 K intact-like target 0.065 and final approximately 5 ns from each of three independent simulations initialized with reassigned velocities were analyzed; the mixed-membrane value is composition-averaged.
RDF analyses for Figure E,F and Figure S9 used 300 K, 0 mV umbrella-window trajectories from the DMPC concentration series. The main-text analysis reports a broader Chol-group/channel-water distribution and uses r* to mark its radial feature position. Figure S9 separately reports atom-specific short-range RDFs for CHL1 O3–channel-water O and DMPC P–channel-water O at the nearest available representative early defect, connected-pore, and expanded-pore windows in the 20 and 50 mol % Chol systems.
Channel-water O denotes the pore-associated water-oxygen selection used in the corresponding calculation. The short-range first peaks in Figure S9 characterize atom-specific local coordination, whereas r* in Figure F marks the radial feature position of the broader Chol-group/channel-water distribution; the joint-coordinate radius term provides the separate pore-radius-like estimate. The shaded envelopes in Figure S9 show the sample standard deviation across three independent simulations.
Relative Electrostatic-Potential and Apparent-Pore-Diameter Analyses
For Figure and Figures S16–S23, relative electrostatic-potential profiles ΔV(z) were calculated with gmx potential from the SYSTEM charge-density distribution along the membrane normal using 1000 slices. Trajectories were centered on the membrane midplane, and each profile was baseline corrected using the mean of the two bulk-water plateaus. These baseline-corrected profiles report the membrane-center potential relative to bulk water, while the applied TMV is defined separately in Section .
The even and odd components were ΔVeven(z) = [ΔV(z) + ΔV(−z)]/2 and ΔVodd(z) = [ΔV(z) – ΔV(−z)]/2, respectively. ΔVcore was the spatial mean of ΔVeven(z) over |z| ≤ 0.50 nm. For each lipid/window condition, Figure B reports the mean ΔVcore and corresponding sample standard deviation across three independent simulations.
Temperature- and TMV-response coefficients were obtained by linear regression within each analyzed window, and Figure F displays TMV-response fits with R2 ≥ 0.50. These profiles represent charge-density-derived relative electrostatic potentials of the molecular system under each prescribed field condition, referenced to the bulk-water region. The nominal applied TMV was defined separately as Unom = EzLz and served as the independent field variable; the reported ΔV(z) was reconstructed from the molecular charge-density profile.
The charge-density integration and reference treatment follow established lipid-bilayer potential methodology.
For Figure and Figures S24–S32, pore-centered r–z maps were generated from species-specific whole-molecule phospholipid and CHL1 mass densities, with z referenced to the membrane midplane and r measured from the frame-specific pore center. Density maps are reported as relative display matrices for spatial-pattern comparison. In Figure A, the difference panel was calculated pointwise from the displayed 1400 and 0 mV relative-density matrices and represents the contrast between their normalized spatial patterns.
Core hydration was the water-oxygen density within |z| ≤ 0.50 nm relative to bulk water. For each analyzed frame, the pore center was assigned from the periodic circular mean of core-water oxygen positions, and the bulk-water reference was estimated from |z| ≥ 2.60 nm. Apparent pore radius was the outward threshold-crossing radius of the connected midplane pore-water density, using the larger of 0.10 times the bulk-water density and 0.20 times the midplane peak density, capped at 0.95 times that peak; apparent diameter was twice this radius.
Where uncertainty bands are shown, they represent the mean and sample standard deviation across three independent simulations. Values reaching the available radial range were treated as lower bounds and marked by white triangles. These threshold-based values characterize restrained-window pore morphology.
Supplementary Material