Strategy to control type I error increases power to identify genetic variation using the full biological trajectory.

Genome-wide association studies have been successful in identifying loci that underlie continuous traits measured at a single time point. To additionally consider continuous traits longitudinally, it is desirable to look at SNP effects at baseline and over time using linear-mixed effects models. Estimation and interpretation of two coefficients in the same model raises concern regarding the optimal control of type I error. To investigate this issue, we calculate type I error and power under an alternative for joint tests, including the two degree of freedom likelihood ratio test, and compare this to single degree of freedom tests for each effect separately at varying alpha levels. We show which joint tests are the optimal way to control the type I error and also illustrate that information can be gained by joint testing in situations where either or both SNP effects are underpowered. We also show that closed form power calculations can approximate simulated power for the case of balanced data, provide reasonable approximations for imbalanced data, but overestimate power for complicated residual error structures. We conclude that a two degree of freedom test is an attractive strategy in a hypothesis-free genome-wide setting and recommend its use for genome-wide studies employing linear-mixed effects models.

IL-6 gene variation is not associated with increased serum levels of IL-6, muscle, weakness, or frailty in older women.

Elevated levels of the inflammatory cytokine IL-6 are associated with the development of disability, frailty, and mortality in older adults. These outcomes are likely mediated through inflammatory activity that alters hormones, skeletal muscle, and the immune system. Polymorphic variants in the IL-6 gene influence IL-6 expression. We hypothesized that IL-6 alleles associate with increased serum of IL-6, decreased muscle strength, and frailty, and tested this in the Women's Health and Aging cohorts. We genotyped 463 participants age 70-79, and identified three common IL-6 haplotype blocks for the Caucasian (n=363) and African American (n=100) subsets. Using linear and logistic regression, and adjusting for age, BMI, race, and osteoarthritis, we identified no significant or clinically meaningful relationship between any single IL-6 single nucleotide polymorphism (SNP) or any IL-6 haplotype and serum IL-6 level, grip, knee, or hip strength, or frailty. Given that the promoter SNP (rs1800795) has been reported to influence IL-6 levels and health outcomes, we performed a similar association study in the In Chianti population (n=266) and confirmed lack of association. These results suggest that IL-6 gene variation may not be an important factor in the determination of elevated IL-6 levels and related phenotypes found in older women.

Paternal transmission of the very common class I INS VNTR alleles predisposes to childhood obesity.

To identify some of the genetic factors that contribute to obesity in children of Central European and North African descent, we studied the parental transmission of alleles at the insulin locus to offspring with early-onset obesity. A variable nucleotide tandem repeat (VNTR) polymorphism upstream of the insulin gene (INS) is associated with variations in the expression of INS and the nearby gene encoding insulin-like growth factor 2 (IGF2). We found an excess of paternal transmission of class I VNTR alleles to obese children: children who inherited a class I allele from their father (but not those inheriting it from their mother) had a relative risk of early-onset obesity of 1.8. Due to the frequency of class I alleles in this population, this risk concerns 65-70% of all infants. These results suggest that increased in utero expression of paternal INS or IGF2 due to the class I INS VNTR allele may predispose offspring to postnatal fat deposition.

Genetic analysis of case/control data using estimated haplotype frequencies: application to APOE locus variation and Alzheimer's disease.

There is growing debate over the utility of multiple locus association analyses in the identification of genomic regions harboring sequence variants that influence common complex traits such as hypertension and diabetes. Much of this debate concerns the manner in which one can use the genotypic information from individuals gathered in simple sampling frameworks, such as the case/control designs, to actually assess the association between alleles in a particular genomic region and a trait. In this paper we describe methods for testing associations between estimated haplotype frequencies derived from multilocus genotype data and disease endpoints assuming a simple case/control sampling design. These proposed methods overcome the lack of phase information usually associated with samples of unrelated individuals and provide a comprehensive way of assessing the relationship between sequence or multiple-site variation and traits and diseases within populations. We applied the proposed methods in a study of the relationship between polymorphisms within the APOE gene region and Alzheimer's disease. Cases and controls for this study were collected from the United States and France. Our results confirm the known association between the APOE locus and Alzheimer's disease, even when the epsilon 4 polymorphism is not contained in the tested haplotypes. This suggests that, in certain situations, haplotype information and linkage disequilibrium-induced associations between polymorphic loci that neighbor loci harboring functional sequence variants can be exploited to identify disease-predisposing alleles in large, freely mixing populations via estimated haplotype frequency methods.

The future of genetic case-control studies.

The case-control study design has been a veritable workhorse in epidemiological research since its inception and acceptance as a valid and valued field of inquiry. The reasons for this owe to the simplicity of the required sampling and the (potential) ease of analysis and interpretation of results. Unfortunately, there are a number of problems that plague the use of the case-control design in assessing relationships between genetic variation and disease susceptibility in the population at large. Many of these problems are entirely analogous to problems that inhere in applications of the case-control design in nongenetic settings. These problems include stratification, the assessment of statistical significance, heterogeneity, and the interpretation of multiple outcomes or phenotypic information. In this chapter we describe 10 problems thought to plague genetic case-control studies and offer potential solutions to each. Many of our proposed solutions require the use of multiple DNA markers to accommodate the genetic background of the individuals sampled as cases and controls. It is hoped that our discussions and proposals will spark further debate about the analysis and ultimate utility of the case-control study in genetic epidemiology research.

The insulin gene VNTR is associated with fasting insulin levels and development of juvenile obesity.

In millions of people, obesity leads to type 2 diabetes (T2D; also known as non-insulin-dependent diabetes mellitus). During the early stages of juvenile obesity, the increase of insulin secretion in proportion to accumulated fat balances insulin resistance and protects patients from hyperglycaemia. After several decades, however,beta-cell function deteriorates and T2D develops in approximately 20% of obese patients. In modern societies, obesity has thus become the leading risk factor for T2D (ref. 5). The factors that predispose obese patients to alteration of insulin secretion upon gaining weight remain unknown. To determine which genetic factors predispose obese patients to beta-cell dysfunction, and possibly T2D, we studied single-nucleotide polymorphisms (SNPs) in the region of the insulin gene (INS) among 615 obese children. We found that, in the early phase of obesity, alleles of the INS variable number of tandem repeat (VNTR) locus are associated with different effects of body fatness on insulin secretion. Young obese patients homozygous for class I VNTR alleles secrete more insulin than those with other genotypes.

Single nucleotide polymorphisms and the future of genetic epidemiology.

In this review, we consider the motivation behind contemporary single nucleotide polymorphism (SNP) initiatives. Many of these initiatives are projected to involve large, population-based surveys. We therefore emphasize the utility of SNPs for genetic epidemiology studies. We start by offering an overview of genetic polymorphism and discuss the historical use of polymorphism in the identification of disease-predisposing genes via meiotic mapping. We next consider some of the unique aspects of SNPs, and their relative advantages and disadvantages in human population-based analyses. In this context, we describe and critique the following six different areas of application for SNP technologies: Gene discovery and mapping. Association-based candidate polymorphism testing. Diagnostics and risk profiling. Prediction of response to environmental stimuli, xenobiotics and diet. Homogeneity testing and epidemiological study design. Physiologic genomics. We focus on key issues within each of these areas in an effort to point out potential problems that might plague the use of SNPs (or other forms of polymorphism) within them. However, we make no claim that our list of considerations are exhaustive. Rather, we believe that they may provide a starting point for further dialog about the ultimate utility of SNP technologies. In addition, although our emphasis is placed on applications of SNPs to the understanding of human phenotypes, we acknowledge that SNP maps and technologies applied to other species (e.g. the mouse genome, pathogen genomes, plant genomes, etc.) are also of tremendous interest.

Linkage disequilibrium analysis of biallelic DNA markers, human quantitative trait loci, and threshold-defined case and control subjects.

Linkage disequilibrium (LD) mapping has been applied to many simple, monogenic, overtly Mendelian human traits, with great success. However, extensions and applications of LD mapping approaches to more complex human quantitative traits have not been straightforward. In this article, we consider the analysis of biallelic DNA marker loci and human quantitative trait loci in settings that involve sampling individuals from opposite ends of the trait distribution. The purpose of this sampling strategy is to enrich samples for individuals likely to possess (and not possess) trait-influencing alleles. Simple statistical models for detecting LD between a trait-influencing allele and neighboring marker alleles are derived that make use of this sampling scheme. The power of the proposed method is investigated analytically for some hypothetical gene-effect scenarios. Our studies indicate that LD mapping of loci influencing human quantitative trait variation should be possible in certain settings. Finally, we consider possible extensions of the proposed methods, as well as areas for further consideration and improvement.

Accuracy of haplotype frequency estimation for biallelic loci, via the expectation-maximization algorithm for unphased diploid genotype data.

Haplotype analyses have become increasingly common in genetic studies of human disease because of their ability to identify unique chromosomal segments likely to harbor disease-predisposing genes. The study of haplotypes is also used to investigate many population processes, such as migration and immigration rates, linkage-disequilibrium strength, and the relatedness of populations. Unfortunately, many haplotype-analysis methods require phase information that can be difficult to obtain from samples of nonhaploid species. There are, however, strategies for estimating haplotype frequencies from unphased diploid genotype data collected on a sample of individuals that make use of the expectation-maximization (EM) algorithm to overcome the missing phase information. The accuracy of such strategies, compared with other phase-determination methods, must be assessed before their use can be advocated. In this study, we consider and explore sources of error between EM-derived haplotype frequency estimates and their population parameters, noting that much of this error is due to sampling error, which is inherent in all studies, even when phase can be determined. In light of this, we focus on the additional error between haplotype frequencies within a sample data set and EM-derived haplotype frequency estimates incurred by the estimation procedure. We assess the accuracy of haplotype frequency estimation as a function of a number of factors, including sample size, number of loci studied, allele frequencies, and locus-specific allelic departures from Hardy-Weinberg and linkage equilibrium. We point out the relative impacts of sampling error and estimation error, calling attention to the pronounced accuracy of EM estimates once sampling error has been accounted for. We also suggest that many factors that may influence accuracy can be assessed empirically within a data set-a fact that can be used to create "diagnostics" that a user can turn to for assessing potential inaccuracies in estimation.

No genetic association between polymorphisms in the Tau gene and Alzheimer's disease in clinic or population based samples.

Mutations in the tau protein gene have recently been found to cause familial fronto-temporal dementia in a number of kindreds demonstrating linkage to chromosome 17. Given that tau pathology is a hallmark of Alzheimer's disease (AD), this raises the possibility that mutations in tau may also be associated with AD. We have investigated the allelic frequencies of polymorphisms in the Tau gene for a possible allelic distortion in Alzheimer's cases, which might suggest a conferred genetic risk. We have genotyped 65 community-based and 200 clinic-based AD cases, and 142 community-based controls at the Tau exon 6 AflIII and BslI polymorphisms and find no independent association with risk for AD in these samples. Further analysis including APOE genotypes from the same samples demonstrated no interaction between either of these polymorphisms and APOE in conferring risk for AD. In addition, haplotype analysis across both sites revealed no difference in haplotype frequencies between cases and controls, nor any interaction with APOE. Therefore our data do not suggest any association between these variations in the Tau gene and Alzheimer's disease.

The -491A/T apolipoprotein E promoter polymorphism association with Alzheimer's disease: independent risk and linkage disequilibrium with the known APOE polymorphism.

The epsilon4 allele of the apolipoprotein E gene (APOE) has repeatedly been associated with increased risk for Alzheimer's disease (AD). Bullido and colleagues recently identified a polymorphism in the promoter region of the APOE gene (-491A/T) and found that -491A homozygosity predicted AD independently of APOE epsilon4. Since the -491A/T polymorphism and the known APOE polymorphism must be in tight linkage disequilibrium, and the later polymorphism is know to be associated with the disease, we wished to determine to what extent this linkage disequilibrium explained the -491A/T polymorphism association with Alzheimer's disease. We genotyped a community-based control sample (n = 132) and a clinic-based Alzheimer's disease sample (n = 190) for the known APOE and -491A/T polymorphisms, and find that, while the -491A/T polymorphism confers some independent risk for AD, linkage disequilibrium between the known APOE and -491A/T polymorphic sites explains most of the -491A association. Furthermore, when considering the known APOE and -491A/T polymorphisms alone, APOE epsilon4 status is the best predictor of the disease.

The butyrylcholinesterase gene is neither independently nor synergistically associated with late-onset AD in clinic- and community-based populations.

The K variant of the butyrylcholinesterase gene (BChE) was recently found to occur at an increased frequency in a late onset Alzheimer's disease (AD) population, specifically in individuals carrying the epsilon4 allele of the apolipoprotein E (APOE) gene. This suggested synergy between these two genes resulting in an increased risk of late-onset AD. We have genotyped 62 community-based and 329 clinic-based AD cases, and 201 community-based controls at BChE and APOE and find no independent association between BChE and AD nor interaction with APOE in risk for AD in either our clinic or community-based samples.

No association between the low density lipoprotein receptor-related protein (LRP) gene and late-onset Alzheimer's disease in a community-based sample.

It is now commonly known that possession of the epsilon4 allele of the apolipoprotein E (APOE) gene confers an increased risk for both familial and sporadic Alzheimer's disease (AD), in a dose-dependent way. Other genes that may play a role in AD, either through independent association with the disease or through modification of the existing APOE risk, have been reported with conflicting results. One such gene, the low density lipoprotein receptor-related protein (LRP) gene, was recently reported by two groups to be associated with AD, although the groups identified different risk-conferring alleles. Both studies were based on clinic-derived AD populations (one American, one French), and both reported only marginally significant results. We have genotyped a community-based AD and control population at this LRP polymorphism and find no association between the variants at that polymorphism and the occurrence of AD. Further, despite the biochemical relationship between LRP and the ApoE protein, we find no significant statistical interaction between the alleles at these loci.

APOE is linked to Alzheimer's disease in a large pedigree.

Although previous association studies have demonstrated that the APOE4 allele is a risk factor for Alzheimer's disease (AD), its value for the prediction of AD in individuals is <100%. The limited predictive value of epsilon4 is also seen in multiply affected families where the epsilon4 allele is not tightly linked to AD. We analyzed a large pedigree multiply affected with AD by lod score linkage analysis at the known loci associated with AD. In this pedigree, the APOE/APOCI gene area was linked to the development of AD, while no linkage was detected to any of the other loci known to be associated with the disease. In this family, then, the inheritance of an epsilon4 allele is highly associated with the early development of the disease (mean age of onset, 62 years), and is a good predictor of disease. However, given the wealth of evidence for association, but not linkage, of APOE4 to AD, we believe this finding suggests that another factor (or factors) interact(s) with APOE to precipitate early disease, and produce positive linkage results. The nature of this factor presently remains unknown.

No interaction between the APOE and the alpha-1-antichymotrypsin genes on risk for Alzheimer's disease.

It is now known that possession of one of the three common forms of the apolipoprotein E gene (allele epsilon 4) confers an increased risk for Alzheimer's disease (AD), both familial and sporadic, and that this risk is dose-dependent. Other genes that may play a role in AD, either through independent association with the disease or through modification of, or interaction with, the existing apolipoprotein E (APOE) risk, are now under investigation including the alpha-1-antichymotrypsin (ACT) gene, the very low density lipoprotein receptor (VLDL-R) gene, and the presenilin-1 (PS-1) gene. Kamboh et al. [1995] reported that a polymorphism in the alpha-1-antichymotrypsin gene could modify the risk for AD conferred by the APOE locus, specifically by increasing the risk for AD among epsilon 4 homozygotes. The ACT gene, which is found on chromosome 14, has previously been proposed as a candidate for AD due to the presence of the ACT protein in senile plaques and the reported elevation of the protein in the cerebro-spinal fluid (CSF) and serum of AD cases. We have investigated this reported association within our familial and sporadic AD dataset, where we find no independent association between ACT and the occurrence of AD. Logistic regression analysis excludes ACT or the interaction between ACT and APOE as significant contributors in the prediction of disease status. By this analysis, ACT genotyping does not provide additional information about an individual's risk of Alzheimer's disease beyond the risk information conferred by APOE genotype alone.

No association between the intronic presenilin-1 polymorphism and Alzheimer's disease in clinic and population-based samples.

Mutations in the Presenilin 1 (PS1) gene on chromosome 14 cause most early-onset familial Alzheimer's disease (AD). An intronic polymorphism in the PS1 gene was recently identified and reported to be associated with late-onset AD [Wragg et al., Lancet 347: 509-512, 1996]. The authors found an excess of homozygotes for the more common allele (allele 1) in AD cases, associated with an approximate doubling of risk. In the present study, we report the PS1 polymorphism distributions in clinic and population-based samples. We were not able to replicate the findings of Wragg et al. [1996]. Our results are consistent with those of other researchers and do not support the conclusion that the PS1 polymorphism is associated with late-onset AD.

No association between the very low density lipoprotein receptor gene and late-onset Alzheimer's disease nor interaction with the apolipoprotein E gene in population-based and clinic samples.

It is now commonly known that possession of one of the three common alleles of the apolipoprotein E (APOE) gene (allele epsilon 4) confers an increased risk for both familial and sporadic Alzheimer's disease (AD), and that this risk is dose-dependent. Other genes that may play a role in AD, either through independent association with the disease or through modification of the existing APOE risk, are under investigation. One such gene, the very low density lipoprotein receptor (VLDL-R) gene, was reported by Okuizumi et al. to be independently associated with AD in a Japanese population, but not interactive with the APOE4 conferred risk. Their clinic-based data set demonstrated a 2-fold increased risk conferred by the 5-repeat allele of a polymorphism in VLDL-R. As recruitment from a clinic rather than a population-based sample may result in a distortion of allele frequencies, as has been shown with APOE allele frequencies, it is important to investigate this association in a population-based study. We have genotyped both population and clinic-based AD data sets at this VLDL-R polymorphism, and we find no independent association between the VLDL-R gene and the occurrence of AD in either sample. Further, despite the biochemical relationship between the VLDL-R and APOE proteins, we find no significant statistical interaction between the alleles at these loci.

Familial and population-based studies of apolipoprotein E and Alzheimer's disease.

The finding of an association between the epsilon 4 allele of the APOE locus and the early expression of late-onset Alzheimer's disease (AD) is robust. However, the estimates of the proportion of AD cases carrying one or more copies of the epsilon 4 allele vary dramatically between studies (highest estimates being 180% of lowest ones). Here we compare the results of association studies in samples drawn from an epidemiologically based study design and samples drawn from families selected for linkage studies. The significant differences between results probably point to the unwitting selection of familial factors other than the APOE locus in the family history positive samples. We conclude that any selection procedure tending to enrich samples for positive family history will also tend to artificially increase APOE epsilon 4 allele frequencies in probands. This is of significance in samples drawn from clinical settings where referral may be influenced by previous known family history. Further work is needed to clarify the nature of the additional factors operating within families. We also report data showing an association between late-onset AD and a polymorphism in an adjacent locus to APOE-the APOCI locus. As no additional risk for AD can be attributed to the APOCI locus, the most likely explanation for the association between AD and APOCI is the disequilibrium between the APOCI and APOE loci. Therefore, there are likely to be other genetic markers in the area that can be used in the same way as APOE as a marker of risk for the disease.