Eur Biophys J (2012) 41:161–175

DOI 10.1007/s00249-011-0768-2


A method for analysis of lipid vesicle domain structure
from confocal image data
Peter Husen • Matthias Fidorra • Steffen Härtel •

Luis A. Bagatolli • John H. Ipsen

Received: 23 March 2011 / Revised: 27 July 2011 / Accepted: 13 October 2011 / Published online: 9 November 2011
Ó European Biophysical Societies’ Association 2011

Abstract Quantitative characterization of the lateral Introduction
structure of curved membranes based on fluorescence
microscopy requires knowledge of the fluorophore distri- In the last decade giant unilamellar vesicles (GUVs)
bution on the surface. We present an image analysis became a very popular membrane model to explore the
approach for extraction of the fluorophore distribution on a lateral structure of diverse artificial and natural lipid mix-
spherical lipid vesicle from confocal imaging stacks. The tures containing membranes (Bagatolli 2006). The cell size
technique involves projection of volumetric image data dimension of GUVs (mean diameter *25 lm) makes this
onto a triangulated surface mesh representation of the versatile membrane model system very appropriate for
membrane, correction of photoselection effects and global applying several fluorescence microscopy-related tech-
motion of the vesicle during image acquisition and seg- niques in order to study membrane lateral structure. This
mentation of the surface into domains using histograms. type of experiment opened for the first time the possibility
The analysis allows for investigation of the morphology to obtain spatially resolved information from distinct free-
and size distribution of domains on the surface. standing membrane regions at the level of single vesicles
(Bagatolli and Gratton 1999; 2000; Korlach et al. 1999;
Keywords Confocal microscopy  Lipid bilayer  Dietrich et al. 2001). This information has been used
Lateral vesicle structure  Triangulated surface  to construct phase diagrams for various lipid mixtures
Image analysis (Veatch and Keller 2005), but also to evaluate lateral het-
erogeneity in compositionally complex mixtures [GUVs
composed of natural lipid extracts or native membranes
(Bernardino de la Serna et al. 2004; Plasencia et al. 2007;
P. Husen  M. Fidorra  J. H. Ipsen (&)
Montes et al. 2007)] as well as to evaluate partitioning of
Department of Physics and Chemistry, MEMPHYS Centre
for Biomembrane Physics, University of Southern Denmark, different membrane proteins into these different membrane
Campusvej 55, 5230 Odense, Denmark regions (Kahya et al. 2005).
e-mail: The distinct membrane domains observed in GUVs are
generally related with equilibrium thermodynamic phases
S. Härtel
Laboratory for Scientific Image Processing (SCIAN-Lab), typically using: (1) the partitioning of different fluorescence
Biomedical Neuroscience Institute (BNI), probes into the distinct coexisting regions in the membrane
Anatomy and Developmental Biology Program, or (2) by performing comparative analysis of different
Faculty of Medicine, University of Chile,
fluorescence parameters (such as fluorescence probe diffu-
Independencia 1027, Independencia, Santiago, Chile
sion coefficient, or fluorescence emission shift) measured in
L. A. Bagatolli the different membrane regions [see, for example, (Korlach
Membrane Biophysics and Biophotonic group/MEMPHYS et al. 1999; Bagatolli and Gratton 2000)]. However, dif-
Centre for Biomembrane Physics, Department of Biochemistry
ferences in the aforementioned parameters between coex-
and Molecular Biology, University of Southern Denmark,
Campusvej 55, 5230 Odense, Denmark isting membrane domains are necessary but not sufficient
e-mail: conditions to demonstrate the presence of equilibrium


In ‘‘Domain charac- that the composition in the whole GUV population is rep. Inc. our results proved projected images will be presented. UK) with an amplitude of 1. and the presented results are proof of min). particular a spherical surface. other lipid mixtures and further validated with information obtained from other biophysical techniques. Thus. sphingomyelin plane near one of the poles to measure correlation lengths (egg. Our method takes into account the characteristics of fluorescent probes dissolved in organic solvent (Cl3CH/ the fluorophores involved and the specific experimental MetOH 2:1 v/v) were deposited on each Pt electrode (4 ll setup.30 . Naphtopyrene was purchased from Sigma. the effects of photoselection are accounted for. fluorescence microscopy experiments pho-topological characteristics of domain structures in quasi-spherical GUVs imaged by confocal fluorescence GUVs of different compositions were prepared following microscopy. 1. 2005) and improved our understanding and 2-(4..P. Once the solution reached 123 . In cal fluorescence microscopy data of GUVs. Similar approaches have been tested by others for closed with a short discussion. The setup can easily be temperature of the different lipid mixtures. characterization of the lateral structure.4a-diaza-sinda- of enzyme coupled lipid domain formation significantly cene-3-pentanoyl)-1-hexadecanoyl-sn-glycero-3-phospho- (Fanani et al. (2008) have used a single confocal toyl-sn-glycero-3-phosphocholine (DPPC). In this article we present a new image analysis approach Preparation of GUVs and confocal laser scanning that facilitates a detailed quantitative analysis of the mor.2-Dioleoyl-sn-glycero-3-phosphocholine (DOPC). ceramide (egg.2°C/ physical phenomena. we introduced a sequential image analysis approach methods’’ section lists the materials used and describes the (involving segmentation and 3D reconstruction) for confo. from the volumetric image Briefly. We showed that ‘‘Projection of confocal data. For free-standing lipid bilayers. After the removal of the organic the fluorescence intensity map. in (1992) using a custom-built chamber (Fidorra et al.4-difluoro-5. Spain) using a temperature ramp (*0. Specific advantages of this solvent the chamber was filled with a sucrose 200 mM method are that it provides identification of domains and solution and an AC field was applied to the chamber using their boundaries (as closed curves). image processing routines have opened access to a purchased from Avanti Polar Lipids.2-dipalmi- Honerkamp-Smith et al. In a previous work (Fidorra et al. The article is GUVs.’’ the techniques of repre- lipid domain area fractions can be retrieved from GUVs and senting confocal image data of a GUV on a sphere are this information can be used to validate predictions from introduced.3. the choline (Bodipy-PC) were purchased from Molecular Probes diverse morphologies observed for solid domains in GUVs (Invitrogen). terization. of 0. but Materials potentially also offer the possibility to obtain additional morpho-topological parameters normally lacking in fluo.162 Eur Biophys J (2012) 41:161–175 thermodynamic phases. 2009). 1. i. 1. cholesterol and along the domain boundary for a single domain. This is achieved by reconstructing an image of the electroformation method described by Angelova et al. extended to quantify new properties of heterogeneous the GUVs chamber was cooled at room temperature in a GUVs. chicken).’’ the tools for equilibrium thermodynamic phase diagrams (such as the analysis of the lateral structure of the GUV based on the lever rule). require refined analysis of the fluorescence images. We demonstrate Stenson. analysis technique rather than new insights in specific bio. procedures for preparation and imaging of GUVs. and it provides a global a function generator (Vann Draper DigimessÒ Fg 100. The article is organized as follows: the ‘‘Materials and 2009). Selecta. Image analysis approaches not only Materials and methods provide a necessary tool to properly determine phase equi- librium in compositionally simple lipid bilayers.2-dioleoyl-sn-glycero-3-phosphoethanolamine-N-(lissamine monolayers observed by time lapse microscopy in Langmuir rhodamine B sulfonyl) (ammonium salt) (Rh-DOPE) were troughs.10 -Dioctadecyl- quantitative description of morpho-topological properties 3.3V and a that a range of surface quantifiers for the domains can be frequency of 10 Hz. auroyl-sn-glycero-3-phosphocholine (DLPC). Barcelona. 2006). In ‘‘Analysis on the surface. For 2D lipid 1. rium conditions in our samples. The last step was done in order to achieve equilib- concept examples of quantitative characterizations. The emphasis in the presentation is put on the image time span of approximately 5 h in an oven (J.e. The electroformation was carried out identified such as vertex angles on polygonal domains and for 90 min at temperatures above the main phase transition curvature of domain boundaries. 2010). the membrane in a two-dimensional representation. Subsequently. NMR (see Juhasz et al. Among other observations. 1.’’ examples of the results from the described resentative of the initial composition utilized to prepare analysis are given for two example vesicles.30 -tetramethylindocarbocyanine perchlorate (DiIC18) (Härtel et al.2 mg/ml lipid stock solution). aliquots of the desired lipid mixture containing the data.7-dimethyl-4-bora-3a. and the solvent was and the domains are identified on the basis of an analysis of evaporated under vacuum. Derby. chicken).2-dil- rescence microscopy experiments involving GUVs.

For clarity Circle fits the scaling factor c = Dz/Dx has been applied to both the ‘‘before’’ and ‘‘after’’ images in this figure.Eur Biophys J (2012) 41:161–175 163 room temperature.2 was used in our experiments. a long mixtures used in this article are: (1) DOPC/DPPC/choles. The laser lines were reflected to the sample Estimator. The density difference between directed to the PMT detectors using a mirror.5 mol% of DiIC18 acquired at a sampling rate of 70 nm for Dx and Dy. the vesicles were transferred to an iso. the detector of DiIC18 and Bodipy-PC (Figs.2 mol% with respect to total lipids of Naphtopyrene and Naphtopyrene and BodipyPC. A C-Apochromat 409 water immersion objective lands). The temperature during image rated in front of the PMT detectors in the two different acquisition was controlled at 20. Hilversum. respectively (Figs. The temperatures for electroformation were 340 nm for Dz. deconvolved by Huygens Scripting Software using an pyPC and DiIC18 respectively) were used as excitation algorithm based on the Classic Maximum Likelihood sources. quency. The image stacks were SM/egg ceramide (70:30) doped with 0. which is slightly above the Nyquist fre- (1) 55°C. and within a few from the laser sources (NFT 545 or NFT 490) in a two minutes the vesicles are ready to be observed using an channel configuration. 1. minimizes vesicle movement and photobleaching. The acquired intensity Rh-DOPE. The GUV channels to measure the fluorescent intensity. DiIC18 glucose ? 50 ll of the GUVs in sucrose in each of the and Bodipy-PC and HFT 458 for Naphtopyrene).5 and and band pass filters of 500 ± 20 nm and 525 ± 25 for 0. (2) 50°C and (3) 70°C. which is then translated so the center of the circle is on a common axis for the whole stack. 4). Additional filters were incorpo- inverted microscope. 6. The eight wells of the plastic chamber used. The circle radii are then used for another circle fit (the thick circle on the top right). Lab-tek Brand fluorescence emission collected through the objective was Products. Generally. 1 A circle is fitted to each Before alignment After alignment slice. essentially fitting a sphere to the whole point cloud. The Nether- scope. obtained confocal raw fluorescence image stacks were rene and Rh-DOPE and 488 nm and 543 nm for Bodi. The sampling above the Nyquist frequency was with a NA 1.5°C. Fig. the interior and exterior of the GUVs induces the vesicles a beam splitter was used to eliminate remnant scatter to sink to the bottom of the chamber. 12). pass filter [560 nm for Rhodamine-DOPE and DiIC18. (Scientific Volume Imaging. 11). and (Figs. 8. (3) egg gain and the detector offset.e. Naperville IL). which image stacks were acquired using multi-track mode.. i. 9. 7. (2) DLPC/DPPC images were checked to avoid PMT saturation and loss of (30:70) doped with 0. The Argon and NeHe lasers (458 and 543 nm for Naphtopy. through the objective using different dichroic mirrors osmolar glucose solution in a special chamber (200 ll of (HFT 488/543/633 for exciting Rhodamine-DOPE. Confocal image stacks were acquired on a Zeiss LSM and *140 nm for Dz with Huygens Scripting Software 510 Meta confocal laser scanning fluorescence micro. terol with various molar fractions doped with 0. 3. Two channel necessary to guarantee sufficient scan speed.25 mol% with respect to total lipids offsets by carefully adjusting the laser power. which was calculated to *40 nm for Dx and Dy. even though it is a fitting parameter Point cloud 123 .0 ± 0. 10.

rÞ pairs. 2 h is the polar angle. The circle fits are calculated as least point cloud. such as spinning-disk confocal microscopy. i. so the image stack can benefit significantly from deconvolution..e. so the res. Fig. each corre- sponding to a constant z. where r is the is seen by displacements between the centers of the circles circle radius and ~z is the slice number. is the motion of domains on the surface. i. the intensities of fluorescence are recorded in separate channels. our images are is carried out by choosing points by intensity thresholds not suitable for such characterizations. The sphere is approximated by a deltahedron. we refer θ to these channels as red. which cannot easily be cor. and the slices y-directions. i. green and blue (RGB) values for φ each pixel. Concerning liquid-liquid coexistence.p as the poles is rather broad (compared to the voxel separation) in the and h ¼ p2 as the equator z-direction. Three-dimensional image data are provided in the form of stacks of bitmap images. however.’’ The motion of the vesicles may be further reduced using an adaption of the immobilization Triangulation of the sphere technique described by Lohse et al. which hedron with triangular faces. a circle is fitted to each slice. The point spread function of the imaging technique takes values in the range (-p. 123 . Typically. might Unfortunately. several minutes.. are then translated to have the circle centers on a common which essentially constitutes a fit of a sphere to the entire axis (see Fig. where R is the sphere radius and zc is the z-coordinate of This does not correct for motion in the remaining the center..e. rotation and translation in the known from the equipment settings during acquisition or z-direction.and displacements.e. as the vesicles included as a fitting parameter. To correct for the voxel depth in the z-direction and the width in the x. i. The scaling parameter c can be provided as degrees of freedom. slices. where c = Dz/Dx is the ratio between the distortion within the individual slices. Faster imaging tech- point cloud to a sphere. the physical separation between slices is greater than the physical separation between neighbor- ing pixels in the xy-plane by a factor of 3–6. We do not see significant is given by z ¼ c~z. a poly- rected for. ‘‘Materials and methods. GUVs with a percolated gel phase are most suitable for morphological characterization of domain boundaries. and plane. y)-coordinates obtained by thresholding the image by intensity. the angle from the x-axis of the projection onto the xy-plane. while / is the azimuth angle. GUVs can undergo significant random enable morphological characterizations of the domain motion while the image stack is acquired. and takes olution in the z-direction is lower than in the xy-focal values in the range (0. 2 for a definition of the equator and the poles. We have in general chosen are resting on the bottom of the chamber because of the to do the latter and check the result with the expected density difference of the sugar solutions described in value. the z-coordinate depicted on adjacent slices.164 Eur Biophys J (2012) 41:161–175 Projection of confocal data Confocal microscopes provide bitmap images of GUVs with fluorophores partitioned into the membrane. These pairs are then used for another circle fit. Translational motion of a spherical vesicle The circle fits provide a set of ð~z. 1). which may take structure on these systems as well. niques.e. Domains near the equator1 are more distorted than domains 1 near the poles. since they span a greater number of slices See Fig. The latter is reduced. In order to project the image onto a sphere.. p] (in radians).. i. i. ð2Þ using a set of (x. and thus are imaged over a longer period of time. p]. For this Alignment of confocal stacks reason. we need to since the gel phase in this case fixes the domain structure in align the model sphere with the vesicle in the image. The vertices are located on causes a distortion in the shapes and sizes of domains. In this article. the angle from the z-axis. If mul- tiple fluorophores are in use. Another type of motion.e. We refer to h = 0. This second fit is the least squares solution of squares solutions (see ‘‘Appendix 1’’) of the equation the equation ð x  xc Þ2 þ ð y  y c Þ2 ¼ r 2 ð1Þ r2 þ ðc~z  zc Þ2 ¼ R2 . so we extracted only applied to the individual channels and fitting the resulting total area information from these.e. a plane parallel to the xy-plane. (2008) to GUVs.. This place.

the point cloud is aligned and the total area though. deltahedron onto the sphere. however.2 R. we cannot use a perfectly mentations of local calculations such as finding gradients regular deltahedron. and of consisting of a triangle and its nearest neighbors (see these the tetrahedron (4 faces). domains. e. on A triangulation with a desired characteristic spacing D is image data on the surface using these small neighborhoods. A triangular cone is defined by the center of the sphere and a triangle on the surface. In because of triangles of vanishing area or with a vanishing terms of the spherical coordinates h and / defined in vertex angle. This should be reasonable as the triangulation of the sphere.g. and the deltahedron corresponds to a triangu. The integral of the intensity over the enclosed volume is assigned to this triangle in the projection. There are a image. e. Projection  hi  vided into 2p sin D discrete values. regular polygons as faces. 3 triangle on the surface and assign it to that triangle (see Having identical. Using a mesh with these properties provides a uniform lation of the sphere. 3 Conceptual diagram of the projection of a point cloud (bottom left) onto a triangulation of a sphere (top left).2 The only convex. 3. It is then ‘‘projected’’ onto the triangular mesh. so the triangles become curvilinear The idea is to integrate the image data that are seen polygons on the sphere.06 radi. which facilitates simple imple- want a much finer triangulation. These need to be divided into triangles to obtain a the radial coordinate. image stack does not provide a true radial resolution of the 123 . few notable defects in the form of triangles that are half the that are confined to the geometry of the surface. a Gaussian or a median filter. 0. The producing a two-dimensional representation of the image. A local neighborhood are the five Platonic solids (Heath and Euclid 1956). In order to area of the majority. constructed by first generating a set of points that are It also avoids the risk of numerical stability problems approximately evenly distributed on a spherical surface. and applying filters. and for each i. optionally by radial coordinate. 2. 1996) was chosen for this purpose.g. and R is the radius of the sphere. These make up a vanishing fraction of produce the surface image. the range of / is subdi. the range of h is first subdivided evenly into Dp discrete values hi. We will. so the individual triangles cannot be seen here the sphere.8 R \ r \ 1. regular3 polyhedra resolution throughout the surface. This reduces the dimensionality by integrating out It happens that four nearby points are coplanar. The triangulation is then calculated as the triangulated4 The objective of the projection is to use the triangles of the convex hull of this set of points.. but only an approximation. scaled as described above. where r is the radial coordinate.. thus (Barber et al. A triangulation with D = 0.Eur Biophys J (2012) 41:161–175 165 Fig. i. 3). resulting in quadrilateral faces. resulting triangulation has triangles with internal angles This representation. is suitable for investigating structures. 4 Fig. the octahedron (8 faces) and Fig. filtered by intensity and ans is shown in Fig. the 2 The triangulation is obtained as the radial projection of the details of which are discussed below. Since we generally throughout the surface. which we will refer to as a surface between 45° and 90° and nearly identical areas. refer to the approximating when looking from the center of the sphere out through a deltahedron directly as a triangulation. The resulting projection on the right is using a finer triangulation than depicted on the left.e. 5) has approximately the same shape and size icosahedron (20 faces) are deltahedra. The software Qhull mesh as pixels for an image of the vesicle surface.   Fig..

Another way of visualizing the surface is a (h. each voxel is treated as a point The voxels are stored as a list of coordinates and located at the voxel center. y and z. For a given channel. some probes are subject to a photoselection effect because of their orientation with respect to the polarization of the ð3Þ incoming laser light (Bagatolli 2006). and the intensity value for each intensities in order to allow for the coordinate displace- channel of the voxel is simply added to the triangle that ments in the aligning step and the removal of individual intersects the line from the the center of the sphere to this voxels in the filtering step. and Fig. possible to correct for this effect if it can be modeled as a grands and bounds in integrals over different parts of the function of the spherical coordinates h and /. F(h. In the first attempts. Say the volume. /Þ: ð4Þ 5 If the vertex order is v1. v3 as we move counterclockwise when This factor can be approximated as a linear combination of looking from outside the sphere. product between the position vectors of the voxel center This approach produces good results for coarse trian. In other words. an equirectangular projection. To prevent this. since it is much thinner than the resolution limit this volume calculation is by far the most computationally of the microscopy technique by a factor of *50. voxel and has intensity data. we can start as a box with side lengths equal to the voxel separation for doing image manipulation and analysis in this geometry. This is partic- ularly the case near the equator. where the gaps between points from adjacent slices can be seen from the center of Analysis on the surface the sphere and result in black bands on the surface. every point in the imaged volume is part of some OOGL used by the program Geomview (Amenta et al. /Þ ¼ Fðh. and there is an Due to the number of branches that typically occur and the orientation specific factor. dependence of the intensity response of the fluorophore. the intensity over the corresponding triangular cone. inequalities. The volume of intersection with the gular meshes. / volumes of the intersections between the box-shaped ) map. and the channel inten- large compared to the voxel depth (Dz). The each dimension rather than just a single point in space. Since the voxels are thus not point. 3). since recursively tested for intersection between their corre- the triangles become small enough that their corresponding sponding cones and the voxel. the probe is correlated to the orientation of the surface (and shaped voxel is described by inequalities that are simply therefore the position on the surface). and the projection is carried cones may completely miss any points from the point cloud out for these as well in case of non-zero intersection. the 1995).e. demanding part of the projection. for finer sities of the voxel are added to the triangle using this meshes. a triangle is at the center of the sphere and the triangle as its base (see located by walking the surface mesh towards increasing dot Fig. 12 shows a derived surface image in six different Calculating this integral involves calculation of the orientations. for each channel the intensity value easily located by their coordinates. r  ðv2  v3 Þ [ 0. the result is an angular constant bounds on x. The algorithm used for this volume calculation (Ong The surface representation of the image data makes it et al. the projection is carried assigned to a triangle is the sum of the intensities of the out by looking at the voxels one at a time and for each voxels with centers inside the triangular cone with its apex locating the triangles to project it to. this approach breaks down. The nearest neighbor triangles are then throughout the surface. /). 2003) recursively constructs polynomials as inte. First. A point r is inside the triangular cone if and only if it is on the internal side of each of the three planes Correction of photoselection effect containing the center of the sphere and one edge of the triangle: A problem with fluorescence microscopy of GUVs is that r  ðv1  v2 Þ [ 0.. This region is a convex Fig.5 The box.166 Eur Biophys J (2012) 41:161–175 membrane. for this molecule. /Þcðh. concentration of a fluorophore is c(h. due to photoselection corresponding number of polynomials that are constructed. r  ðv3  v1 Þ [ 0. and thus not get any intensity information. i. where the side lengths of the triangles are cone of this triangle is calculated. each voxel of the point cloud is instead treated Once the image is represented on a surface. v2. Figure 3 shows an example of such a visualization. Since the orientation of where vis are the position vectors of the vertices. In image is stored in the OFF format of the geometry language this way. /). such that the measured intensity becomes iðh. which are needed to utilize the available resolution volume as a weight. However. 4. This has the advantage that the whole surface can be polyhedron and can be represented by a set of linear seen at once at the cost of shapes being distorted. as exemplified in voxels and the triangular cone. and the triangle center. while the inequalities are regarded one at a time. a number of spherical harmonics (see ‘‘Appendix 2’’): 123 . which can be used to visually inspect the surface intensity value assigned to a triangle is now the integral of image.

‘‘Edge map’’) the noise is amplified hFi ‘ [ 0.g.m f‘m Y‘m ðh. it may lead to many very small domains ‘. the surface image has channel intensities (see Fig. The non-normalized convolution mask used for a Gaussian filter is shown in Fig. Spherical harmonics with (‘. ±4) have been included in this case X Fðh. from which we get Histograms and segmentation im f‘m ¼ ‘ : ð8Þ i00 Y00 Information about the distribution of the fluorophores on Figure 4 shows an example of correction for such a photo. we can e. In segmentation and domain 123 . ð5Þ identification (see Sect. For correct for the photoselection effect: simplicity these are implemented by only looking at very ~ /Þ  hFicðh. The images are (h. the representation of the structure in these regions on the surface Fig. (4. /ÞY‘m ðh. /Þ . In edge X F‘m ¼ hFi 1 þ f‘m Y‘m ðh. consisting of four triangles showing the non-normalized Noise reduction convolution kernel used for an approximately Gaussian filter A noisy image can be problematic for the various image analysis techniques. ð7Þ filter can be applied multiple times. m) = (0.m consisting of only a few triangles to be erroneously iden- ! tified.. where we have taken the mean value to be given by The effect of noise can be reduced by applying filters. hF i ¼ F00 Y00 :If we can determine the factors f‘m . After the correction. (4. 5. P : ð6Þ 1 þ ‘ [ 0. near the edges of the proper domains. Before the correction. 0). e. 2). 0). /Þ small neighborhoods including a triangle and its three iðh. f‘m ¼ . 6). and data represented on a spherical surface. 0). 4 An example of correction of a photoselection effect. (2. ±2). 5 A small neighborhood image is significantly improved.. In order to increase the spread. the membrane can be gained from histograms of the selection effect. a Gaussian or a median filter. /Þ  F‘m Y‘m ðh. /Þ  iðh.m by taking gradients and may dominate the signal from the actual edges.g. /Þ nearest neighbors. (2. thus gradually ‘ S smoothing the image. the histogram can the structure in the green channel characteristic for the rest of be calculated over the surface rather than the imaged the surface cannot easily be seen. (4. the cF‘m ¼ im  iðh. This way of The coefficients f‘m can be determined using a reference applying the filter is directly applicable to general (not vesicle that is expected to have a constant concentration c of necessarily spherical) triangular meshes as long as the the fluorophore by taking the inner product of the measured shapes and areas of the triangles are within a satisfyingly intensity with the corresponding spherical harmonics: ZZ narrow distribution. /Þ identification. /) maps of the surface with the poles at the left and right (see Fig.Eur Biophys J (2012) 41:161–175 167 Before After φ θ Fig. Since we have the image two regions where the intensity in the red channel is high. /ÞdX. ±2). to the surface image.

8 1 1. coexistence region in the phase diagram can be characterized The image can now be segmented into phases by fixing a in detail. The local minimum indicated at / = 0.679 evaluated by visual inspection of the resulting segmenta- Area / (intensity unit) φ = π/4 450 tion of the surface image.2 1.4 0. structure of the membrane. A histogram of this value over the Area / radian 20 8 Total area surface is also shown in Fig. the angle from the red axis in and sizes of the peaks. it is defined by only one parameter: the φ = 0. Once a choice has been made. This method provides a systematic means of segmen- Fig. Area/radian Area/radian (smoothed) If for simplicity the curve is chosen as a straight line Total red area 30 12 through the origin.6 0 photoselection effect that may lead to position specific Red intensity (ir) variations of the recorded fluorescence intensity. Finally. Thus. 6. composition). It can. for comparison of a number of vesicles. solid angle taken to keep the experimental conditions equal except for the varied parameter (e.3 400 Ideally the phases give rise to sharp peaks such that the 0. however. such that small differences in this region may lines (possible choices of threshold) on the two-channel histogram. The areas are on the unit sphere. 100 including noise. where care is of / is shown.5 0. so the threshold has to be chosen with 0. typical 0. this value for a range of compositions. it can be vides a way to determine the phase of each triangle and 123 . ig)-space into two regions.4 0. By measuring two-phase system we expect to see two peaks in the his.1 0. also be unstable in the two-dimensional histogram. This value as well as / ¼ p4 is also represented as two peaks.679 on the / histogram is found using the cases where there is a wide non-zero ‘‘floor’’ between the smoothed curve. A number of effects may cause such broadening. volume.2 250 histograms show rather broad peaks that overlap in the tail 0. the area near domain boundaries (if the 0. However. the lipid membrane.168 Eur Biophys J (2012) 41:161–175 2 φ = 0. and the local minimum between 10 4 them is a good candidate for a threshold that satisfies the 5 2 above considerations.. i.4 1.679 φ = π/4 i 25 10 angle / ¼ tan1 igr from the red axis.. On this one-dimensional 15 6 histogram. choice of threshold is guided by the observed behavior of the histogram and the criterion that the threshold curve should have one peak on either side and preferably pass Domain characterization through a low-valued region of the histogram so that the sensitivity of the segmentation to small variations of the The segmentation described in the previous section pro- curve is low. 6 Top A two-dimensional histogram of the channel intensities of tation for a variety of vesicles with these general properties a two-channel surface image (the one  shown in Fig. This results in the phases occupying The area fraction of the red phase is also shown as a function different positions in the intensity histogram. the fluorophores are expected to have dif. 6. The information in the histogram along with a choice of In general. the anatomy of the togram as is the case in the example shown in Fig. Such a study is being worked on by the authors for threshold.6 kernel and then walking the curve in the direction of φ decreasing value.15 200 regions as in Fig. the total red area as a function of the chosen threshold value Thus. threshold yields the total area of one of the phases and hence ferent affinities to the different thermodynamic phases of provides a measure of the degree of transition between them.e. 6. Bottom A ig with a reasonable tolerance to differences in the positions histogram of the value / ¼ tan1 ir .2 0.6 0. over the surface. for a of the threshold in the / histogram in Fig. 6.1 150 care. cause the threshold to end up in vastly different locations. 3). it may be a better choice to simply fix the threshold value of / to a constant. which for a two-channel image takes the form of the liquid-ordered/liquid-disordered coexistence region of a curve that divides the (ir.g. 0.3 0.25 350 threshold curve can easily be chosen to pass through a Green intensity (ig ) 300 zero-valued region of the histogram. The the DOPC/DPPC/cholesterol mixture. This is desirable when investigating the lateral since this approach is simpler and more stable. two peaks corresponding to the two coexisting phases are easily seen. where ir and ig are the red and green intensities.2 0. This minimum is identified by first smoothing the histogram by convolution with a Gaussian 0 0 0 0. if these 35 14 effects are not perfectly corrected.05 50 transition across them is smooth) or effects such as the 0 0 0.

Domain morphology Figure 7 also shows histograms of the domain areas (for domains with A [ 0. ð12Þ interfacial energy.197 A) surface image into the two phases. Despite the limited number of domains. along with the distribution from Eq.01) using various spreads for a Gaussian Inverting this function gives kernel compared to the hyperbola corresponding to exponential fit above 1 a nðAÞ ¼ ln : ð10Þ b A The area distribution of the domains can now be of a domain picked at random from an ensemble of vesicles approximated by with the same size and composition. AðnÞ ¼ aebn : ð9Þ Bottom Kernel density histograms of domain areas (excluding domains with area A \ 0.. Apart from domain sizes. as it is not normalizable for the range A 2 ½0 : 4p. Equation 12 pro. 7 the areas of the green domains are sorted 20 by decreasing area and plotted on a logarithmic scale. This is achieved by finding the connected regions of the same phase on the surface. The dashed line is a fit of an exponential function. it appears we 0 can still say something about the area distribution.05 this analysis. while textures in more ordered phases 2pr2 i can give rise to different shapes (Bernchou et al.6 0. A proper statistical analysis of domain areas would This distribution can only apply for a limited range of require a large sample of vesicles with the same size and areas.197 n vidual phases coexisting on the membrane. 7 Top domain area vs. we need a way to identify the individual domains. but it is clearly  dn   ¼ 1 : ð11Þ dA bA a very small sample of domains to produce a reliable his- togram. 11. number when domains are sorted by decreasing area.10 mum in the / histogram was used for segmentation of the 80 σ = 0. If we are interested in more specific information about the lateral structure. The threshold determined by the local mini. The histogram seems   to follow the model distribution somewhat. Since the histograms show a 100 clear separation into two phases.2 0. 6. σ = 0. In Fig. i. we have not been able to reliably 123 . the example is suitable for σ = 0. i.Eur Biophys J (2012) 41:161–175 169 thus trivially a way to calculate the total area of the indi. 0. Due to the high mobility of domains in liquid-liquid sep- vides an estimate of the probability distribution of the area arated systems.1 domains. 40 division into many very small domains of only a few tri- angles.857 e-0..01) generated by kernel density esti.e. composition. The segmentation pro- duces a set of 286 domains in total (red and green domains 60 | dn / dA | combined). in a lipid bilayer with liquid-liquid phase separation are 1 X ðAA2i Þ2 expected to be circular because of minimization of the ^ðAÞ ¼ pffiffiffiffiffiffiffiffiffiffi q e 2r . In this section we present 0.4 0.e. 0 0. our method also allows investi- mation (Rosenblatt 1956. 3) are shown in Fig. but many of these are most likely due to noise or below-resolution structures leading to an incorrect sub. Fig. such as the size distribution or morphology of the Domain area (A) 0. Parzen 1962) using a Gaussian gating the morphology of the domain boundaries.20 1 / (0. 3.8 1 Apparently the areas of the first 20 domains follow an Domain area (A) exponential decay when plotted against their index in the ordering.01 examples of quantitative characterization of the domains on the vesicle in Fig. Domains kernel. 1 A = 0.001 Domain size distribution 0 10 20 30 40 50 60 70 The histograms of the surface image of the example vesicle Domain number (when sorted by decreasing area) (n) (Fig. 2009). and an upper bound of the domain area is given by the total area of the phase.

and we wish to identify the vertices of this polygon. 9 The local curvature of the domain boundary depicted in Fig. However. for longer segments it can provide good results and has been chosen for the blue curve in the figure. 8 determined by fitting of small segments (7 points) by parabolas.e. while the appear to be curvilinear polygons (i. triangular domain. This characterization requires distance of the edge to the osculating plane of the circle. 2005). where the deviation from a tangent plane due to properties should be measured over sufficiently large the curvature of the sphere is small compared to the noise. the boundaries of many of the domains curvature.e. characterizing the edges connecting two points on the edge by a geodesic. and rc is the vector from the center of domain boundaries can then be obtained as lists of vertices the sphere to the center of the circle and hence also defines from the mesh.170 Eur Biophys J (2012) 41:161–175 45 Curvature 40 Threshold 35 30 Curvature 54. (Lewiner et al. geometric segments. The boundary segment. since it clearly provides a better fit than the geodesic. their geodesic curvature) and mea. circles) to the edges.6° 10 65. the domains are provided as sets where {ri} is the collection of position vectors of the of triangles of the same phase on the triangular mesh. The resulting curvature estimate is which is a non-geodesic circle on the sphere. This A geodesic approximating an edge is determined by simply involves identifying the vertices.. in which the red phase is a percolated gel Figure 8 also shows approximations of the edges by phase. Therefore. to give examples of boundary characterizations. while the remaining two shown in Fig. Since we are interested in characterizing edges cannot be concluded to deviate from the geodesics from this the curve as a curvilinear polygon (in this case a triangle). The blue between them (e.56 done by applying a threshold to the curvature estimate and characterize the boundary morphology from confocal identifying the positions of peaks of large curvature as images of these systems. i.g. and therefore tends and from the original voxel lattice. blue curve is a non-geodesic circle on the spherical surface. The three green curves are geodesics. This is hence its geodesic curvature is jg = 0.87. The green curves are fitting a parabola to a small segment surrounding the point geodesics (great circles) approximating the edges of the triangle. which gives rise to to favor a near-tangential plane when applied to short staircase-like curves near the equator. circle was found by least-squares minimization of the suring the vertex angles.8° 25 20 15 96. 8 is quite adequate for this purpose. The triangular domain shown in jrc j2 Fig. vesicle in Fig. length scales to overcome these effects.5° 5 0 0 50 100 150 200 Vertex number Fig. the least squares solution of the system ously identify the vertices and to reliably fit simple curves rc ri  ¼ 1... data. The leftmost edge is apparently better approximated by the blue curve. After the segmentation. The threshold is used to identify the vertices and edges of the curvilinear triangle We can estimate the curvature at a point of the curve by Fig. 9. 8 A rather simple. ð13Þ (e. The radius of the blue curve (on the unit sphere) is r = 0. For the 6 This might be possible using faster imaging techniques. defining the boundaries as closed space the osculating plane by r  rc ¼ jrc j2 : curves. These curves suffer from noise as well as artifacts The latter method suffers from not minimizing the dis- from the mesh (they follow the triangle edges of the mesh) tance to the model circle but to a plane. they have vertices).. simple curves and estimates of vertex angles and geodesic On this vesicle. so we would like to characterize them as such. remaining two edges of the triangular domain the geodesics 123 .6 We will thus continue with the vertices and the segments between them as edges. 3.g. good fidelity in the domain shape in order to unambigu.

15 +/. The lengths and areas are on the unit sphere dimensions Dp and DA for the perimeter and area respec- tively by p ¼ Cp LDp were found to be the best fits. Also shown are the identified domain boundaries. the image has only parameter is easy to measure using our method and pro- one channel. 10 Surface image of a GUV prepared from a mixture of egg sphingomyelin and egg ceramide. however.069 +/.05. Figure 11 shows these these variables 0. The vesicle approximately the characteristic value for compact two- shows an intriguing domain structure with flower-shaped dimensional regions.0.0.15 10 chosen such that the segmentation captures much of the 0. i.0. If we define the fractal Fig. for which the fractal dimensions are domains and is thus a good candidate for morphological Dp = 1 and DA = 2. 1 A simple characterization facilitated by our method is Domain area (A) the relationship between the domain perimeter p and domain area A.09 morphology of the domain boundaries while producing a 0. the characterizations of domain boundaries.0.003 p1. 10. 123 . The measured angles are ð14Þ between the tangent vectors of the model curves at their A ¼ CA LDA .005 p reasonable identification of domains. intersections. for this channel.01 adding up the areas of the triangles constituting a domain.0 = 0.001 triangular mesh. our identified power law behavior provides an estimate of the Area and perimeter ratio DDAp : Figure 10 shows a surface image of a vesicle prepared from a For the small domains. For the larger domains. this ratio is DDAp  1:93. which apply at two different regimes Domain perimeter (p) of domain sizes with a cross-over at a perimeter around Fig.e. which is mixture of egg sphingomyelin and egg ceramide.Eur Biophys J (2012) 41:161–175 171 Fig. 11 Domain perimeters and areas on the vesicle depicted in p & 1 or equivalently A & 0.0. ir. where L is measure of the linear extent of a domain. This fluorophore was used in the preparation.. vides a means of testing models of the growth dynamics stant threshold value. The two straight lines indicate separate 0. The domain areas were determined by simply 0.1 1 10 power law behaviors. the boundaries were first smoothed in order to avoid the artifact imposed by the 0.1 plotted against each other in a double-logarithmic coordi- nate system. The value is and underlying textures of the domains. when using the threshold ir.04 1.93 +/. a non-compact behavior. and the segmentation is achieved using a con. For the perimeter calculation.063 +/. Since only one ratio is only DDAp  1:15.

which will basis of the plane of the flattened surface patch. this field:  The edge map is now given by 2 1 1 ig e ¼ jrgj . and the local ori. g~ is the original greenness function. jrgj2 dence in the proper location of the threshold. where vi is the distance vector between the centers of the borhood of a small boundary segment. takes high values near domain boundaries and low values The difference in the scalar field between the triangle elsewhere. This reduces the gradients due to noise or structure within we get the individual domains. 6. the four-triangle neighborhood is first flattened by rotating the outer triangles around their Another means of investigating the domain morphology is common edge with the center triangle. In order to make this Since function change more rapidly when crossing the threshold ! line compared to changing within either side. To complete the definition of an edge map of the vesicle we avoid the need to transform them to a two-dimensional in Fig. Pairwise these determine the We have presented a methodology to perform detailed gradient vector. For two directions i and j. The position of a Conclusion triangle is here taken as its barycenter. This trick requires some confi. g ¼ tan ð5ð/  0:718ÞÞ. the ‘‘greenness.. For this ð20Þ example. For a two. ð21Þ where k is a constant. so an average is taken of the square analysis of the lateral domain characteristics of GUVs norms obtained using the three possible pairs. The Arg  b. but since the edges in the edge map. we get a system of entation of the boundary can thus be determined as the linear equations principal axis with the smallest moment of inertia. i. as it can make domains appear smaller or larger on the edge jvj j2 Dg2i þ jvi j2 Dg2j  2ðvi  vj ÞDgi Dgj  eij  : ð22Þ map depending on the value of g~0 : A Gaussian filter can jvi j2 jvj j2  ðvi  vj Þ2 also be used to reduce noise at the cost of broadening the The distance vectors are coplanar 3-vectors. such that the three an edge map. vjx vjy Dgj One approach for defining an edge map is to first define a suitable scalar field (say. ‘‘Histograms and segmentation’’). 3 we have used the square norm of the gradient of orthonormal basis.e. ð19Þ domain morphology. For the imaged by confocal fluorescence microscopy. / ¼ tan : ðe12 þ e23 þ e31 Þ ir e¼ : ð23Þ 3 ð16Þ An example edge map is shown in Fig.’’ g) on the where the vectors have been represented in an orthonormal surface as a map from the channel intensities. the mean of the three vertex positions. it can be jvi j2 vi  vj AA> ¼ wrapped in the inverse tangent function: vi  vj jvj j2 1 ! g ¼ tan ðkð~ g  g~0 ÞÞ ð15Þ 1 jvj j2 vi  vj > 1 ) AA ¼ . ð18Þ edge map might also be used for setting up an active contour method (Kass et al. This value at a triangle is calculated by first evaluating the components of the gradient along the vectors from the triangle to to its three nearest neighbors. result is expressed in terms of dot products of the vectors. solution of the system is channel image with two phases as in the histogram in rg  A1 b Fig. we will use the latter. This map can be used to overcome the lattice and its ith nearest neighbor is to first order artifact when determining tangent vectors along the Dgi  vi  rg. this could be the signed distance to the threshold > 1 line in the two-channel histogram or the angle from the red ) jrgj2 ¼ rg>rg  b> A1 A1 b ¼ b> AA> b: axis (/ in Sect. detð AA>Þ vi  vj jvi j2 and g~ ¼ g~0 represents the threshold line in the histogram. though. 12. Xu and Prince 1997) on where   the spherical surface as another means of investigating vix viy Dgi A¼ and b¼ . two triangles. 1988. a moment of inertia tensor can be constructed in a neigh. which is any scalar field on the surface that distance vectors become coplanar. The have a large gradient near the edges of domains. ð17Þ boundary: By taking the edge map as the density. A starting 123 .172 Eur Biophys J (2012) 41:161–175 Edge map purpose of this calculation.

The surface image is triangulated to provide a discretiza- tion that allows for analysis of morphological characteris- tics of lateral domain structures. lateral motion of the acknowledge Jesus Sot and Felix Goni (Biophysics Unit. University of experimental setup. e. set of m data points.H. the parameters to be determined. e. Research residuals can be written as 123 . domain size distri. i. where m [ n. Denmark. tion of vesicles) continues to improve the quality of con. data used to perform our image analysis. FONDEF (D07I1019). For this image.B. rðqÞ ¼ b  Aq: ð25Þ Acknowledgments MEMPHYS Center for Biomembrane Physics The problem of minimizing the sum of squares of the is supported by the Danish National Research Foundation. The vector q vesicles based on the stacks and thus helps close the contains the unknowns. Center for Biomembrane Physics. which faithfully repre. it can be written in matrix form as focal imaging stacks. CSIC-UPV/ EHU) and Laura Rodriguez Arriaga for providing some of the raw vesicle during image acquisition and photoselection effect. fixa. the greenness is chosen as g ¼ tan1 ð5ð/0:718ÞÞ. 3.and microscopy techniques with respect to stack acquisition left-hand sides of each equation. 12 An edge map of the vesicle shown in Fig. We show that the approach provides characterization of global Least squares is a method for finding an approximate properties of the lateral structure.. the least squares solution to the system is the one angles and geodesic curvatures of domain boundaries.g. that minimizes the sum of the squares of the residuals. fluorophore fitting a model equation with n adjustable parameters to a development and GUV preparation techniques (e. lengths. The authors want to stacking distance and orientation. The same vesicle is depicted in six different orientations point for the analysis is the construction of a surface image in SCIAN-Lab (S. Odense. solution to an overdetermined system of equations. which provide effective Appendix 1: Linear least squares fitting determinations of domain areas and boundaries.e. and S. spinning disk confocal microscopy). laboratories is supported by FONDECYT (1090246) membrane. ig where / ¼ tan1 ir is the angle from the red axis in the histogram. The rapid development in confocal fluorescence which are defined as the difference between the right.. Our method facilitates a very detailed Aq ¼ b.g.. ð24Þ characterization of the lateral structure of near-spherical where A is an m 9 n matrix with m [ n. cifically.) is funded by FONDECYT (1090246) and from the confocal imaging stack..e. Spe- bution. i.H. This construction takes into account the and MEMPHYS.. and single domain characterization.g. The collaboration between the sents the variations of fluorophore concentrations over the L. both CONICYT (Chile) and the Millennium Scientific Initiative (ICM P09-015-F). The method is used for time (e. optical resolution.g.. The domains are identified using image intensity histograms.Eur Biophys J (2012) 41:161–175 173 Fig. If the system is linear. confocal Southern Denmark.A. increasing gap between the level of detail available in The vector of residuals for a given choice of q is given by confocal imaging of vesicles and the quantitative charac- terization the lateral structure obtained from it.

A 1 2xN 2yN y c 2 x þy 2 N N If the coefficients f‘m decrease rapidly with ‘. the function ð30Þ can be approximated as a linear combination of only a few spherical harmonics... /Þ  f‘ Y‘ ðh. Appendix 2: Spherical harmonics j j j ð26Þ The spherical harmonics are defined on the unit sphere using spherical coordinates (see Fig. the parameters are: harmonic: 123 . the with respect to the coefficients f‘m : overdetermined system is ZZ !2 o X 0 0 0¼ m f ðh. They are orthonormal. m ¼ ‘. /ÞdX or  ZZ 0 10 1 0 r2 1 1 2~ z1 z21 ~ R2  c 2 ~ z2c 1 ¼2 f‘m  f ðh. Circle fits where Pm ‘ is the associated Legendre polynomial of degree For the circle fits. C ‘¼0 m¼‘ @. . /ÞdX rewritten as ‘0 . . The least squares fit of a linear defining A. ... X ZZ 0 0 and R is the radius of the sphere. ð28Þ 0 Y‘m ðh. .. /Þ dX ri2 þ ðcð~ z c Þ Þ 2 ¼ R2 . The system can be ¼2 f‘m0 Y‘m ðh. the desired parameters can be found as found by minimizing the integral of the square of the qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi residual xc ¼ q2 .. . A zc A ¼ @ . zi  ~ ð32Þ of‘ ‘0 . 1. any such function can be written as or 0 1 0 2 1 X 1 X ‘ 1 2x1 2y1 0 r 2  x2  y2 1 x1 þ y21 f ðh. /ÞY‘m0 ðh. /ÞdX ¼ d‘‘0 dmm0 . . the overdetermined system is ‘ and order m.m For fitting a sphere using the radii ri of the circle fits. A: c2 ~ ð34Þ c2 So the coefficient for a particular spherical harmonic is 1 2~ zN ~z2N r2 N simply the inner product of the function with the spherical From the least squares solution.. /Þ dX ð39Þ Sphere fit ‘. /Þ ¼ f‘m Y‘m ðh. 4p ð‘ þ mÞ! ‘ ð36Þ A>Aq ¼ A>b ð27Þ ‘ ¼ 0. . ZZ ðxi  xc Þ2 þ ðyi  yc Þ2 ¼ r 2 .m0 where c = Dz/Dx is the scaling factor for the z-direction. /Þ  f‘m0 Y‘m0 ðh. ~zc ¼ and R ¼ q1 þ 2 ð35Þ q3 q3 o >> ¼ q A Aq þ b>b  b>Aq  q>A>b oqi X X X ¼2 A>A ij qj  2 A>ij bj ¼ 2 A>Aq  A>b i ..e. A xc @ . . i. ‘  1. . ð37Þ which can be rewritten as where dX ¼ sin hdhd/:They form a complete basis for the x2i þ y2i ¼ r 2  x2c  y2c þ 2xc x þ 2yc y. . /ÞdX ð40Þ B .. /Þ: ð38Þ c c B . /ÞY‘m ðh. /ÞY‘m ðh.. C @ A¼B . ð29Þ set of square-integrable functions on the unit sphere. yc ¼ q3 and r ¼ q1 þ q22 þ q23 ð31Þ ZZ !2 X m m f ðh. . From the least squares combination of some number of spherical harmonics is solution for q.e. ‘ þ 1..174 Eur Biophys J (2012) 41:161–175 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi o > 0¼ rr pffiffiffiffiffi q2 q2 oqi c ¼ q3 .. /Þ ¼ P ðcoshÞeim/ .m0 |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} d‘‘0 dmm0 ZZ ! ri2 ¼R 2 c2 ~ z2c 2 þ 2c ~ zi  zc ~ c2 ~ z2i ð33Þ  f ðh. . . i. q and b in equation (24). C@ B C @.. . /ÞY‘m0 ðh. 2) as so the least squares solution to the overdetermined system sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi is the solution to the corresponding ‘‘normal equations’’: 2‘ þ 1 ð‘  mÞ! m Y‘m ðh. ‘.

Lsche M. Biophys J 78(1):290–305 Graph 29(5):641–655 Barber CB. Garcia A. Huin W (2003) Finding the exact volume of a Simonsen AC (2009) Texture of lipid bilayer domains. pp 127–131. Den Nijs M. Dobkin DP. pp 10–13 Honerkamp-Smith A. Brown D. Schwille P. Levi M. Faucon F. Webb W. and system for geometric visualization. Witkin A. Biochim Kass M. Biochim Biophys Acta density function. ACM Trans Math Softw 22(4): measured on single small unilamellar vesicles. Levy S. /ÞY‘m ðh. Biophys J 80(3):1417–1428 domains in human skin stratum corneum’s lipid membranes: Fanani M. Davis JH (2009) Quantitative characterization of Angelova M. Maggio B (2005) Shape transitions and lattice structuring of ceramide-enriched domains generated by sphingomyelinase in lipid monolayers. Schwille P (2005) Raft partitioning and dynamic 10. Bagatolli L (2007) Direct visualization of lipid membranes. Biophys J 95(1):236–246 New York. Simonsen A. In: SCG ’95: proceedings of critical exponents in lipid membranes near critical points. Fidorra M. Berlin. Collins M. De Tullio L. In: CVPR. Volovyk Z. Mhwald H (eds) magnetic resonance. Springer. Maggio B.geomview. Comput binary phospholipid mixtures. Gratton E (1999) Two-photon fluorescence microscopy Korlach J. Prince J (1997) Gradient vector flow: a new external force for 90(12):4437–4451 snakes. Munzner T. Lopes H. ACM. Biophys J Xu C. Bothorel P (1992) coexisting phases in DOPC/DPPC/cholesterol mixtures: com- Preparation of giant vesicles by external AC electric fields. Jara J. Ipsen JH. Huang H. Keller S (2008) Line tensions. Ong H. pp 412–413. Gomes JD Jr. Int J Comput Vis 1(4):321–331 Bagatolli L. Dover.1021/ja903375m Parzen E (1962) On estimation of a probability density function and Dietrich C. doi:10. Bagatolli L (2007) Giant unilamellar (2004) Cholesterol rules: direct observation of the coexistence of vesicles electroformed from native membranes and organic lipid two fluid phases in native pulmonary surfactant membranes at mixtures under physiological conditions. Laura Fanani M. Leidy C. Amenta N. Biophys J 88(1):287–304 Heath TL (eds) (1956) The thirteen books of the elements. Härtel S. /ÞdX: ð41Þ Acta Biomembr 1788(10):2142–2149 Härtel S. http://www. 2nd edn. Keller S (2005) Seeing spots: complex phase behavior in Fidorra M. doi: Kahya N. Biophys J 77(4):2090– correlation spectroscopy.qhull.Eur Biophys J (2012) 41:161–175 175 ZZ fluorescence microscopy imaging Juhasz J. Goñi F. Gratton E (2001) Lipid rafts reconstituted in model Plasencia I. Adv Eng Softw 34(6):351–356 Chem Soc 131(40):14130–14131. progress in colloid 1788(12):2541–2552 and polymer science. Craizer M (2005) Curvature and of coexisting lipid domains in giant unilamellar vesicles of torsion estimators based on parametric curve fitting. Biophys J 93(9):3142–3155 R (2010) The action of sphingomyelinase in lipid monolayers as Rosenblatt M (1956) Remarks on some nonparametric estimates of a revealed by microscopic image analysis. Olmos F. Sharom FJ. Ann Math Stat 33(3):1065–1076 K. New York. Bagatolli L (2006) simple membranes. Ann Math Stat 27(3):832–837 Biomembr 1798(7):1309–1323 Veatch S. Biochim Biophys Acta Mole Cell Res Absence of fluid-ordered/fluid-disordered phase coexistence in 1746(3):172–185 ceramide/popc mixtures containing cholesterol. Biochim Biophys f‘m ¼ f ðh. Oliveira effect of pH and temperature. Phillips M (1995) Geomview: a Schick M. Proc Natl Acad Sci USA 96(15):8461– 2101 8466 Bagatolli L. Cicuta P. Bagatolli LA Montes L. Simonsen AC. Feigenson G (1999) Characterization observation of shape changes at the phase transition in of lipid bilayer phases by confocal microscopy and fluorescence phospholipid giant unilamellar vesicles. Jacobson mode. vol 89. J Biol Chem 279(39):40715–40722 3548–3554 Bernchou U. Norlén L. Huhdanpaa H (1996) The quickhull Lohse B. Biochim Biophysica Acta Biomembr Trends in colloid and interface science VI. Ipsen J. Terzopoulos D (1988) Snakes: active contour Biophys Acta Biomembr 1758(10):1541–1556 models. Bagatolli LA. Alonso A. J Am polyhedron. paring confocal fluorescence microscopy and deuterium nuclear kinetics and applications. References vol 3. Bolinger P. Stamou D (2008) Encapsulation efficiency algorithm for convex hulls. Published by the IEEE Computer Society. 130(44):14372–14373 Bernardino de la Serna J. Bagatolli L (2009) Lipid p 66 domains in giant unilamellar vesicles and their correspondence with equilibrium thermodynamic phases: a quantitative 123 . Bagatolli L. Härtel S. Mlard P. Solau S.1007/BFb0116295 behavior of human placental alkaline phosphatase in giant Bagatolli L (2006) To see or not to see: lateral organization of unilamellar vesicles. Biochemistry 44(20):7479–7489 biological membranes and fluorescence microscopy. Perez-Gil J. In: Helm C. correlation lengths. Midtiby HS. Gratton E (2000) Two photon fluorescence microscopy Lewiner T. Veatch S. Biophys J 93(10): physiological temperatures. Brewer J. J Am Chem Soc 469–483. Duelund L. the 11th annual symposium on computational geometry. Thompson N.

. download. users may print.V.Copyright of European Biophysics Journal is the property of Springer Science & Business Media B. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However. or email articles for individual use.

Master your semester with Scribd & The New York Times

Special offer for students: Only $4.99/month.

Master your semester with Scribd & The New York Times

Cancel anytime.