The Human Skin Barrier Is Organized as StackedBilayers of Fully Extended Ceramides withCholesterol Molecules Associated with theCeramide Sphingoid MoietyIchiro Iwai1,2, HongMei Han1,8, Lianne den Hollander1,8, Stina Svensson1,3, Lars-Goran Ofverstedt4,Jamshed Anwar5, Jonathan Brewer6, Maria Bloksgaard6, Aurelie Laloeuf1, Daniel Nosek1,7, Sergej Masich1,Luis A. Bagatolli6, Ulf Skoglund4 and Lars Norlen1,7

The skin barrier is fundamental to terrestrial life and its evolution; it upholds homeostasis and protects against theenvironment. Skin barrier capacity is controlled by lipids that fill the extracellular space of the skin’s surfacelayer—the stratum corneum. Here we report on the determination of the molecular organization of the skin’s lipidmatrix in situ, in its near-native state, using a methodological approach combining very high magnification cryo-electron microscopy (EM) of vitreous skin section defocus series, molecular modeling, and EM simulation. Thelipids are organized in an arrangement not previously described in a biological system—stacked bilayers offully extended ceramides (CERs) with cholesterol molecules associated with the CER sphingoid moiety. Thisarrangement rationalizes the skin’s low permeability toward water and toward hydrophilic and lipophilicsubstances, as well as the skin barrier’s robustness toward hydration and dehydration, environmental temperatureand pressure changes, stretching, compression, bending, and shearing.

Journal of Investigative Dermatology (2012) 132, 2215–2225; doi:10.1038/jid.2012.43; published online 26 April 2012

INTRODUCTIONThe skin constitutes a barrier between the body and theenvironment (Elias and Friend, 1975). By preventing waterloss via evaporation it upholds homeostasis, and by prevent-ing penetration of exogenous substances it protects againstthe environment. The skin’s barrier capacity is a functionof the molecular architecture of the lipid structure in theextracellular space between the cells of the stratum corneum(Bouwstra et al., 1991). The lipids consist of a heterogeneous

mixture of saturated, long-chain ceramides (CERs), free fattyacids (FFAs), and cholesterol (CHOL) in a roughly 1:1:1 molarratio (Wertz and Norlen, 2003). In the CER fraction alone,more than 300 different species have been identified(Masukawa et al., 2009).

Ever since the discovery in the early 1970s (Breathnachet al., 1973; Elias and Friend, 1975) that the stratum corneumextracellular space is filled with lipid material, the skin lipidshave been a subject of much activity. However, themolecular organization of these lipids remains unresolved.By using conventional electron microscopy (EM) on ruthe-nium tetroxide–stained mouse skin, Madison et al. (1987)reported that the stratum corneum lipid material, or ‘‘lipidmatrix’’, displays 13-nm repeating units of broad:narrow:broad electron lucent bands. Using small-angle X-raydiffraction on isolated human stratum corneum, Garsonet al. (1991) reported the presence of one 4.5-nm and one6.5-nm diffraction peak related to lipids. Concomitantly,White et al. (1988) and Bouwstra et al. (1991) reported thepresence of a 13-nm repeating unit in mouse and humanisolated stratum corneum, respectively. Later, McIntosh(2003) observed an asymmetric distribution of CHOL withinmodel systems composed of reconstituted stratum corneumlipids extracted from pig skin. On the basis of these data,as well as on data obtained from different in vitro lipidmodel systems, six theoretical models for the moleculararrangement of the extracellular lipid matrix have been

See related commentary on pg 2131

& 2012 The Society for Investigative Dermatology www.jidonline.org 2215

ORIGINAL ARTICLE

Received 12 May 2011; revised 12 December 2011; accepted 17 December2011; published online 26 April 2012

1Department of Cell and Molecular Biology (CMB), Karolinska Institutet,Stockholm, Sweden; 2Shiseido Research Center, Yokohama, Japan; 3Centerfor Image Analysis, Swedish University of Agricultural Sciences, Uppsala,Sweden; 4Structural Cellular Biology Unit, Okinawa Institute of Science andTechnology, Okinawa, Japan; 5Institute of Pharmaceutical Innovation,University of Bradford, Bradford, United Kingdom; 6Department ofBiochemistry and Molecular Biology, Membrane Biophysics andBiophotonics group/MEMPHYS–Center for Biomembrane Physics, Universityof Southern Denmark, Odense, Denmark and 7Dermatology Clinic,Karolinska University Hospital, Stockholm, Sweden

Correspondence: Lars Norlen, Department of Cellular and Molecular Biology(CMB), Karolinska Institutet, von Euler’s v 1, 171 77 Stockholm, Sweden.E-mail: lars.norlen@ki.se

8These authors contributed equally to this work.

Abbreviations: CEMOVIS, cryo-electron microscopy of vitreous skin section;CER, ceramide; CHOL, cholesterol; EM, electron microscopy; FFA, freefatty acid; GP, generalized polarization; LAURDAN, 6-dodecanoyl-2-dimethylaminonaphthalene

proposed. These are based on either triple-band repeatingunits and/or bands with the same width (SupplementaryFigure S1 online).

We have used a novel methodological approach based oncryo-electron microscopy of vitreous skin section (CEMOVIS)defocus series combined with molecular modeling and EMsimulation to investigate the molecular organization of thehuman skin’s lipid matrix in its near-native state. In CEMOVIS,the native tissue is preserved down to the molecular level, andthe micrograph pixel intensity is directly related to the localelectron density of the specimen (Dubochet et al., 1988; Al-Amoudi et al., 2004, 2007; Norlen et al., 2009). Biomoleculesgenerally possess small intermolecular and intramoleculardifferences in electron density, as they are essentiallycomposed of atoms with similar atomic weight (carbon,nitrogen, and oxygen). However, for orderly arrangedmolecular assemblies, such as lipid tails and headgroups inmembranes, even small differences in shape and atomiccomposition may be amplified because of interference effectsthat appear in the image phase contrast. During imageacquisition, phase contrast is made visible using defocus(compare e.g., Fanelli and Oktem, 2008).

Here we show that the human skin barrier’s CEMOVISpattern is characterized by an asymmetric B11-nm repeatingunit consisting of alternating narrow (B4.5 nm) and broad(B6.5 nm) bands. Further, at, and only at, very highmagnification (pixelsize p3.31 A), complex interferencepatterns can be resolved in the CEMOVIS micrographs.Exploiting this, CEMOVIS micrographs were recorded repeat-edly at very high magnification at the same position of the skinsample while increasing stepwise the microscope’s defocus,thus ensuring that differences in the recorded micrographswere due exclusively to the different defocuses used.Simulated electron micrographs were then generated atcorresponding defocuses for different skin lipid models, andcompared with the CEMOVIS micrographs. This approach wasfound to be highly discriminating between different models.

The lipid organization that emerges from the analysis is astacked bilayer structure of CERs in the fully extended(splayed chain) conformation with CHOL associated withthe CER sphingoid moiety.

RESULTSSkin lipid’s CEMOVIS patternThe CEMOVIS data reported here represent over 1,000original observations obtained from the left volar forearmand abdomen of five Caucasian men in the age group of40–50 years with no history of skin disease.

The CEMOVIS data demonstrate that the stratum corneumextracellular lipid matrix is composed of a stack of layers(Figure 1a). They display a meandering pathway (Figure 1a) andfolds/unfolds on hydration/dehydration (Supplementary FigureS2A–C online). This implies that these layers are malleable. Athigh magnification (Figure 1d), the distinct dark lines of thelamellar stacks (Figure 1b) are revealed to be not entirelycontinuous. There are regular breaks in intensity, implyingsome disorder at the microstructural level (Figure 1d, whitesolid arrows; Supplementary Figure S4A online). At these

breaks, the intensity is displaced away from and perpendicularto the mainstream intensity bands, giving rise to thin dark‘‘cross-stripes’’ (Figure 1d, white open arrows; SupplementaryFigure S4A online).

The averaged, image intensity profile for the stacked lipids(Figure 1e) was obtained by fuzzy distance-based imageanalysis and is shown in Figure 1b and c. Lamellar regionswith 2, 4, 6, 8, 10, or 12 dark lines (Figure 2a–f, left column)can be observed between adjacent stratum corneum cellplasma membranes (also referred to as ‘‘lipid envelopes’’).The dark lines are arranged in a periodic pattern with anasymmetric B11-nm (10–12 nm) repeating unit consisting ofalternating narrow (B4.5 nm) and broad (B6.5 nm) bands,where the 4.5-nm bands express a higher averaged imageintensity (i.e., they are darker) than the 6.5-nm bands (Figure2a–f, right column). A low-intensity region (i.e., a light band)is present in the mid-plane of the 6.5-nm bands (Figure 1b,solid black arrows; Supplementary Figure S4A online). Small‘‘shoulders’’ centrally on each slope of the correspondingaveraged intensity profiles delimit its extension (Supplemen-tary Figure S4B online, open arrows). No interindividual orintersite (forearm vs. abdomen) variation was recorded(Supplementary Figure S5 online).

Power spectra extracted along the lamellar stacks revealedthat the 6.5-nm region, around its low-intensity mid-plane, issomewhat ordered, whereas the other regions appear to beunstructured (Figure 3a–g). The 6.5-nm region is character-ized by peaks equating to repeat distances of 2.3–3.0 nm and1.5–1.7 nm (Figure 3d).

Generalized polarization (GP) function measurementsobtained from multi-photon excitation images of 6-dodeca-noyl-2-dimethylaminonaphthalene (LAURDAN) labeled near-native skin (Figure 4a and b) show that the stratum corneumlipid matrix is in a condensed state and contains little, if any,water at the lipid headgroup/hydrocarbon interface regions.Furthermore, the LAURDAN GP function was not affected byhydration (Figure 4a and b). Similarly, the CEMOVIS intensityprofiles of the extracellular lipid matrix remained unalteredupon hydration (Figure 4c and d), whereas the intracellularkeratin intermediate filament network at the same locationsswelled extensively (Figure 4c; compare SupplementaryFigure S2A online) compared with normal stratum corneum(Figure 1a; compare Supplementary Figure S2B online).

Modeling and simulationThe key aspect of the lipid matrix’ CEMOVIS image intensityprofile is the asymmetric B11-nm repeating unit. This consistsof two peaks with a peak–peak distance of B4.5 nm separatedby a shallow trough of low intensity (Figure 2). The twin-peakrepeats are separated by a deeper (low intensity) trough with apeak–peak distance of B6.5 nm. The peaks are expected tocorrespond to the lipid headgroup regions, which arecharacterized by the heavier polar atoms nitrogen andoxygen. The low-intensity troughs are expected to correspondto the carbon-dominated lipid alkyl chains.

Given these characteristics of the data, we decided tocombine molecular modeling and EM simulation (Figure 5) toidentify a lipid organization that would be consistent not only

2216 Journal of Investigative Dermatology (2012), Volume 132

I Iwai et al.The Human Skin Barrier

with the observed intensity profile but also with theCEMOVIS intensity interference pattern changes observedon changing the microscope’s defocus during image acquisi-tion (Figure 6).

Rather than considering the full complexity of the stratumcorneum lipid composition, we focused on the three maincomponents (Wertz and Norlen, 2003; Masukawa et al.,2009), namely CER NP (C18-phytosphingosine-non-hydroxy-C24:0), CHOL, and lignoceric acid (C24:0), because these willlikely determine the main features of the lipid organization,and variation in chain lengths will serve only to modulate thestructure. Polyunsaturated o-hydroxyacid CERs were notincluded, as these correspond to o15% of the total lipidmass extracted from the stratum corneum and are structurallydifferent from the remaining stratum corneum lipids.

Further, water was not included in the models as neitherthe LAURDAN GP (Figure 4a and b) nor the CEMOVIS(Figure 4c and d) swelling experiments expressed signs of thepresence of water in the lipid matrix.

For each of the three lipid components, we built molecularmodels with the atoms located at their ideal bond distances,angle, and torsions. The torsions in the headgroup region of theCER molecules were chosen to yield either a fully extended

(the splayed chain conformation; compare SupplementaryFigures S6–S7 online) or a fully folded (the hairpin conforma-tion; compare Supplementary Figure S8 online) structure.

EM image intensity patterns were then simulated for avariety of molecular models at three radically differentdefocuses (!0.5, !2, and !5mm; Figure 6d–f; SupplementaryFigures S6–S8 online) using a newly developed EM simulator(Rullgard et al., 2011). The different stratum corneum lipidmodels were thus discriminated by how closely theircorresponding simulated EM patterns mirrored the CEMOVISpatterns of near-native skin, and how closely they reflectedthe CEMOVIS interference pattern changes observed uponradically changing the microscope’s defocus during imageacquisition (Figure 6a–f; Supplementary Figures S6–S8 online).For the best-fitting model (Supplementary Figure S6D online),EM images were simulated at !1, !2, and !3-mm defocus(Figure 6j–l) and compared with a series of CEMOVISmicrographs obtained sequentially at the same locationat !1, !2, and !3-mm defocus (Figure 6g–i).

Remarkably, of the various molecular organizationsconsidered, a model based on bilayers of fully extendedCERs with CHOL selectively localized to the CER sphingoidpart was found to account for all the major features of the

5 nm

4.5 nm

6.5 nm

4.5 nm

6.5 nm 5 nm

4.5 nm

a b

d

e

c

Figure 1. The cryo-electron microscopy of vitreous skin section (CEMOVIS) intensity pattern of the stratum corneum extracellular lipid matrix consists offolded stacked layers. (a) Medium-magnification CEMOVIS micrograph of the interface between two cells in the mid-part of the stratum corneum. Note that inCEMOVIS the tissue is unstained, and that the pixel intensity is directly related to the local electron density of the sample. The stacked lamellar pattern representsthe extracellular lipid matrix. Dark B10-nm dots represent keratin intermediate filaments filling out the intracellular space. Note that the extracellularlipid matrix locally expresses extensive folding. (b) High-magnification CEMOVIS micrograph of the extracellular space in the mid-part of the stratum corneum.The intensity profile of the lipid matrix was obtained by fuzzy distance-based image analysis. The red stars in (b) represent the manually chosen start andend points for fuzzy distance-based path growing. (c) The red line represents the traced-out path. Stacked lines mark extracted intensity profiles. (d) Enlarged areaof the central part of (b). Note that the electron-dense bands are composed of dark 1- to 3-nm dots from which thin, weak lines protrude 2–3 nm into thelucent areas. The same pattern is also evident in (b). (e) Reversed averaged pixel intensity profile obtained from the extracted area in (c). Peaks in (e) correspondto dark bands and valleys to lucent bands in (d). Black arrows in (b) denote electron lucent narrow bands at the center of the 6.5-nm bands. Sectionthickness: B50 nm (a–d). Bar (a): 100 nm. Pixel size in (a–d): 6.02 A.

www.jidonline.org 2217

I Iwai et al.The Human Skin Barrier

CEMOVIS images at all defocuses (Figure 6). None of theother lipid models could account for any major feature of theCEMOVIS images at any defocus. For reference andcomparison, the simulated electron micrographs for all thelipid models tested are given in Supplementary Figures S6–S8online. From these simulated images, it is clear that onlybilayer models in which the CER molecules are in the fullyextended conformation show any similarity to the observedCEMOVIS data (Supplementary Figure S6 online). The folded(hairpin) CER conformation and stacked monolayers ofextended CERs are readily discounted (SupplementaryFigures S7 and S8 online).

We specifically explored whether the simulated imagesare sensitive to how CHOL is distributed within the lipidmodel, given the reported asymmetric distribution of CHOLwithin model systems composed of reconstituted stratumcorneum lipids (McIntosh, 2003). We considered models

(i) without CHOL (Supplementary Figure S6E online), (ii) withCHOL selectively localized to the CER sphingoid part(Supplementary Figure S6D online), (iii) with CHOL selec-tively localized to the CER fatty acid part (SupplementaryFigure S6F online), and (iv) with CHOL homogenouslydistributed between the CER fatty acid and sphingoid parts(Supplementary Figure S6G–J online). The simulations revealthat the CEMOVIS patterns are remarkably sensitive to CHOLdistribution and suggest that CHOL is selectively localized tothe CER sphingoid part (Supplementary Figure S6D online).

DISCUSSIONThe lipid organization that emerges from the analysis is abilayer structure of fully extended CERs with the sphingoidmoieties interfacing. Both CHOL and the FFA appearto be selectively distributed: CHOL at the CER sphingoidend and the FFA at the CER fatty acid end. The characteristic

4.5 nm

4.5 nm 4.5 nm

4.5 nm4.5 nm 4.5 nm

4.5 nm 4.5 nm 4.5 nm

4.5 nm 4.5 nm4.5 nm4.5 nm4.5 nm

4.5 nm 4.5 nm 4.5 nm 4.5 nm 4.5 nm 4.5 nm

4.5 nm

6.5 nm

6.5 nm

6.5 nm

6.5 nm

6.5 nm 6.5 nm 6.5 nm 6.5 nm 6.5 nm

6.5 nm 6.5 nm 6.5 nm

6.5 nm 6.5 nm

6.5 nm

Figure 2. An asymmetric B11-nm repeating unit characterizes the cryo-electron microscopy of vitreous skin section (CEMOVIS) intensity profile of thelipid matrix. Left column: high-magnification CEMOVIS micrographs of the extracellular space in the mid-part of the stratum corneum of vitrified epidermis.Right column: corresponding intensity profiles obtained from fuzzy distance-based image analysis. (a) Two high-intensity (dark) lines between stratumcorneum cell borders. (b) Four high-intensity lines. (c) Six high-intensity lines. (d) Eight high-intensity lines. (e) Ten high-intensity lines. (f) Twelve high-intensitylines. Section thickness: B50 nm (a–f). Pixel size in (a): 3.31 A, and in (b–f): 6.02 A.

2218 Journal of Investigative Dermatology (2012), Volume 132

I Iwai et al.The Human Skin Barrier

6.5-nm distance is consistent with two apposing C24 amide-bound fatty acid chains plus one CER headgroup (48C!1.27 A"5.1 A# 66 A; C24 fatty acids dominate in stratumcorneum (Masukawa et al., 2009)). The 4.5-nm distance isconsistent with two apposing C18 sphingoid backbonechains plus one CER headgroup (36C! 1.27 A"5.1 A#51 A; C18 sphingoid backbones dominate instratum corneum (Masukawa et al., 2009)).

The proposed lipid organization is optimal in terms ofpacking, with the fatty acid chain lengths matching the CERfatty acid chains and CHOL matching the CER sphingoidchains. This ‘‘hydrophobic matching’’ minimizes the poten-tial energy of the molecular configuration. The ‘‘segregation’’of CHOL and FFAs to the respective ends of the CERs may beexpected, as CHOL and stratum corneum FFAs do not mix inmodel systems (Norlen et al., 2007). CHOL, however, does

600

400

Pow

er

Pow

er

Pow

er

200

0

600

400

200

0

Pow

er

600

400

400

600

800

1,000

200

200

0

0 0.02 0.04 0.06 0.08 0.10

0 0.02 0.04 0.06 0.08 0.10

0 0.02 0.04 0.06 0.08 0.10

0

0

0.02 0.04 0.06 0.08 0.10

1 Å–1

1 Å–1

1 Å–1

1 Å–1

Power

Strip

num

ber

30

20

10

1,000

800600400200

0

0.02

0.04

0.06

0.08

1 Å –1

Figure 3. Cryo-electron microscopy of vitreous skin section (CEMOVIS) micrograph power spectra analysis indicates different molecular packing ordersin different regions within the lamellar structure. (a) Surface plot of all power spectra obtained along the lamellar pattern of a CEMOVIS micrograph(b, c, Figure 5a) of a unit of one central 4.5-nm band with two adjacent 6.5-nm bands. The power spectra of the 6.5-nm bands, besides their mid-plane,show recurring characteristics, represented by one 2.3- to 3.0-nm peak and one 1.5- to 1.7-nm peak (d). The 6.5-nm bands, besides their low-intensitymid-plane, are thus somewhat ordered. The power spectra extracted from the remaining regions are more amorphous with nonreproducible characteristics (e–g).(e) Power spectra obtained from four strips located within the mid-plane of the 6.5-nm bands. (f) Power spectra obtained from five strips located within thecentral 4.5-nm band. (g) Power spectra obtained from four strips located within the high-intensity (dark) headgroup bands. The 4.5-nm-thick band, including itsmid-plane, and the headgroup bands are thus amorphous but have, at the same time, some intrinsic order with recurring characteristics of low reproducibility(f, g). The mid-plane low-intensity region of the 6.5-nm-thick bands is amorphous (e). In (d–g) the extracted power spectra are connected to their respectivesampling locations (two-pixel-wide strips) in the CEMOVIS micrograph (c). Pixel size in (b, c): 3.31 A.

www.jidonline.org 2219

I Iwai et al.The Human Skin Barrier

mix with sphingosine and its derivatives (Garmy et al., 2005).The hydrophobic match between CHOL and saturated alkylchains is optimized between 14 and 18C chain lengths,which correspond to the length of the CHOL molecule(Ouimet and Lafleur, 2004). Further, wide-angle X-raydiffraction experiments on isolated stratum corneum indicatecrystalline-like hydrocarbon chain packing with character-istic distances of 3.7–3.8 A and 4.1–4.2 A (Wilkes et al., 1973;White et al., 1988; Garson et al., 1991; Bouwstra et al., 1992;Hatta et al., 2006), which would not appear if CHOL weredistributed throughout both lipid hydrocarbon chains of theCERs. However, a stratum corneum lipid matrix in whichCHOL and FFAs are segregated into different bands doesallow for crystalline-like hydrocarbon chain packing on thefatty acid sides of the stacked extended CER bilayer system. Itthus rationalizes how a biological lipid system composed of30% CHOL could express crystalline-like wide-angle X-raydiffraction patterns without signs of lateral crystal domainformation detectable by high-resolution CEMOVIS.

The power spectra obtained along the lamellar planereflect the lateral lipid packing at a resolution of 1–2 nm,which characterizes our CEMOVIS micrographs. The follow-ing comments therefore refer to the microstructure rather thanthe molecular-level packing of headgroups or alkyl chains.Overall, the microstructure is predominantly disordered,but there are some recurring characteristics detectable inthe power spectra (Figure 3). These characteristics weremost evident around the mid-plane of the 6.5-nm bands, i.e.,the CER fatty acid moiety region, and corresponded toperiodicities of 2.3–3.0 nm and 1.5–1.7 nm (Figure 3d). Theycorrespond roughly to the distance between (2–4 nm) andwidth of (1–1.5 nm) the ‘‘cross stripes’’ observed withinthe 6.5-nm bands of the CEMOVIS micrographs (Figure 1d,open white arrows). These periodicities may thereforereflect the average repeat distance and width of coherentlipid clusters separated by defect regions. Remarkably,the perpendicular cross-stripe pattern could be recreatedin the simulated electron micrographs by randomly rotating

Max 1

GP GP

0.8

0.6

0.4

0.2

0.0Normal Hydrated

–1

20 µm

5 nm 6.5 nm 6.5 nm

4.5 nm 4.5 nm

5 nm

4.5 nm

FI

Min

Figure 4. The stratum corneum lipid matrix is in a condensed state and remains unaltered upon hydration. (a) 6-Dodecanoyl-2-dimethylaminonaphthalene(LAURDAN) fluorescence intensity (left) and LAURDAN generalized polarization (GP) function images (right) of normal near-native skin (top panel) and of skinexposed to hydration for 2 hours in vivo (bottom panel). The images were obtained at the level of mid-part of the stratum corneum using a multi-photonexcitation microscope. (b) Bar graphs (average±SD) showing the LAURDAN GP values of normal and hydrated skin. The high (40.5) LAURDAN GP valuesshow that the extracellular lipid matrix is in a condensed state and displays little, if any, water dipolar relaxation at the lipid headgroup interface regions.(c) High-magnification cryo-electron microscopy of vitreous skin section (CEMOVIS) micrograph of the extracellular space in the mid-part of the stratumcorneum after hydration. (d) Intensity profile obtained from fuzzy distance-based path growing of the area marked by a white box in (c). Note that the intensityprofile after skin hydration (d) is identical to that of normal skin (compare Figure 1e). Also note the loose arrangement of keratin intermediate filaments(open arrows) in the swollen stratum corneum cells as compared with their condensed arrangement in normal stratum corneum (compare Figure 1a).The CEMOVIS intensity profile of the extracellular lipid matrix thus remains unaltered on hydration. Pixel size in (c): 3.31 A.

2220 Journal of Investigative Dermatology (2012), Volume 132

I Iwai et al.The Human Skin Barrier

CEMOVIS micrograph

Averaged intensity profilealong line pattern

Schematic 2D model

Atomic 3D model repeating unit

Electron scattering potential 3D mapof model subunit composed ofone extended ceramide,one free fatty acid, andone cholesterol molecule

One out of 20 stacked layers of the total modelelectron scattering potential 3D map used forthe model EM simulation

Simulated electron micrograph

Figure 5. Electron microscopy (EM) simulation of the stratum corneum extracellular lipid matrix. (a) High-magnification cryo-electron microscopy of vitreousskin section (CEMOVIS) micrograph of the extracellular space in the mid-part of the stratum corneum. (b) Corresponding intensity profile obtained by fuzzydistance-based path growing (compare Figure 1b and c). (c) Schematic two-dimensional (2D) illustration of ceramides (CERs) (tetracosanylphytosphingosine(C24:0)) in fully extended conformation with cholesterol (CHOL) associated with the CER sphingoid part, and free fatty acids (FFAs) (lignoceric acid(C24:0)) associated with the CER fatty acid part. (d) Atomic three-dimensional (3D) model of the repeating unit composed of two mirrored subunits, eachcomposed of one fully extended CER, one CHOL, and one FFA molecule. (e) Calculated electron scattering potential of one model subunit. (f) Calculatedelectron scattering potential 3D map of the topmost layer out of 20 superimposed layers used to generate the simulated electron micrograph (g). Defocus(a, g): !2.5mm. Pixel size in (a, g): 3.31 A.

www.jidonline.org 2221

I Iwai et al.The Human Skin Barrier

the individual lipid molecules along their length axesand displacing them ±1 A in the x, y, and z directions(Figure 5g).

A particularly notable feature of the proposed lipidstructure is that it comprises a bilayer rather than anarrangement of stacked monolayers. Energetically, eitherarrangement should be feasible, given that leaflets of extendedCER structures present alkyl chains at both ends, and thereforeboth organizations exclusively involve hydrocarbon interac-tions between leaflets. The preference for the bilayerorganization in the stratum corneum lipid matrix could result

from the biological processes involved in its formation. Thelipid matrix is formed from stacked bilayers of glycosylcer-amides in the hairpin conformation in a hydrated environ-ment, after deglycosylation and dehydration (Holleran et al.,1993; Caspers et al., 2001; Norlen, 2001; Al-Amoudi et al.,2005). During its formation, the extended CER bilayerorganization will therefore require a transformation fromhairpin to splayed chain conformation, involving a flip of oneof the two CER alkyl chains. For glycosylceramides, whichbind 5–10 water molecules per lipid molecule (Bach et al.,1982; Bach and Miller, 1998), flip-flop is slow (half-time of

UF: –0.5 µm

CEMOVISfirst exposuremicrographs

CEMOVISunderfocus series

micrographs

Model simulation

Model simulation

UF: –2 µm UF: –5 µm

UF: –1 µm UF: –2 µm UF: –3 µm

Figure 6. Electron microscopy (EM) simulation of alternating fully extended ceramides (CERs) with selective localization of cholesterol to the CERsphingoid part accurately accounts for the observed cryo-electron microscopy of vitreous skin section (CEMOVIS) intensity pattern, as well as for theinterference pattern changes observed in sequential CEMOVIS micrograph defocus series obtained at very high magnification (p1.88 A pixel size).(a–c) High-magnification CEMOVIS micrographs (first-exposure images) of the extracellular space in the mid-part of the stratum corneum obtained at(a) !0.5 mm, (b) !2 mm, and (c) !5 mm defocus. At very low defocuses (!0.5 mm) (a) CEMOVIS intensity patterns can only be observed at very high magnification(p1.88 A pixel size). At very high defocuses (!5 mm) (c) image resolution is low but still allows for resolution of the B11-nm repeating unit. The slightlylarger lamellar repeat distance in (b) (B12 nm) compared with (a) and (c) (B11 nm) may be due to more pronounced compression of the vitreous skin sectionduring cryo-sectioning along the lamellar plane in (b) compared with that in (a) and (c). (d–f) represents corresponding atomic three-dimensional (3D)model (compare Figure 5) EM simulation images recorded at (d) !0.5 mm, (e) !2 mm, and (f) !5 mm defocus. (g–i) Sequential CEMOVIS micrographdefocus series obtained at very high magnification (1.88 A pixel size) at (g) !1 mm, (h) !2 mm, and (i) !3 mm defocus. Note the fine changes in interferencepatterns caused by gradually increasing the microscope’s defocus during repeated image acquisition at a fixed position. Owing to electron beam damage afterrepeated electron exposure, the image contrast is lower in micrographs (h, i) compared with micrograph (g), which was acquired first. In micrograph (i), someshrinkage can be observed. This is probably due to mass loss during repeated electron exposure. In addition, the curvature of the lamellar pattern is slightlyincreased in micrographs (h, i) compared with that in micrograph (g), which may similarly be caused by nonhomogeneous mass loss during repeated electronexposure. (j–l) Represents corresponding atomic 3D model (compare Figure 5) EM simulation images recorded at (j) !1 mm, (k) !2 mm, and (l) !3 mm defocus. Itis shown that the atomic 3D model in Figure 5 accurately accounts not only for the major features of the CEMOVIS micrographs (a–f) but also for the interferenceintensity pattern changes observed upon varying the microscope’s defocus during image acquisition at very high magnification (g–l). UF, under focus. Pixel sizein (c, f): 3.31 A, in (b, e): 6.02 A, and in (a, d and g–l): 1.88 A.

2222 Journal of Investigative Dermatology (2012), Volume 132

I Iwai et al.The Human Skin Barrier

hours (Buton et al., 2002)). For CERs, which only bind 0–1water molecules per lipid molecule (Faure et al., 1998), theflip-flop movement is considerably faster (half-time o1minute (Lopez-Montero et al., 2005)). For CHOL, flip-flop israpid (half-time o1 second (Steck et al., 2002)). The initialhairpin glycosylceramide bilayer organization may thereforecarry over into the extended CER bilayer organization by a flipof one of the two CER alkyl chains together with CHOL. Thedriving force could be the deglycosylation and dehydrationencountered during the formation process. Once the stronglyhydrated glycosyl groups of the glycosylceramides have beenremoved (Holleran et al., 1993) and the water evacuated fromthe extracellular space (Al-Amoudi et al., 2005), the extensionof the two hydrocarbon chains in opposite directions aids theseparation of both the ill-matched CHOL and FFAs (Norlenet al., 2007), and the ill-matching CER C24 fatty acid and C18sphingoid moities, into different bands within an optimallyclose-packed stratum corneum bilayer structure. Specula-tively, being a rapid membrane flipper, CHOL may facilitatethe extension of the CERs by dragging the sphingoid moietywith it during the lipid matrix’ reorganization from a hairpin toan extended bilayer structure.

The major physiological consequence of a condensed,fully extended CER bilayer organization is that the lipidmatrix will be largely impermeable to water, as well as to bothhydrophilic and lipophilic substances, because of thecondensed structure and the presence of alternatinglipophilic (alkyl chain) and hydrophilic (headgroup) regions(compare Figure 5c). It will be resistant toward bothhydration and dehydration because of the absence ofexchangeable water between lipid leaflets (compare Figure 4).Further, the proposed structure accounts for stratumcorneum cell cohesion without advocating desmosomalcell-adhesion structures, and hence allows for the possibilityof sliding of stratum corneum cells to accommodate skinbending. Finally, as the interaction between the individuallayers of the lipid structure involves only hydrocarbons, theindividual layers are relatively free to slide with respect toeach other, making the matrix pliable. The lipid matrix thusmeets the barrier needs of skin by being simultaneously robustand impermeable.

How does our CEMOVIS data reconcile with earlierstudies? The B11-nm repeating unit fits with the 12- to13-nm repeating unit observed in small-angle X-ray diffractionby White et al. (1988), Bouwstra et al. (1991), Hou et al.(1991), and Schreiner et al. (2000) in isolated mouse andhuman stratum corneum, as well as with the 13-nm repeatingunit observed by Madison et al (1987) and Hou et al. (1991) inruthenium tetroxide–stained mouse skin, given a preparation(dehydration)-induced slight straightening of lipid alkyl chainsin these studies. Further, in ruthenium tetroxide–stainedstratum corneum (compare Supplementary Figure S9A online),the electron intensity pattern of the extracellular lipid matrixexpresses a broad:narrow:broad electron lucent band pattern(Madison et al., 1987), which is consistent with our data giventhat ruthenium tetroxide not only associates with lipidheadgroups but also penetrates to some extent between thelipid leaflets when facilitated by local liquid-like disordering

(Supplementary Figure S9C online). In the CEMOVIS micro-graphs, we observed a mid-plane low-intensity region in the6.5-nm bands but not in the 4.5-nm bands (Figure 1b and d;Supplementary Figure S4A, B online. This translates to anincreasingly lower atomic density toward the distal parts of theCER fatty acid and FFA alkyl chains, which would suggest aliquid-like disordered region close to terminal methyl groups.This is further supported by molecular dynamic simulations,which point at a skin lipid hydrocarbon chain arrangementinvolving a gradual change between condensed and liquidalong the chain axes (Das et al., 2009).

We are unable to comment on previous wide-angle X-raydiffraction and attenuated total reflection-Fourier transforminfrared spectroscopy studies (Bouwstra et al., 1992; Damienand Boncheva, 2010), particularly with respect to alkyl chainpacking as to whether it is hexagonal or orthorhombic orboth, as the CEMOVIS data are limited to a resolution ofaround 1–2 nm.

To summarize, the stratum corneum lipid matrix is,according to our analysis, organized as stacked bilayersof fully extended CERs with CHOL molecules associatedwith the CER sphingoid moiety. Further, despite its crystal-line-like character, it is malleable. This condensed but pliablefully extended CER bilayer organization with asymmetricCHOL distribution rationalizes the skin’s low permeabilitytoward water and toward hydrophilic and lipophilic sub-stances, as well as the skin barrier’s robustness towardhydration and dehydration, environmental temperatureand pressure changes, stretching, compression, bending,and shearing.

Final remarksUntil now, progress in skin barrier replacement, skin barrierrepair, skin permeability enhancement, and skin protectionhas largely resulted from empirical efforts, treating thestratum corneum lipid matrix as a black box. The moleculardescription of the lipid matrix presented here may constitutea molecular platform for the above research goals, and inparticular for in silico approaches such as molecularsimulations, as well as for in vitro modeling, to underpininteractions of the lipid matrix with drugs and otherchemicals. For example, it is foreseeable that this knowledgeof the lipid organization will now enable in silico screeningto identify molecules for enhancing skin penetration for drugdelivery and to find ‘‘tightening’’ or ‘‘repair’’ molecules forthe stratum corneum lipid matrix. Further, it may constitute areference for engineering artificial skin for the treatment ofwounds and burn injuries.

More generally, the novel methodological approach wehave used, that combines very high magnification CEMOVISdefocus series with molecular modeling and EM simulation,may open the way for determining the near-native molecularorganization of other structures in normal and diseased cellsand tissues.

MATERIALS AND METHODSA more detailed description of experimental procedures is available

in Supplementary Information.

www.jidonline.org 2223

I Iwai et al.The Human Skin Barrier

Skin biopsy specimens were collected from the human volar

forearm and abdomen of five Caucasian males in their 40s–50s with

no history of skin disease. The samples were immediately vitrified

using a Leica EMPACT2 high-pressure freezer and cryosectioned at

!1401C. Images were recorded at !0.5 to !5 mm defocus, at

120 kV, at a dose of 1,000–17,000 electrons nm!2 per image and a

pixel size of 6.02, 3.31, and 1.88 A, respectively, in a FEG CM200

FEI microscope equipped with a cooled slow-scan 2048" 2048

TVIPS TemCam-F224 HD CCD camera.

The averaged intensity profiles of the cryo-electron micrographs

were obtained by fuzzy distance-based image analysis. A path was

automatically extracted based on the shortest intensity-weighted

distance between two points. Along the path, perpendicular intensity

profiles were extracted (Figure 1c) and then averaged (Figure 1e).

To accurately estimate the distance between intensity peaks,

numerical analysis was used, based on least-squares fitting between

the central part of the measured intensity peaks and a Gaussian

model (Supplementary Figure S3 online).

To check the degree of the packing order of the 6.5- and 4.5-nm

bands, power spectra of 2-pixel-wide strips parallel to the bands in

the micrographs were extracted (Figure 3).

To evaluate the native phase state and the hydration behaviour of

the stratum corneum lipid matrix, multi-photon excitation micro-

scopy LAURDAN GP measurements were performed on excised

forearm skin before and after hydration in vivo (Figure 4a and b).

Simulated CEMOVIS images for the different atomic models

(Supplementary Figures S6D3-J5, S7D3-J5, and S8D3-E5) were created

with a computer program simulating the interaction between the

potential map and the electron beam, the optical transformation

effect of the lens system of the microscope, and the image formation

on the detector (Rullgard et al., 2011). The software is freely

available at http://tem-simulator.sourceforge.net/.

CONFLICT OF INTERESTThe authors state no conflict of interest.

ACKNOWLEDGMENTSWe thank Bertil Daneholt for unwavering support and encouragement for morethan a decade. We also thank Ozan Oktem, Jacques Dubochet, Massimo Noro,Chinmay Das, Emma Sparr, Johan Engblom, Philip Wertz, Ichiro Hatta, JonathanHadgraft, Richard Guy, Marek Haftek, Richard Mendelsohn, Stephen Hoath,David Moore, Jennifer Thewalt, Walter Holleran, Gopi Menon, Michael Roberts,Fred Frasch, Gabriel Wittum, Jim Riviere, Samir Mitragotri, Jesus Perez Gil,Thomas McIntosh, Reinhard Neubert, and our reviewers for invaluablecomments on the manuscript. The present work was made possible by thegenerous support from the Swedish Medical Society, Swedish Research Council,Oriflame, Shiseido Japan, LEO Pharma, Wenner-Gren foundation, WelanderFoundation, European community (3D-EM Network of Excellence), VisualizationProgram by Knowledge Foundation, Vardal Foundation, Foundation for StrategicResearch, VINNOVA, Invest in Sweden Agency, Forskningsradet for Sundhed ogSygdom (FSS, Denmark), and the Danish National Research Foundation. TheKnut and Alice Wallenberg Foundation is acknowledged for support of theelectron microscope facility. II developed and performed high-resolution(pixelsize p3.31 A; nominal section thickness 20–30 nm) CEMOVIS anddesigned the in vivo hydration/dehydration experiment. HH and LdH performedCEMOVIS. SS developed the fuzzy distance-based analysis method. L-GOperformed the EM model simulations. JA constructed the atomic three-dimensional models, formulated the alternative models with LN, and contributedto the writing of the paper. JB, MB, and LAB performed the multi-photonexcitation microscopy LAURDAN GP analysis. AL and DN assisted theCEMOVIS data analysis. SM supported the high-resolution CEMOVIS develop-ment and data collection. US performed the numerical curve fitting analysis andthe power spectra analysis. LN designed the study, analyzed the data, formulatedthe model, and wrote the paper.

SUPPLEMENTARY MATERIAL

Supplementary material is linked to the online version of the paper at http://www.nature.com/jid

REFERENCES

Al-Amoudi A, Chang J-J, Leforestier A et al. (2004) Cryo-electron microscopyof vitreous sections. EMBO J 23:3583–8

Al-Amoudi A, grave;ez DC, Betts MJ et al. (2007) The molecular archi-tecture of cadherins in native epidermal desmosomes. Nature 450:832–837

Al-Amoudi A, Dubochet J, Norlen L (2005) Nanostructure of the epidermalextracellular space as observed by cryo-electron microscopy of vitreoussections of human skin. J Invest Dermatol 124:764–77

Bach D, Miller IR (1998) Hydration of phospholipid bilayers in thepresence and absence of cholesterol. Biochim Biophys Acta 1368:216–24

Bach D, Sela B, Miller IR (1982) Compositional aspects of lipid hydration.Chem Phys Lipids 31:381–94

Bouwstra JA, Gooris GS, Van der Spek JA et al. (1991) Structural investigationsof human stratum corneum by small-angle X-ray scattering. J InvestDermatol 97:1005–12

Bouwstra JA, Gooris GS, Salmon-de Vries MA et al. (1992) Structure of humanstratum corneum as a function of temperature and hydration: a wide-angle x-ray diffraction study. Inter J Pharmaceut 84:205–16

Breathnach AS, Goodman T, Stolinski C et al. (1973) Freeze fracture repli-cation of cells of stratum corneum of human epidermis. J Anat 114:65–81

Buton X, Herve P, Kubelt J et al. (2002) Transbilayer movement ofmonohexosylsphingolipids in endoplasmatic reticulum and Golgimembranes. Biochemistry 41:13106–15

Caspers PJ, Lucassen GW, Carter EA et al. (2001) In vivo confocal ramanmicrospectroscopy of the skin: noninvasive determination of molecularconcentration profiles. J Invest Dermatol 116:434–42

Damien F, Boncheva M (2010) The extent of orthorhombic lipid phases in thestratum corneum determines the barrier efficiency of human skin in-vivo.J Invest Dermatol 130:611–4

Das C, Olmsted PD, Noro M (2009) Simulation studies of stratum corneumlipid mixtures. Biophys J 97:1941–51

Dubochet J, Adrian M, Chang J-J et al. (1988) Cryo electron microscopy ofvitrified specimens. Q Rev Biophys 21:129–228

Elias PM, Friend DS (1975) The permeability barrier in mammalian epidermis.J Cell Biol 65:180–91

Fanelli D, Oktem O (2008) Electron tomography: a short review with anemphasis on the absorption potential model for the forward problem.Inverse Probl 24:013001 (51pp)

Faure C, Tranchant J-F, Duforc EJ (1998) Interfacial hydration of ceramide instratum corneum model membrane measured by 2H-NMR of D2O.J Chim Phys 95:480–6

Garmy N, Taieb N, yahi N et al. (2005) Interaction of cholesterol withsphingosine: physicochemical characterization and impact on intestinalabsorption. J Lipid Res 46:36–45

Garson JC, Doucet J, Leveque J-L et al. (1991) Oriented structure in humanstratum corneum revealed by X-ray diffraction. J Invest Dermatol 96:43–49

Hatta I, Ohta N, Inoue K et al. (2006) Coexistence of two domains inintercellular lipid matrix of stratum corneum. Biochim Biophys Acta1758:1830–6

Holleran WM, Takagi Y, Menon G et al. (1993) Processing of epidermalglycosylceramides is required for optimal mammalian cutaneouspermeability barrier function. J Clin Invest 91:1656–64

Hou SYE, Mitra AK, White SH et al. (1991) Membrane structures in normaland essential fatty acid-deficient stratum corneum: characterization byruthenium tetroxide staining and X-ray diffraction. J Invest Dermatol96:215–23

Lopez-Montero I, Rodriguez N, Cribier S et al. (2005) Rapid transbilayermovement of ceramides in phospholipids vesicles and in humanerytrocytes. J Biol Chem 280:25811–9

2224 Journal of Investigative Dermatology (2012), Volume 132

I Iwai et al.The Human Skin Barrier

Madison KC, Swartzendruber DC, Wertz PW et al. (1987) Presence of intactintercellular lamellae in the upper layers of the stratum corneum. J InvestDermatol 88:714–8

Masukawa Y, Narita H, Sato H et al. (2009) Comprehensive quantificationof ceramide species in human stratum corneum. J Lipid Res 50:1708–19

McIntosh TJ (2003) Organization of skin stratum corneum extracellularlamellae: diffraction evidence for asymmetric distribution of cholesterol.Biophys J 85:1675–81

Norlen L (2001) Skin barrier formation: the membrane folding model. J InvestDermatol 117:823–9

Norlen L, Oktem O, Skoglund U (2009) Molecular cryo-electron tomo-graphy of vitreous tissue sections: current challenges. J Microsc 235:293–307

Norlen L, Plasencia I, Simonsen A et al. (2007) Human stratum corneum lipidorganization as observed by atomic force microscopy on Langmuir-Blodgett films. J Struct Biol 158:386–400

Ouimet J, Lafleur M (2004) Hydrophobic match between cholesterol andsaturated fatty acid is required for the formation of lamellar liquidordered phases. Langmuir 20:7474–81

Rullgard H, Ofverstedt L-G, Masich S et al. (2011) Simulation of transmissionelectron microscope images of biological specimens. J Microsc 243:234–56

Schreiner V, Gooris GS, Pfeiffer S et al. (2000) Barrier characteristics ofdifferent human skin types investigated with X-ray diffraction, lipidanalysis, and electron microscopy imaging. J Invest Dermatol 114:654–60

Steck TL, Ye J, Lange Y (2002) Probing red cell membrane cholesterolmovement with cyclodextrine. Biophys J 83:2118–25

Wertz P, Norlen L (2003) ‘‘Confidence Intervals’’ for the ‘‘true’’ lipidcompositions of the human skin barrier?. In: Forslind B, Lindberg M (eds)Skin, Hair, and Nails Structure and Function. Marcel Dekker, New York,pp 85–106

White SH, Mirejovsky D, King GI (1988) Structure of lamellar lipid domainsand corneocyte envelopes of murine stratum corneum. An x-raydiffraction study. Biochemsitry 27:3725–32

Wilkes GL, Nguyen AL, Wildnauer R (1973) Structure-property relations ofhuman and neonatal rat stratum corneum. I. Thermal stability of thecrystalline lipid structure as studied by X-ray diffraction and differentialthermal analysis. Biochim Biophys Acta 304:265–75

This work is licensed under the Creative CommonsAttribution-NonCommercial-No Derivative Works

3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by-nc-nd/3.0/

www.jidonline.org 2225

I Iwai et al.The Human Skin Barrier

MATERIALS AND METHODS All human studies were approved by the authors' Institutional Review Board and

followed the Declaration of Helsinki protocols. All subjects involved gave their written

informed consent.

Skin sample preparation

Skin biopsies (area, 1x1 mm2; thickness, 100-150 µm) for CEMOVIS were collected,

before as well as after hydration in-vivo, from the left volar forearm and abdomen of 5

male Caucasians in their 40s-50's with normal skin and no history of skin disease, as

judged from dermatological evaluation. The skin area used had not been exposed to any

detergents, treatments or skin care products for one month prior to experimentation. To

avoid sample dehydration, the biopsies were instantly immersed in 1-hexadecane

(MERCK). Subsequently, the samples were placed in the cavity of a cylindrical gold cup

prefilled with 1-hexadecane, and immediately vitrified using a Leica EMPACT2 high

pressure freezer (Leica, Wien, Austria).

Skin hydration and dehydration

To swell the skin, two filter papers (2 cm x 2 cm, Whatman no.1001090, Maidstone,

England) soaked with 300 µl deionised water were put on the volar forearm and occluded

for two hours with Tegaderm (3M Health Care, Germany). Biopsies were then taken and

high-pressure frozen either immediately or after 24 hours acclimatization at 21°C at 93%

RH, in air in equilibrium with a saturated solution of KNO3 (cf. Middleton, 1968).

Cryo sectioning

The vitreous skin samples were trimmed with a trimming diamond blade (Diatome, Biel,

Switzerland) and cryosectioned at -140°C with a nominal section thickness of 25-50 nm

using a 35° diamond knife (Diatome, Biel, Switzerland) with a clearance angle of 6°.

Cutting speed was set to 0.2-1.0 mm/s. The sections were transferred to copper grids with

1000 mesh, using an eyelash glued to a wooden stick. Subsequently they were pressed

with a stamping tool and stored in liquid nitrogen. A facemask was used throughout the

cutting and section transfer procedure to minimize ice-crystal contamination. Air flow,

temperature (21°C) and humidity (<25%RH) were controlled in the work room.

CEMOVIS

Two major advantages of CEMOVIS are that the native, hydrated tissue is preserved

down to the molecular level and that the image pixel intensity is directly related to the

local electron density of the sample. For more detailed accounts of CEMOVIS, see

Dubochet et al., (1988), Al-Amoudi et al. (2004) and Norlén et al. (2009).

We used a GATAN model 626 cryo-holder (GATAN, Pleasanton, CA) at -180°C.

Images were collected at 120kV in a FEG CM200 FEI microscope, equipped with a

cooled slow scan 2048x2048 TVIPS TemCam-F224 HD CCD camera (pixel size 14µm).

Images were generally recorded at a dose of 1.000-17.000 electrons/nm2 per image and a

total magnification of 27.500 X (pixel size: 6.02 Å), 50.000 X (pixel size: 3.31 Å), and

88.000 X (pixel size: 1.88 Å), respectively. During image acquisition, a defocus of about

-2 µm was regularly employed as it yields a good compromise between relatively high

image resolution (low defocus) and relatively high image contrast (high defocus). To

discriminate different molecular lipid models, images were also acquired at very low

defocus (-0.5 µm), yielding images with high resolution but with very low contrast (cf.

Fig. 6A), and at very high defocus (-5 µm), yielding images with high contrast but with

very low resolution (cf. Fig. 6C). In addition, defocus series were acquired at very high

magnification (pixel size: 1.88 Å) (cf. Fig. 6G-I). In this way, fine changes in interference

patterns caused by gradually increasing the microscope's defocus during repeated image

acquisition at a fixed position could be used to verify the selected model (Fig. 6J-L).

Image analysis

For the image analysis, we developed a semi-automatic program using MATLAB and

C++. As a preprocessing step, we used the edge preserving smoothing algorithm

anisotropic diffusion filtering (Perona and Malik, 1990). In this way small variations,

largely corresponding to noise, are suppressed while significant features are left

untouched. We then treated the image locally as a fuzzy membership map, where the

membership corresponds to degree of belongingness to a lucent band of the lamellar

intensity pattern. The procedure is as follows. Two points are marked on one lucent band.

A path is then automatically extracted from one point to the other based on the shortest

intensity-weighted distance between the two points. We used the intensity-weighted

distance introduced by Levi and Montanari (1970), for which the mean of the intensities

of two neighbouring points is multiplied by the spatial distance between them. Intensity-

weighted distance is described in a theoretical framework and denoted fuzzy distance by

Saha et al (2002). The result is that a path is placed along the highest ridge between the

two points. It thus becomes centrally located on the lucent band. For each point along the

path, perpendicular intensity profiles were extracted using the improfile method in

MATLAB. Finally, the profiles were averaged.

Numerical curve fitting

Averaged perpendicular intensity profiles (cf. Fig. 1C) were used for the numerical curve

fitting. The goals were to accurately estimate the distance between intensity peaks and to

detect systematic deviation from Gaussian distribution. In figure S3, seventeen profiles

were used, each with 201 grid points. The grid unit was 3.31 Å and corresponded to the

image pixel size. As seen in Figure 1D, the electron micrographs vary systematically in

intensity in lucent regions (a fine ‘stripe’ pattern perpendicular to the lamellar plane),

while lamellar peak intensity positions remain stable. The average intensity is thus

suitable for accurate determination of peak distance. The least squares based fit to the

central part of the measured intensity peaks was excellent (Fig. S3). This indicates that

Gaussian modelling is sufficient to establish peak positions accurately. Two distances

emerge, one of ~ 4.5 nm and one of ~ 6.5 nm. Deviations from Gaussian distribution

were confined to the intensity peak peripheries (Fig. S3). The bilateral broadening of the

peaks towards the valley mid planes is largely an effect of deviation from the Gaussian

model. As the deviation is equally pronounced on the 4.5 nm valley side and the 6.5 nm

valley side of the intensity peaks (Fig. S3), the higher intensity of the 4.5 nm valley mid-

planes with respect to the 6.5 nm valley mid-planes is partly due to more pronounced

intensity curve overlapping between the more closely separated peaks (4.5 nm) compared

to between the more widely separated peaks (6.5 nm).

Power spectra analysis

To check the degree of order or amorphous character of the ~6.5 nm and ~4.5 nm bands

we analysed power spectra of 2 pixel wide strips parallel to the bands in the micrographs

(Fig. 3A). Individual power spectra obtained from different regions of the layered

structure are presented in figures 3D-G.

The power spectra show that each strip-region has some intrinsic order with

recurring characteristics. However, at the same time the intrinsic order seems to vary

considerably when comparing the spectra from other strips in a similar region. The only

region with somewhat reproducible spectra across its sampled strips is seen in (D). The

conclusion is that the (D) region, besides its low intensity mid-plane (E), is somewhat

ordered, while the other regions are more amorphous.

The ~4.5 nm thick band including its mid-plane (F), as well as the head-group

bands (G), are amorphous but have at the same time some intrinsic order with recurring

characteristics of low reproducibility.

LAURDAN GP before and after hydration in-vivo

The LAURDAN generalized polarization (GP) function reflects the wavelength

dependence of LAURDAN’s emission spectrum (Parasassi et al., 1998). The LAURDAN

emission spectrum is sensitive to the extent of the water dipolar relaxation process that

occurs in the probe’s microenvironment (this probe is insoluble in water and shows a

high partition to membranes). Since the extent of water dipolar relaxation at the

membrane interface (where LAURDAN is located) is highly connected with the packing

of the host membrane, this probe is able to discriminate different lipid phases (Bagatolli,

2006). For example, a 50 nm shift is observed in the fluorescence emission spectrum of

LAURDAN when phospholipid membranes undergo solid ordered/liquid disordered

phase transition (Parasassi et al., 1998). The GP function was defined analogously to the

fluorescence polarization function as:

RB

RB

IIIIGP

+

−= (1)

where IB and IR correspond to the intensities at the blue and red edges of the emission

spectrum (440 and 490 nm) using a given excitation wavelength (Parasassi et al., 1990).

In lipid bilayers high LAURDAN GP values (0.45 – 0.60) correspond to ordered phases

whereas low LAURDAN GP values (below 0.15) correspond to fluid disordered-like

phases (Bagatolli, 2006). Ordered phases have been reported previously in mixtures

composed of purified human stratum corneum lipid extracts (Plasencia-Gil et al., 2007)

and in pig skin ex-vivo (Carrer et al., 2008).

A total of 96 GP values were calculated from normal and hydrated volar forearm

skin (one from each of 6 images, for each of 4 samples, for each of 2 preparations, for

each of 2 individuals) (Fig. 4A-B). The subjects were both male Caucasians in their 40s

with normal skin and no history of skin disease, as judged from dermatological

evaluation. The skin area used had not been exposed to any detergents, treatments or skin

care products for one month prior to experimentation. The GP value obtained per image

was an average GP value obtained from over 20 regions using a ROI routine. After

excision, the skin samples were immediately immersed in a hexadecane solution of

LAURDAN (1.4 mM) for 2 hours to avoid sample dehydration and ensure proper

LAURDAN incubation. As a control, experiments were also performed after 5 and 30

minutes of LAURDAN incubation. After LAURDAN incubation, the samples were

rinsed in pure hexadecane and mounted in a microscope slide for obtaining the GP

images. A custom-built multi-photon excitation microscope (Brewer et al., 2010) was

used in our experiments. This microscope is controlled by Globals for Images SimFCS

(Laboratory for fluorescence dynamics, University of Irvine). The excitation light source

was a femtosecond*Ti:Sa*laser*(Mai Tai DeepSeeTM,*Spectra*Physics,*Mountain*View,*CA)* operated* at* a*wavelength of 780 nm. For the LAURDAN GP measurements the

total fluorescence signal was divided between two detection channels (each of them

equipped with a Hamamatsu H7422P-40 photomultiplier) using a dichromatic mirror

(splitting below and above 475 nm). One detection channel contain a bandpass filter of

438 ± 12nm whereas the other contain a bandpass filter of 494 ± 10 nm (corresponding

respectively to IB and IR in equation 1). The calculation of the GP function was

performed from Globals for images software and a GP standard was used as previously

described (Brewer et al., 2010).

Atomic 3D modelling

The atomic model comprised the 3 key components of the known lipid composition,

namely a C24 ceramide NP (the predominant ceramide in the stratum corneum),

cholesterol, and a C24 free fatty acid (the predominant free fatty acid in the stratum

corneum). Models of the individual molecules were generated using a molecular-model

building procedure with the atoms being located at their ideal bond distances, angle and

torsions. The ceramide molecule was rotated about the bonds –

CH(OH)−CH(CH2OH)NH– and –CH(CH2OH)NH–COCH– to yield a fully extended

conformation. The three component molecules were then rotated to lie lengthwise along

the z-axis and translated relative to each other to give a close packed assembly with the

cholesterol molecule lying adjacent to the sphingoid chain of ceramide NP and the fatty

acid adjacent to the ceramide fatty chain. The full assembly containing fully extended

ceramides gave a unit subcell with dimensions of approximately 1.0 x 1.1 x 5.25 nm.

Cryo-EM simulation

Simulated images from generated atomic models were created with a TEM simulator

program recently developed by H. Rullgård, L.-G. Öfverstedt and O. Öktem (Rullgård et

al., 2011). The first part of the program is a phantom generator that can read one or more

atomic models in the RCSB Protein Data Bank (PDB) format and construct a model

scenario with molecules at defined positions. An electron scattering potential map is then

generated with a background structure and potential corresponding to that of vitrified

water. The second part of the program simulates the interaction between the potential

map and the electron beam, the optical transformation effect of the lens system of the

microscope and the image formation on the detector. Parameters defining optical

properties of the microscope, e. g. acceleration voltage, aberration constants and defocus,

and the point spread function of the detector can be set to mimic the conditions in a real

experiment. Thus, simulated images (Fig. 5J, 6D-F, 6J-L, S6D3-J5, S7D3-J5, and S8D3-

E5) were generated with conditions corresponding to the CEMOVIS images.

REFERENCES

Al-Amoudi, A. Chang, J-J. Leforestier, A. McDowall, A. Michel Salamin, L. Norlén, L.

Richter, K. Sartori Blanc, N. Studer, D. Dubochet, J (2004) Cryo-electron microscopy of

vitreous sections. EMBO J. 15;23(18):3583-8

Bagatolli, LA (2006) To see or not to see: lateral organization of biological membranes

and fluorescence microscopy. Biochim Biophys Acta 1758:1541-1556

Brewer, J. Bernardino de la Serna, J. Wagner, K. and Bagatolli, LA (2010) Multiphoton

excitation fluorescence microscopy in planar membrane systems, Biochim. Biophys. Acta

1798:1301-8

Carrer, DC. Vermehren, C. Bagatolli, LA (2008) Pig skin structure and transdermal

delivery of liposomes: A two photon microscopy study. J. Controlled Release, 132(1):12-

20

Dubochet, J. Adrian, M. Chang, J-J. Homo, J-C. Lepault, J. McDowall, AW. Schultz, P

(1988) Cryo electron microscopy of vitrified specimens. Q Rev Biophys. 21 (2): 129-228.

Levi, G., and Montanari, U (1970) A grey-weighted skeleton. Information and Control,

17, 62-91

Middleton, JD (1968) The mechanism of water binding in stratum corneum. Br. J.

Dermatol. 80:437-450

Norlén, L. Öktem, O, Skoglund, U (2009) Molecular cryo-electron tomography of

vitreous tissue sections: current challenges. J. Microscopy 235:293-307

Parasassi, T. De Stasio, G. d'Ubaldo, A. Gratton, E (1990) Phase fluctuation in

phospholipid membranes revealed by Laurdan fluorescence, Biophysical journal

57:1179-1186

Parasassi, T. Kranowska, E. Bagatolli, LA. Gratton, E (1998) Laurdan and Prodan as

polarity-sensitive fluorescent membrane probes, J. Fluorescence 8:365-373

Perona, P. and Malik, J (1990) Scale-space and edge detection using anisotropic

diffusion. IEEE Transactions on Pattern Analysis, 12(7):629-639.

Plasencia-Gil, I. Norlén, L. Bagatolli, LA (2007) Direct visualization of lipid domains in

human skin stratum corneum’s lipid membranes: effect of pH and temperature. Biophys.

J 93(9):3142-3155

Rullgård, H. Öfverstedt, L-G. Masich, S. Daneholt, B. Öktem, 0 (2011) Simulation of

transmission electron microscope images of biological specimens. J. Microscopy

243(3):234-256

Saha, PK. Wehrli, FW. Gomberg, BR (2002) Fuzzy distance transform: theory,

algorithms, and applications. Computer Vision and Image Understanding. 86:171–190.

*

SUPPLEMENTARY FIGURE LEGENDS

Figure S1. Molecular models for the lipid organization of the stratum corneum

extracellular space. Models proposed by Schröter et al. (2009) (A), by McIntosh (2003)

(B), by Hill and Wertz (2003) (C), by Bouwstra et al., (2001) (D), by Forslind (1994) (E), by

Swartzendruber et al., (1989) (F). Six different models for the molecular organization of the

stratum corneum extracellular lipid matrix have been proposed. Swartzendruber et al. (1989)

presented the first model (F). It was based on the broad:narrow:broad electron lucent band

pattern observed in RuO4 stained stratum corneum (cf. Suppl. fig. 8A), and consisted of

triple-band units with ceramides in fully extended conformation (i.e., with the ceramide

sphingoid- and the fatty acid parts pointing in opposite directions). Each triple-band unit

was composed of one narrow central band and two broad peripheral bands. The central band

expressed hydrocarbon chain interdigitation and contained no cholesterol. The peripheral

bands expressed no chain interdigitation but contained cholesterol (F). Later, Forslind

(1994) presented a model (E) based on lateral lipid domain segregation. It consisted of

stacked lipid bilayers composed of ceramides in folded conformation (i.e., with the ceramide

sphingoid- and the fatty acid parts pointing in the same direction). Each bilayer expressed

thickness changes due to laterally segregated crystalline and liquid crystalline domains (E).

Combining data derived from conventional electron microscopy and SAXD, Bouwstra et al.

(2001) presented a model (D) consisting of triple-band units with a 13 nm repeat period.

Each triple-band unit was composed of one narrow central band and two broad peripheral

bands. The central band expressed interdigitating omega-esterified polyunsaturated fatty

acids of ceramide EOS, short chain ceramides in folded conformation and cholesterol. The

two peripheral bands expressed long chain ceramides in folded conformation and cholesterol

(D). Reinterpreting the band patterns of RuO4 stained stratum corneum, Hill and Wertz

(2003) proposed a model (C) similar to that of Bouwstra et al (2001), but with uniform

thicknesses of the three bands constituting the triple-band units. Based on SAXD

experiments on skin lipid model systems in-vitro, McIntosh (2003) presented a model (B)

consisting of twin-band (double bilayer) units with a 13 nm repeat period, with an

asymmetric cholesterol distribution, and with equal thicknesses of the constituent bands.

Note that McIntosh's model does not state whether the constituent ceramides are all in

folded conformation or, alternatively, as depicted in (B), both in folded and fully extended

conformation. Recently, Schröter et al. (2009) presented a model (A) based on neutron

! 2!

scattering experiments on skin lipid model systems in-vitro. It consisted of lamellar units of

equal thickness, composed of mixed ceramides in folded and extended conformation, and

with a homogenous cholesterol distribution. All bands were spanned by the fatty acid chains

of ceramide EOS (A). LE: 'lipid envelope' (or stratum corneum cell plasma membrane).

Note that the schematically sketched molecular organization of the lipid envelopes is not

part of the models (A-E) and is included only to put the models in their proper context.

Figure S2. The lipid matrix is, despite its crystalline-like character, malleable. The

stratum corneum extracellular lipid matrix is folded locally. The folding decreases on

hydration and increases on dehydration. Low magnification CEMOVIS micrographs of

stratum corneum after hydration in-vivo (A), at normal in-vivo conditions (B), and after

hydration in-vivo followed by dehydration ex-vivo (C). The lower panel illustrates the

folded pattern of the extracellular space. Image side lengths: 5 µm.

Figure S3. Numerical curve analysis was performed to accurately estimate the distance

between intensity peaks. (A) Averaged electron intensity line (grey) obtained from 17

profiles extracted from figure S4A. Gaussian curve peaks (red) numerically least squares

fitted to the electron intensity peaks (black). Intensity difference (green) between measured

(black) and calculated (red) peaks. The least squares based fit to the central part of the

measured intensity peaks was excellent. This indicates that Gaussian modelling is sufficient

to establish peak positions accurately. Two alternating distances emerge, one of

approximately 4.5 nm and one of approximately 6.5 nm.

Figure S4. The 6.5 nm bands express a mid-plane low intensity region. (A) High

magnification CEMOVIS micrograph of the extracellular space in the mid-part of stratum

corneum. (B) Intensity profile obtained from fuzzy distance based path growing of the area

shown in (A). Note the "shoulders" (open arrows) centrally on each slope delimiting the

mid-plane low intensity region of the 6.5 nm bands. Black arrows (A) denote their

corresponding positions in the cryo-electron micrograph. (C) Schematic representation of a

stacked ceramide alternating monolayer organization with cholesterol localized to the

regions corresponding to the 4.5 nm bands and with FFA localized to the regions

! 3!

corresponding to the 6.5 nm bands. Square brackets mark regions with presumed liquid-like

disordering. Pixel size in (A): 3.31 Å.

Figure S5. No inter-individual or inter-site (forearm versus abdomen) variation was

recorded. Left column: high magnification CEMOVIS micrographs of the extracellular

space in the midpart of stratum corneum of vitrified epidermis obtained from left volar

forearm (A, C-E) and abdomen (B). Right column: corresponding intensity profiles obtained

from fuzzy distance based image analysis. Section thicknesses ~50 nm (A-E). Pixel size in

(A): 6.02 Å, and in (B-E): 4.35 Å.

Figure S6. Electron microscopy simulation results from 7 fully extended ceramide

bilayer models with varying cholesterol distribution. (A-C) CEMOVIS micrographs of

the stratum corneum extracellular lipid matrix acquired at -5 µm (A), -2 µm (B), and -0.5

µm (C) defocus. (D3-J5) Corresponding simulated electron micrographs obtained from 7

fully extended ceramide models. (D1-J1) Repeating units for each simulated model. (D2-J2)

Calculated electron scattering potential 3D maps of the topmost layer out of 20

superimposed layers used to generate each individual simulated micrograph (D3-J5). In

model (D), cholesterol is selectively localized to the ceramide sphingoid part. In model (E),

cholesterol has been removed to evaluate whether the simulation method could discriminate

the presence (D) or absence (E) of cholesterol. In model (F), cholesterol is selectively

localized to the ceramide fatty acid part. In models (G-J), cholesterol is homogenously

distributed between the ceramide sphingoid and fatty acid parts. Contrary to models (G, J),

models (H, I) express axial headgroup displacement of cholesterol and free fatty acids.

Models (H, I) differ in that model (H) expresses a pair-wise lateral distribution of ceramides

while model (I) expresses a homogeneous lateral distribution of ceramides. Note that except

for the position of the lipid headgroups, the localization of cholesterol within the fully

extended ceramide structure largely determines the electron scattering properties of the

models.

Figure S7. Electron microscopy simulation results from 7 fully extended ceramide

stacked monolayer models with varying cholesterol distribution. (A-C) CEMOVIS

! 4!

micrographs of the stratum corneum extracellular lipid matrix acquired at -5 µm (A), -2 µm

(B), and -0.5 µm (C) defocus. (D3-J5) Corresponding simulated electron micrographs

obtained from 7 stacked fully extended ceramide models. (D1-J1) Two repeating units for

each simulated model. (D2-J2) Calculated electron scattering potential 3D maps of the

topmost layer out of 20 superimposed layers used to generate each individual simulated

micrograph (D3-J5). In model (D) cholesterol is selectively localized to the ceramide

sphingoid part. In model (E) cholesterol has been removed to evaluate whether the

simulation method could discriminate the presence (D) or absence (E) of cholesterol. In

model (F), cholesterol is selectively localized to the ceramide fatty acid part. In models (G-

J), cholesterol is distributed homogenously between the ceramide sphingoid and fatty acid

parts. Contrary to models (G, J), models (H, I) express axial headgroup displacement of

cholesterol and free fatty acids. Models (H, I) differs in that model (H) expresses a pair-wise

lateral distribution of ceramides while model (I) expresses a homogeneous lateral

distribution of ceramides. Note that except for the position of the lipid headgroups, the

localization of cholesterol within the fully extended ceramide structure largely determines

the electron scattering properties of the models.

Figure S8. Electron microscopy simulation results from 2 folded ceramide bilayer

models with and without the presence of cholesterol. (A-C) CEMOVIS micrographs of

the stratum corneum extracellular lipid matrix acquired at -5 µm (A), -2 µm (B), and -0.5

µm (C) defocus. (D3-E5) Corresponding simulated electron micrographs obtained from 2

folded ceramide models. (D1-E1) Repeating units for each simulated model. (D2-E2)

Calculated electron scattering potential 3D maps of the topmost layer out of 20

superimposed layers used to generate each individual simulated micrograph (D3-E5). In

model (D), cholesterol is present. In model (E), cholesterol has been removed to ascertain if

the simulation method could distinguish the presence (D) or absence (E) of cholesterol. Note

that the presence of cholesterol within the folded ceramide structure largely determines the

electron scattering properties of the models.

Figure S9. Electron intensity pattern in ruthenium tetroxide stained neonatal mouse

skin. (A) RuO4 staining pattern of the stratum corneum extracellular space. (B) Enlarged

view of the area marked by a white box in (A). (C) Schematic model for the RuO4 staining

! 5!

pattern in (B). (D) RuO4 staining pattern of lamellar body inside cell in stratum granulosum.

(E) Enlarged view of the area marked by a white box in (D). (F) Schematic model for the

RuO4 staining pattern in (E). A transition from folded to stretched ceramide bilayer

conformation during skin barrier formation can explain the ruthenium tetroxide staining

pattern of both 'lamellar bodies' (i.e., the precursor system of the stratum corneum

extracellular lipid matrix) and the stratum corneum lipid matrix. Inside lamellar bodies in

topmost viable epidermis just beneath the stratum corneum, glycosylceramides (i.e., the

precursors of the stratum corneum ceramides) sometimes form aggregates of irregularly

stacked bilayers (Norlén et al., 2003; Al-Amoudi et al., 2005). The corresponding electron

density pattern in ruthenium tetroxide stained skin is characterised by alternating broad

(strong) and thin (weak) dark lines. The broad dark lines correspond to lipid headgroups, as

they can be seen to change in thickness locally, reflecting local bilayer separation in the

aqueous protein-enriched environment inside the lamellar bodies (D-F). On the contrary, the

thin dark lines are constant in thickness throughout (D-E). They therefore likely correspond

to the mid-plane of the lipid bilayer, reflecting a change in the hydrocarbon chain region

between liquid-like disordered (closer to the bilayer mid-plane) and ordered (closer to the

lipid headgroups) carbons. This indicates that ruthenium tetroxide not only associates with

lipid headgroups but also penetrates to some extent between lipid leaflets, especially when

facilitated by local liquid-like hydrocarbon chain disordering. This makes the interpretation

of the broad:narrow:broad electron lucent band pattern in ruthenium tetroxide stained

stratum corneum (A) straightforward. The central narrow electron lucent band then

corresponds to the mid-plane of the cholesterol-enriched sphingoid side (C), and the broad

electron lucent bands to the proximal part of the fatty acid side (C), of the extended

ceramide bilayers. The broad (strong) dark lines thus correspond to lipid headgroups and the

thin (weak) dark lines to the fatty acid mid-plane (C). The staining mechanism in the

ceramide-based stretched-out bilayer system of the stratum corneum extracellular space (A-

C) thus becomes identical to that in the glycosylceramide-based bilayer system of the

lamellar bodies (D-F). Figures (A-B, D-E) are adapted from Madison et al. (1987), with

permission.

!