Leveraging High-Dimensional Neuroimaging Data in Genetic Studies of Neuropsychiatric Disease

Abstract

The current state of biomedical science is such that both the number and sophistication of methods available to investigate the genetic determinants of disease is unprecedented. For example, the introduction of high-throughput technologies such as DNA microarrays, allow researchers to comprehensively assess the human genome for single nucleotide polymorphisms that confer genetic susceptibility. Indeed, while these, and other similarly sophisticated methods, have yielded notable findings with regard to identification of risk variants in diseases such as diabetes, obesity, and glaucoma, similar studies of neuropsychiatric diseases such as schizophrenia and bipolar disorder have been somewhat less successful in producing strong findings. The reasons why this is the case are numerous, but likely refl ect the very complex genetic architecture of neuropsychiatric conditions. In this chapter, we consider an approach to addressing this complexity that involves the use of what are termed ‘endophenotypes’ (or alternatively ‘intermediate phe-notypes’) in genetic studies of neuropsychiatric disorders. Endophenotypes are biological changes, such as brain structural differences, that are thought to represent underlying molecular, physiologic, or otherwise subclinical changes resulting directly from the genetic variations that mediate susceptibility to overt clinical disease. Furthermore, neuroimaging phenotypes are, for a variety of reasons, thought to represent good candidate endophenotypes for genetic association studies of neuropsychiatric disease. Like high-dimensional genome-wide data, however, these phenotyping technologies can produce hundreds to thousands of data points or more, when all neuroanatomic regions and tissue types of interest are considered. The question then becomes, how can two or more high-dimensional data types (i.e., in this case genomic and neuroimaging) be leveraged, integrated, and analyzed in order to make valid inferences about the genetic basis of neuropsychi-atric disease? We comment on the analytic issues that arise when trying to leverage both genome-wide genetic data and neuroimaging data (e.g., problems related to multiple comparisons and false positives, as well as small sample sizes), and discuss four general approaches, each with its own set of advantages and disadvantages, that can be used in the analysis of imag-ing-genetics data. Finally, we provide a brief review of some of the recent studies that combine imaging and genetics, but note that the field, as a whole, is still very much in its infancy. We also provide suggestions for future directions.