7 What Have We Learned from GWAS?
161
genome-wide approach has clearly added value in the analysis of genetic data and
the understanding of complex traits.
Based on the studies that have been initiated, there are a number of clear directions to continue the process of uncovering the architectural organization, biological
insight, and evolutionary history of the genes and pathways that contribute to these
traits. Given some of the clues suggested by GWAS that have yet to be fully and
systematically enumerated, I conclude with a description of several future research
threads that might be anticipated in the coming years.
7.2
Lessons over Eight Years of Association Studies
(2006–2013)
Looking back to the discussion leading up to the first published genome-wide
associations for traits in 2006/2007 (The Diabetes Genetics Initiative 2007; Wellcome Trust Case Control Consortium 2007), one can be reminded of the uncertainty and subsequent debate waged around the scientific merit and technical
feasibility of these studies. (Interested readers can consult a prescient review
with the rationale and issues of the subject written just before this time by
Hirschhorn and Daly (2005) or a historical perspective from Bodmer and Bonilla
(2008).)
While the design of such studies appeared tractable, concerns surrounding the
details of how such studies would be technically implemented were raised. A first
intellectual hurdle was overcome in 1999, with a prediction based on empirical
observation and simulations that the selection of as many as ∼500,000 wellchosen markers would be required to survey common variation throughout the
entire human genome (Kruglyak 1999); a number of SNPs were within the costs
of the technologies used to ascertain them. However, the scope of such experiments
raised concerns about the technical capability of genotyping technologies to provide
accurate and unbiased genotypes in that number of markers in thousands of
people. Further still, concerns about the appropriate statistical thresholds by which
significance could be declared, how observations would be replicated and validated,
and if studies that perform statistical analysis accounting for effects known to induce
confounding (Pritchard and Rosenberg 1999) would work in practice. Today, the
application of these analyses and technical quality control of data produced by these
technologies are routine, owing to the careful and systematic groundwork laid down
by researchers that contributed to the first studies published on the subject.
A further point of the discussion focused on the question if the architecture of
common complex traits and diseases could even be dissected by such a design.
Proponents who argued in favor hypothesized that common genetic variants likely
contribute to common disease, motivating the international Hapmap Project (International HapMap Consortium 2005, 2007; International HapMap 3 Consortium
2010) that formed the basis of the design of array technologies that facilitate
high-throughput genetics underpinning GWAS. One alternative model to this view
suggested that common disease could indeed be influenced by rare variants rather
161
genome-wide approach has clearly added value in the analysis of genetic data and
the understanding of complex traits.
Based on the studies that have been initiated, there are a number of clear directions to continue the process of uncovering the architectural organization, biological
insight, and evolutionary history of the genes and pathways that contribute to these
traits. Given some of the clues suggested by GWAS that have yet to be fully and
systematically enumerated, I conclude with a description of several future research
threads that might be anticipated in the coming years.
7.2
Lessons over Eight Years of Association Studies
(2006–2013)
Looking back to the discussion leading up to the first published genome-wide
associations for traits in 2006/2007 (The Diabetes Genetics Initiative 2007; Wellcome Trust Case Control Consortium 2007), one can be reminded of the uncertainty and subsequent debate waged around the scientific merit and technical
feasibility of these studies. (Interested readers can consult a prescient review
with the rationale and issues of the subject written just before this time by
Hirschhorn and Daly (2005) or a historical perspective from Bodmer and Bonilla
(2008).)
While the design of such studies appeared tractable, concerns surrounding the
details of how such studies would be technically implemented were raised. A first
intellectual hurdle was overcome in 1999, with a prediction based on empirical
observation and simulations that the selection of as many as ∼500,000 wellchosen markers would be required to survey common variation throughout the
entire human genome (Kruglyak 1999); a number of SNPs were within the costs
of the technologies used to ascertain them. However, the scope of such experiments
raised concerns about the technical capability of genotyping technologies to provide
accurate and unbiased genotypes in that number of markers in thousands of
people. Further still, concerns about the appropriate statistical thresholds by which
significance could be declared, how observations would be replicated and validated,
and if studies that perform statistical analysis accounting for effects known to induce
confounding (Pritchard and Rosenberg 1999) would work in practice. Today, the
application of these analyses and technical quality control of data produced by these
technologies are routine, owing to the careful and systematic groundwork laid down
by researchers that contributed to the first studies published on the subject.
A further point of the discussion focused on the question if the architecture of
common complex traits and diseases could even be dissected by such a design.
Proponents who argued in favor hypothesized that common genetic variants likely
contribute to common disease, motivating the international Hapmap Project (International HapMap Consortium 2005, 2007; International HapMap 3 Consortium
2010) that formed the basis of the design of array technologies that facilitate
high-throughput genetics underpinning GWAS. One alternative model to this view
suggested that common disease could indeed be influenced by rare variants rather