When formulating skin care emulsions, ingredient suppliers and formulators often focus on the technical requirements for the shelf-life stability of the end product. However, consumers consider it a given that the product will remain stable for months, even under harsh conditions. The sensory perception of the product and belief that the product delivers on its promise are far more important. Even performance claims supported by objective clinical evidence are strongly influenced by the aesthetic properties of the product both on the shelf and especially during use.1 Thus, the tactile sensory properties of a cosmetic product intended for application to the skin—the largest and most sensitive organ—are crucial to the product’s commercial success.
It is not easy for formulators to get a product’s aesthetics right. Usually the relationship between sensory perception and formulation acceptance by consumers is far from straightforward and depends on many uncontrollable parameters. Detailed and systematic consumer research is necessary to understand the sensory preferences of target market segments. Further, in trying to express their preferences, consumers often use unclear definitions and product descriptors.
However, if the sample size is large enough, one can group these descriptors and look for a relationship with the chemistry and physics of the formulation. Many such attempts have been made in the past 40 or more years and can be found in literature,2 but there remains a need for a bridge between qualitative consumer language and clearly defined sensory attributes based on the chemistry and physics of a formulation. Understanding the quantitative or even qualitative impact of ingredients is a complex task, especially given the complexity of formulations and enormous choice of ingredients.
Navigating regulatory and brand-specific restrictions, formulators tend to choose emulsifiers conservatively because they are viewed mainly as technical ingredients to keep oils and aqueous solutions together for a given length of time. The amounts of other ingredients, especially emollient oils, are considered less risky to adjust in a formula when fine-tuning it for skin feel. Here, the authors describe an approach to investigate the intrinsic effects of emulsifiers and quantify them before attempting to translate them into regional consumer sensory preferences.
ASTM Standard 1490-03, a long-established descriptive analysis technique,3 was used to qualify and quantify model and market formulations. This involved a trained panel of 10 (± 2) individuals, who assessed emulsions under strictly controlled, standardized conditions. Numerical scores on 30 sensory attributes, e.g., spreadability and gloss, were collected during all steps of the product’s use—from appearance in the package, pick-up on the fingertips and rub-out, to after-feel on the skin up to 20 min after application. These attributes were scored on a scale from 0–100, each with internal reference standards, such as mineral oil or petrolatum, to reduce panel variability. The sets of numbers generated by this protocol represent the sensory chromatograms of formulas but these profiles use a technical language that, while understood by trained formulators, does not express quality or preference judgment in a language understood by consumers.
Data was collected in this manner for more than 10 years, allowing the authors to make comparisons within a particular set of samples or even between experiments conducted years apart. This constancy of method, devoid of cultural preference or bias, also allows comparable data to be added at any point later in time. A database on hundreds of emulsions was accumulated, and the authors processed this large sample size to identify patterns of emulsion characteristics or ingredient-related sensory properties and translate them into consumer language, as described next.
Mapping Sensory Space
As noted, 30 attributes were assessed for each formulation, and each attribute could be considered one dimension of a sensory space in which the tested products could be placed. Hence, each test sample could be placed in a 30-dimensional (30D) map based on its individual panelist ratings of sensory attribute data, which relate to its ingredients and their concentration. However, to recognize patterns in these 30 dimensions, this space must be simplified to a more user-friendly, two- (2D) or three-dimensional (3D) map, which requires a mathematical technique to manipulate the data. Here, the authors use an approach referred to as Principal Component Analysis (PCA) to reduce the 30D sensory space to a 2D modela. This mathematical technique seeks correlations between attribute values.4
For example, in an extreme case where all 30 sensory attributes are perfectly correlated to one another, the 30D space could be reduced to one dimension because each attribute would be represented by the same data point. This would result in one so-called principal component (PC) that explains 100% of the data. The reality, of course, is not so simple. While the PCA model can identify correlations between attributes such as integrity of shape and firmness, it does not give any information about why they are correlated. In the current set of data, PCA analysis was used to help identify what sensory attributes are affected by a given emulsifier selection and to what degree. Ultimately, the goal was to achieve a 2D representation of the simple emulsion systems tested. Interestingly, more than 60% of the data explanation was found within just the first two PCs. It is important for this percentage of explanation to be high, ideally over 66%, otherwise significant amounts of modeling information could be missed and flaw the 2D representation of the 30D sensory reality. In another sensory analysis, for example, a 4D representation was required to produce a satisfactory level of data explanation.5
This percentage can also be improved by comparing emulsions of the same type, such as simple o/w emulsions consisting of one emollient and one emulsifier at one concentration. In doing so, more than 80% of the data explanation can be obtained from PC1 and PC2. The decryption of information contained in a PC can be visually represented in a PCA “loading plot” (see Figure 1), in which the sensory attributes are positioned with PC1 and PC2 functioning as the x and y axes, respectively, creating a 2D projection of each attribute vector in 30D space. Figure 1 shows the loading plot for all 30 attributes in the 2D sensory space of all simple one-oil, one-emulsifier formulas.
Attributes that have a connection with formulation rheology—i.e., spreadability, cohesion, firmness, peaking and shape—are observed roughly along the PC1 axis. Along the PC2 axis, lubrication and friction attributes, such as oily, greasy or waxy effects, can be seen. The clutter around the origin can mean two things: either the spread on the data is too big and the attribute cannot be sufficiently determined accurately, or the attributes sit in another dimension of sensory space. It turns out to be a bit of both. Each point is a specific formulation with a specific ingredient composition. The surprise is that the samples are not randomly dispersed over the sensory map but clustered. This clustering is especially driven by the emulsifier used, and less so by the chosen emollient. This finding stresses how, in order to create a significantly different texture, changing the emulsifier has a greater impact than changing the emollients.
PCA Loading Plot
After the PC1 and PC2 are calculated, they are set as the x and y axes on a graph and the remaining attributes are plotted based on their variation from these axes. As a result, those attributes that strongly correlate with PC1 and/or PC2 lie closer to them, and those that show an inverse correlation are on the diametrically opposite sides. Also, those having a strong impact on the test emulsion’s position in this redefined 2D space will lie close to, or on, the PC axis. Further, although attributes may lie close to the origin, they are still not explained by PC1 or PC2, and attributes that lie close to the PC1 axis do not correlate at all with attributes located close to PC2.
Using the same PC1 and PC2 axis, interesting observations can also be made by plotting several individual emulsions (see Figure 2). The position of the points provides condensed sensory information on the emulsions relative to each other, the consequences of which are fundamental. First, it allows for the quantitative classification of emulsions based on their sensory (dis)similarities. It also assists formulators in investigating the contribution of chassis ingredients to the overall sensory profile of the emulsion. For example, if an ingredient’s effects are sufficiently independent, they could serve as a basis to create specific ingredient-controlled sensory signatures. The clustering of simple emulsions all made with different emollients but the same emulsifier implies that the emulsifier selection is the dominant variable, and the sensory properties are independent from the emollient choice. This approach, along with the described data analysis model, could also be used to match or reverse-engineer sensory profiles.
Limitations: It is important to recognize that this approach takes a complex 30D concept and simplifies it to a tangible 2D representation, thus it has some limitations. For instance, the approach looks for linear correlations between attribute values and assumes that measurement errors are normally distributed. To reduce the impact of these assumptions in building a meaningful model, similar formulation types should be compared and tested by an identical protocol; e.g., only leave-on skin care emulsion formulations.
Map as a matching tool: To interpret the meaning of the proximity between points on a sensory map, one must know the statistical quality and sensitivity of the data provided by human sensory analysis. A meta-analysis of the raw data, i.e., the individual panelist scores on each attribute for the complete database, shows that with a fairly homogeneous set of emulsions, the attributes tested in the beginning of the protocol are more statistically significant than the attributes tested after 20 min. Specifically, the initial pick-up and rub-out attribute values were fairly reliable within 10 scale points, but the 20-min after feel attributes were less reliable. The typical error estimate for each attribute was therefore incorporated into the modela to determine how significantly two sample points on the PCA map differ. By doing this, it was established that points closely situated to one another on the PCA sensory map often show a close resemblance in the descriptive analysis results.
Emulsifier Case Studies
Utilizing this approach, three stabilization mechanisms were examined by testing simple formulations made with the recommended use level of emulsifier and 10% of very different emollients. PCA sensory maps of these simple emulsions were generated, revealing clusters of formulas.
Biopolymer stabilized emulsions: In the first example, biopolymer stabilized emulsions made with 1% emulsion stabilizerb and 10% emollient were mapped and appeared in a cluster (see Figure 3). This polymeric emulsion stabilizer functions by a suspension mechanism that is theoretically HLB- and therefore oil-independent. The emollients used varied from very light to very heavy. Despite the fact that emulsions were quick-breaking, the clustering of the dots representing the emulsions on the 2D map seems to indicate the choice of emollient plays less of a sensory role than the choice of emulsifier; had the emollients played a larger role in defining sensory properties, one would expect to see more scattering of those dots on the map.
This impression is reinforced when the total profile PCA sensory space is split into separate PCA plots for initial feel and after feel (see Figure 4). The clustering remains in the initial phase, but the points spread out in the after feel PCA plot. This can be explained as follows. The initial phase sensory properties are dominated by the characteristic rheology induced by the emulsion stabilizing system. Since this is constant for a given emulsifier, the sample points cluster. The after feel sensory properties are defined by the friction properties of residual film residues of oils, polymers and waxes on skin moisturized to different degrees, depending on the chosen ingredients. It is therefore not surprising that the after feel sensory attribute values are subject to a much wider spread. In this case of quick-breaking emulsions with very low emulsifier content, the after feel will be oil-dominated.
Hydrosome-forming emulsifiers: Hydrosome-forming emulsifiers are self-emulsifying, waxy materials usually added to the external water phase of a formula to form an aqueous lyotropic liquid crystal phase. These are capable of suspending and emulsifying oil droplets, which prevents the emulsion oil droplets from coalescing. Also, this emulsion stabilizing concept is essentially independent of the required HLB of the oil phase. Subsequently, emulsions incorporating 10% emollient and 2% of a sorbitol-based hydrosome formerc were mapped (see Figure 5).
A similar clustering to the polymer in example one is seen here for the initial application phase; however, because the liquid crystalline phase slows the evaporation of water from the emulsion, and because it is a wax that is present at a higher concentration, one would expect the after feel to be determined by a complex oil-wax interaction. It turns out that the waxy nature of this emulsifier reduces the differentiation between emollients.
Polymeric w/o emulsifier: In a third, w/o example, the emollient oil was the external phase. Hence the nature of the oil was expected to play a much bigger role with less of an impact from the emulsifier on the sensory properties. This is more difficult to test since few w/o emulsifiers can work with a broad variety of oils, and at a low use concentration.
Figure 6 shows the results of a polymeric w/o emulsifierd used at 2% with 20% emollient. Here, the map is spread out across the PCA plot. Increasing oil viscosity moves the sample data points from the left to the right in the diagram, confirming that the impact of emollient choice is more important in w/o systems than in o/w systems.
Emollient impact: The secondary impact of emollients on the skin feel of o/w emulsions is shown in Figure 7, which reveals the ratings of test samples incorporating a frequently used estere. In contrast to clustering of data points for emulsions with the same emulsifier, data points for emulsions containing the same emollient are not clustered. If the emollient had a dominant effect on sensory properties, one would expect it to shift emulsions into an emollient-specific cluster. However, this is not apparent in Figure 7. This confirms an earlier study of a smaller set of simple o/w emulsions showing that the emulsifier is responsible for roughly 80% of the emulsions’ sensory signature.3
Of course, the emollient does have a sensory effect, but it is not the primary factor in the sensory properties of an o/w emulsion in the early phases of product application.
The emollient influence on sensory properties would be expected to increase with the sensitivity of the emulsifier to the nature of the oil phase; i.e., when matching the required HLB of the oil phase becomes necessary, and oil and emulsifier are no longer independent variables in the model.
Nevertheless, from the examples given, it is clear that a change of emulsifier can modify the sensory properties much more effectively.
Translating to Consumer Language
Careful examination of the complete set of emulsion samples in the database and their position in the PCA sensory space reveals a pattern that can be captured qualitatively by the universally recognizable consumer-language descriptors of “light/heavy” and “fluid/viscous.” The fairly abstract PC sensory map can therefore be transposed into an easy-to-understand, four-quadrant grid with matching consumer language descriptors of the x and y axes (see Figure 8).
Provided with simple emulsion samples where only the emulsifier varies, formulators can use this approach to choose the desired sensory properties of an emulsion at the beginning of the formulating process. Descriptors can be plotted on a selector gridf that can be modified to help focus on the key sensory descriptors for a specified project brief. This method also allows for variations in consumer descriptors, which will vary by region and language. Understanding that the emulsifier system is the dominant variable in a product’s sensory attributes allows for a more direct approach to engineering formulations.
Here, the authors present a methodology to construct simplified 2D sensory maps from complex sensory data. The PCA mapping technique can be used to match sensory properties of emulsions and ingredients to sensory benchmarks. Systems with oil in the external phase, such as w/o emulsions, were notably less defined and require further study. Regardless, while the obtained PCA data representation is abstract, as shown here, it can be converted into a practical tool for o/w emulsion formulation.
- L Rigano, Sensory in cosmetics, Cosm & Toil 127(9) 628–634 (2012)
- BW Barry and AJ Grace, Sensory testing of spreadability: Investigations of rheological conditions operative during application of topical preparations, J Pharm Sci 61, 335 (1972)
- G Civille, The Spectrum Descriptive Analysis Method, ASTM International (May 1992)
- Principal component analysis mathematics function explanation, available at http://cosmic.mse.iastate.edu/library/pdf/pcainterpretation.pdf (Accessed May 6, 2013)
- ME Parente, A Gámbaro and G Solana, Study of sensory properties of emollients used in cosmetics and their correlation with physicochemical properties, J Cosmet Sci 56(3)175 (2005)