5 Methods for Association Studies
93
5.2.2 Genome-Wide Association Studies
5.2.2.1 Background
Increased throughput, scalability, and speed have enabled investigators to undertake
GWAS (Claussnitzer et al. 2020)—research that would have been far too complex
to consider even 20 years ago. It has become possible to simultaneously measure
hundreds of thousands of SNPs due to technological advances in array-based
genotyping (Wang et al. 1998). The number of variants that may be assayed by these
SNP arrays rapidly increased at the same time that array prices steadily decreased.
At present, arrays can directly measure millions of SNPs while providing relatively
high coverage of common genetic variation across the human genome (Jorgenson
and Witte 2006; Lindquist et al. 2013; Nelson et al. 2013; Xing et al. 2016; Wojcik
et al. 2018).
The genetic content of such arrays was facilitated by the development of
technology that allows for large-scale sequencing efforts in combination with the
sequencing of the human genome (Lander et al. 2001; Venter et al. 2001). Beginning
in 2002, the International Haplotype Map (HapMap) Project undertook an effort
to catalog the common genetic variants that occur in human beings. It was also
determined that a substantial portion of this variation can be efficiently captured by a
subset of “tag” SNPs via the phenomenon of LD among neighboring SNPs (Daly et
al. 2001; Gabriel et al. 2002; International HapMap Consortium 2003; International
HapMap Consortium 2005) and that this structure varies across ancestral populations (International HapMap Consortium 2003; International HapMap Consortium
2005; Frazer et al. 2007).
Unlike candidate gene studies, GWAS are not hypothesis-driven; they do not
require a priori specification of the genes or polymorphisms that are conjectured to
be associated with the phenotype of interest. Rather, they quantify DNA sequence
variations from across the entire human genome in an attempt to pinpoint genetic
risk factors for common diseases. In designing an array for genome-wide assessment, a primary objective should thus be to capture as much common variation in
the human genome as possible.
5.2.2.2 Multistage Study Designs
When GWAS first became popular, the high cost of SNP arrays and necessity for
large sample sizes (to achieve sufficient statistical power to detect the anticipated
modest associations among hundreds of thousands of SNPs) (Witte et al. 2000)
motivated the development and use of multistage GWAS designs (Thomas et al.
2005). Decreasing SNP array costs have made multistage designs for GWAS less
essential, but we choose to briefly describe them for two reasons: (1) an overview
is important for understanding historical studies, and (2) as we move into the
post-GWAS era, next-generation sequencing of entire genomes may be sufficiently
expensive to once again make multistage designs relevant.
In the initial discovery stage of a multistage design, a subset of the study sample
is genotyped using genome-wide SNP arrays. Then the most strongly associated
93
5.2.2 Genome-Wide Association Studies
5.2.2.1 Background
Increased throughput, scalability, and speed have enabled investigators to undertake
GWAS (Claussnitzer et al. 2020)—research that would have been far too complex
to consider even 20 years ago. It has become possible to simultaneously measure
hundreds of thousands of SNPs due to technological advances in array-based
genotyping (Wang et al. 1998). The number of variants that may be assayed by these
SNP arrays rapidly increased at the same time that array prices steadily decreased.
At present, arrays can directly measure millions of SNPs while providing relatively
high coverage of common genetic variation across the human genome (Jorgenson
and Witte 2006; Lindquist et al. 2013; Nelson et al. 2013; Xing et al. 2016; Wojcik
et al. 2018).
The genetic content of such arrays was facilitated by the development of
technology that allows for large-scale sequencing efforts in combination with the
sequencing of the human genome (Lander et al. 2001; Venter et al. 2001). Beginning
in 2002, the International Haplotype Map (HapMap) Project undertook an effort
to catalog the common genetic variants that occur in human beings. It was also
determined that a substantial portion of this variation can be efficiently captured by a
subset of “tag” SNPs via the phenomenon of LD among neighboring SNPs (Daly et
al. 2001; Gabriel et al. 2002; International HapMap Consortium 2003; International
HapMap Consortium 2005) and that this structure varies across ancestral populations (International HapMap Consortium 2003; International HapMap Consortium
2005; Frazer et al. 2007).
Unlike candidate gene studies, GWAS are not hypothesis-driven; they do not
require a priori specification of the genes or polymorphisms that are conjectured to
be associated with the phenotype of interest. Rather, they quantify DNA sequence
variations from across the entire human genome in an attempt to pinpoint genetic
risk factors for common diseases. In designing an array for genome-wide assessment, a primary objective should thus be to capture as much common variation in
the human genome as possible.
5.2.2.2 Multistage Study Designs
When GWAS first became popular, the high cost of SNP arrays and necessity for
large sample sizes (to achieve sufficient statistical power to detect the anticipated
modest associations among hundreds of thousands of SNPs) (Witte et al. 2000)
motivated the development and use of multistage GWAS designs (Thomas et al.
2005). Decreasing SNP array costs have made multistage designs for GWAS less
essential, but we choose to briefly describe them for two reasons: (1) an overview
is important for understanding historical studies, and (2) as we move into the
post-GWAS era, next-generation sequencing of entire genomes may be sufficiently
expensive to once again make multistage designs relevant.
In the initial discovery stage of a multistage design, a subset of the study sample
is genotyped using genome-wide SNP arrays. Then the most strongly associated