Steven Abbott, Steven Abbott TCNF Ltd, 7 Elsmere Road, Ipswich, IP1 3SZ UK. Tel.: 01473 232813; e-mail: firstname.lastname@example.org
The academic literature on skin delivery provides countless examples of scientific insights into specific aspects of the overall process. For the practical formulator, however, it is difficult to know how to combine such insights in a way that fits into the realities of commercial formulations. In this study, five key principles are combined into an integrated approach that can be applied to real-world formulations. Given the complexities of skin science, the integrated approach cannot be expected to be highly precise. Instead, it is intended as a way for a formulation team to balance the many conflicting issues. The predictions are sufficiently specific to be examined by those with the appropriate analytical resources and data on formulation efficacy. It is hoped that such challenges will allow the approach to be refined for the future.
There are many ways to get an active into or through the skin . This study is concerned only with the approach whereby a (somewhat) soluble active is expected to go into the skin with the help of ‘solvents’ or, as the industry prefers to call them, excipients, emollients, vehicles or plain ‘ingredients’. In many cases, the active/solvent system is applied onto the skin with the aid of vehicles that are dynamic in nature. Ethanol, for example, evaporates so rapidly, and its effects are so transitory that its presence in the initial formulation is of no further concern for this study. Similarly, an oil-in-water emulsion rapidly becomes an oil-on-skin formulation, and the effect of the water, which has mostly evaporated after a few minutes, is relatively small, especially on skin that is already well hydrated. That these emulsions contain surfactants is a complication outside the scope of this study. Although it would be highly desirable to take into account the effects of ethanol, water and surfactants, that is a challenge for the future if and when it is shown that neglecting these effects poses a serious problem. However, decades of work on ‘ome’ systems (liposomes  and their many variants) containing levels of surfactants higher than in normal emulsion formulations have not indicated that the effects of lower levels of ‘safe’ surfactants are particularly dramatic in terms of dermal delivery.
Given the scope outlined in the previous paragraph, an integrated approach to skin delivery should take into account at least five principles, which should be uncontroversial. Although there may be additional principles to be included in future approaches, any model that does not take into account these principles must surely be of more limited value.
1 Any active has a maximum ‘ideal solubility’ (the solubility in a perfectly compatible solvent) that cannot be exceeded by any solvent combination.
2 Actives and solvents will partition into various parts of the system according to activity coefficients that can, in principle, be calculated a priori.
3 Diffusion through skin can be modelled on the basis of concentration gradients and on diffusion coefficients that depend on molecular shape/size and also on the concentration of (swelling) solvents at each point in the skin.
4 For many reasons (regulatory, cost, formulation constraints), practical formulations will contain multiple ingredients, each of which will have an effect on the overall behaviour of the system.
5 The formulation will be delivered as a finite dose, rather than the infinite dose regime commonly used in skin permeation experiments. There are some exceptions to this principle, for example transdermal patches that act as occlusive sources of a liquid formulation, but the majority of cosmetics and many pharmaceuticals are delivered by a non-occluded formulation.
Each of these principles will now be discussed in turn.
An active will, by definition, be insoluble in a bad solvent. Thermodynamically, a bad solvent is one with a high activity coefficient. A perfect solvent is one with an activity coefficient of 1, which means that neither solvent nor active notice the presence of the other.
There are cases of super-perfection where the solvent can positively drag the active inside, to give an activity coefficient below 1. An obvious example is that an amine solvent can react with an acidic active to give a soluble salt. For example, adding ethanolamine to salicylic acid simply shifts the problem from the solubility of salicylic acid to that of ethanolammonium salicylate. Instead of the solubility of the active, the question is about the solubility of the salt. Similarly, adding a cyclodextrin shifts the problem to the solubility of a cyclodextrin complex. For simplicity, therefore, the definition of a perfect solvent will remain as one with an activity coefficient of 1.
What, then, controls the ideal solubility? Thermodynamically, it is the energy required to melt the crystal into a virtual liquid at room temperature. This energy depends on the enthalpy of fusion, various specific heats and the melting point of the active. The papers by Yalkowsky and Wu  give a full account of these factors along with the conclusion that unless full data are available the ideal solubility can be estimated directly from the melting point via the formula Log10 (Solubility) = −0.01(MPt−25), although Yalkowsky uses ‘crystal liquid fugacity ratio’ rather than ‘solubility’ in his published formula. Put simply, the higher the melting point, the lower the solubility.
Armed with this estimate, the formulator can quickly find out whether any solvent system will ever be able to dissolve enough of the active to solubilize at least the minimum effective dose. If the answer is ‘no’, then there is no point in trying to deliver the active via a solubility route. If the ideal solubility is sufficiently high, then the project can move on to the next stage where finding an actual solvent blend for the active is merely one of the issues.
Solubility and partition
Assume, for a moment, that the skin is a polymer. There is a well-known technique, used over many years, for calculating the solubilities of the components of a mixture and how they might partition into the polymer or skin. This technique uses the 3D Hansen solubility parameters  (HSP) to characterize each component of the formulation as well as the polymer/skin. The parameters are as follows: δD that is the general ‘dispersion’ or Van der Waals component of a molecule, δP that is the ‘polar’ component and δH that is the ‘hydrogen-bonding’ component. These three parameters provide a rich, numerical way to describe molecules. They are more informative than loose notions such as ‘polar’. Both acetonitrile and methanol are ‘polar’ but they behave very differently. Acetonitrile has a high δP value (it is polar in the sense of having a high dipole moment) and methanol has a high δH value and is polar in the sense of possessing strong hydrogen-bonding abilities. Similarly, ‘lipophilic’ molecules such as hexane and toluene are differentiated because toluene has a higher δD value, which reflects the greater polarizability of its aromatic electrons.
Because ‘like dissolves like’, it is important to have a measure of ‘like’. HSP provide an objective ‘distance’ measure for this. Given two chemicals with HSP [δD1, δP1, δH1] and [δD2, δP2, δH2], the HSP distance is given by
The smaller the distance, the more alike the chemicals are so the higher their mutual solubility. Partition coefficients can be calculated from the ratio of distances (Hansen ).
Finally, the HSP values of a mixture of solvents are simply the volume-weighted average of the HSP of the individual solvents. One important aspect of this simple fact is that two non-solvents can, if the averaging works out correctly, produce a mixture that is a good solvent. This is of great importance for skin delivery where there are many individual poor solvents and few good ones – the formulator can create good ones via these mixing rules.
Continuing with the assumption that the skin can be considered a polymer with a set of HSP values, and knowing the HSP of the active, it is possible to manipulate the blend of solvents to gain a good match (small distance between …) for the active, a good match for the skin or somewhere in between, according to choice. A good solvent for the skin will readily enter the skin, swell it and (as discussed below) increase the diffusion of the active. One hallmark of a good solvent (providing it is small) is that it quickly disappears from the surface because of its solubility and the increased diffusion coefficient that comes from this. A good solvent for the active will allow a high concentration in the vehicle if that is required for efficacy. A good balance ensures that the active is encouraged to enter the skin via the extra solubility provided by the solvents and helped to diffuse by that higher concentration.
The word ‘balance’ is a reminder that partitioning is a balance of solubility in two phases. Although there have been attempts to use low-solubility formulations to provide a thermodynamic driving force into the skin, having insoluble actives on the surface is often not a pleasant experience for the user. So, it is usual to require at least sufficient solubility for the active to be fully in solution, whilst not wanting the solubility to be too high in case the active prefers to remain outside the skin. Such worries tend to concern those who focus on infinite doses. If the solvent blend is not too far in HSP terms from the skin, then it will tend to enter the skin and generate a welcoming environment for the active, and as it enters the skin, the active has less solvent on the outside so will be encouraged to enter. Numerical ways to think through these issues can be a great help to the formulation team and HSP, with its mixing rules (even into the skin) provide a way to do these calculations.
These simple principles are powerful, rational tools for optimizing penetration. But they depend on the key assumption that the skin acts somehow like a polymer with a defined HSP value. To cosmetic scientists, this sounds overly simple. Some believe that skin is a ‘bricks and mortar’ construction with impermeable corneocyte bricks and a permeation pathway via the lipid. Others de-emphasize the lipid route and say that chemicals can indeed permeate via the bulk of the corneocytes. An overview of this debate can be found in the work of Kasting’s group . It is often also stated that skin is a formidable barrier with unusual properties that require special explanations.
To an outsider, this is bizarre. Skin (considered just as a barrier – it has many other functions) is a rather poor barrier to most chemicals, compared with any normal standard of barrier film used for, say, the food industry. A thin layer of polyethylene is far tougher to cross for most chemicals than the stratum corneum. The fact that the skin is a composite material is also nothing special to the outsider. Many polymers contain complex domains of chains of varying crystallinity and chemical affinities. One outsider who had to help make major decisions about skin as a barrier was Hansen  whose concerns were to do with safety when handling solvents. Applying principles that work very well for analysing the safety of gloves, which are often complex polymers, he and co-workers were able to show that skin behaved like just a (not particularly good) polymer with respect to standard solvents. From this work, he was able to derive the HSP values for skin and could make reliable predictions about what chemicals would and would not tend to penetrate skin, just as could be made for gloves. It is unfortunate that the skin permeation work of Sloan  and others has been based on Hildebrand solubility parameters, which are the sum of the three HSP; Hildebrand’s theory is inappropriate to the skin permeation because it explicitly applies only to systems containing no polar or hydrogen-bonding interactions.
Building on Hansen’s work, and adopting the three principles described in this study, Abbott  was able to further refine the HSP for skin incorporating the Jmax ideas of Cross discussed below. Independent work by Wiechers had reached a rather similar conclusion. For the rest of this study, it will be assumed that the HSP value of skin is around [17, 8, 8]. This is not far from solvents such as DMSO and DMI, which are known to penetrate the skin quickly without significant damage. Interestingly, if a rough estimate of the HSP of the lipid part of the skin is made, then it is not as ‘lipidic’ as it first seems. The 20+% of cholesterol provides a large boost from the modest [16, 3, 3] that might be expected from the basic lipids, taking values up to something like [17, 6, 6].
The claims made about HSP are modest. They certainly cannot explain everything connected to solubility, and they cannot be of academic accuracy in such complex systems. The HSP of skin cannot be an exact number like [17, 8, 8]. The methodology described in this study uses HSP as a powerful guide to thinking through solubility and partitioning issues. The power gained by the use of the mixing effect is particularly impressive. If others can develop a tool with comparable ease of use but greater precision [e.g. one based on COSMOtherm (http://www.cosmologic.de)], it can readily be fitted in to this overall methodology.
The speed of diffusion at any point in the skin depends on just two factors: the concentration gradient and the diffusion coefficient at that point. A high concentration in the outer nanometre of the stratum corneum is an absolute requirement for fast diffusion as that sets the maximum possible concentration gradient across the SC. In HSP terms, this means as close a match as possible between the HSP of the skin and that of the permeating species – active or solvent.
The diffusion coefficient depends first of all on the molecular shape and size. Larger and more branched molecules diffuse more slowly than smaller, linear molecules. A convenient way to estimate relative diffusion coefficients is via molar volume, which happens to be an integral part of HSP calculations. Over a variety of polymers, it has been found that the diffusion coefficient depends on (molar volume)x where x is approximately 1 for open polymers and approximately 3 for closed, crystalline polymers – in other words, the more rigid the polymer, the stronger the dependence on the molar volume. For skin, the dependence on molar volume seems to have a value of x∼2. Other functional forms can also be used for skin, for example the classic Potts and Guy  exponential dependency of permeation coefficient on molecular weight D = D0 Exp(−0.006*MWt). Cross and co-workers  showed that the maximum flux through skin, Jmax, has a stronger exponential dependency than the permeation coefficient: J = J0Exp(−0.019*MWt). As users tend to be more interested in Jmax, this stronger dependency on MWt may be the more relevant, and this dependency is a reasonable match to the x∼2 approximation found for normal polymers.
Often missing from these well-known aspects of diffusion is the fact that the diffusion coefficient depends strongly on the concentration of other components in the polymer or skin. A solvent that swells the skin will increase the diffusion coefficient both for that solvent and for any other component, such as the active.
So, there is an important feedback system. The greater the solubility in the skin, the greater the diffusion coefficient, and the faster the solvent can enter deeper into the skin which, in turn, increases the diffusion coefficient further down. Any integrated approach needs to take these important effects into account.
When it comes to the difficult real-world formulation decisions discussed in the next section, there is a key trick available to the formulator. Suppose that the solvent blend contains a mix of high and low molar volume molecules. The smaller ones will tend to enter the skin faster, leaving a very different solvent blend on the skin. If this new blend is less favourable to the active, it can provide a driving force to take the active into the skin. But if it is too unfavourable, the active will tend to fall out of solution and become useless or, perhaps, irritating. By understanding how the overall HSP change as the smaller components migrate, the formulator can create a rational balance between competing effects. This is a very powerful formulation strategy that seems to be, at least in the published literature, little used.
The number of skin ingredients authorized to be used at levels of 100% is small and diminishing. Regulatory agencies and suppliers are reluctant to guarantee that a high level of any single ingredient is safe. At the same time, many ingredients that are included for, say, sensorial reasons, have very poor solubility profiles so need to be matched by ingredients that can improve solubility. Therefore, the formulator generally has little choice but to work with multiple ingredients. Pure research can focus on individual components, an integrated approach must take the reality of multiple ingredients into account. The HSP approach very naturally copes with the solubility behaviour of mixtures of ingredients.
The gold standard of academic skin permeation work has classically been the Franz cell experiment. To the unwary, the results of such experiments can be misleading. Take, for example, the oft-repeated statement that terpenes are good permeation enhancers (see, e.g.). On HSP grounds, this makes no sense. Terpenes (limonene’s HSP, for example, are [17.2, 1.8, 4.3]) are a poor match for skin and, incidentally, for many actives. Looking more closely at the details, it is clear where this idea comes from. Compared with an ethanol-only formulation, one with terpenes added to the ethanolic formulation can indeed give faster permeation. The HSP explanation is that the HSP of a 65 : 35 limonene : ethanol mix [16.7, 4.3, 9.6] is a closer match to that of skin than either on their own. Using infinite dose experiments in Franz cells, interesting terpene effects can therefore be found.
In practice, however, most formulations will be delivered as a finite dose. The ethanol will evaporate rapidly, leaving a large concentration of the terpene, which will probably be a poor solvent and provide little benefit in terms of permeation. Also, ethanol is known to extract lipids, and this is likely to be more significant under infinite dose conditions vs. finite dose conditions.
Practical formulators need to focus their precious resources on systems that recognize the tough requirements of finite dose conditions. And they need to make a choice between two extremes:
1 Do they want to help the active into the skin via solvents that greatly increase the rate by swelling the skin (increasing the diffusion coefficient) and creating a better HSP environment (on average) and by entering the skin and providing no choice to the active but to go with the solvent?
2 Do they want the active to penetrate more slowly, driven out of a relatively hostile formulation that remains mostly on the surface of the skin and which risks being lost from the skin when the user touches it and transfers it to clothes and the environment?
For a formulator, the first choice makes a lot of sense. Regulatory agencies might not be so sympathetic. For those who choose the latter, the big question is whether the active has a significant chance to enter the skin before being lost from the skin. It is astonishing that this elementary question about loss from the surface of the skin is so little addressed in the skin permeation literature.
The probability is that a formulation will strike a balance between these extremes. It is therefore important that an integrated approach can model what happens during diffusion of a complex mixture so that the relative balance of effects can be plausibly simulated.
The integrated approach
In the following discussions, the integrated package formulating for efficacy (FFE)  is used to illustrate what can happen with a systematic attempt to integrate all five elements. It is important to emphasize that other ways to integrate this approach can be readily imagined and should be positively encouraged. Also, it should be made clear that FFE claims only that the models are reasonable and a way to think through complex issues. In the Summary, it will be made explicit that FFE (and other packages) needs to be challenged by academics and real-world formulators. The importance is the fact that the user interacting with FFE is required to take into account all five principles. Although success is not guaranteed using the package, arguably focussing in on just one or two of the factors will guarantee suboptimal performance.
The formulator needs some basics. The HSP and melting points (for predicting ideal solubilities) for a number of well-known actives are provided, along with the HSP and molar volumes for over 100 ingredients typically used in cosmetic formulations, see Figs 1 and 2, for examples.
Built in to FFE is the ability to automatically estimate the HSP of new actives and ingredients. For actives, however, it is strongly recommended that the HSP be measured rather than estimated. This can be performed in-house via a simple procedure or performed by contract organizations for which this is routine.
Once an active is selected, that immediately defines one target for optimizing solubilities. The other target is that of skin, set to [17, 8, 8] as default. The HSP distance for each ingredient from these targets is automatically calculated and presented as ‘gaps’– the ingredient active gap and the ingredient skin gap, see Fig. 3. The larger the gap, the smaller the solubility. By sorting the solvents according to either of the gaps, it is easy to find individual ingredients that might provide optimized solubility.
If the formulator has a specific formulation in mind, then the quantities can be entered and the gap between the active and formulation (AFG) and skin and formulation (SFG) can be calculated. If the formulator wishes the programme to provide an optimized formulation (smallest gap), this can be performed with respect to the active or the skin.
To open up fresh possibilities, the formulator can ask the programme to suggest sets of two or three ingredients that are close to either target. This request can often produce mixtures of ‘bad’ ingredients that, by the law of HSP averages, form a ‘good’ solvent. As these formulations are unlikely to be suggested by intuition, this can be a spur to innovation. Sometimes, two suggested ingredients are too different to be mutually soluble; the software makes some attempts to avoid such incompatibilities, but is not too restrictive because other ingredients might make them mutually compatible. Figure 4 shows the results of a request for the best two ingredients to match the skin-whitening ingredient, octadecendioic acid.
In all cases, the actual solubility (important for delivery!) is estimated based on the ideal solubility (from the Yalkowsky approximation based on melting point) and the activity coefficient based on the HSP distance of the formulation from the active.
Assuming that the formulator is happy with the balance of solubilities as well as costs and regulatory issues (not included in the package), then three of the five principles have been followed. Now it is time to use the other two.
The fate of the formulation on the skin
The diffusion modeller within FFE keeps track of the active and the ingredients during whatever timescale the user chooses. Over time, the smaller ingredients tend to enter the skin, leaving the larger ones at the surface. This in turn changes the HSP of the mixture on the skin, which affects the solubility of the active and the concentration of the ingredients in the surface of the skin and, therefore, the overall concentration gradient. No attempt is made to model the loss from the surface of the skin by general contact with skin and clothes although this could, in principle, be added.
Although the calculations are complex (see Fig. 5), they are not too slow, and the user rapidly builds up a picture of what is happening to the active and the individual ingredients. The calculation shows results in terms of the thickness of the original applied dose in μm. In the example shown, a layer of 10 μm thickness (approximately 1 mg cm−2) is applied and after 12 h 4.3 μm remain on the skin, 0.45 μm are in the SC and 5.2 μm have passed through the skin. For the typically complex formulations normally used, it is not helpful to quote amounts in molar units. Changes in the original formulation can be made to see how that changes the balance of behaviours. For example, if two molecules have similar HSP (and therefore solvency characteristics) but one is low and the other high molar volume, the two different formulations can be compared. This allows the formulator to make a rational choice between, say, a caprylate and a palmitate.
When the smaller molecules enter the skin, the diffusion coefficients rise. As they pass through, the skin contains lower overall concentrations of solvents, so diffusion coefficients fall. These effects can have profound impacts on the fate of the active: stranded on the surface, sitting within the SC or passed through into the more gel-like regions of the dermis. Although the software can indicate the fate of the active, the decisions about how best to balance the ingredients have to be reached by the formulator using the results of multiple runs of the modeller. The software has no way to know, for example, whether the formulator would prefer rapid initial delivery with lower overall delivery to a slower, but overall higher delivery.
Another aspect of the package (not shown) allows estimates of the ‘skin delivery gap’– the gap between the rate at which the active arrives in the dermis, the rate at which it is removed from the dermis and the concentration (usually determined in vitro) required for efficacy.
The arguments for an integrated approach to the finite dose delivery from complex cosmetic and pharmaceutical formulations seem to be powerful. There are multiple ways to implement each of the five elements and no special claim is made that FFE has all five elements correct. In order for such models to be further developed, they need to be specifically challenged. In particular, the models make specific predictions about diffusion of different elements of the formulation into the SC. By making predictions in systems that are representative but sufficiently simple, academics with suitable analytical tools will be able to test the models in relevant, finite dose experiments. Other academics with the ability to data-mine the relevant literature will be able to test the package against historical real-world data. Cosmetics and pharmaceutical companies can rapidly do virtual tests of their current formulations, develop hypotheses about how these could be improved and test the predictions.
In fact, for FFE, all three activities are under way, and readers of this publication are actively encouraged to contact the author to progress such exercises. Informal feedback so far is encouraging, showing that the integrated approach not only works, but allows a change of mindset from the simplistic approaches. Specific examples of using FFE for optimization of delivery of octadecenedioic acid have been presented at conferences by Wiechers. Alternatives to FFE will be greatly welcomed as they can only add to our knowledge of how best to model the complexities of skin delivery, to the benefit of those who provide the formulations and those who use them.
The author of this study should have been the well-known industry expert, Professor Dr Johann Wiechers. His sudden and unexpected death whilst attending a conference where he was to teach the principles of FFE was a great shock to his many friends and colleagues. No single reference to Wiechers’ work on FFE is provided because it has been disseminated widely in the cosmetics literature, at conferences, and especially in his inspired teaching courses. This study is dedicated to his memory.
Work for this study has been funded by the author. There are no conflicts of interest.