A Microscopic Multiphase Diffusion Model of Viable Epidermis Permeability

Abstract

A microscopic model of passive transverse mass transport of small solutes in the viable epidermal layer of human skin is formulated on the basis of a hexagonal array of cells (i.e., keratinocytes) bounded by 4-nm-thick, anisotropic lipid bilayers and separated by 1-μm layers of extracellular fluid. Gap junctions and tight junctions with adjustable permeabilities are included to modulate the transport of solutes with low membrane permeabilities. Two keratinocyte aspect ratios are considered to represent basal and spinous cells (longer) and granular cells (more flattened). The diffusion problem is solved in a unit cell using a coordinate system conforming to the hexagonal cross section, and an efficient two-dimensional treatment is applied to describe transport in both the cell membranes and intercellular spaces, given their thinness. Results are presented in terms of an effective diffusion coefficient, ${\overline{D}}^{\mathrm{epi}}$, and partition coefficient, ${\overline{K}}^{\mathrm{epi}/w}$, for a homogenized representation of the microtransport problem. Representative calculations are carried out for three small solutes—water, L-glucose, and hydrocortisone—covering a wide range of membrane permeability. The effective transport parameters and their microscopic interpretation can be employed within the context of existing three-layer models of skin transport to provide more realistic estimates of the epidermal concentrations of topically applied solutes.

Introduction

Quantitative understanding of the biophysics of drug/chemical diffusion though the skin (1,2) is crucial to the effective development of topically applied and transdermally delivered drugs (3,4), as well as assessment of risks associated with chemical exposures (5,6). Theoretical models of dermal absorption (7) can strongly support advancement in these areas by providing mechanistic insights into solute flux pathways, reasonable predictions of transdermal transport rates and subsurface concentration levels, and a quantitative framework to correlate experimental data and guide measurements.

The most comprehensive modeling approach is based on a two-scale strategy. At a macroscopic level of description, the skin is broken down into stratum corneum (SC, barrier), viable epidermis, and dermis layers (8–10). Each layer is imbued with an effective partition coefficient $\overline{K}$ quantifying solute affinity for the layer relative to an aqueous reference solution, and an effective diffusion coefficient $\overline{D}$. The dermis is also described by one or more parameters quantifying the rate of clearance from the tissue into the dermal vasculature (8–12). Depending on the solute, all layers may additionally be characterized in terms of an equilibrium constant (or sorption isotherm) and/or rate constants for reversible solute binding to tissue constituents (for instance, keratin and other corneocyte constituents in the SC (13,14), or albumin (11,12,15) and other binding/transporter proteins (16,17) in the epidermis and dermis). Solution of the macroscopic multilayer transport equations generally requires the application of numerical techniques (7,10,14). The resulting predictions are only as good as the effective tissue properties input to the calculation.

Quantitative estimates of the effective tissue properties $\overline{K}$ and $\overline{D}$ for a given solute can come, in turn, from microscopic diffusion models that explicitly account for partitioning and diffusion (and possibly also binding) within each phase of some more or less idealized geometrical representation of the heterogeneous microstructure (7,13,18–20). Ideally such theoretical efforts should be calibrated against experimental studies. In this regard, targeted measurements of microscopic phase-specific properties (relatively few in the literature) are ultimately more decisive than macroscopic observations of overall absorption rates, in which a multitude of factors express themselves in convoluted form. Given its role as the primary and outermost barrier layer, it is not surprising that the SC has attracted the most attention in the form of microscopic brick-and-mortar models of SC permeability (7,13,18–20). Cleek and Bunge (21) developed a useful and influential description of the additional mass transfer resistance contributed by viable tissue below the SC, but it constitutes a lumped parameter (effective macroscopic) approach. A useful model has recently been developed to estimate partition and diffusion coefficients specifically in the dermis from solute properties including molecular weight (MW) and octanol/water partition coefficient (K^oct/w) (11). There seems to be no detailed microscopic diffusion model for the viable epidermis (henceforth distinguished here by the superscript epi) comparable to the brick-and-mortar models of the SC. Formulation of such a model yielding ${\overline{\mathit{K}}}^{\mathrm{epi}/\mathrm{w}}$ and ${\overline{D}}^{\mathrm{epi}}$ is the objective of this work.

Less is known about effective transport properties of the epidermis than about those of the SC or dermis, because it is difficult to isolate (8,22). Until now, a reasonable pragmatic approach has been to treat the viable epidermis as aqueous tissue roughly equivalent to dermis without any vascular clearance in dermal diffusion models (8). This approximation has minimal impact for the prediction of systemic absorption rates because the resistance of the viable epidermis is usually low relative to that of the SC. However, it falls short in cases where strong penetration into deeper tissue occurs (e.g., transdermal drug delivery when the SC barrier has been seriously compromised by a penetration enhancer (23)). It also does not suffice to model actual solute concentrations within the epidermis, which is critical for the quantitative mechanistic understanding of epidermal bioavailability in the areas of topically applied drugs (24) and contact allergy (9). Due to its viable cellular structure, epidermis below the barrier layer presents a set of obstacles to solute diffusion inherently different from those of the SC and dermis. Transport through it involves a combination of phenomena including:

• 1.Hindered permeant diffusion within cytoplasm relative to bulk aqueous diffusion (25–27);
• 2.Cell wall permeation by transbilayer mass transfer (19,28) in parallel with intercellular transfer via epidermal gap junctions (GJ) (26,27,29–32);
• 3.Diffusion through extracellular fluid in the interstitial space;
• 4.Barrier properties of tight junctions (TJ) (33–36);
• 5.Permeant binding to cellular structures, as well as to albumin (11,12,15) and other binding/transporter proteins (16,17) in both cytoplasm and extracellular fluid; and
• 6.Permeant metabolism (37,38).

Our analysis develops a basic microscopic diffusion model that explicitly incorporates elements 1–4 within a realistic unit cell of epidermal structure, and defines the basic passive transport properties of viable epidermal tissue for small solutes that are not excessively hydrophobic and therefore not overly susceptible to protein binding. We consciously choose to limit the scope of physical phenomena addressed by presently not including items 5 and 6, for two reasons. The first is that they are highly permeant-specific. The second is that their inclusion for any specific permeant requires a preliminary understanding of passive transport per se, upon which these phenomena are superposed and from which they must be distinguished. Thus, the first priority is to understand passive diffusion through the cellular structure.

Our model is very similar in spirit to existing brick-and-mortar models of the SC (7,13,18–20). Nothing precludes this type of analysis from being applied to viable tissue, and indeed, somewhat similar models exist for other tissues of the body (27,39). The problem for analysis of epidermal penetration is that the unit cells in such a model must be carefully constructed to accurately reflect the specific attributes of the real tissue, and this work just has not been undertaken for the viable epidermis. It is important to distinguish this microscopic approach from coarse-grained (macroscopic) models of epidermal transport (21) that employ effective tissue properties but do not directly derive their values from the microstructural physics.

The presentation is organized as follows. First we define the model microstructure and equations governing solute transport, and formulate a reasonable set of estimates for a host of physicochemical parameters appearing therein. Subsequent sections describe the analytical and numerical procedures used to solve the model equations and derive the effective transport coefficients, and analyze the results to understand the key microscopic contributors to solute flux. We close with a discussion of the elements needed for broad practical application of the model, including experimental studies for comprehensive parameterization across a spectrum of permeants, and more permeant-specific extensions to the effects of binding and metabolism.

Method of Solution

Actual calculations are performed using a system of horizontal coordinates ξ and ϕ defined by the relations

$$\begin{array}{l}x=\xi ,\\ y=\xi \tan \varphi ,\end{array}$$ (1)

which form the nonorthogonal mesh shown in Fig. 1 c. The computation domain is defined by the intervals 0 ≤ ξ ≤ x_max, 0 ≤ ϕ ≤ ϕ_max, and z_min ≤ z ≤ z_max. Direct use of polar coordinates r = (x² + y²)^1/2, ϕ = arctan(y/x) would be cumbersome because the cellular membrane is situated at a variable (ϕ-dependent) radius. Use of ξ and ϕ makes the membrane coincide with the coordinate surface ξ = x_max.

Figure 1.

Assumed microstructure for the spinous layer. (A) Representation of a small section of tissue with two unit cells cut open. Node points for a coarse finite difference discretization of the interior (cytoplasmic) domain are shown within one unit cell; surface discretization points for the bounding lipid membranes and layers of extracellular fluid are not shown to avoid clutter. (B and C) Cross sections of one unit cell as seen from top and side, respectively. The thickness of the lipid membranes is exaggerated (not to scale) for visibility in all parts of the figure.

The unit cell diffusion problem, cast in terms of the variables ξ, ϕ, and z, and written in a dimensionless form, is solved by the method of finite differences coded in Fortran 95. An illustrative coarse discretization of the computational domain is shown from two perspectives in Fig. 1, b and c. The integrals in Eq. 10 are evaluated numerically using the calculated solute concentrations at the nodes of the finite difference mesh. Significant aspects of the mathematical formulation and computational scheme are relegated to the Supporting Material for brevity here. The computer code is available from the authors upon request.

Problem Formulation

Fig. 1 shows the assumed microstructural geometry. The epidermal cell population (keratinocytes, melanocytes, Langerhans cells) is represented by a single repeated structure considered to be a keratinocyte. This simplification suffices for a model limited to passive transport as there are (to our knowledge) no documented differences in cellular permeability and the keratinocyte is the predominate cell type. Keratinocytes are idealized as hexagonal prisms arranged in a hexagonal array within a given cellular layer, and aligned in columns vertically. This assumption squares with the roughly polyhedral shape evident in immunofluorescence microscopy images of keratinocytes (29,32,35), and their hexagonal shape as seen in cross section upon migration to the surface (40). Based on the tightness of the observed structure (32,35) and recent published conceptualizations (see Fig. 1 in Brandner (36) and Fig. 2 in Tsukita and Furuse (33)), the amount of extracellular fluid seems to be small, and so is represented in terms of thin aqueous films separating adjacent keratinocytes. Tight junctions appear in Fig. 1 as effective barriers interrupting these liquid films, asymmetrically located nearer the tops than the bottoms of the keratinocytes (33,36). We focus on neutral permeant molecules, or the unionized (neutral) form of ionizable permeants, because the approach to treating ionization is well established (41).

Overall approach

Our objective is a rigorous calculation of the effective diffusivity ${\overline{D}}^{\mathrm{epi}}$ for vertical (transdermal or −z-directed) motion through the structure shown. For purposes of this calculation, the structure is regarded as an infinite medium extending indefinitely in all directions including the vertical. Clearly, in subsequently using the effective diffusivity in an absorption calculation at the macroscopic scale, the epidermis has a finite thickness.

Substrata of the viable epidermis differ significantly in terms of the aspect ratio of keratinocytes, metabolic and other processes, and occurrence of tight junctions. Therefore, we develop two variants of the model, respectively geared toward representing tissue of the spinous and granular layers, and ultimately envision use of the model in a two-layer macroscopic representation of the epidermis. The basal layer is combined with the spinous layer in this description as the higher aspect ratio for the single layer of basal cells (42) is inconsequential to passive permeability, as will be shown.

Use of the effective (coarse-grained) value of ${\overline{D}}^{\mathrm{epi}}$ calculated for each substratum constitutes a reasonable approximation in a macroscopic model, because the substratum comprises several layers of keratinocytes—typically three for the granular layer, and three or more for the spinous layer (33,42).

Geometry

The side length (ℓ_side) and height (ℓ_height) of the hexagonal prisms representing the cytoplasm of keratinocytes are regarded as adjustable parameters, as are the thicknesses of the cellular membrane (s^lip) and the layer of extracellular fluid separating neighboring keratinocytes (s^ext). Nominal values of these (and other) parameters are listed in Table 1. Given the spatial periodicity of the structure, it suffices to consider only one unit cell (say that centered at (x, y, z) = (0,0,0) for definiteness), and considerations of symmetry indicate that the unit-cell transport problem need only be posed and solved in 1/12th of this unit cell (i.e., a sector of the horizontal cross section spanning an angle of ϕ_max = 30°). The computational domain in the xy plane ($\mathcal{D}$_xy) considered for the cytoplasm is therefore as shown in Fig. 1 B. The coordinate x runs from 0 to x_max (with $x_{\max }=\left(\text{cos}{\mathit{\varphi }}_{\text{max}}\right){\ell }_{\mathrm{side}}=\left(\sqrt{3}/2\right){\ell }_{\mathrm{side}}$), and z runs from z_min to z_max (with z_max = − z_min = ℓ_height/2). The vertical period of the structure,

$${\ell }_{\mathrm{period}}={\ell }_{\mathrm{height}}+2{\mathit{s}}^{\mathrm{lip}}+s^{\mathrm{ext}},$$ (2)

differs little from the height ℓ_height of the cytoplasm of one keratinocyte owing to thinness of the lipid and extracellular fluid phases.

Table 1.

Assumed dimensions quantifying keratinocyte geometry, and formulas providing provisional values of microscopic input parameters for the diffusion model

Microscopic tissue attribute, phase, or element	Formula
Geometry	${\ell }_{\text{side}}=\{\begin{array}{ll}12.4\mu \text{m}\hfill & \left(\text{stratum}\text{spinosum}\right)\hfill \\ 15.3\mu m\hfill & \left(\text{stratum}\text{granulosum}\right)\hfill \end{array}$
	${\ell }_{\text{height}}=\{\begin{array}{ll}23\mu \text{m}\hfill & \left(\text{stratum}\text{spinosum}\right)\hfill \\ 14\mu m\hfill & \left(\text{stratum}\text{granulosum}\right)\hfill \end{array}$
	$s^{\text{lip}}=0.004\mu \text{m}$
	$s^{\text{ext}}=1.0\mu \text{m}$
Cytoplasm	$K^{\text{cyt}/\text{w}}=0.6$
	$D^{\text{cyt}}=D^{\text{aq}}/H^{\text{cyt}},H^{\text{cyt}}=3\left(\text{hindrance}\text{factor}\text{for}\text{cytoplasm}\right)$
Extracellular fluid	$K^{\text{ext}/\text{w}}=0.6$
	$D^{\text{ext}}=D^{\text{aq}}/H^{\text{ext}},H^{\text{ext}}=2\left(\text{hindrance}\text{factor}\text{for}\text{extracellular}\text{fluid}\right)$
Lipid (see the Supporting Material)	$K^{\text{lip}/\text{w}}=1.25K^{\text{oct}/\text{w}}$
	$D^{\text{lip}}=\left(1.58\times {10}^{-3}{\text{cm}}^2/\text{s}\right){\text{MW}}^{-1.69}+8.34\times {10}^{-8}{\text{cm}}^2/\text{s}$
	$k^{\text{trans}}=2P^{\text{DMPC}/\text{w}}/K^{\text{lip}/\text{w}}$
Gap junctions (see the Supporting Material)	$k^{\text{GJ}}=\{\begin{array}{lll}5\times {10}^{-7}\text{cm}/\text{s}\left(\text{lower}\text{bound}\right),\hfill & 5\times {10}^{-4}\text{cm}/\text{s}\left(\text{upper}\text{bound}\right)\hfill & \left(\text{stratum}\text{spinosum}\right)\hfill \\ 5\times {10}^{-8}\text{cm}/\text{s}\left(\text{lower}\text{bound}\right),\hfill & 5\times {10}^{-5}\text{cm}/\text{s}\left(\text{upper}\text{bound}\right)\hfill & \left(\text{stratum}\text{granulosum}\right)\hfill \end{array}$
Tight junctions	$k^{\text{TJ}}=\{\begin{array}{ll}\infty \hfill & \left(\text{no}\text{seal};\text{appropriate}\text{for}\text{stratum}\text{spinosum}\right)\hfill \\ 0\hfill & \left(\text{perfect}\text{seal};\text{maximum}\text{barrier}\text{for}\text{stratum}\text{granulosum}\right)\hfill \end{array}$

MW denotes molecular weight and K^oct/w denotes octanol/water partition coefficient of the solute. Required transport properties are the aqueous diffusivity D^aq of the solute, and its permeability coefficient P^DMPC/w through DMPC bilayers. A full explanation of the formulas is given in the Supporting Material.

For the spinous layer the numerical values of ℓ_side and ℓ_height in Table 1 imply a keratinocyte diameter of ∼23 μm, which seems to be broadly consistent with various microscopic images (29,32,35,40). They are determined as the unique pair of values consistent with both a unit aspect (height/width) ratio and a nominal ∼2500 μm² membrane area. This area estimate is based on the notion of approximate conservation of surface area as keratinocytes terminally differentiate (because they had to build the scaffold for the cornified cell envelope while living), together with knowledge of the final shapes (which are roughly hexagonal corneocytes with a diameter of ∼40 μm and a thickness of 0.8–1 μm). Keratinocytes in the stratum granulosum tend to be partially flattened, as illustrated nicely in Fig. 2 of Tsukita and Furuse (33). The nominal dimensions in Table 1 decrease the aspect ratio to 0.5 for this layer.

The value s^lip = 4 nm agrees with a recent visualization of viable epidermis by cryo-electron microscopy (43), in which “[t]he plasma membranes appear as ∼4 nm thick high-density double-layer patterns.”

The average thickness s^ext = 1 μm of the layers of extracellular fluid separating keratinocytes may ultimately be fixed partly by matching permeability data for hydrophilic permeants (see Discussion). The value assumed here is the order of magnitude that is consistent with apparent thinness relative to the keratinocyte diameter, yet also produces a significant (∼12%) extracellular fluid volume fraction (compare to an anecdotal estimate of 20% (44)).

Governing transport equations

Calculation of the effective diffusivity requires the solution of a well-defined steady-state diffusion problem in the unit cell. Solute transport within the cytoplasm is described by the equation (19,26,27)

$$\nabla ·\left(D^{\mathrm{cyt}}\nabla C^{\mathrm{cyt}}\right)=0$$ (3)

governing the solute concentration C^cyt, in which appears the diffusion coefficient D^cyt (cm²/s) quantifying hindered aqueous diffusion through the cytoplasm. For lack of more detailed information about transport properties of the cytoplasm (see below), we regard the keratinocyte interior as a homogenous effective medium characterized by a suitable average value of D^cyt, following common practice (25–27), which reduces Eq. 3 to the Laplace equation.

Transport in the lipid phase occurs by processes illustrated in Figs. 2 and 3. Given its thinness, we idealize the solute distribution within this phase in terms of effectively two-dimensional distributions of concentration representing averages over the bilayer cross section (C^lip(y,z) for the vertical section of membrane in Fig. 3, and C^lip(x,y) for the horizontal section). This surface description has precedent in some brick-and-mortar models of SC permeability (19), and can be put on a rigorous asymptotic foundation (J. M. Nitsche, unpublished result). Motion parallel to the plane of the bilayer occurs by lateral diffusion, quantified by a diffusivity D^lip. Perpendicular solute transfer to or from the bilayer is described by a mass transfer coefficient k^trans referred to lipid-phase concentrations (19,28). It represents the constant of proportionality between flux past the headgroups and through the barrier (ordered acyl-chain) region (45) to the center of the bilayer, and the lipid-phase concentration difference driving this flux. (The coefficient k^trans as defined here is equivalent to 2k′ and 2k_trans in the notation of Johnson et al. (28) and Wang et al. (19), respectively.) The surface transport equation for the lipid phase necessarily involves the cytoplasm on one side and the extracellular fluid on the other side, because these phases represent the sources from which solute enters the bilayer perpendicularly, as described by the equation (19)

$$-s^{\mathrm{lip}}{D}^{\mathrm{lip}}{\nabla }_{\mathrm{s}}^2{C}^{\mathrm{lip}}=k^{\mathrm{trans}}\left[\left(\frac{C^{\mathrm{cyt}}}{K^{\mathrm{cyt}/\mathrm{lip}}}-C^{\mathrm{lip}}\right)+\left(\frac{C^{\mathrm{ext}}}{K^{\mathrm{ext}/\mathrm{lip}}}-C^{\mathrm{lip}}\right)\right]$$ (4)

(see Fig. 3). Here K^cyt/lip and K^ext/lip, respectively, denote the cytoplasm/lipid and extracellular fluid/lipid partition coefficients. We adopt the convention of defining partition coefficients for each phase relative to a reference aqueous solution w, i.e., K^lip/w, K^cyt/w, and K^ext/w. In terms of them, the two partition coefficients appearing in Eq. 4 are given by the ratios K^cyt/lip = K^cyt/w/K^lip/w and K^ext/lip = K^ext/w/K^lip/w. The symbol ${\nabla }_{\mathrm{s}}^2$ denotes the two-dimensional (surface) Laplacian operator, i.e., ${\nabla }_{yz}^2={\partial }^2/\partial y^2+{\partial }^2/\partial z^2$ for the vertical section of membrane in Fig. 3, and ${\nabla }_{xy}^2={\partial }^2/\partial x^2+{\partial }^2/\partial y^2$ for the horizontal section. The two terms in parentheses on the right-hand side of Eq. 4 represent solute influx to the bilayer from the cytoplasm and the extracellular fluid, indicated with dotted arrows and underlines in the figure.

Figure 2.

Mechanisms of transport along and through interfaces of the structure shown in Fig. 1. (Solid arrows) Lateral diffusion along the lipid cell membrane in both horizontal (from point A to B) and vertical (from B to C) sections of the membrane. (Dotted arrows) Perpendicular transfer between the cytoplasm or extracellular fluid and the lipid membrane. (Outlined open arrow) Direct intercellular diffusion pathway mediated by gap junctions.

Figure 3.

Illustration of Eqs. 4 and 6. (Dotted arrows/underlines) Solute influx to the bilayer from the cytoplasm and the extracellular fluid for an application of Eq. 4 at the vertical boundary of the cytoplasm in the computational domain. (Dotted and outlined white arrows/underlines) Solute flux out of the cytoplasm through the membrane and through gap junctions, respectively, for an application of Eq. 6 at the upper horizontal boundary.

There is no literal belief that the cellular membrane actually possesses crisp edges (sharp corners seen in cross section) in transitioning from horizontal to vertical sections, i.e., the actual membrane turns with a finite radius of curvature. We ensure physical realism by stipulating that the corners in the model do not impede lateral diffusion of the solute, indicated with solid arrows in Fig. 2. Thus, for instance, the flux from point A to B must match that from B to C. The mathematical statement of this requirement is given in the Supporting Material.

Thinness of the layers of extracellular fluid separating apposed sections of cellular membranes implies that they can be treated by the same type of two-dimensional formalism used for the lipid phase, albeit with markedly different properties. The equation analogous to Eq. 4 governing the solute concentration field C^ext is given in the Supporting Material.

Tight junctions produce a zone of kissing points between apposed vertical sections of cellular membranes that interrupts the vertical extracellular route of transport with a more or less tight (impermeable) seal. This zone is located near the apical (upper) extreme of the lateral membrane (33,36). We do not attempt to describe literally the associated undulations in the membrane contour. The overall effect is represented rather by an effective barrier (depicted as a cross-hatched plug in Fig. 2) located at an elevation z^TJ and characterized by a mass transfer coefficient k^TJ dependent on the effective permeability of the microscopic structure. This coefficient appears in the interfacial condition

$$D^{\mathrm{ext}}\frac{\partial C^{\mathrm{ext}}}{\partial z}\left(y,z_{+}^{\mathrm{TJ}}\right)=D^{\mathrm{ext}}\frac{\partial C^{\mathrm{ext}}}{\partial z}\left(y,z_{-}^{\mathrm{TJ}}\right)=k^{\mathrm{TJ}}\left[C^{\mathrm{ext}}\left(y,z_{+}^{\mathrm{TJ}}\right)-C^{\mathrm{ext}}\left(y,z_{-}^{\mathrm{TJ}}\right)\right],$$ (5)

in which the symbols $z_{+}^{\mathrm{TJ}}$ and $z_{-}^{\mathrm{TJ}}$ are used to denote limiting values as one approaches the barrier from above and below.

Aside from transbilayer transport, an additional mechanism exists for intercellular solute transfer, namely the direct route provided by epidermal gap junctions (26,27,29–32); see outlined open arrows in Figs. 2 and 3. The level of detail of our analysis does not support a literal accounting for variations in the thickness of the extracellular fluid layers resulting from the touching of cellular membranes at junctions or in junctional plaques. Thus, our assumed layer thickness s^ext represents an average value. The overall expression of junctional permeability is a surface-average mass transfer coefficient k^GJ.

Both k^trans and k^GJ appear in the boundary condition on C^cyt quantifying normal solute flux at the cellular membrane, namely

$$-D^{\mathrm{cyt}}\mathbf{n}·\nabla C^{\mathrm{cyt}}=k^{\mathrm{trans}}\left(\frac{C^{\mathrm{cyt}}}{K^{\mathrm{cyt}/\mathrm{lip}}}-C^{\mathrm{lip}}\right)+k^{\mathrm{GJ}}\left(C^{\mathrm{cyt}}-C^{\mathrm{cyt,a}}\right),$$ (6)

in which C^cyt,a denotes the local cytoplasmic concentration of solute in the neighboring keratinocyte. The outward normal derivative n · ∇ is equivalent to ∂/∂x on the vertical boundary of the cytoplasmic domain in Fig. 3, ∂/∂z on the upper horizontal boundary, and −∂/∂z on the lower horizontal boundary (not included in the figure).

Effective transport properties

For convenience, we introduce the symbol K^/w(x,y,z) to denote the piecewise constant, periodic function of position equal to K^α/w for positions (x, y, z) lying in phase α (α = cyt, lip, or ext) (19). The macroscopically observable effective partition coefficient is simply its unit-cell (volume) average (18,19),

$${\overline{K}}^{\text{epi}/\text{w}}=\langle K^{/\text{w}}\rangle =\frac{V^{\text{cyt}}{K}^{\text{cyt}/\text{w}}+V^{\text{lip}}{K}^{\text{lip}/\text{w}}+V^{\text{ext}}{K}^{\text{ext}/\text{w}}}{V^{\text{cyt}}+V^{\text{lip}}+V^{\text{ext}}},$$ (7)

in which V^cyt, V^lip, and V^ext denote the volumes of the cytoplasmic, lipid membrane, and extracellular fluid phases in one unit cell.

We utilize a classical steady-state formalism (19,46) to calculate the effective diffusivity ${\overline{D}}^{\text{epi}}$. As discussed in the Supporting Material, this method is equivalent to homogenization methods that have been applied in microscopic diffusion models of the SC (20,47,48). The overall idea is to calculate the transdermally (−z-) directed flux of solute resulting from the imposition of a z-directed gradient in its concentration. This procedure yields ${\overline{D}}^{\text{epi}}$ as the constant of proportionality between average flux and concentration gradient. The diffusion problem embodied in Eqs. 3–6 (together with additional equations in the Supporting Material) is posed subject to the periodicity-type condition (19,46)

$$C\left(x,y,z+{\ell }_{\text{period}}\right)=C\left(x,y,z\right)+C_0{K}^{/\text{w}}\left(x,y,z\right).$$ (8)

Here C(x, y, z) is a collective symbol denoting C^α(x, y, z) for positions (x, y, z) lying in phase α, and C₀ is an arbitrary constant with dimensions of concentration (which can be set to unity without loss of generality). Equation 8 corresponds to a macroscopically linear variation in volume-average concentration possessing a gradient (19,46)

$$\overline{\partial C/\partial z}=\frac{C_0{\overline{K}}^{\text{epi}/\text{w}}}{{\ell }^{\text{period}}}.$$ (9)

The average transdermally (−z-directed) solute flow through any horizontal cross section of the unit cell must be the same by mass conservation. Dividing the flow through the plane z = 0, say, by the cross-sectional area $\left|{\mathcal{D}}_{\text{xy}}\right|$ gives this macroscopically observable solute flux:

$${\overline{J}}_{-z}={\left|{\mathcal{D}}_{xy}\right|}^{-1}[{\iint }_{{\mathcal{D}}_{xy}^{\text{cyt}}}{D}^{\text{cyt}}\frac{\partial C^{\text{cyt}}}{\partial z}\left(x,y,0\right)dxdy+s^{\text{lip}}{\int }_0^{{\ell }_{\text{side}}/2}{D}^{\text{lip}}\frac{\partial C^{\text{lip}}}{\partial z}\left(y,0\right)dy+\frac{s^{\text{ext}}}{2}{\int }_0^{{\ell }_{\text{side}}/2}{D}^{\text{ext}}\frac{\partial C^{\text{ext}}}{\partial z}\left(y,0\right)dy].$$ (10)

As written, this formula is formulated in terms of the reduced computational domain $\mathcal{D}$_xy shown in Fig. 1 B. The effective diffusivity is calculable from the results of Eqs. 9 and 10 as the ratio

$${\overline{D}}^{\text{epi}}=\frac{{\overline{J}}_{-z}}{\partial C/\partial z}.$$ (11)

Model Parameters

Innumerable microscopic structural details of the epidermis are known and have been presented in exquisite detail (42) (see also the various microscopy studies cited above (29,30,32,35,36)). In contrast, we are unaware of any reported data on partition and diffusion coefficients of individual microscopic phases of viable human epidermis for any solute. All of the quantities K^lip/w, D^lip, k^trans, D^cyt, K^cyt/w, D^ext, K^ext/w, k^GJ, and k^TJ are measureable in principle (in many cases with considerable difficulty), either directly or indirectly, but are presently unknown. Such detailed, microscopic, phase-specific information is distinct from:

• 1.Effective solute partition and diffusion coefficients for viable epidermis, and
• 2.Overall dermal absorption rates.

Item 1 represents an average over the microscopic quantities, which is precisely what we undertake to calculate theoretically in this article. Even such data are extremely scarce. Nitsche and Kasting (8) review some available information, which is anecdotal at best. It is this type of information with which results of our model should ultimately be compared. Item 2 represents a further convolution of epidermal properties within the overall outcome of diffusion through the epidermis in series with the SC and dermis.

In the face of this dearth of information, we formulate in Table 1 reasonable provisional estimates based on contemporary knowledge of hindered aqueous mobility in cytoplasm and extracellular fluid (25–27), and mobility properties of phospholipid bilayer membranes (49), as is explained in detail in the Supporting Material. The required inputs for any given solute are its molecular weight (MW), octanol/water partition coefficient (K^oct/w), aqueous diffusivity at 37°C (D^aq), and permeability coefficient for passive diffusion through pure dimyristoylphosphatidylcholine (DMPC) bilayers at 37°C (P^DMPC/w). Considerable effort went into the development of general equations to estimate K^lip/w, D^lip, and k^trans, based on a comprehensive assessment of the literature on partition and permeability coefficients as well as fatty-acid chain density of phospholipid bilayers (49). Use of DMPC as the basis for modeling keratinocyte membranes is justified in detail in the Supporting Material.

We ultimately judge the estimates in Table 1 by comparison of model results with approximate expectations for the average properties of epidermal tissue (8). Part of the value of a detailed microscopic model is that it lays bare all the areas where measurements are needed. It can also help to guide and prioritize experiments by revealing which parameters seem to be the most important determinants of solute flux or epidermal concentration levels based on preliminary estimates.

The level of ambiguity inherent in some parameters is so great that they are best quantified in terms of upper and lower bounds. The data and logical arguments leading to the bounds on the gap junctional permeability coefficient k^GJ listed in Table 1 are presented in detail in the Supporting Material. Available knowledge on tight junctions (33–36) simply does not support any quantitative microscopic pore model or specific numerical values of the permeability coefficient k^TJ. Therefore, we adopt the strategy of restricting numerical calculations to the two extremes of no seal (k^TJ → ∞) or a perfect seal (k^TJ → 0) between adjacent keratinocytes. Tight junctions occur primarily in the granular layer (33,36). The perfect-seal limit provides an upper bound on the barrier properties of this layer. The no-seal limit offers a representation of the spinous layer.

Calculations are carried out for three illustrative permeants, namely water (regarded as a tracer species), L-glucose, and hydrocortisone. (L-Glucose is a largish hydrophilic molecule behaving as a nonmetabolized, passively transported solute. As such, it is on the same footing as the other two permeants. It characterizes the intrinsic mobility properties of the physiologically relevant D stereoisomer with all biochemical mechanisms for metabolism and active transport effectively switched off.) Numerical values of all model parameters for these three compounds, resulting from the formulas in Table 1, are listed in Table 2.

Table 2.

Physical input properties (MW, log₁₀K^oct/w, D^aq, P^DMPC/w) and estimated model parameters for water, L-glucose, and hydrocortisone

Property	Water	L-glucose	Hydrocortisone
MW	18.015	180.16	362.46
log10Koct/w	−1.38	−3.24	1.61
Daq	3.3 × 10−5 cm2/s	1.0 × 10−5 cm2/s	5.8 × 10−6 cm2/s
PDMPC/w	3.7 × 10−3 cm/s	4.0 × 10−11 cm/s	5.0 × 10−4 cm/s
Kcyt/w	0.60	0.60	0.60
Dcyt	1.10 × 10−5 cm2/s	3.33 × 10−6 cm2/s	1.93 × 10−6 cm2/s
Kext/w	0.60	0.60	0.60
Dext	1.65 × 10−5 cm2/s	5.00 × 10−6 cm2/s	2.90 × 10−6 cm2/s
Klip/w	5.21 × 10−2	7.19 × 10−4	5.09 × 101
Dlip	1.20 × 10−5 cm2/s	3.27 × 10−7 cm2/s	1.58 × 10−7 cm2/s
ktrans	1.42 × 10−1 cm/s	1.11 × 10−7 cm2/s	1.96 × 10−5 cm/s
kGJ	See Table 1
kTJ	See Table 1
Kcyt/wDcyt/(20 μm)	3.30 × 10−3 cm/s	1.00 × 10−3 cm/s	5.80 × 10−4 cm/s
Klip/wktrans/4	1.85 × 10−3 cm/s	2.00 × 10−11 cm/s	2.50 × 10−4 cm/s
Kext/wDext/(1 μm)	9.90 × 10−2 cm/s	3.00 × 10−2 cm/s	1.74 × 10−2 cm/s

Results

Results stated here have the significance of rigorously calculated outcomes of the model predicated on the assumed microscopic geometry and physicochemical parameter values.

Effective partition coefficient

Table 3 presents a breakdown of ${\overline{K}}^{\text{epi}/\text{w}}$ into contributions from occupancy of the cytoplasmic (cyt), membrane (lip), and extracellular fluid (ext) phases. The results are understandable from the structure, which comprises 87.9% cytoplasm, 0.1% lipid, and 12.0% extracellular fluid by volume for stratum spinosum. (The fractions differ little from these values for stratum granulosum.) On top of the small volume fraction of lipid, unfavorable partitioning makes lipid-phase holdup truly negligible for hydrophilic solutes like water and L-glucose. In contrast, high lipophilicity can compensate partly for the small lipid volume fraction, to the extent that ∼8% of hydrocortisone partitions into the lipid phase. It is fair to say that ${\overline{K}}^{\text{epi}/\text{w}}$ will usually be a number of the order of unity, excepting highly lipophilic species.

Table 3.

Tissue-average (effective) partition coefficient ${\overline{K}}^{\text{epi}/\text{w}}$ and phase-specific breakdown of solute holdup

Stratum	Solute	${\overline{K}}^{\text{epi}/\text{w}}$	% of Holdup
			Cyt	Lip	Ext
Spinous	Water	0.599	88.0	8 × 10−3	12.0
	L-glucose	0.599	88.0	1 × 10−4	12.0
	Hydrocortisone	0.648	81.3	7.6	11.1
Granular	Water	0.599	87.2	9 × 10−3	12.8
	L-glucose	0.599	87.2	1 × 10−4	12.8
	Hydrocortisone	0.652	80.2	8.0	11.8

The abbreviations cyt, lip, and ext denote cytoplasm, lipid membrane, and extracellular fluid, respectively.

Effective diffusivity

Calculated results for ${\overline{D}}^{\text{epi}}$ are presented in Tables 4–6. The first two are geared toward comparisons of ${\overline{D}}^{\text{epi}}$ among the three representative solutes considered, and among the four possible cases considered (involving choices of upper- or lower-bound values of k^GJ and k^TJ). The last offers a more detailed mechanistic breakdown of ${\overline{D}}^{\text{epi}}$ for the spinous layer, with k^TJ → ∞, for both upper- and lower-bound values of k^GJ.

Table 4.

Values of the effective (tissue-average) diffusion coefficient ${\overline{D}}^{\text{epi}}$ calculated for various scenarios with different values of k^GJ and k^TJ, expressed in absolute terms in units of cm²/s

Stratum	Solute	${\overline{D}}^{\text{epi}}$ in cm2/s for case kGJ high, kTJ → ∞	${\overline{D}}^{\text{epi}}$ in cm2/s for case kGJ high, kTJ = 0	${\overline{D}}^{\text{epi}}$ in cm2/s for case kGJ low, kTJ → ∞	${\overline{D}}^{\text{epi}}$ in cm2/s for case kGJ low, kTJ = 0
Spinous	Water	6.8 × 10−6	6.3 × 10−6	6.5 × 10−6	6.0 × 10−6
	L-glucose	1.3 × 10−6	8.2 × 10−7	4.5 × 10−7	1.1 × 10−9
	Hydrocortisone	1.2 × 10−6	1.1 × 10−6	9.6 × 10−7	8.7 × 10−7
Granular	Water	5.0 × 10−6	4.3 × 10−6	5.0 × 10−6	4.3 × 10−6
	L-glucose	4.5 × 10−7	6.9 × 10−8	3.8 × 10−7	7.0 × 10−11
	Hydrocortisone	7.5 × 10−7	6.3 × 10−7	7.2 × 10−7	6.0 × 10−7

The identifiers k^GJ high and k^GJ low, respectively, represent the upper- and lower-bound values of k^GJ listed in Table 1 for each stratum. The cases k^GJ high, k^TJ → ∞ and k^GJ low, k^TJ → ∞ for the spinous layer are singled out in boldface for more detailed consideration in Table 6. The cases with k^TJ = 0 for the spinous layer are hypothetical (i.e., not physically representative) because tight junctions occur primarily in the granular layer (33,36).

Table 5.

Values of the effective (tissue-average) diffusion coefficient ${\overline{D}}^{\text{epi}}$ calculated for various scenarios with different values of k^GJ and k^TJ, expressed relative to the aqueous diffusivity, i.e., given as the ratio ${\overline{D}}^{\text{epi}}/D^{\text{aq}}$

Stratum	Solute	${\overline{D}}^{\text{epi}}/D^{\text{aq}}$ for case kGJ high, kTJ → ∞	${\overline{D}}^{\text{epi}}/D^{\text{aq}}$ for case kGJ high, kTJ = 0	${\overline{D}}^{\text{epi}}/D^{\text{aq}}$ for case kGJ low, kTJ → ∞	${\overline{D}}^{\text{epi}}/D^{\text{aq}}$ for case kGJ low, kTJ = 0
Spinous	Water	0.20	0.19	0.20	0.18
	L-glucose	0.13	0.082	0.045	1.1 × 10−4
	Hydrocortisone	0.21	0.19	0.17	0.15
Granular	Water	0.15	0.13	0.15	0.13
	L-glucose	0.045	0.0069	0.038	7.0 × 10−6
	Hydrocortisone	0.13	0.11	0.12	0.10

The identifiers k^GJ high and k^GJ low, respectively, represent the upper- and lower-bound values of k^GJ listed in Table 1 for each stratum. The cases k^GJ high, k^TJ → ∞ and k^GJ low, k^TJ → ∞ for the spinous layer are singled out in boldface for more detailed consideration in Table 6. The cases with k^TJ = 0 for the spinous layer are hypothetical (i.e., not physically representative) because tight junctions occur primarily in the granular layer (33,36).

Table 6.

Mechanistic breakdown of ${\overline{D}}^{\text{epi}}$ for the spinous layer, with k^TJ → ∞, for upper- and lower-bound values of k^GJ (5 × 10⁻⁴ cm/s and 5 × 10⁻⁷ cm/s, respectively)

Solute	kGJ, cm/s	% of total solute flow through cross section halfway up cell (z = 0)			% of solute flow into top of keratinocyte (z = ℓheight/2)
		Cyt	Lip	Ext	from ktrans	from kGJ
Water	5 × 10−4 (high)	85	0.007	15	85	15
	5 × 10−7 (low)	85	0.007	15	99.98	0.02
L-glucose	5 × 10−4 (high)	65	≈0	35	≈0	100
	5 × 10−7 (low)	0.2	≈0	99.8	0.004	99.996
Hydrocortisone	5 × 10−4 (high)	84.6	0.5	14.9	43	57
	5 × 10−7 (low)	82.5	0.5	17	99.9	0.1

Diffusion through the assumed periodic structure is an inherently three-dimensional process not amenable to representation by any simple analytical formulas (which is precisely the reason necessitating a detailed numerical model). Nevertheless, for physical understanding it is worthwhile to introduce a very approximate decomposition in terms of two parallel pathways considered additively:

• 1.A hypothetical transcellular pathway that alternates between ∼20-μm-thick cytoplasmic layers, ∼1-μm-thick layers of extracellular fluid, and cellular membranes; and
• 2.A pathway that involves diffusion vertically through thin-walled hexagonal cylinders of extracellular fluid, which cover ∼9% of the area fraction in a typical cross section.

We shall refer to these hypothetical pathways as the transcellular and extracellular pathways, respectively, with the understanding that they represent approximate concepts. (An analogous lipid pathway for diffusion along the cell membranes is not worth introducing because of its truly negligible contribution to the total solute flow—see below.) The combinations of parameters included at the bottom of Table 2 represent mass transfer coefficients (operative in series, and referred to concentrations in aqueous solution w) that would characterize the cytoplasmic, extracellular, and lipid phases of the transcellular pathway. They are useful for subsequent discussion of ${\overline{D}}^{\text{epi}}$.

One of the two breakdowns presented in Table 6 is the distribution among the contributions to vertical diffusive flow of solute through the horizontal cross section at z = 0 from the three phases (cyt, lip, and ext), corresponding to the three respective terms in Eq. 10. For hydrocortisone it shows that the majority of solute flow occurs in the aqueous (cyt, ext) compartments. The lipid phase is so thin that diffusion in the plane of the membrane accounts for only 0.5% of the total flow, despite significant occupancy of this phase (∼8%). For hydrophilic compounds like water and L-glucose, the lipid contribution is even smaller. Thus, the lateral diffusivity D^lip is essentially irrelevant to transepidermal transport. The other breakdown in Table 6 applies only to transport into the cytoplasm at the top of the cell (z = ℓ_height/2). Solute from the cytoplasm above can enter directly via the gap junctional pathway, controlled by k^GJ. It can also follow a transmembrane pathway passing through a phospholipid barrier region four times (passing through two cell membranes) controlled by k^trans.

Lipophilic compounds tend to be adept at passing through lipid membranes, owing primarily to favorable partitioning (which elevates in-membrane concentration levels, and thereby, flux). The intrinsic membrane permeability of the moderately lipophilic compound hydrocortisone is already sufficiently high for:

• 1.Even the upper-bound value of k^GJ to provide little more than half of the solute flow into the top of the cytoplasm; and
• 2.The two cell membranes (i.e., four phospholipid barrier regions) in one unit cell to present only about twice as much mass transfer resistance as diffusion through the cytoplasm (specifically, (K^lip/wk^trans/4)⁻¹ = 2.3 [K^cyt/wD^cyt/(20 μm)]⁻¹ based on the numbers in Table 2).

For more highly lipophilic compounds, the cytoplasm would, in fact, present the dominant mass transfer resistance in the transcellular pathway. It follows that ${\overline{D}}^{\text{epi}}$ for highly lipophilic compounds reflects mostly D^cyt (equivalent to D^aq modified by a hindrance factor here called H^cyt), a conclusion that accords with the concept enunciated long ago (50) that tissue below the stratum corneum presents a primarily aqueous barrier to dermal penetration for such species. The extracellular pathway is a minor contributor to ${\overline{D}}^{\text{epi}}$ because it operates over a minority (∼9%) of the tissue area. For this reason, even a complete choking off of this pathway (k^TJ = 0) has a minor effect (Tables 4 and 5).

In contrast, a large hydrophilic compound like L-glucose has such a low intrinsic membrane permeability that gap junctions represent the dominant means of intercellular transfer. Membrane permeation by this junctional mechanism is the rate-limiting step in the transcellular pathway, because the mass transfer resistance for this step (3.3 × 10³ s/cm < (K^cyt/wk^GJ)⁻¹ < 3.3 × 10⁶ s/cm) is larger than that for diffusion through the cytoplasm ([K^cyt/wD^cyt/(20 μm)]⁻¹ = 1.0 × 10³ s/cm). Thus, the transcellular pathway is controlled by k^GJ. At the upper-bound value of k^GJ, it provides 65% of the L-glucose flow, whereas at the lower-bound value, it is effectively shut down. In the latter case, the extracellular pathway assumes responsibility for essentially all (99.8%) of the L-glucose flow through the spinous layer, ${\overline{D}}^{\text{epi}}$ coming out to 4.5 × 10⁻⁷ cm²/s (Table 4). (This value is, in fact, predictable a priori. Indeed, the operative diffusivity is D^ext, and diffusion occurs over ∼9% of the area of a typical cross section. The numerical value 0.09 × D^ext exactly matches ${\overline{D}}^{\text{epi}}$.) If the extracellular pathway is now choked off by a hypothetically perfect tight junctional seal (k^TJ = 0), then the effective diffusivity drops to the meager level (1.1 × 10⁻⁹ cm²/s) afforded by the low k^GJ.

Discussion

The average epidermal partition coefficient offers no real surprises because it is primarily a reflection of bookkeeping on the volume fraction and hydro/lipophilicity of each constituent phase (Eq. 7). The dynamics of diffusion is considerably more complex (see Eqs. 3–6 and 8–11, and additional equations in the Supporting Material). Broadly speaking, the most significant finding is that calculated values of ${\overline{D}}^{\text{epi}}$ for water and hydrocortisone are typically 10–20% of D^aq:

$${\overline{D}}^{\text{epi}}=\frac{D^{\text{aq}}}{H^{\text{epi}}},H^{\text{epi}}\approx 5\text{to}10.$$ (12)

Interestingly, this provisional theoretical conclusion accords with the limited anecdotal evidence that is available (see, e.g., the brief review given in Nitsche and Kasting (8)). Thus, for instance, Bunge and Cleek (51) suggested a hindrance factor of about an order of magnitude relative to bulk aqueous diffusion based on measurements by Tojo and Lee (52). We did not adjust any model parameters with the aim of arriving at what seems to be the correct order of magnitude for ${\overline{D}}^{\text{epi}}$. Rather, we made the best possible a priori estimates of all unknown parameters (provisional values in Table 1), and then observed the logical outcome from a rigorous analysis of the microscopic multiphase transport equations.

Mechanistic conclusions

It is worthwhile to highlight the dominant transport mechanisms underlying the quantitative outcomes for ${\overline{D}}^{\text{epi}}$, and the priorities they suggest for experimental clarification. The theoretical considerations presented above regarding lipophilic permeants suggest the approximate relation

$${\overline{D}}^{\text{epi}}\approx D^{\text{cyt}}=\frac{D^{\text{aq}}}{H^{\text{cyt}}}\left(\text{highly}\text{lipophilic}\text{permeants}\right),$$ (13)

which speaks for further understanding of cytoplasmic mobility, given its role as the primary determinant of average epidermal mobility. The extracellular pathway is a minor player, which makes the whole issue of tight junctional barrier properties a moot point. High transmembrane permeability tends to make also k^GJ a largely irrelevant factor. For highly lipophilic permeants (i.e., those with log K^oct/w ≥ 3), however, the role of binding to diffusible proteins both intracellularly and extracellularly cannot be ignored (12). Thus, a further theoretical development accounting for binding would be needed to actually use this model for such permeants.

For large hydrophilic, passively-transported permeants like L-glucose (which are very membrane impermeable), we conclude that

$$\begin{array}{r}{\overline{D}}^{\text{epi}}\approx \{\begin{array}{cc}\left(20\mu \text{m}\right)k^{\text{GJ}}=1\times {10}^{-6}{\text{cm}}^2/\text{s}& \left(k^{\text{GJ}}\text{high}\right)\\ \left(0.09\right)D^{\text{ext}}=\left(0.09\right)D^{\text{aq}}/H^{\text{ext}}& \left(k^{\text{GJ}}\text{low}\right)\end{array}\\ \left(\text{large}\text{hydrophilic}\text{permeants}\right).\end{array}$$ (14)

At the upper bound of gap junctional permeability, the value of k^GJ directly controls ${\overline{D}}^{\text{epi}}$ as the rate-limiting element of the transcellular pathway. At the lower bound of k^GJ, the primary pathway seems to be extracellular in the absence of a significant tight junctional barrier (spinous and basal layers). This finding underscores the importance of:

• 1.Quantifying solute mobility in extracellular fluid;
• 2.Carefully calibrating the volume fraction of extracellular fluid, which underlies the numerical coefficient 0.09; and
• 3.Characterizing tight junctions, which more or less completely choke off the extracellular pathway in the granular layer.

Equation 14 is consistent with the results of Khalil et al. (22), who present three determinations of ${\overline{D}}^{\text{epi}}$ for glucose, namely (3.7 ± 1.9) × 10⁻⁸ cm²/s, (7.5 ± 5.0) × 10⁻⁸ cm²/s, and (1.0 ± 0.6) × 10⁻⁶ cm²/s. They ultimately conclude that “the true value [is] most likely toward the upper end of this range,” because the lower values likely reflect the effects of residual stratum corneum (SC) remaining on tape-stripped epidermis.

Small hydrophilic permeants like water and moderately lipophilic permeants like hydrocortisone lie at the intersection of all the phenomena discussed. Both transmembrane and cytoplasmic resistance are significant in the transcellular pathway, and the extracellular route also makes a discernible contribution to solute flux. Moreover, transbilayer and gap junctional diffusion can both contribute significantly to transmembrane transport. Thus, everything is important, and one must resort to the full computational model to calculate ${\overline{D}}^{\text{epi}}$. A model is only as reliable as the parameters entering it, and experiments targeting microscopic phase-specific properties of the epidermis are ultimately the only means of determining them definitively, much more so than measurements of overall (average) epidermal permeabilities (which reflect these parameters only in an aggregate form). Although Table 1 is consistent with available anecdotal information on average properties (8,51), it represents a provisional statement subject to revision.

Use of the model

The ultimate outcome of our analysis is a computer code that estimates ${\overline{K}}^{\text{epi}/\text{w}}$ and ${\overline{D}}^{\text{epi}}$ for the mobile fraction of any solute based on its values of MW, K^oct/w, D^aq, and P^DMPC/w. The results can be slotted directly into the viable epidermis layer of rigorous transient diffusion models of dermal absorption (8–10). Thus, the two-scale program can now be supported with a microscopic basis for predicting mobile solute concentrations at specific sites within the epidermis (e.g., in the extracellular fluid bathing the plasma membrane of Langerhans cells in contact allergy applications). Because means of estimating P^DMPC/w are available (49), our model can be applied generally, and not only to solutes for which this permeability coefficient has been measured.

We distinguish spinous and granular layers in terms of aspect ratio and occurrence of tight junctions, although a macroscopic model might employ a single average layer for simplicity. The basal (mono)layer differs fundamentally from the spinous layer above it in terms of metabolism and cell division, among a host of biological processes not occurring elsewhere. However, at this level of description, differences in passive permeability properties would owe primarily to the higher aspect ratio of the cells of the former (42). In the limit k^TJ → ∞, which is representative of these layers, the calculated ${\overline{D}}^{\text{epi}}$ is quantitatively affected by, but not overly sensitive to, keratinocyte aspect ratio (see Tables 4 and 5). Thus, little error in predictions of passive permeability would likely result from subsuming the basal layer into a thickened spinous layer.

Scope of the model, and future directions

We consciously limited the scope of this investigation to defining the basic passive transport properties of viable epidermal tissue for the mobile fraction of permeating solutes. As it stands, the theory applies only to small solutes that are not excessively hydrophobic and therefore not overly susceptible to protein binding. The goal was to establish the level of mobility attributable to passive diffusion through the cellular structure. This is the baseline upon which binding (and also metabolic) phenomena are superposed.

The next logical step is the addition of solute binding to elements of the microstructure, which include cytoplasmic structures as well as albumin and other binding/transporter proteins in both cytoplasm and extracellular fluid. Although there is hope of developing general correlations for nonspecific binding properties along the lines of a recent theoretical model of dermis (11), in many cases the treatment of binding (and particularly metabolism) will likely be highly permeant-specific.