Page 219 - Biobehavioral Resilence to Stress

196 Biobehavioral Resilience to Stress

to avoid false positive or spurious findings, the researcher must somehow

correct for the large number of statistical tests that are necessary to identify
associations in a large dataset (Taylor & Tibshirani, 2006). Th e researcher
must also carefully consider the manner in which effects of individual vari-

ations are to be modeled for the phenotype of interest. For example, each
individual locus and its variations can be tested for their association with
the phenotype of interest, but this strategy ignores linkage disequilibrium
relationships that might exist between variations at adjacent loci. Alter-
natively, the researcher might consider testing haplotypes composed of
variations at multiple loci. However, this strategy raises questions con-
cerning the biological meaningfulness of the haplotypes tested.* Th ere
are also questions about how best to model gene × gene interactions, gene
× environment interactions, and variant × variant interactions within a
gene. With ∼400,000 loci and perhaps 15–20 environmental risk factors
for consideration, there are a potentially enormous number of resulting
possible interactions that might be tested. Fortunately, there has been recent
progress toward the development of rational and compelling approaches to
the analysis of genome-wide association data (see de Bakker et al., 2005;
Lin et al., 2004; Schork et al., 2001).
Genome-wide association studies require large sample sizes to provide
appropriate statistical power and to account for the potential heterogeneity
of factors that may contribute to phenotype expression. For example, if it is

believed that a particular phenotype may be influenced by demographic or
behavioral variables of relevance, these variables must be represented and

controlled for by including a sufficiently large number of individuals in each
of multiple groups of experimental subjects. This method can be very dif-

ficult to achieve if the phenotype in question is rare or hard to diagnose,
or if the researcher is interested in an endophenotype that is particularly

expensive and difficult to gather (e.g., functional imaging-derived pheno-

types). The genetic analysis of large human samples also poses the issue of
genetic background heterogeneity, that is, the fact that individual subjects
may have unique genetic predispositions to disease or phenotypes associ-
ated with ancestry (Campbell et al., 2005; Freedman et al., 2004; Marchini,
Cardon, Phillips & Donnelly, 2004). When the entire subject sample is ana-
lyzed, genetic background heterogeneity may obscure the contribution of
another gene or genetic variation to the phenotype (see Berger et al., 2006).
Fortunately, this problem can be overcome by the use of sophisticated statis-
tical methods (Pritchard, Stephens, Rosenberg & Donnelly, 2000; Pritchard
& Rosenberg, 1999; Schork et al., 2001).

* In the process of defining a haplotype and testing it for association with a particular

phenotype, one typically assumes that there exist additional, unobserved variations for
each haplotype that influence the phenotype. This assumption may be incorrect.

1/22/2008 9:36:19 PM
CRC_71777_Ch008.indd 196 1/22/2008 9:36:19 PM
CRC_71777_Ch008.indd 196