Next generation modeling in GWAS: comparing different genetic architectures

Genomic prediction of growth in a commercially, recreationally, and culturally important marine resource, the Australian snapper (Chrysophrys auratus)

Growth is one of the most important traits of an organism. For exploited species, this trait has ecological and evolutionary consequences as well as economical and conservation significance. Rapid changes in growth rate associated with anthropogenic stressors have been reported for several marine fishes, but little is known about the genetic basis of growth traits in teleosts. We used reduced genome representation data and genome-wide association approaches to identify growth-related genetic variation in the commercially, recreationally, and culturally important Australian snapper (Chrysophrys auratus, Sparidae). Based on 17,490 high-quality single-nucleotide polymorphisms and 363 individuals representing extreme growth phenotypes from 15,000 fish of the same age and reared under identical conditions in a sea pen, we identified 100 unique candidates that were annotated to 51 proteins. We documented a complex polygenic nature of growth in the species that included several loci with small effects and a few loci with larger effects. Overall heritability was high (75.7%), reflected in the high accuracy of the genomic prediction for the phenotype (small vs large). Although the single-nucleotide polymorphisms were distributed across the genome, most candidates (60%) clustered on chromosome 16, which also explains the largest proportion of heritability (16.4%). This study demonstrates that reduced genome representation single-nucleotide polymorphisms and the right bioinformatic tools provide a cost-efficient approach to identify growth-related loci and to describe genomic architectures of complex quantitative traits. Our results help to inform captive aquaculture breeding programs and are of relevance to monitor growth-related evolutionary shifts in wild populations in response to anthropogenic pressures.

Identification of functional mutations associated with environmental variance of litter size in rabbits

Background

Environmental variance (V_E) is partly under genetic control and has recently been proposed as a measure of resilience. Unravelling the genetic background of the V_E of complex traits could help to improve resilience of livestock and stabilize their production across farming systems. The objective of this study was to identify genes and functional mutations associated with variation in V_E of litter size (LS) in rabbits. To achieve this, we combined the results of a genome-wide association study (GWAS) and a whole-genome sequencing (WGS) analysis using data from two divergently selected rabbit lines for high and low V_E of LS. These lines differ in terms of biomarkers of immune response and mortality. Moreover, rabbits with a lower V_E of LS were found to be more resilient to infections than animals with a higher V_E of LS.

Results

By using two GWAS approaches (single-marker regression and Bayesian multiple-marker regression), we identified four genomic regions associated with V_E of LS, on chromosomes 3, 7, 10, and 14. We detected 38 genes in the associated genomic regions and, using WGS, we identified 129 variants in the splicing, UTR, and coding (missense and frameshift effects) regions of 16 of these 38 genes. These genes were related to the immune system, the development of sensory structures, and stress responses. All of these variants (except one) segregated in one of the rabbit lines and were absent (n = 91) or fixed in the other one (n = 37). The fixed variants were in the HDAC9, ITGB8, MIS18A, ENSOCUG00000021276 and URB1 genes. We also identified a 1-bp deletion in the 3′UTR region of the HUNK gene that was fixed in the low V_E line and absent in the high V_E line.

Conclusions

This is the first study that combines GWAS and WGS analyses to study the genetic basis of V_E. The new candidate genes and functional mutations identified in this study suggest that the V_E of LS is under the control of functions related to the immune system, stress response, and the nervous system. These findings could also explain differences in resilience between rabbits with homogeneous and heterogeneous V_E of litter size.

Genomic regions influencing intramuscular fat in divergently selected rabbit lines

Intramuscular fat (IMF) is one of the main meat quality traits for breeding programmes in livestock species. The main objective of this study was to identify genomic regions associated with IMF content comparing two rabbit populations divergently selected for this trait, and to generate a list of putative candidate genes. Animals were genotyped using the Affymetrix Axiom OrcunSNP Array (200k). After quality control, the data involved 477 animals and 93 540 SNPs. Two methods were used in this research: single marker regressions with the data adjusted by genomic relatedness, and a Bayesian multiple marker regression. Associated genomic regions were located on the rabbit chromosomes (OCU) OCU1, OCU8 and OCU13. The highest value for the percentage of the genomic variance explained by a genomic region was found in two consecutive genomic windows on OCU8 (7.34%). Genes in the associated regions of OCU1 and OCU8 presented biological functions related to the control of adipose cell function, lipid binding, transportation and localisation (APOLD1, PLBD1, PDE6H, GPRC5D and GPRC5A) and lipid metabolic processes (MTMR2). The EWSR1 gene, underlying the OCU13 region, is linked to the development of brown adipocytes. The findings suggest that there is a large component of polygenic effect behind the differences in IMF content in these two lines, as the variance explained by most of the windows was low. The genomic regions of OCU1, OCU8 and OCU13 revealed novel candidate genes. Further studies would be needed to validate the associations and explore their possible application in selection programmes.

A genomewide association study in divergently selected lines in rabbits reveals novel genomic regions associated with litter size traits

Uterine capacity (UC), defined as the total number of kits from unilaterally ovariectomized does at birth, has a high genetic correlation with litter size. The aim of our research was to identify genomic regions associated with litter size traits through a genomewide association study using rabbits from a divergent selection experiment for UC. A high-density SNP array (200K) was used to genotype 181 does from a control population, high and low UC lines. Traits included total number born (TNB), number born alive (NBA), number born dead, ovulation rate (OR), implanted embryos (IE) and embryo, foetal and prenatal survivals at second parity. We implemented the Bayes B method and the associations were tested by Bayes factors and the percentage of genomic variance (GV) explained by windows. Different genomic regions associated with TNB, NBA, IE and OR were found. These regions explained 7.36%, 1.27%, 15.87% and 3.95% of GV, respectively. Two consecutive windows on chromosome 17 were associated with TNB, NBA and IE. This genomic region accounted for 6.32% of GV of TNB. In this region, we found the BMP4, PTDGR, PTGER2, STYX and CDKN3 candidate genes which presented functional annotations linked to some reproductive processes. Our findings suggest that a genomic region on chromosome 17 has an important effect on litter size traits. However, further analyses are needed to validate this region in other maternal rabbit lines.

Rare Variants in Known Susceptibility Loci and Their Contribution to Risk of Lung Cancer

BACKGROUND

Genome-wide association studies are widely used to map genomic regions contributing to lung cancer (LC) susceptibility, but they typically do not identify the precise disease-causing genes/variants. To unveil the inherited genetic variants that cause LC, we performed focused exome-sequencing analyses on genes located in 121 genome-wide association study-identified loci previously implicated in the risk of LC, chronic obstructive pulmonary disease, pulmonary function level, and smoking behavior.

METHODS

Germline DNA from 260 case patients with LC and 318 controls were sequenced by utilizing VCRome 2.1 exome capture. Filtering was based on enrichment of rare and potential deleterious variants in cases (risk alleles) or controls (protective alleles). Allelic association analyses of single-variant and gene-based burden tests of multiple variants were performed. Promising candidates were tested in two independent validation studies with a total of 1773 case patients and 1123 controls.

RESULTS

We identified 48 rare variants with deleterious effects in the discovery analysis and validated 12 of the 43 candidates that were covered in the validation platforms. The top validated candidates included one well-established truncating variant, namely, BRCA2, DNA repair associated gene (BRCA2) K3326X (OR = 2.36, 95% confidence interval [CI]: 1.38-3.99), and three newly identified variations, namely, lymphotoxin beta gene (LTB) p.Leu87Phe (OR = 7.52, 95% CI: 1.01-16.56), prolyl 3-hydroxylase 2 gene (P3H2) p.Gln185His (OR = 5.39, 95% CI: 0.75-15.43), and dishevelled associated activator of morphogenesis 2 gene (DAAM2) p.Asp762Gly (OR = 0.25, 95% CI: 0.10-0.79). Burden tests revealed strong associations between zinc finger protein 93 gene (ZNF93), DAAM2, bromodomain containing 9 gene (BRD9), and the gene LTB and LC susceptibility.

CONCLUSION

Our results extend the catalogue of regions associated with LC and highlight the importance of germline rare coding variants in LC susceptibility.

Variable selection in omics data: A practical evaluation of small sample sizes

In omics experiments, variable selection involves a large number of metabolites/ genes and a small number of samples (the n < p problem). The ultimate goal is often the identification of one, or a few features that are different among conditions- a biomarker. Complicating biomarker identification, the p variables often contain a correlation structure due to the biology of the experiment making identifying causal compounds from correlated compounds difficult. Additionally, there may be elements in the experimental design (blocks, batches) that introduce structure in the data. While this problem has been discussed in the literature and various strategies proposed, the over fitting problems concomitant with such approaches are rarely acknowledged. Instead of viewing a single omics experiment as a definitive test for a biomarker, an unrealistic analytical goal, we propose to view such studies as screening studies where the goal of the study is to reduce the number of features present in the second round of testing, and to limit the Type II error. Using this perspective, the performance of LASSO, ridge regression and Elastic Net was compared with the performance of an ANOVA via a simulation study and two real data comparisons. Interestingly, a dramatic increase in the number of features had no effect on Type I error for the ANOVA approach. ANOVA, even without multiple test correction, has a low false positive rates in the scenarios tested. The Elastic Net has an inflated Type I error (from 10 to 50%) for small numbers of features which increases with sample size. The Type II error rate for the ANOVA is comparable or lower than that for the Elastic Net leading us to conclude that an ANOVA is an effective analytical tool for the initial screening of features in omics experiments.

Genetics of body fat mass and related traits in a pig population selected for leanness

Obesity is characterized as the excessive accumulation of body fat and has a complex genetic foundation in humans including monogenic high-risk mutations and polygenic contributions. Domestic pigs represent a valuable model on an obesity-promoting high-caloric diet while constantly evaluated for body characteristics. As such, we investigated the genetics of obesity-related traits, comprising subcutaneous fat thickness, lean mass percentage, and growth rate, in a pig population. We conducted genome-wide association analyses using an integrative approach of single-marker regression models and multi-marker Bayesian analyses. Thus, we identified 30 genomic regions distributed over 14 different chromosomes contributing to the variation in obesity-related traits. In these regions, we validated the association of four candidate genes that are functionally connected to the regulation of appetite, processes of adipogenesis, and extracellular matrix formation. Our findings revealed fundamental genetic factors which deserves closer attention regarding their roles in the etiology of obesity.

Toward the integration of Omics data in epidemiological studies: still a "long and winding road"

Primary and secondary prevention can highly benefit a personalized medicine approach through the accurate discrimination of individuals at high risk of developing a specific disease from those at moderate and low risk. To this end precise risk prediction models need to be built. This endeavor requires a precise characterization of the individual exposome, genome, and phenome. Massive molecular omics data representing the different layers of the biological processes of the host and the nonhost will enable to build more accurate risk prediction models. Epidemiologists aim to integrate omics data along with important information coming from other sources (questionnaires, candidate markers) that has been proved to be relevant in the discrimination risk assessment of complex diseases. However, the integrative models in large-scale epidemiologic research are still in their infancy and they face numerous challenges, some of them at the analytical stage. So far, there are a small number of studies that have integrated more than two omics data sets, and the inclusion of non-omics data in the same models is still missing in most of studies. In this contribution, we aim at approaching the omics and non-omics data integration from the epidemiology scope by considering the "massive" inclusion of variables in the risk assessment and predictive models. We also provide already available examples of integrative contributions in the field, propose analytical strategies that allow considering both omics and non-omics data in the models, and finally review the challenges imbedding this type of research.

Inflammatory-Related Genetic Variants in Non-Muscle-Invasive Bladder Cancer Prognosis: A Multimarker Bayesian Assessment

BACKGROUND

Increasing evidence points to the role of tumor immunologic environment on urothelial bladder cancer prognosis. This effect might be partly dependent on the host genetic context. We evaluated the association of SNPs in inflammation-related genes with non-muscle-invasive bladder cancer (NMIBC) risk-of-recurrence and risk-of-progression.

METHODS

We considered 822 NMIBC included in the SBC/EPICURO Study followed-up >10 years. We selected 1,679 SNPs belonging to 251 inflammatory genes. The association of SNPs with risk-of-recurrence and risk-of-progression was assessed using Cox regression single-marker (SMM) and multimarker methods (MMM) Bayes A and Bayesian LASSO. Discriminative abilities of the models were calculated using the c index and validated with bootstrap cross-validation procedures.

RESULTS

While no SNP was found to be associated with risk-of-recurrence using SMM, three SNPs in TNIP1, CD5, and JAK3 showed very strong association with posterior probabilities >90% using MMM. Regarding risk-of-progression, one SNP in CD3G was significantly associated using SMM (HR, 2.69; P = 1.55 × 10(-5)) and two SNPs in MASP1 and AIRE, showed a posterior probability ≥80% with MMM. Validated discriminative abilities of the models without and with the SNPs were 58.4% versus 60.5% and 72.1% versus 72.8% for risk-of-recurrence and risk-of-progression, respectively.

CONCLUSIONS

Using innovative analytic approaches, we demonstrated that SNPs in inflammatory-related genes were associated with NMIBC prognosis and that they improve the discriminative ability of prognostic clinical models for NMIBC.

IMPACT

This study provides proof of concept for the joint effect of genetic variants in improving the discriminative ability of clinical prognostic models. The approach may be extended to other diseases. Cancer Epidemiol Biomarkers Prev; 25(7); 1144-50. ©2016 AACR.

Rare Variants in Transcript and Potential Regulatory Regions Explain a Small Percentage of the Missing Heritability of Complex Traits in Cattle

The proportion of genetic variation in complex traits explained by rare variants is a key question for genomic prediction, and for identifying the basis of “missing heritability”–the proportion of additive genetic variation not captured by common variants on SNP arrays. Sequence variants in transcript and regulatory regions from 429 sequenced animals were used to impute high density SNP genotypes of 3311 Holstein sires to sequence. There were 675,062 common variants (MAF>0.05), 102,549 uncommon variants (0.01<MAF<0.05), and 83,856 rare variants (MAF<0.01). We describe a novel method for estimating the proportion of the rare variants that are sequencing errors using parent-progeny duos. We then used mixed model methodology to estimate the proportion of variance captured by these different classes of variants for fat, milk and protein yields, as well as for fertility. Common sequence variants captured 83%, 77%, 76% and 84% of the total genetic variance for fat, milk, and protein yields and fertility, respectively. This was between 2 and 5% more variance than that captured from 600k SNPs on a high density chip, although the difference was not significant. Rare variants captured 3%, 0%, 1% and 14% of the genetic variance for fat, milk and protein yields, and fertility respectively, whereas pedigree explained the remaining amount of genetic variance (none for fertility). The proportion of variation explained by rare variants is likely to be under-estimated due to reduced accuracies of imputation for this class of variants. Using common sequence variants slightly improved accuracy of genomic predictions for fat and milk yield, compared to high density SNP array genotypes. However, including rare variants from transcript regions did not increase the accuracy of genomic predictions. These results suggest that rare variants recover a small percentage of the missing heritability for complex traits, however very large reference sets will be required to exploit this to improve the accuracy of genomic predictions. Our results do suggest the contribution of rare variants to genetic variation may be greater for fitness traits.

The genetics of feed conversion efficiency traits in a commercial broiler line

Individual feed conversion efficiency (FCE) is a major trait that influences the usage of energy resources and the ecological footprint of livestock production. The underlying biological processes of FCE are complex and are influenced by factors as diverse as climate, feed properties, gut microbiota, and individual genetic predisposition. To gain an insight to the genetic relationships with FCE traits and to contribute to the improvement of FCE in commercial chicken lines, a genome-wide association study was conducted using a commercial broiler population (n = 859) tested for FCE and weight traits during the finisher period from 39 to 46 days of age. Both single-marker (generalized linear model) and multi-marker (Bayesian approach) analyses were applied to the dataset to detect genes associated with the variability in FCE. The separate analyses revealed 22 quantitative trait loci (QTL) regions on 13 different chromosomes; the integration of both approaches resulted in 7 overlapping QTL regions. The analyses pointed to acylglycerol kinase (AGK) and general transcription factor 2-I (GTF2I) as positional and functional candidate genes. Non-synonymous polymorphisms of both candidate genes revealed evidence for a functional importance of these genes by influencing different biological aspects of FCE.

Performance of a blockwise approach in variable selection using linkage disequilibrium information

Background

Genome-wide association studies (GWAS) aim at finding genetic markers that are significantly associated with a phenotype of interest. Single nucleotide polymorphism (SNP) data from the entire genome are collected for many thousands of SNP markers, leading to high-dimensional regression problems where the number of predictors greatly exceeds the number of observations. Moreover, these predictors are statistically dependent, in particular due to linkage disequilibrium (LD).

We propose a three-step approach that explicitly takes advantage of the grouping structure induced by LD in order to identify common variants which may have been missed by single marker analyses (SMA). In the first step, we perform a hierarchical clustering of SNPs with an adjacency constraint using LD as a similarity measure. In the second step, we apply a model selection approach to the obtained hierarchy in order to define LD blocks. Finally, we perform Group Lasso regression on the inferred LD blocks. We investigate the efficiency of this approach compared to state-of-the art regression methods: haplotype association tests, SMA, and Lasso and Elastic-Net regressions.

Results

Our results on simulated data show that the proposed method performs better than state-of-the-art approaches as soon as the number of causal SNPs within an LD block exceeds 2. Our results on semi-simulated data and a previously published HIV data set illustrate the relevance of the proposed method and its robustness to a real LD structure. The method is implemented in the R package BALD (Blockwise Approach using Linkage Disequilibrium), available from http://www.math-evry.cnrs.fr/publications/logiciels.

Conclusions

Our results show that the proposed method is efficient not only at the level of LD blocks by inferring well the underlying block structure but also at the level of individual SNPs. Thus, this study demonstrates the importance of tailored integration of biological knowledge in high-dimensional genomic studies such as GWAS.

Electronic supplementary material

The online version of this article (doi:10.1186/s12859-015-0556-6) contains supplementary material, which is available to authorized users.