Curvature-induced migration of small domains on lipid membranes

We consider a smooth membrane surface and an isolated fluid domain of area $A$. Domain and background have bending rigidities $\kappa_{d}$ and $\kappa$, Gaussian moduli $\bar{\kappa}_{d}$ and $\bar{\kappa}$, and spontaneous curvatures $m_{d}$ and $m$. Their interface has a positive line tension $\lambda$. We use the convention

where the principal curvatures $k_{1}$ and $k_{2}$ are positive on a sphere. Strong adhesion to an external scaffold is assumed to keep $M$ and $K$ fixed as the domain moves. The energy, defined relative to that of the background membrane in the absence of the domain, is

where $P$ is the perimeter of the domain, which is a topological disk with a narrow interface compared with its radius.

We now take the limit of a small domain. Define the area-equivalent radius $r\equiv\sqrt{A/\pi}$. The small-domain limit requires $r|k_{1}|\ll 1$ and $r|k_{2}|\ll 1$, together with slow variation of the curvatures across the domain. For such a small domain with area $A$ centered at a point with curvature $K$, Nardulli’s local isoperimetric expansion (specialized to two dimensions) gives the leading curvature correction to the minimal perimeter as (18)

the remainder depends on the local curvature and its spatial derivatives, Note that a geodesic disk has the same displayed correction, but an exactly perimeter-minimizing boundary need not be a geodesic circle on a variable-curvature surface. As expected, at a flat point with $K=0$ we recover the perimeter of a flat disk, $P_{\mathrm{flat}}=2\sqrt{\pi A}=2\pi r$. Equation 3 has a direct physical interpretation (Fig. 2). At the same small area, a domain needs a longer boundary on a negatively curved region (saddle) and a shorter boundary on a positively curved region than in a flat plane. Positive line tension therefore favors larger Gaussian curvature when the bulk elastic contributions are held equal. We note as well that, within the domain of applicability of Eq. 3, with $|K|A\ll 1$, coalescence of two domains to form a larger one always reduces the total perimeter, i.e. $P(K,2A)-2P(K,A)<0$, implying that coalescence is favored independently of the local Gaussian curvature $K$.

Additionally, in the limit of a small domain the mean and Gaussian curvatures are approximately constant over the domain, and the integral in Eq. 2 becomes trivial, with the next order of corrections being quadratic in $A$. Using this result and Eq. 3 then gives our central result, asymptotically valid in the limit of small $A$:

where $\Delta_{\kappa}\equiv\kappa_{d}-\kappa$, $\Delta_{\kappa m}\equiv\kappa_{d}m_{d}-\kappa m$, and $\Delta_{\bar{\kappa}}\equiv\bar{\kappa}_{d}-\bar{\kappa}$ are the bending rigidity contrast, rigidity-weighted spontaneous curvature contrast, and Gaussian modulus contrast, respectively. Terms independent of the local curvature have been omitted, as they only contribute a baseline to the curvature-dependent energy landscape. Because the energy in Eq. 4 depends on the local curvature of the membrane, the energy depends on the position of the domain along a membrane with non-constant curvature, resulting in domain migration toward points that are local minima of this energy. The mechanical force experienced by the domain will be $-\nabla E$, where $\nabla$ is to be interpreted as the gradient operator in curvilinear coordinates. On an axisymmetric surface, this force has the direction of $-dE/ds$, where $s$ is the arclength (distance along the meridian).

In Section 3.2, we will illustrate our results using axisymmetric spheroids. These can be parameterized as

where $a$ and $c$ are the equatorial and polar semi-axes, respectively. Defining the function

for conciseness, the principal curvatures can be written as

Consequently, the mean and Gaussian curvatures are given by

In particular, at the poles the two curvatures are $M_{\mathrm{p}}=c/a^{2}$ and $K_{\mathrm{p}}=c^{2}/a^{4}$, while at the equator they are $M_{\mathrm{e}}=(a/c^{2}+1/a)/2$ and $K_{\mathrm{e}}=1/c^{2}$.

The total surface area $S$ of the spheroid can be calculated as

Defining the aspect ratio $q\equiv c/a$, we find that for a prolate spheroid with $q>1$, the area is given by

and, for an oblate spheroid with $q<1$, by

For comparison between shapes, we define the area-equivalent radius $L\equiv\sqrt{\frac{S}{4\pi}}$ which will be used below. We will use one prolate example with aspect ratio $q=1.25$ and one oblate example with $q=0.8$. For spheroids, the mean and Gaussian curvatures vary monotonically with latitude. For the prolate spheroid, both $M$ and $K$ are lowest at the equator and increase toward the poles, while the opposite is true for the oblate spheroid, with $M$ and $K$ largest at the equator and decreasing toward the poles.

The universal diagram classifies which combinations of mean and Gaussian curvatures (or, equivalently, mean and deviatoric curvature) are energetically preferred when the local curvatures are treated as independent variables subject only to the geometric constraint $K\leq M^{2}$. A concrete membrane, however, samples only the one-dimensional (if axisymmetric) or two-dimensional subset of $(M,K)$ values actually realized on that surface. The universal optimum (regime I) may therefore be absent, while a direction where energies are formally unbounded from below (regimes II, III, IV) can be cut off by the finite range of accessible curvatures. Consequently, the universal regimes above should be interpreted as local curvature preferences rather than direct predictions of domain position on a concrete membrane shape: the actual localization pattern is obtained only after restricting the energy to the curvature landscape of the chosen membrane shape. We investigate this below, for a prolate and an oblate spheroid.

In the following, we choose the area-equivalent radius $L\equiv\sqrt{S/(4\pi)}$ of the spheroid as the characteristic length scale. For a finite bending rigidity contrast $\Delta_{\kappa}\neq 0$, we can define two dimensionless parameters

together with dimensionless curvatures $\hat{M}\equiv ML$ and $\hat{K}\equiv KL^{2}$. The dimensionless energy landscape $\hat{E}\equiv L^{2}E/(|\Delta_{\kappa}|A)$ as a function of the curvatures then depends only on $\hat{\Delta}_{\kappa m}$ and $\hat{\Delta}_{\bar{\kappa}}$, as well as on the sign of $\Delta_{\kappa}$, via

The factor $L^{2}$ is needed because $E/(|\Delta_{\kappa}|A)$ alone is not dimensionless. For each chosen shape, two planar phase diagrams (one for each sign of $\Delta_{\kappa}$) therefore describe all possible energy landscapes experienced by a small domain (Fig. 4). Importantly, we note that $\hat{\Delta}_{\bar{\kappa}}$ depends on the domain size via the line tension contribution, and in particular decreases toward more negative values with increasing domain size. Thus, trajectories of increasing domain size correspond to decreasing $\hat{\Delta}_{\bar{\kappa}}$ at constant $\hat{\Delta}_{\kappa m}$.

For the prolate spheroid, both $\hat{M}$ and $\hat{K}$ increase monotonically from the equator to the poles. Because growth lowers $\hat{\Delta}_{\bar{\kappa}}$, we would expect that the preferred location of the domain should progress toward the poles as it grows. Figure 4(a,b) shows that this is indeed the case. Large enough $\hat{\Delta}_{\bar{\kappa}}$ selects the equator, while sufficiently small $\hat{\Delta}_{\bar{\kappa}}$ selects the poles. The opposite is true for the oblate spheroid, for which $\hat{M}$ and $\hat{K}$ are largest at the equator and decrease toward the poles. Here we find that, as $\hat{\Delta}_{\bar{\kappa}}$ decreases during domain growth, the preferred location changes from the poles toward the equator, as seen in Figure 4(c,d).

Interestingly, in both cases the transition can be continuous or discontinuous. Taking the prolate spheroid as an example, in a continuous transition, the domain moves progressively from the equator to one of the two poles as it grows, as the energy minimum progressively shifts from the equator to the poles. In a discontinuous transition, local minima at the poles and at the equator coexist. A growing domain may remain at the equator even after this minimum has become a metastable, higher energy state relative to the lower energy minimum at the pole. In the absence of a thermally-activated escape, only beyond a critical domain size does the local minimum at the equator lose stability and the domain migrates directly to one of the two poles. Bifurcation diagrams for both the continuous and discontinuous case are shown in Figure 4(e,f).

In both cases, we find that whether the transition is continuous or discontinuous depends primarily on the spontaneous curvature contrast $\hat{\Delta}_{\kappa m}$. Above a critical value of $\hat{\Delta}_{\kappa m}$, the transition is continuous, while below this critical value the transition is discontinuous. Note that very close to the critical value, near the boundary between these behaviors, a discontinuous jump involving an intermediate latitude can be followed or preceded by continuous migration. The critical value of $\hat{\Delta}_{\kappa m}$ is negative for $\Delta_{\kappa}>0$ (domain stiffer than background), and positive for $\Delta_{\kappa}<0$ (domain softer than background). For the particular case of zero spontaneous curvature for both domain and background (which corresponds to $\Delta_{\kappa m}=0$), this implies that stiff domains undergo a continuous transition whereas soft domains undergo a discontinuous one.

Taken together, the two spheroid examples show that the small-domain theory developed here does not require geometrically elaborate shapes to produce rich behavior. Smooth, monotonic curvature profiles are already enough to realize both continuous and discontinuous migration. On more complex shapes, coexistence between multiple local minima, as well as a multitude of bifurcations and thus domain migration between these minima can be expected.

In order to estimate whether the effects predicted here are observable in experiment, be it with model membranes or in cells, it is important to place bounds on the key parameters and to estimate their experimental values. Rigorous bounds can be put on the Gaussian modulus contrast $\Delta_{\bar{\kappa}}$ based on general considerations. In particular, local nonnegativity of the Helfrich energy density (9) with respect to arbitrary principal curvatures requires (5) $-2\kappa\leq\bar{\kappa}\leq 0$. In the present context, these inequalities (applied to both the domain and the surrounding membrane) imply that the Gaussian modulus contrast must satisfy the bounds

With these constraints, any of the four universal regimes of curvature preference in Fig. 3 are accessible in principle. Experimentally, however, it is often observed that $\bar{\kappa}\approx-\kappa$ (11, 5). Again applied to both the domain and the surrounding membrane, this suggests that we should typically expect $\Delta_{\bar{\kappa}}\approx-\Delta_{\kappa}$, i.e. the Gaussian modulus and bending rigidity contrasts will be of opposite sign and similar order of magnitude. With this constraint, only regimes I, II, and IV of Fig. 3 would be accessible, with regime III remaining out of reach.

To be more concrete, let us consider model membranes with coexisting liquid-disordered ($L_{d}$) and liquid-ordered ($L_{o}$) phases as a potential experimental realization. The $L_{d}$ phase is generally substantially softer than the $L_{o}$ phase, with bending rigidities typically of order tens of $k_{\mathrm{B}}T$ for $L_{d}$ and values reaching hundreds of $k_{\mathrm{B}}T$ for the $L_{o}$ phase, depending on composition and measurement method. (2, 24). Taking $20\,k_{\mathrm{B}}T$ and $100\,k_{\mathrm{B}}T$ as representative values, we would expect $\Delta_{\kappa}\approx\pm 80\,k_{\mathrm{B}}T$ and $\Delta_{\bar{\kappa}}\approx\mp 80\,k_{\mathrm{B}}T$, where the upper and lower signs correspond to small $L_{o}$ domains in a $L_{d}$ matrix ($L_{o}$ in $L_{d}$), or to small $L_{d}$ domains in a $L_{o}$ matrix ($L_{d}$ in $L_{o}$), respectively. Line tensions between coexisting $L_{o}$ and $L_{d}$ domains are typically in the sub-pN to few-pN range, making $\lambda\approx 1\,\mathrm{pN}\approx 0.25\,k_{\mathrm{B}}T/\mathrm{nm}$ a representative value (3, 24).

The line tension and the bending rigidity contrast (together with the constraint $\Delta_{\bar{\kappa}}=-\Delta_{\kappa}$) define a characteristic domain radius $r_{*}=4|\Delta_{\kappa}|/\lambda$, see Eq. 16. Interestingly, for the estimates above, we obtain $r_{*}\approx 1.3\,\mu$m, which is in the typical range of domain sizes achievable in $L_{o}/L_{d}$ experiments with model membranes. In particular, Fig. 3 predicts that domains larger than this characteristic radius are found in regime II. Domains smaller than this critical radius, on the other hand, will be in regime I in the case of $L_{o}$ in $L_{d}$ (as $\Delta_{\kappa}>0$ in this case), while they will be in regime IV in the case of $L_{d}$ in $L_{o}$ (as $\Delta_{\kappa}<0$ in this case).

For these same parameters, we can rewrite the definition of $\hat{\Delta}_{\bar{\kappa}}$ in Eq. 20 as $\hat{\Delta}_{\bar{\kappa}}=\mp 1-r/r_{*}$, where the minus and plus signs correspond to $L_{o}$ in $L_{d}$ ($\Delta_{\kappa}>0$) and $L_{d}$ in $L_{o}$ ($\Delta_{\kappa}<0$), respectively. Comparing the accessible ranges as the domain radius increases from $r=0$ with the diagrams in Fig. 4, it appears that in the special case of zero spontaneous curvature ($m=m_{d}=\hat{\Delta}_{\kappa m}=0$) experiments would be able to test the transition equator-pole transition only in the case of $L_{o}$ in $L_{d}$ ($\Delta_{\kappa}>0$), as in the case of $L_{d}$ in $L_{o}$ ($\Delta_{\kappa}<0$) even small domains are already already have the same preferred configuration as large domains.

The magnitude of spontaneous curvature depends on the extent to which the membrane is asymmetric, and the cause of this asymmetry. For asymmetric exposure of small molecules such as sugars to the membrane, values as low as $m\approx 0.01\,\mu$m^-1 have been reported, whereas for asymmetric adsorption of BAR domain proteins, values as large as $m\approx 50\,\mu$m^-1 are possible (15). Taking $m\approx m_{d}\approx 1\,\mu$m^-1 as a typical value, and $L=10\,\mu$m as a characteristic lengthscale of the curved membrane, we see from Eq. 20 that we can expect values of $|\hat{\Delta}_{\kappa m}|\approx|m|L\approx 10$, well within the range that allows exploration of the different behaviors predicted in the diagrams in Fig. 4. In fact, in order to explore values of $\hat{\Delta}_{\kappa m}$ of order one it is likely more convenient to use smaller spontaneous curvatures, such as those induced by asymmetric exposure to small molecules such as sugars and salts.

The above considerations show that experimentally accessible parameters could probe the different bifurcations that we predict regarding the spontaneous migration and preferred localization of small domains on the membrane. Importantly, however, lipid membranes are thermal systems subject to thermal fluctuations. Thus, it is important to ensure that the typical energy differences that cause domain migration are large relative to the thermal energy, $k_{\mathrm{B}}T$. Indeed, from Boltzmann statistics we expect that the ratio of probabilities of finding a domain at a point $A$ versus at a point $B$ is given by $p_{A}/p_{B}=\exp{[(E_{B}-E_{A})/(k_{\mathrm{B}}T)]}$, and energy differences $|E_{B}-E_{A}|\ll k_{\mathrm{B}}T$ can be washed out by thermal fluctuations. From Eq. 4, we have

Let us focus on the case of zero spontaneous curvature for simplicity, $\Delta_{\kappa m}=0$. Additionally, for the typical parameters and domain sizes described above, with $|\Delta_{\kappa}|\sim|\Delta_{\bar{\kappa}}|\sim\lambda r$, we see that the typical energies are of order $A|\Delta_{\kappa}(M^{2}_{B}-M^{2}_{A})|$ and $A|\Delta_{\kappa}(K_{B}-K_{A})|$, with $A=\pi r^{2}$ the area of the domain. Let us take the points $A$ and $B$ to be the equator and the poles. For a strongly prolate shape, with $c\gg a$, we have that $M_{\mathrm{p}}^{2}-M_{\mathrm{e}}^{2}\approx K_{B}-K_{A}\approx c^{2}/a^{4}$. Using as an example $a=10\,\mu$m, $c=50\,\mu$m, $r=1\mu m$, $|\Delta_{\kappa}|=80\,k_{\mathrm{B}}T$, we find energies with a typical scale of $\pi r^{2}|\Delta_{\kappa}|c^{2}/a^{4}\approx 60k_{B}T$. Thus, the energy differences involved are significantly stronger than thermal fluctuations, and we expect the predicted effects to be observable in appropriately designed experiments.

We have shown that a small mobile lipid domain on a fixed curved membrane experiences a local energy landscape whose shape is controlled by the bending rigidity contrast, spontaneous curvature contrast, Gaussian modulus contrast, line tension and domain size. This local energy yields a very simple universal classification for the curvature-preference of small domains, whose regime boundaries are independent of the spontaneous curvature contrast. Further, the local energy prediction of its preferred localization (or coexisting localizations) on any chosen shape, to leading order in the small size of the domain. Calculations for prolate and oblate spheroids show that even very simple closed surfaces can display both continuous and discontinuous migration, and that spontaneous-curvature contrast remains an essential control parameter in this case. Typically, we predict relocation of the domains upon growth by coalescence beyond a critical size, and comparison with experimental values suggests this to be a testable prediction.

Our work has various limitations, of which the most salient one is that it is restricted to the case where the membrane shape is fixed, e.g. by strong adhesion to a solid substrate, cortex, or scaffold. If adhesion or scaffolding does not suppress local shape change, the domain can alter the curvatures that attracted it, and line tension can drive budding (12). Of course, a second important limitation is that our results are constrained to the regime in which the domains are much smaller than the local curvature radii of the membrane, as well as the characteristic lengthscale over which this curvature varies. As the domains grow beyond the realm of applicability of our approximation, the determination of preferred shapes and locations of membrane domains becomes a rather involved exercise (6, 7). Finally, our work is based on a quasistatic approach, focused on determining the energy landscape experienced by the domain. Understanding the dynamics of domain migration will require dealing with the effects of membrane hydrodynamics and substrate friction, for which classical membrane-inclusion diffusion provides a starting point (21). These are left as avenues for future work.