How Reliable is the Analysis of Complex Cuticular Hydrocarbon Profiles by Multivariate Statistical Methods?

Numerous recent studies have correlated cuticular hydrocarbon profiles with a wide range of behaviors, particularly in social insects. These findings are wholly or partly based on multivariate statistical methods such as discriminate analysis (DA) or principal component analysis (PCA). However, these methods often provide limited insight into the biological processes that generate the small differences usually detected. This may be a consequence of variability in the system due to inadequate sample sizes and the assumption that all compounds are independent. A fundamental problem is that these methods combine rather than separate the effects of signal components. By using cuticular hydrocarbon data from previous social insect studies, we showed that: (1) in 13 species of Formica ants and seven species of Vespa hornets, at least one group of hydrocarbons in each species was highly (r ² > 0.8) correlated, indicating that all compounds are not independent; (2) DA was better at group separation that PCA; (3) the relationships between colonies (chemical distance) were unstable and sensitive to variability in the system; and (4) minor compounds had a disproportionately large effect on the analysis. All these factors, along with sample size, need to be considered in the future analysis of complex chemical profiles.