202 JOURNAL OF THE SOCIETY OF COSMETIC CHEMISTS However, selection of preservatives for a cosmetic is complex. The ideal characteristics of a biocide are that it be safe, stable, and compatible with both the product and the container, be inexpensive, readily available, approved by appropriate regulatory agen- cies, have a positive consumer perception, and be environmentally friendly. Raw ma- terial quality, container and cap design, expected shelf life and exposure conditions, and even how the consumer will use and misuse the product are additional considerations in choosing the preservative system (30,33). Compatibility of the biocide with other ingredients in the product requires the micro- biologist to have knowledge of the art of formulation. Suspended solids in a formulation (e.g., carbonates, silicates, talc, metal oxides, cellulose, and starch) may adsorb pre- servatives (34). Minor pH changes inactivate other preservatives (35-37). Minor shifts in ionic strength or changes in the buffering system in a product can also alter a bacterium's susceptibility to a biocide or affect how a preservative partitions between the water matrix and the microbial cell (38,39). Parabens provide unique formulation challenges for water-in-oil emulsions because they have an affinity for the oil phase while the microbes live in the water phase (40). Even the surfactant system used can affect biocide performance (41-43). In fact, nonionic surfactants are used to neutralize some preservatives (44-46). However, these same surfactants enhance quaternary ammonium compounds (47). Finally, protein (often used in conditioner and lotions) may also reduce the antimicrobial activity of many preservatives (48-50) the presence of hydrophilic polymers will affect others (51). Even simply choosing a container requires a microbiologist to check compatibility with the preservative (52,53). The preservative may either be absorbed into the container material in the case of lipid-soluble preservatives, inactivated because of complexation of the preservative with the dyes used in the plastic, or lost because of the volatility of the preservative (e.g., phenoxyethanol, formaldehyde, and ethanol). When considering containers, one should also not overlook the impact that dispensing closures have in preventing microbial contamination, especially during consumer use. Some closures provide more protection of products than others (30). Alternatively, some closures may inactivate the preservative (54). PRESERVATIVE EFFICACY TESTING DEFINING THE PURPOSE OF THE PET Test protocols for determining preservation efficacy in cosmetics vary (55-58). The logic and arguments that go into establishing these protocols are primarily based on consen- sus. These compendial efforts, such as those developed by the CTFA, are "state-of-the- art," but they are not rigidly controlled protocols subjected to multiple laboratory replication and statistical analysis. Nevertheless, they have been useful. The CTFA/ AOAC/FDA collaborative program mentioned previously may fill this gap despite not being a method that has been validated to predict consumer contamination potential. To develop such predictive tests, a company must employ a microbiologist who conducts a validated "in-house protocol" that is specific for the company's products. Alterna- tively, the protocol developed by the company may be contracted out to laboratories capable of conducting PETs.

COSMETIC PRESERVATION 203 A major difference between PETs is due to a lack of understanding of the purpose of the PET. Defining the purpose of the test is critical. The entire experimental design for validating the PET will differ depending on the definition of purpose. The experimental design for validating use of a PET as a predictor of the potential for consumer contam- ination is different from that for validating use of a PET to demonstrate the presence of the preservative or as a predictor of potential for manufacturing contamination. To validate a PET as a predictor of consumer contamination requires prospective correlative consumer studies or retrospective validation, based on lack of consumer complaints, to corroborate the PET laboratory results. Regardless of which philosophy one adopts to define the purpose of a PET, at a minimum the goal should be to develop a data base to rank the antimicrobial hostility of the company's products. The purpose of a PET as viewed by the FDA is to predict consumer contamination potential (59,60). The FDA has tried several times to develop a PET for this purpose without success. Products failing such a test would be subject to recall. Despite the collaborative work between CTFA, AOAC, and FDA to develop a standardized PET complete with multi-lab comparisons and statistical analysis, the method has not yet been demonstrated to have the ability to predict a product's ability to withstand mi- crobial insult that may occur during intended use, since no correlative consumer studies were conducted using the same products for which the PET was conducted. Such an omission is fortunate. If such a standard PET were developed that was predictive of consumer contamination, then it could be used to enforce a recall on those products that fail it. One could counter the recall by pointing out that a PET does not account for consumer use and packaging parameters. One might also counter this argument with the observation that if the PET is done on freshly made product, then the PET data would only apply to freshly made product. Since most companies conduct PETs on shelf-aged product, such a statement would be admitting that they are out of line with the majority of reputable companies and have products that become a risk over time. Several publications have already shown that a modified version of the CTFA preserva- tive efficacy test is a valid predictive model of the risk of consumer contamination (17-20), but these all used proprietary "in-house" organisms unavailable to others. Thus, the CTFA test described by these publications does not provide a standard PET that could be used to enforce a recall as described above. Another study has compared several PETs for the ability to predict "in-use" contamination (61). The major criticism of this work is that the in-use test was merely simulated. The subjects dabbled with the product after rubbing their underarms with their fingers. The significance of ranking PETs against their ability to predict how well a product can withstand simulated consumer use does not represent validation against true in-use conditions. Nevertheless, since all the PETs were ranked against a single standard, one can still derive considerably useful information. For example, nearly all the compendial tests adequately separated poorly preserved from well-preserved products. Some of the more conservative tests classified marginally preserved products the same as poorly preserved ones, while the more liberal tests allowed marginally preserved products to rank with well preserved ones. The CTFA test exhibited the tendency to rank all three (poor, marginal, and well) correctly against the flawed but useful standard of a simulated in-use test. This study does not, however, support the use of the CTFA test to enforce recalls, since the comparison was against an invalid simulated in-use test.