The complex genetics of multiple sclerosis: pitfalls and prospects

Abstract

The genetics of complex disease is entering a new and exciting era. The exponentially growing knowledge and technological capabilities emerging from the human genome project have finally reached the point where relevant genes can be readily and affordably identified. As a result, the last 12 months has seen a virtual explosion in new knowledge with reports of unequivocal association to relevant genes appearing almost weekly. The impact of these new discoveries in Neuroscience is incalculable at this stage but potentially revolutionary. In this review, an attempt is made to illuminate some of the mysteries surrounding complex genetics. Although focused almost exclusively on multiple sclerosis all the points made are essentially generic and apply equally well, with relatively minor addendums, to any other complex trait, neurological or otherwise.

Fig. 1

Posterior Odds that a result is true, assuming risk alleles with a frequency of 10% and a Genotype Relative Risk (GRR) of 1.3 and a multiplicative model. This figure indicates the posterior odds that a result is true (plotted on a log scale on the y-axis) against the significance of the result (plotted as -log of the P-value on the x-axis). Five sample sizes are listed in the legend, in each the number of cases and controls are equal, the 200 line thus indicates the posterior odds for a study involving 200 cases and 200 controls and so on. Power was calculated using the on-line genetic power calculator (Purcell et al., 2003).

Fig. 2

Posterior Odds that a result is true, assuming risk alleles with a frequency of 10%, a GRR of 1.2 and a multiplicative model. The axes and samples sizes are as in Fig. 1. Power was calculated using the on-line genetic power calculator (Purcell et al., 2003).

Fig. 3

Required sample size as a function of the GRR conferred by susceptibility allele. The sample size required is plotted on a log scale. The sample size indicates the number of cases required to ensure that results with a P-value of <5 × 10⁻⁷ are twice as likely to be true as false, assuming an equal number of controls, a multiplicative model and risk allele frequency of 10%. Sample sizes were calculated using the on-line genetic power calculator (Purcell et al., 2003).

Fig. 4

Posterior odds that a result is true, assuming candidate risk alleles with a frequency of 10%, a GRR of 1.2, a multiplicative model and prior odds of 1000 : 1. The axes and samples sizes are as in Fig. 1. Power was calculated using the on-line genetic power calculator (Purcell et al., 2003).

Fig. 5

Influence of risk allele frequency (RAF) on power to identify significant association (P-value <5 × 10⁻⁷). Power was calculated under the assumption that the susceptibility alleles have a GRR of 1.3 and a multiplicative model. Sample sizes are indicated in the legend. Power was calculated using the on-line genetic power calculator (Purcell et al., 2003).

Fig. 6

Influence of phenocopy rate on power to identify significant association (P-value < 5 × 10⁻⁷). The phenocopy rate indicates the proportion of cases which have been misdiagnosed as having the disease when in fact they are controls. A phenocopy rate of 25% thus indicates that 1 in 4 of the cases is a misdiagnosis (or heterogeneity). Since the prevalence of multiple sclerosis is 1 per 1000 the impact of using unselected controls is minimal; for example in a cohort of 2000 unselected controls only two would be expected to be cases and thus the impact on power and posterior odds is imperceptible. Power was calculated under the assumption that the susceptibility alleles have a GRR of 1.3 and a multiplicative model. Sample sizes are indicated in the legend, note these are not all the same as in earlier figures. Power was calculated using the on-line power for association with error (PAWE) calculator (Gordon et al., ; Edwards et al., 2005).

Fig. 7

Influence of linkage disequilibrium (LD) between the tested and causative variant on the power to identify significant association (P-value <5 × 10⁻⁷). There is a simple relationship between the extent of linkage disequilibrium and effective sample size (Wall and Pritchard, 2003), such that the product of r² (see Supplementary material section 2) and actual sample size indicates the sample size which would have yielded the same power if the causative variant had been directly typed. This simple relationship was used to calculate the effective sample size at each level of r² and thereby calculate the power recorded in the figure. In these calculations, the susceptibility allele was assumed to have a frequency of 10%, a GRR of 1.3 and a multiplicative model. Actual samples sizes in which the test variant is typed are listed in the legend. Power was calculated using the on-line genetic power calculator (Purcell et al., 2003).

Fig. 8

Expected P-value in follow-up studies of rs6897932, the IL7R-associated SNP. The red line indicates the expected P-value and the dotted lines the 95% confidence intervals on this estimate (plotted as the negative log). It can thus be expected that 95% of the time the observed P-value with fall in this space. The blue dots indicate the studies already reported concerning this locus (the first two studies did not consider this variant directly but it is expected that the observed signal was due to LD with rs6897932). From left to right the studies are Teutsch (2003), Zhang (2005), IMSGC (screening phase) (2007), WTCCC (2007), Lundmark (2007) and Gregory (2007). Expected P-values and confidence intervals were calculated using the on-line genetic power calculator (Purcell et al., 2003).