We examined whether developmental instability can be used as a bio-monitoring tool in the endangered Taita thrush (Turdus helleri L.) through the measurement of individual levels of fluctuating asymmetry in tarsus length. Because estimates of the association between developmental instability, stress and fitness derived from traditional regression are biased, we compared parameter estimates obtained from likelihood based analysis with those obtained from a Bayesian latent variable model.
Taita thrushes were captured and measured in three isolated cloud forest fragments located in the Taita Hills of south-east Kenya.
We applied mixed-effects regression with Restricted Maximum Likelihood parameter estimation (performed with SAS version 8.0) and Bayesian latent variable modelling (performed with WINBUGS version 1.3 and CODA version 0.4) to estimate unbiased levels of developmental instability and to model relationships between developmental instability and body condition in 312 Taita thrushes.
Likelihood and Bayesian analyses yielded highly comparable results. Individual levels of developmental instability were strongly inversely related to body condition in the subpopulation with the lowest average condition. In contrast, both variables were unrelated in two other subpopulations with higher average condition. Such heterogeneity in association was in the direction expected by developmental theory, given that higher condition suggests more benign ambient conditions. The estimated levels of body condition in the three subpopulations did not support their presumed ranking in relation to environmental stress. Developmental instability and body condition are therefore believed to reflect different aspects of individual fitness.
Variation in developmental homeostasis, either modelled as observable variable (fluctuating asymmetry) or latent variable (developmental instability), appears a useful indicator of stress effects in the Taita thrush. Because relationships between environmental stress and developmental instability may depend on the extent to which stress-mediated changes in other components of phenotypic variation are correlated, the study of trait asymmetry should preferably be combined with that of other measures of trait variability, such as trait size or organismal condition.