Post-GWAS: where next? More samples, more SNPs or more biology?

The power of genome-wide association studies (GWAS) rests on several foundations: (i) there is a significant amount of additive genetic variation, (ii) individual causal polymorphisms often have sizable effects and (iii) they segregate at moderate-to-intermediate frequencies, or will be effectively 'tagged' by polymorphisms that do. Each of these assumptions has recently been questioned. (i) Why should genetic variation appear additive given that the underlying molecular networks are highly nonlinear? (ii) A new generation of relatedness-based analyses directs us back to the nearly infinitesimal model for effect sizes that quantitative genetics was long based upon. (iii) Larger effect causal polymorphisms are often low frequency, as selection might lead us to expect. Here, we review these issues and other findings that appear to question many of the foundations of the optimism GWAS prompted. We then present a roadmap emerging as one possible future for quantitative genetics. We argue that in future GWAS should move beyond purely statistical grounds. One promising approach is to build upon the combination of population genetic models and molecular biological knowledge. This combined treatment, however, requires fitting experimental data to models that are very complex, as well as accurate capturing of the uncertainty of resulting inference. This problem can be resolved through Bayesian analysis and tools such as approximate Bayesian computation-a method growing in popularity in population genetic analysis. We show a case example of anterior-posterior segmentation in Drosophila, and argue that similar approaches will be helpful as a GWAS augmentation, in human and agricultural research.

The gene expression profile of cardia intestinal metaplasia is similar to that of Barrett's esophagus, not gastric intestinal metaplasia.

The etiology and significance of cardia intestinal metaplasia (CIM) is disputed. CIM may represent a form of Barrett's esophagus due to reflux or could reflect generalized gastric intestinal metaplasia due to Helicobacter pylori. The aim of this study was to utilize gene expression data to compare CIM to Barrett's and gastric intestinal metaplasia. Endoscopic biopsies were classified by endoscopic and histologic criteria as CIM (n= 33), Barrett's (n= 25), or gastric intestinal metaplasia of the antrum or body (n= 18). The squamocolumnar and gastroesophageal junctions were aligned in CIM patients and patients with diffuse gastric intestinal metaplasia were excluded. H. pylori was tested for in the biopsies of all patients. After laser-capture microdissection, quantitative reverse transcription-polymerase chain reaction (RT-PCR) was used to measure the mRNA expression of a panel of nine genes that has been shown to differentiate Barrett's from other foregut mucosa. Cluster analysis with linear discriminant analysis of the expression data was used to classify each sample into groups based solely on similarity of gene expression. Cluster analysis was performed for three groups (CIM vs. Barrett's vs. gastric intestinal metaplasia) and two groups (CIM + Barrett's vs. gastric intestinal metaplasia). There was no difference in H. pylori infection among groups (P= 0.66). Clustering into three groups resulted in frequent misclassification between CIM and Barrett's while misclassification of gastric intestinal metaplasia was uncommon. The CIM and Barrett's groups were then combined for two group clustering and linear discriminant analysis correctly predicted 95% of CIM and Barrett's samples and 83% of gastric intestinal metaplasia samples based on gene expression alone. In conclusion, the gene expression profiles of CIM and Barrett's esophagus were similar in 95% of biopsies and differed significantly from that of gastric intestinal metaplasia. The indistinguishable gene expression profile of CIM and BE suggests that they may share a common etiology in the majority of patients with a similar biology, and calls into question the perception that CIM is an innocuous process.

The effects of rate variation on ancestral inference in the coalescent.

We describe a Markov chain Monte Carlo approach for assessing the role of site-to-site rate variation in the analysis of within-population samples of DNA sequences using the coalescent. Our framework is a Bayesian one. We discuss methods for assessing the goodness-of-fit of these models, as well as problems concerning the separate estimation of effective population size and mutation rate. Using a mitochondrial data set for illustration, we show that ancestral inference concerning coalescence times can be dramatically affected if rate variation is ignored.

The age of a unique event polymorphism.

We develop a Markov chain Monte Carlo approach for estimating the distribution of the age of a mutation that is assumed to have arisen just once in the history of the population of interest. We assume that in addition to the presence or absence of this mutation in a sample of chromosomes, we have DNA sequence data from a region completely linked to the mutant site. We apply our method to a mitochondrial data set in which the DNA sequence data come from hypervariable region I and the mutation of interest is the 9-bp region V deletion.

Analysis of germline mutation spectra at the Huntington's disease locus supports a mitotic mutation mechanism.

Trinucleotide repeat disease alleles can undergo 'dynamic' mutations in which repeat number may change when a gene is transmitted from parent to offspring. By typing >3500 sperm, we determined the size distribution of Huntington's disease (HD) germline mutations produced by 26 individuals from the Venezuelan cohort with CAG/CTG repeat numbers ranging from 37 to 62. Both the mutation frequency and mean change in allele size increased with increasing somatic repeat number. The mutation frequencies averaged 82% and, for individuals with at least 50 repeats, 98%. The extraordinarily high mutation frequency levels are most consistent with a mutation process that occurs throughout germline mitotic divisions, rather than resulting from a single meiotic event. In several cases, the mean change in repeat number differed significantly among individuals with similar somatic allele sizes. This individual variation could not be attributed to age in a simple way or to ' cis ' sequences, suggesting the influence of genetic background or other factors. A familial effect is suggested in one family where both the father and son gave highly unusual spectra compared with other individuals matched for age and repeat number. A statistical model based on incomplete processing of Okazaki fragments during DNA replication was found to provide an excellent fit to the data but variation in parameter values among individuals suggests that the molecular mechanism might be more complex.

Ancestral inference from samples of DNA sequences with recombination.

The sampling distribution of a collection of DNA sequences is studied under a model where recombination can occur in the ancestry of the sequences. The infinitely-many-sites model of mutation is assumed where there may only be one mutation at a given site. Ancestral inference procedures are discussed for: estimating recombination and mutation rates; estimating the times to the most recent common ancestors along the sequences; estimating ages of mutations; and estimating the number of recombination events in the ancestry of the sample. Inferences are made conditional on the configuration of the pattern of mutations at sites in observed sample sequences. A computational algorithm based on a Markov chain simulation is developed, implemented, and illustrated with examples for these inference procedures. This algorithm is very computationally intensive.

Pairwise comparisons of mitochondrial DNA sequences in subdivided populations and implications for early human evolution.

We consider the effect on the distribution of pairwise differences between mitochondrial DNA sequences of the incorporation into the underlying population genetics model of two particular effects that seem realistic for human populations. The first is that the population size was roughly constant before growing to its current level. The second is that the population is geographically subdivided rather than panmictic. In each case these features tend to encourage multimodal distributions of pairwise differences, in contrast to existing, unimodal datasets. We argue that population genetics models currently used to analyze such data may thus fail to reflect important features of human mitochondrial DNA evolution. These may include selection on the mitochondrial genome, more realistic mutation mechanisms, or special population or migration dynamics. Particularly in view of the variability inherent in the single available human mitochondrial genealogy, it is argued that until these effects are better understood, inferences from such data should be rather cautious.