Automated detection of informative combined effects in genetic association studies of complex traits

Identification of secreted phosphoprotein 1 gene as a new rheumatoid arthritis susceptibility gene

OBJECTIVE

To evaluate the contribution of the SPP1 rs11439060 and rs9138 polymorphisms, previously reported as autoimmune risk variants, in the rheumatoid arthritis (RA) genetic background according to anti-citrullinated protein antibodies (ACPAs) status of RA individuals.

METHODS

We analysed a total of 11,715 RA cases and 26,493 controls from nine independent cohorts; all individuals were genotyped or had imputed genotypes for SPP1 rs11439060 and rs9138. The effect of the SPP1 rs11439060 and rs9138 risk-allele combination on osteopontin (OPN) expression in macrophages and OPN serum levels was investigated.

RESULTS

We provide evidence for a distinct contribution of SPP1 to RA susceptibility according to ACPA status: the combination of ≥3 SPP1 rs11439060 and rs9138 common alleles was associated mainly with ACPA negativity (p=1.29×10(-5), ORACPA-negative 1.257 (1.135 to 1.394)) and less with ACPA positivity (p=0.0148, ORACPA-positive 1.072 (1.014 to 1.134)). The ORs between these subgroups (ie, ACPA-positive and ACPA-negative) significantly differed (p=7.33×10(-3)). Expression quantitative trait locus analysis revealed an association of the SPP1 risk-allele combination with decreased SPP1 expression in peripheral macrophages from 599 individuals. To corroborate these findings, we found an association of the SPP1 risk-allele combination and low serum level of secreted OPN (p=0.0157), as well as serum level of secreted OPN correlated positively with ACPA production (p=0.005; r=0.483).

CONCLUSIONS

We demonstrate a significant contribution of the combination of SPP1 rs11439060 and rs9138 frequent alleles to risk of RA, the magnitude of the association being greater in patients negative for ACPAs.

Role of genetic heterogeneity and epistasis in bladder cancer susceptibility and outcome: a learning classifier system approach

BACKGROUND AND OBJECTIVE

Detecting complex patterns of association between genetic or environmental risk factors and disease risk has become an important target for epidemiological research. In particular, strategies that provide multifactor interactions or heterogeneous patterns of association can offer new insights into association studies for which traditional analytic tools have had limited success.

MATERIALS AND METHODS

To concurrently examine these phenomena, previous work has successfully considered the application of learning classifier systems (LCSs), a flexible class of evolutionary algorithms that distributes learned associations over a population of rules. Subsequent work dealt with the inherent problems of knowledge discovery and interpretation within these algorithms, allowing for the characterization of heterogeneous patterns of association. Whereas these previous advancements were evaluated using complex simulation studies, this study applied these collective works to a 'real-world' genetic epidemiology study of bladder cancer susceptibility.

RESULTS AND DISCUSSION

We replicated the identification of previously characterized factors that modify bladder cancer risk--namely, single nucleotide polymorphisms from a DNA repair gene, and smoking. Furthermore, we identified potentially heterogeneous groups of subjects characterized by distinct patterns of association. Cox proportional hazard models comparing clinical outcome variables between the cases of the two largest groups yielded a significant, meaningful difference in survival time in years (survivorship). A marginally significant difference in recurrence time was also noted. These results support the hypothesis that an LCS approach can offer greater insight into complex patterns of association.

CONCLUSIONS

This methodology appears to be well suited to the dissection of disease heterogeneity, a key component in the advancement of personalized medicine.

A faster pedigree-based generalized multifactor dimensionality reduction method for detecting gene-gene interactions

We proposed a faster pedigree-based generalized multifactor dimensionality reduction algorithm, called PedG-MDR II (PII), to detect gene-gene interactions underlying complex traits. Inherited from our previous framework of PedGMDR (PI), PII can handle both dichotomous and continuous traits in pedigree-based designs and allows for covariate adjustment. Compared with PI, this faster version can theoretically halve the computing burden and memory requirement. To evaluate the performance of PII, we performed comprehensive simulations across a wide variety of experimental scenarios, in which we considered two study designs, discordant sib pairs and mixed families of varying size, and, for each study design, we considered five common factors that may potentially affect statistical power: minor allele frequency, missing rate of parental genotypes, covariate effect, gene-gene interaction, and scheme to adjust phenotypic outcomes. Simulations showed that PII gave well controlled type I error rates against population admixture. Under a total of 4,096 scenarios simulated, PII, in general, had a higher average power than PI for both dichotomous and continuous traits, and the advantage was more pronounced for continuous traits. PII also appeared to be less sensitive than PI to changes in the other four factors than the magnitude of genetic effects considered in this study. Applied to the Mid-South Tobacco Family study, PII detected a significant interaction with a p value of 5.4 × 10(-5) between two taster receptor genes, TAS2R16 and TAS2R38, responsible for nicotine dependence. In conclusion, PII is a faster supplementary version of our previous PI for detecting multifactor interactions.

Use of pathway information in molecular epidemiology

Candidate gene studies are generally motivated by some form of pathway reasoning in the selection of genes to be studied, but seldom has the logic of the approach been carried through to the analysis. Marginal effects of polymorphisms in the selected genes, and occasionally pairwise gene-gene or gene-environment interactions, are often presented, but a unified approach to modelling the entire pathway has been lacking. In this review, a variety of approaches to this problem is considered, focusing on hypothesis-driven rather than purely exploratory methods. Empirical modelling strategies are based on hierarchical models that allow prior knowledge about the structure of the pathway and the various reactions to be included as 'prior covariates'. By contrast, mechanistic models aim to describe the reactions through a system of differential equations with rate parameters that can vary between individuals, based on their genotypes. Some ways of combining the two approaches are suggested and Bayesian model averaging methods for dealing with uncertainty about the true model form in either framework is discussed. Biomarker measurements can be incorporated into such analyses, and two-phase sampling designs stratified on some combination of disease, genes and exposures can be an efficient way of obtaining data that would be too expensive or difficult to obtain on a full candidate gene sample. The review concludes with some thoughts about potential uses of pathways in genome-wide association studies.

Identification of gene-gene interactions in the presence of missing data using the multifactor dimensionality reduction method

Gene-gene interaction is believed to play an important role in understanding complex traits. Multifactor dimensionality reduction (MDR) was proposed by Ritchie et al. [2001. Am J Hum Genet 69:138-147] to identify multiple loci that simultaneously affect disease susceptibility. Although the MDR method has been widely used to detect gene-gene interactions, few studies have been reported on MDR analysis when there are missing data. Currently, there are four approaches available in MDR analysis to handle missing data. The first approach uses only complete observations that have no missing data, which can cause a severe loss of data. The second approach is to treat missing values as an additional genotype category, but interpretation of the results may then be not clear and the conclusions may be misleading. Furthermore, it performs poorly when the missing rates are unbalanced between the case and control groups. The third approach is a simple imputation method that imputes missing genotypes as the most frequent genotype, which may also produce biased results. The fourth approach, Available, uses all data available for the given loci to increase power. In any real data analysis, it is not clear which MDR approach one should use when there are missing data. In this article, we consider a new EM Impute approach to handle missing data more appropriately. Through simulation studies, we compared the performance of the proposed EM Impute approach with the current approaches. Our results showed that Available and EM Impute approaches perform better than the three other current approaches in terms of power and precision.

Lack of association between polymorphisms of the IL18R1 and IL18RAP genes and cardiovascular risk: the MORGAM Project

Background

Interleukin-18 is a pro-inflammatory cytokine suspected to be associated with atherosclerosis and its complications. We had previously shown that one single nucleotide polymorphism (SNP) of the IL18 gene was associated with cardiovascular disease (CVD) through an interaction with smoking. As a further step for elucidating the contribution of the IL-18 pathway to the etiology of CVD, we here investigated the association between the genetic variability of two IL-18 receptor genes, IL18R1 and IL18RAP, with the risk of developing CVD.

Methods

Eleven tagging SNPs, 5 in IL18R1 and 6 in IL18RAP, characterizing the haplotypic variability of the corresponding genes; were genotyped in 5 European prospective CVD cohorts including 1416 cases and 1772 non-cases, as part of the MORGAM project. Both single-locus and haplotypes analyses were carried out to investigate the association of these SNPs with CVD.

Results

We did not find any significant differences in allele, genotype and haplotype frequencies between cases and non-cases for either of the two genes. Moreover, the search for interactions between SNPs located in different genes, including 5 IL18 SNPs previously studied in the MORGAM project, and between SNPs and environmental factors remained unfruitful.

Conclusion

Our analysis suggests that the variability of IL18R1 and IL18RAP genes are unlikely to contribute to modulate the risk of CVD.

Clique-finding for heterogeneity and multidimensionality in biomarker epidemiology research: the CHAMBER algorithm

BACKGROUND

Commonly-occurring disease etiology may involve complex combinations of genes and exposures resulting in etiologic heterogeneity. We present a computational algorithm that employs clique-finding for heterogeneity and multidimensionality in biomedical and epidemiological research (the "CHAMBER" algorithm).

METHODOLOGY/PRINCIPAL FINDINGS

This algorithm uses graph-building to (1) identify genetic variants that influence disease risk and (2) predict individuals at risk for disease based on inherited genotype. We use a set-covering algorithm to identify optimal cliques and a Boolean function that identifies etiologically heterogeneous groups of individuals. We evaluated this approach using simulated case-control genotype-disease associations involving two- and four-gene patterns. The CHAMBER algorithm correctly identified these simulated etiologies. We also used two population-based case-control studies of breast and endometrial cancer in African American and Caucasian women considering data on genotypes involved in steroid hormone metabolism. We identified novel patterns in both cancer sites that involved genes that sulfate or glucuronidate estrogens or catecholestrogens. These associations were consistent with the hypothesized biological functions of these genes. We also identified cliques representing the joint effect of multiple candidate genes in all groups, suggesting the existence of biologically plausible combinations of hormone metabolism genes in both breast and endometrial cancer in both races.

CONCLUSIONS

The CHAMBER algorithm may have utility in exploring the multifactorial etiology and etiologic heterogeneity in complex disease.

An information-theoretic analysis of genetics, gender and age in cancer patients

Germline genetics, gender and hormonal-signaling pathways are all well described modifiers of cancer risk and progression. Although an improved understanding of how germline genetic variants interact with other cancer risk factors may allow better prevention and treatment of human cancer, measuring and quantifying these interactions is challenging. In other areas of research, Information Theory has been used to quantitatively describe similar multivariate interactions. We implemented a novel information-theoretic analysis to measure the joint effect of a high frequency germline genetic variant of the p53 tumor suppressor pathway (MDM2 SNP309 T/G) and gender on clinical cancer phenotypes. This analysis quantitatively describes synergistic interactions among gender, the MDM2 SNP309 locus, and the age of onset of tumorigenesis in p53 mutation carriers. These results offer a molecular and genetic basis for the observed sexual dimorphism of cancer risk in p53 mutation carriers and a model is proposed that suggests a novel cancer prevention strategy for p53 mutation carriers.

Large recursive partitioning analysis of complex disease pharmacogenetic studies. II. Statistical considerations

Identifying genetic variations predictive of important phenotypes, such as disease susceptibility, drug efficacy, and adverse events, remains a challenging task. There are individual polymorphisms that can be tested one at a time, but there is the more difficult problem of the identification of combinations of polymorphisms or even more complex interactions of genes with environmental factors. Diseases, drug responses or side effects can result from different mechanisms. Identification of subgroups of people where there is a common mechanism is a problem for diagnosis and prescribing of treatment. Recursive partitioning (RP) is a simple statistical tool for segmenting a population into non-overlapping groups where the response of interest, disease susceptibility, drug efficacy and adverse events are more homogeneous within the segments. We suggest that the use of RP is not only more technically feasible than other search methods but it is less susceptible to multiple-testing problems. The numbers of combinations of gene-gene and gene-environment interactions is potentially astronomical and RP greatly reduces the effective search and inference space. Moreover, the certain reliance of RP on the presence of marginal effects is justifiable as was found by using analytical and numerical arguments. In the context of haplotype analysis, results suggest that the analysis of individual SNPs is likely to be successful even when susceptibilities are determined by haplotypes. Retrospective clinical studies where cases and controls are collected will be a common design. This report provides methods that can be used to adjust the RP analysis to reflect the population incidence of the response of interest. Confidence limits on the incidence of the response in the segmented subgroups are also discussed. RP is a straightforward way to create realistic subgroups, and prediction intervals for the within-subgroup disease incidence are easily obtained.

Polymorphisms in 33 inflammatory genes and risk of myocardial infarction--a system genetics approach

The hypothesis of a causal link between inflammation and atherosclerosis would be strengthened if variants of inflammatory genes were associated with disease. Polymorphisms of 33 genes encoding inflammatory molecules were tested for association with myocardial infarction (MI). Patients with MI and a parental history of MI (n = 312) and controls from the UK (n = 317) were genotyped for 162 polymorphisms. Thirteen polymorphisms were associated with MI (P values ranging from 0.003 to 0.041). For three genes, ITGB1, SELP, and TNFRSF1B haplotype frequencies differed between patients and controls (P values < 0.01). We further assessed the simultaneous contribution of all polymorphisms and relevant covariates to MI using a two-step strategy of data mining relying on Random Forest and DICE algorithms. In a replication study involving two independent samples from the UK (n = 649) and France (n = 706), one interaction between the ITGA4/R898Q polymorphism and current smoking status was replicated. This study illustrates a strategy for assessing the joint effect of a large number of polymorphisms on a phenotype that may provide information that single locus or single gene analysis may fail to uncover. Overall, there was weak evidence for an implication of inflammatory polymorphisms on susceptibility to MI.

Association Rule Discovery Has the Ability to Model Complex Genetic Effects

Dramatic advances in genotyping technology have established a need for fast, flexible analysis methods for genetic association studies. Common complex diseases, such as Parkinson's disease or multiple sclerosis, are thought to involve an interplay of multiple genes working either independently or together to influence disease risk. Also, multiple underlying traits, each its own genetic basis may be defined together as a single disease. These effects - trait heterogeneity, locus heterogeneity, and gene-gene interactions (epistasis) - contribute to the complex architecture of common genetic diseases. Association Rule Discovery (ARD) searches for frequent itemsets to identify rule-based patterns in large scale data. In this study, we apply Apriori (an ARD algorithm) to simulated genetic data with varying degrees of complexity. Apriori using information difference to prior as a rule measure shows good power to detect functional effects in simulated cases of simple trait heterogeneity, trait heterogeneity and epistasis, and moderate power in cases of trait heterogeneity and locus heterogeneity. Also, we illustrate that bootstrapping the rule induction process does not considerably improve the power to detect these effects. These results show that ARD is a framework with sufficient flexibility to characterize complex genetic effects.

A survey of data mining methods for linkage disequilibrium mapping

Data mining methods are gaining more interest as potential tools in mapping and identification of complex disease loci. The methods are well suited to large numbers of genetic marker loci produced by high-throughput laboratory analyses, but also might be useful for clarifying the phenotype definitions prior to more traditional mapping analyses. Here, the current data mining-based methods for linkage disequilibrium mapping and phenotype analyses are reviewed.

Genetic analysis of the interleukin-18 system highlights the role of the interleukin-18 gene in cardiovascular disease

BACKGROUND

Interleukin (IL)-18 plays a key role in atherosclerosis and its complications. The present study investigated the genetic variability of 4 genes of the IL-18 system-IL18, IL18R1, IL18RAP, and IL18BP-in relation to circulating IL-18 levels and cardiovascular mortality.

METHODS AND RESULTS

Twenty-two polymorphisms were genotyped in 1288 patients with coronary artery disease prospectively followed up during a median period of 5.9 years. The end point was death from cardiovascular causes (n=142). Baseline IL-18 levels were predictive of cardiovascular deaths occurring during < or =4 years of follow-up (HR=2.96, 95% CI 1.54 to 5.70, P=0.001 for the top compared with the bottom quartile) but not of later deaths. Haplotypes of the IL18 gene were associated with IL-18 levels (P=0.002) and cardiovascular mortality (P=0.006) after adjustment for cardiovascular risk factors. The same haplotype was associated with both a 9% lowering effect on IL-18 levels and a protective effect on risk (HR=0.57, 95% CI 0.36 to 0.92). IL18 haplotypes explained only 2% of IL-18 variability. Adjustment for baseline IL-18 levels abolished the association of haplotypes with cardiovascular risk. The haplotype associated with phenotypes was the only one carrying the minor allele of the IL18/A+183G polymorphism located in the 3'untranslated region and potentially affecting mRNA stability. The other genes of the system were not related to IL-18 levels or cardiovascular outcome.

CONCLUSIONS

Variations of the IL18 gene consistently influence circulating levels of IL-18 and clinical outcome in patients with coronary artery disease, which supports the hypothesis of a causal role of IL-18 in atherosclerosis and its complications.

Thematic review series: The pathogenesis of atherosclerosis. Toward a biological network for atherosclerosis

The goal of systems biology is to define all of the elements present in a given system and to create an interaction network between these components so that the behavior of the system, as a whole and in parts, can be explained under specified conditions. The elements constituting the network that influences the development of atherosclerosis could be genes, pathways, transcript levels, proteins, or physiologic traits. In this review, we discuss how the integration of genetics and technologies such as transcriptomics and proteomics, combined with mathematical modeling, may lead to an understanding of such networks.