Using Variance Components Modeling with Twin Data to Examine Impact of Traumatic Events

Karestan C. Koenen, Columbia University, New York, New York, USA

In the Traumatic StressPoints Winter 2001 article, "Using Twins in Trauma Studies to Determine Genetic Influences," the twin method of behavioral genetics research was described briefly, and one variant on the twin method, the co-twin control study, was discussed. The article below introduces another research approach: the use of variance components modeling with twin data.

Traditionally, variance components modeling has been used to decompose the observed variance in a set of symptoms or disorders, such as PTSD, into 1) additive or nonadditive genetic effects; 2) shared, common or family environmental effects; and/or 3) nonshared or unique environmental effects. Additive genetic effects reflect the actions of a large number of genes, each of small effect, whose influences combine in an additive fashion to produce differences at the phenotypic or observable level.

When the correlation between monozygotic twins is more than twice as large as that for dizygotic twins, it suggests that at least some of the genetic variance in the phenotype is due to nonadditive genetic factors. These factors could be a significant influence from a dominant gene at one locus or the interactive or synergistic effects of multiple genes at more than one locus (epistasis). Of course, common environmental effects are influences on both members of the twin pair, and unique or nonshared environmental effects are influences specific to individuals (rather than the pair) and to random error. These latter influences promote dissimilarity between twins. For example, one twin's exposure to combat in Vietnam, and not the other, would be a nonshared environmental effect.

The full model including all three effects (genetic, shared environmental and nonshared environmental) is contrasted against reduced or restricted models that delete either additive genes or shared environment. By comparing the full model with more restricted models, it is possible to determine which model best fits the data. Models are evaluated using model-fitting statistical software packages (such as Mx, developed by Michael Neale at Virginia Commonwealth University: http://views.vcu.edu/mx/).

Variance components modeling can be used to disentangle the influence of genetic, common environmental and unique environmental factors among traumatic events, PTSD and other psychiatric disorders. To illustrate, consider the highly comorbid disorders of PTSD and major depression. It is possible that they are comorbid because they share an underlying genetic vulnerability; for example, the genetically influenced personality trait of neuroticism has been shown to increase risk for PTSD and major depression independently. Variance components modeling can be used to test whether the genetic influence on PTSD also is related to major depression. If this is the case, the results imply that some of the same genes or sets of genes might influence risk for developing both disorders.

A word of caution, however: Modeling techniques are most effective when they test a prior hypothesis about the relationships among variables. Without such hypotheses, the best-fitting model may not make sense in the context of previous knowledge or theory regarding the phenomenon studied.

If you are interested in further information about statistical methods in behavioral genetics, see the Behavioural Genetics Interactive Modules developed by Shaun Purcell (http://statgen.iop.kcl.ac.uk/bgim/), a series of freely available interactive online resources and computer programs designed to accompany Statistical Methods in Quantitative Genetics, the appendix to the 4th edition of Behavioral Genetics by Robert Plomin, John DeFries, Gerald McClearn and Peter McGuffin (New York: Worth Publishers, 2001).