ABSTRACT
To investigate whether the 21q21 region may contain a quantitative trait locus (QTL) with pleiotropic effect on % predicted FEV1 (forced expiratory volume in 1 second) and SPTQ (number of positive skin test responses to 11 allergens), in 295 EGEA families ascertained through asthmatic probands, we conducted a bivariate linkage analysis using two approaches:
(1) a bivariate variance components (VC) analysis and (2) A combined principal components (CPC) analysis, with 13 microsatellites covering the whole chromosome 21. To identify the genetic variants associated with these traits, we performed family-based association analysis (FBAT) for the second principal component (PC2) using two microsatellites and 27 SNPs belonging to three candidate genes, located in the vicinity of the linkage peak. Univariate linkage analyses showed suggestive evidence of linkage to %FEV1 and SPTQ at two positions. Bivariate VC linkage analysis of these phenotypes led to an increase in linkage signals as compared to univariate analysis at the same positions. The peaks obtained by the CPC led to similar results as the full bivariate VC approach; evidence for linkage mainly coming from PC2. The strongest association signal, using single marker analysis for PC2, was obtained with D21S1252 (p=0.003 for global test; p=0.004 for allele 2 and p=0.001 for allele 11) and rs2834213 of IFNGR2 (p=0.003), these two loci being 3 Mb apart. Multi-marker analysis further strengthened this finding. These results indicate that at least two genetic factors may be involved in SPTQ and %FEV1 variability but further genotyping is needed to better understand these findings.
- Linkage analysis
- bivariate
- asthma
- pleiotropy
- principal components
- variance components
- association analysis
- multi-marker
- asymptotic distribution
INTRODUCTION
According to the World Health Organization, more than 339 million people are living with asthma globally and the numbers are increasing, especially in the Western countries (Douwes and Pearce, 2014; Douwes et al., 2011). Asthma is one of the major non-communicable diseases and it is the most common chronic disease of childhood, affecting 14% children (WHO; www.who.int, Fong MK et al, 2020). Asthma is a multifactorial and complex disease associated with intermediate phenotypes involved in immune response to inflammation and lung function. Although the genetic component of asthma and asthma-related phenotypes has long been established [Vercelli 2008], the extent to which the genetic factors involved are common or specific to these phenotypes is unclear. To date, multiple genetic factors have been identified, with more than 100 loci linked to asthma and asthma-related phenotypes of potential interest in genome-wide-association studies (GWAS) [Hernandez-Pacheco N et al., 2019; Kabesh et al., 2020; Zhu Z et al, 2021]. One interesting feature of these published genome screens is that a given chromosomal region is often linked to various asthma-associated phenotypes across studies, suggesting that these phenotypes may share genetic determinants. However, the formal characterization of pleiotropic effects of genes underlying two or more asthma-related phenotypes has received little attention [Ziegler, et al, 2001; Bouzigon, et al. 2007; Ferreira, et al. 2006; Aschard H, et al. 2009; Dixon and Poytner, 2016].
A previous genome-wide scan conducted in 295 French families from the Epidemiological study on the Genetic and Environmental factors of Asthma (EGEA) for asthma and seven asthma-associated phenotypes detected a linkage signal at the same marker position in the 21q21 region for two asthma-related phenotypes: forced expiratory volume in 1 second percent predicted (%FEV1) and a measure of polysensitization to allergens (SPTQ), suggesting that these traits may share genetic determinants [Bouzigon, et al. 2004].
Multivariate genetic linkage analyses and genome-wide association analyses of correlated phenotypes have been shown to improve power of detecting genes with small effects where these genes may be missed with univariate analyses [Allison, et al. 1998; Marlow, et al. 2003; Kochunov P et al, 2010; Wu T et al, 2010; Kochunov P et al, 2011; Yang Q and Wang Y 2012; Deliu M et al, 2016; Salidas et al, 2018; Fatumo S et al, 2019]. Several methods have been proposed to conduct bivariate linkage analysis, among which are the bivariate variance component (VC) approach [Amos, et al. 2001; Bauman, et al. 2005] and a combined analysis of principal components (PCs) [Mangin, et al. 1998]. Using a model with fixed genetic effects in a single phase known cross, [Mangin, et al. 1998] showed the asymptotic equivalences between the combined principal component approach and the bivariate test for linkage. However, these methods were not compared analytically in a VC model. A previous limited simulation study showed similar power for VC and CPC methods [Gorlova, et al. 2002]. We have further studied the asymptotic distribution of the bivariate VC test statistic under the null hypothesis of no linkage and defined upper and lower bounds of this statistic, since the asymptotic distribution of this test is more complex in VC model than in linear model.
In the present study, we tested for pleiotropic genetic determinants that may influence both %FEV1 and SPTQ on chromosome 21q21 in the French EGEA families using the bivariate VC analysis and combined analysis of principal components. This linkage analysis was followed by an association analysis with candidate genes in the linkage region.
METHODS
FAMILY SAMPLE
The protocol of the EGEA data collection has been described elsewhere [Kauffmann, et al. 2001; Kauffmann, et al. 1997]. The sample examined by the present study consisted of 291 nuclear families ascertained trough at least one asthmatic subject. The inclusion criteria for asthma have been described in details elsewhere [Kauffmann, et al. 1997]. Subjects answered a detailed questionnaire regarding respiratory symptoms and treatment based on international standardized questionnaires. Biological and physiological tests were performed on each participant. Written informed consent was obtained from all subjects participating to the study under an Institutional Review Board-approved protocol.
PHENOTYPES ANALYZED
Skin-prick tests were performed for 11 allergens (including moulds, indoors and outdoors allergens). A positive response was defined as a wheal size exceeding that of the negative control by ≥ 3 mm [Maccario, et al. 2003] A quantitative score (SPTQ) was constructed by counting the number of positive responses to allergens and thus measuring the degree of polysensitization. SPTQ originally included 12 classes (from 0 to 11) but classes 4 to 11 were combined into the last category because of small sample size. Prior to the analysis, SPTQ was adjusted for relevant covariates including age and sex using multiple regression. These regression models, including main effects and interaction terms, were built separately in three groups (parents, children offspring < 16 years of age, offspring ≥16 years of age) as discussed in Bouzigon et al. [Bouzigon, et al. 2004].
Spirometric measures were carried out for adults and children separately. A survey for adults were conducted according to the European Respiratory Health Survey protocol [Burney, et al. 1994] and [Quanjer 1983] and a survey for children were conducted according to Polgar and Weng [Polgar and Weng 1979]The best of three pre-bronchodilator FEV1 measures was used to calculate a percentage of predicted FEV1 values (%FEV1) based on age, height and gender [Polgar and Weng 1979; Quanjer 1983].
Since, %FEV1 and SPTQ showed departures from normality with significant kurtosis for both traits (p < 6.10−7), a probit transformation was applied to each phenotype prior to linkage analysis to normalize their distribution [Peng, et al. 2007].
GENOTYPING
Genotyping of chromosome 21 was done at CNG (Centre National de Génotypage at Evry), with five microsatellites from the original scan and eight additional markers for fine-mapping. The markers were distributed at an average distance of 3 cM and had an average heterozygosity of 75%. After rigorous genotype quality control, the final sample for the present analysis included 291 families (1301 subjects) with at least one asthmatic subject, comprising 566 genotyped parents (97.3% of all parents) and 718 genotyped sibs. In addition to the microsatellite data, three candidate genes genotyped for a total of 27 SNPs and spanning a 33 - 42 Mb region around the 43 cM linkage peak of 21q21 were available in the EGEA study (genotyping done at CNG with the TaqMan® SNP technology). These genes included: 4 SNPs in interferon (alpha and beta) receptor 2 (IFNAR2), 15 SNPs in the interleukin 10 receptor, beta (IL10RB) and 10 SNPs in the interferon gamma receptor 2 (IFNGR2) (see Supplementary Table S1). Additionally, supplementary table S2 shows the microsatellites and SNPs also belonging to the ADAMTS1, ADAMTS5 and ICOSL genes.
STATISTICAL ANALYSES
LINKAGE ANALYSIS
Genetic and environmental components of variance for each trait and genetic and environment correlations between the two traits were estimated by variance decomposition using maximum likelihood methods implemented in the multipoint quantitative trait linkage software package (ACT) (http://www.epigenetic.org/Linkage/act.html) [de Andrade 1998].
We first conducted univariate linkage analysis of %FEV1 and SPTQ (adjusted and probit transformed values) using the VC method. We then carried out bivariate linkage analysis using bivariate VC and combined analysis of principal components (CPC) of the two phenotypes.
The univariate VC method separates the total variation of a trait into genetic and environmental components and evaluates linkage by comparing a model incorporating both a genetic additive variance at a putative QTL linked to marker(s) and a polygenic component with a purely polygenic model (QTL variance, , being set to zero) by a likelihood ratio test (LRT). Minus twice the natural logarithm of this likelihood ratio follows asymptotically a one-sided chi-square with one degree of freedom. This chi-square divided by 2ln10 is a LOD score.
The univariate and bivariate variance component models are described in details in Amos et al [Amos, et al. 2001] and Williams et al [Williams and Blangero 1999]. For the univariate model, under the null hypothesis, the parameters estimated are the mean, μ, the additive genetic variance, , the polygenic variance , and the environmental variance . For the bivariate test, under the null hypothesis, the marker linked to the QTL parameters are restricted to be equal to zero and being the additive genetic variance related respectively to trait1 and trait2 and σa1,a2 the additive genetic covariance between these traits. Under the alternative hypothesis, the three parameters are estimated with the constraints: , and . The test for genetic linkage is constructed by a LRT. The distribution of the bivariate linkage test statistic depends whether a constraint is imposed on the genetic correlation between the traits. When the correlation is unconstrained, the asymptotic distribution of the bivariate test statistic, under the null hypothesis that the linked-genetic components and covariance are zero, is the supremum of a process (See Appendix). However, computing the probability distribution function of this supremum, in order to get quantiles or P-values, is a difficult task, so we used a mixture of , and as an approximation of the bivariate test statistic distribution. This approximation was shown, by simulations, to be an upper limit of the distribution [Amos, et al. 2001]. When the genetic correlation between the traits is constrained to be zero, or any other value, the distribution of the bivariate linkage test statistic follows, under the null hypothesis, a mixture of , and and we define this as the lower bound of the test statistic distribution (See Appendix).
The idea of considering PC analysis of phenotypes prior to linkage analysis to obtain a set of uncorrelated PCs and obtaining a combined test statistic was proposed by Mangin et al [Mangin, et al. 1998]. These authors showed that the likelihood-ratio test used to test for the presence of a pleiotropic QTL modelled as a fixed genetic effect for a given mating type is asymptotically equivalent to the sum of likelihood-ratio tests of univariate analyses applied to the principal components of the phenotypes. Here, we applied this method in a VC model and called this approach as the CPC method. In a first step, a principal components (PC) analysis was applied to %FEV1 and SPTQ, to obtain two uncorrelated principal components. These components were then subjected to independent univariate linkage analyses based on the VC method. In a second step, a combined test statistic, CPC test, was constructed by summing the univariate VC likelihood-ratio (LR) test statistics obtained for each of the two , where the distribution of each LRi test, under the null hypothesis of no linkage, is a mixture of . Since the two principal components are independent, the asymptotic distribution of the combined PC test, CPC, is a mixture of .
FAMILY-BASED ASSOCIATION TEST
Association between quantitative phenotypes (%FEV1, SPTQ, PC1 and PC2) and genetic polymorphisms (microsatellites and SNPs) was assessed using the family-based association test (FBAT) method [Lange, et al. 2002]. This method was applied to test for association in the presence of linkage. We tested each marker under additive, dominant and recessive models and the best fitting model was selected. Following single marker analysis, we carried out multi-marker analyses using FBAT-LC [Xu, et al. 2006]. This method allows testing multiple markers simultaneously without haplotype reconstruction. In brief, the FBAT-LC method proposed by Xu et al, is based on a linear combination of single-marker FBAT test statistics using data-driven weights, where marker weight derivation is based on the conditional mean model [Lange, et al. 2003].
RESULTS
DESCRIPTIVE STATISTICS
The phenotypic characteristics of 718 genotyped siblings belonging to 291 families are shown in Table I. Among these siblings, fifty-three percent were males and their mean age was 16.0 ± 7.7 (SD) years. The proportion of asthmatic siblings was 54.2%. The proportion of siblings having at least one positive skin prick test was 67.4%. The mean of %FEV1 was 96.6 (SD=13.3) and SPTQ was 2.6 (SD=1.4) before adjustment on age and sex. The total number of sibs with phenotypic information was 694 for %FEV1, 705 for SPTQ, and 681 when both phenotypes were jointly considered.
LINKAGE ANALYSIS
Heritability estimates, (h2), were significant for both probit transformed and age, sex adjusted phenotypes (P < 1.0×10−4) and were equal to 38% (SD=3.3) for SPTQ and 57% (SD=3.2) for %FEV1. The genetic correlations between %FEV1 and SPTQ was positive (ρG = 0.2), whereas the environmental correlation was negative (ρE = -0.2).
Univariate and bivariate linkage analysis results for SPTQ and %FEV1 are shown in Figure 1 and Table II. Univariate VC linkage analyses showed evidence for linkage to %FEV1 at two positions: D21S265 at 25.5 cM (LOD = 2.2, p = 0.0008) and D21S1252 at 43 cM (LOD=1.9, p=0.002). Univariate VC linkage analysis of SPTQ showed lower evidence for linkage than for %FEV1 but at two positions adjacent to the two latter ones; D21S1914 at 24.4 cM (LOD = 1.07, p = 0.01) and D21S1895 at 41.1 cM (LOD = 1.10, p = 0.01).
Bivariate VC linkage analysis of %FEV1 and SPTQ showed an increase in LOD scores as compared to univariate analyses with highest LOD scores obtained at the two previous positions (Figure 1): D21S265 (25.5 cM) with LOD scores ranging between 3.3 (p = 2.0×10-4) and 4.2 (p = 7.0 ×10−5) for the lower and upper bound of test statistics respectively and D21S1252 (43.0 cM) with LOD scores ranging between 3.48 (p = 0.0001) and 4.3 (p = 6.0×10−5). Parameter estimates at D21S265 and D21S1252 linked QTL are presented in Table III. Using the unconstrained model, the QTL covariance (σ a1, a2) for each of the two markers was positive (0.52-0.55) whereas the polygenic covariance (σ p1, p2) was of opposite sign, varying from -0.85 to -0.98. The proportion of total variance explained by the QTL varied from 33% to 36% for %FEV1 and from 23% to 29% for SPTQ.
The principal components (PC) analysis of %FEV1 and SPTQ showed that the first and second PCs contributed respectively to 51% and 49% of the overall phenotypic variance. On the first PC (PC1), the coefficients for the two phenotypes were of opposite sign, (0.71 for %FEV1 and –0.71 for SPTQ), while on PC2, the coefficients were positive. CPC analysis showed highest evidence for linkage at the same locations as the bivariate method, at 25.5 cM at D21S265 (p = 1.0×10-4) and at 43 cM for D21S1252 (p = 3.9×10−5) and at 41.1 cM position at D21S1895 (p = 3.6×10−5).
The bivariate VC and CPC methods led to similar results but with a slightly higher evidence for pleiotropic QTLs at 25.6 cM with bivariate VC and at 41.1 and 43 cM with CPC. As can be seen in Table II, for the combined CPC, evidence for linkage came almost solely from PC2 at (p=7.0×10-5 at 25.5 cM and p=2.0×10-5 at 41.1 and 43 cMs) while no significant lod score for PC1 was observed. This led us to further study PC2 for association analysis.
ASSOCIATION ANALYSIS
The list of 27 SNPs belonging to the three candidate genes spanned a 0.185 Mb region from 33.55 to 33.73 Mb (Supplementary Table S1) between the two linkage peaks. None of these SNPs showed departure from Hardy-Weinberg equilibrium.
Univariate association of PC2 with the microsatellite markers located at the highest linkage peaks and the 10 SNPs of the candidate genes showed association signals with D21S1252 (p=0.003 with a global test under a dominant model) and two SNPs within IFNGR2 (p=0.02 with rs9976971 and p=0.06 with rs2284553 under an additive model) (See Table IV and Supplementary Table S2 for all markers). When examining each allele separately at D21S1252, allele 2 was positively associated with PC2 (p=0.004) while allele 11 was negatively associated (p=0.001). Interestingly, this microsatellite is located within the CLDN14 gene.
This single-marker association analysis was extended to two marker analysis using FBAT_LC by examining jointly each of the SNPs belonging to IFNGR2 with either allele 2 or 11 from D21S1252. Higher evidence of association was consistently observed for each of IFNGR2 SNPs in combination with allele 11 as compared with allele 2 of D21S1252. The strongest signal was obtained when considering jointly allele 11 of D21S1252 and IFNGR2_rs2834213 (p=0.0001).
Examination of the LD pattern between D21S1252 alleles and IFNGR2 SNPs showed no LD (LD coefficient r2 being < 0.2). The 10 SNPs within IFNGR2 showed also low pairwise LD except for three pairs of SNPs (see Table V and Figure 2).
DISCUSSION
The present study is the first one that examined allergy polysensitization and lung function in a bivariate fashion. Bivariate linkage analysis revealed two potential pleiotropic QTLs at 25 cM and 43 cM. Further, association between a combination of lung function values and measures of sensitization was formed with genetic variants at two loci, D21S1252 and IFNGR2, 3 Mb distant from each other, close to the 43 cM linkage peak. To our knowledge, this study is also the first attempt to test association between IFNGR2 gene and %FEV1 and allergen polysensitization. These results indicate that two genes in the 21q21 region influence lung function and allergen polysensitization.
Bivariate linkage analysis of %FEV1 and SPTQ was carried out using two different methods: a bivariate variance component approach and a combined principal component analysis. Both bivariate linkage analyses led to an increase in linkage signals of the same magnitude at the two positions previously detected by univariate analyses. Simulation studies have shown that VC multivariate approaches are more powerful than univariate analyses when the traits are not very highly positively correlated, this power being highest when correlation relative to QTL and polygenic components underlying two phenotypes are of opposite sign [Amos, et al. 2001; Gorlova, et al. 2002]. Thus, bivariate analysis of %FEV1 and SPTQ, which show small phenotypic correlation, may be particularly useful to detect genes with pleiotropic effects on these phenotypes. One disadvantage of the VC approach is that it may be difficult to apply to more than two phenotypes since the number of VC parameters to be estimated increases geometrically as the number of phenotypes increases which can create convergence problems when maximizing the likelihood. Therefore, a potential advantage of the CPC approach is that it can be easily applied to more than two phenotypes although covariance and variances cannot be estimated. Moreover, this approach can be used with other test statistics [Bouzigon, et al. 2007; Aschard, et al, 2009].
Both bivariate VC and CPC approaches assume multivariate normality of the phenotypes and departure from normality may lead to inflation of type I error [Allison, et al. 1999]. However, probit transformation of each phenotype was used here to ensure the validity of the normality assumption.
Replication of linkage results across studies are important in supporting the actual involvement of linkage regions. We have considered all previously reported linkage peaks with P ≤ 0.01 by published genome scans performed to date in more than 20 different populations. Five different genome screens have reported linkage signals to 21q. At 21q21, around the first linkage peak for lung function, atopic asthma and specific skin prick tests (Postma et al, 2005, Pillai et al, 2006 and Blumenthal et al 2006). The 21q22 region around our second linkage peak was also reported linked to asthma and several atopy related phenotypes (specific IgE, specific skin prick tests) in three different populations (CSGA, Hutterites and German families) [The Collaborative Study on the Genetics of Asthma, 1997; Blumenthal, et al. 2006; Ober, et al. 1999; Wjst 1999]. This is in agreement with our results.
The QTL detected on 21q22, IFNGR2 gene, encodes the non-ligand-binding beta chain of the gamma interferon (IFNG) receptor. IFNG has been implicated to be associated to several asthma related phenotypes and its role on immune and allergic responses has been established [Malerba and Pignatti 2005; Teixeria LK et al, 2005]. Human interferon-gamma receptor is a heterodimer of IFNGR1 and IFNGR2. [Nakao, et al. 2001] investigated the association of a coding variant, Gln64Arg of the IFNGR2 gene, with atopic asthma in Japanese child population but did not find any significant associations whereas [Gao, et al. 1999] found an association of Gln64Arg variant with total serum IgE levels in the British population. Interestingly, the Gln64Arg is also represented in our analysis sample as rs9808753 but no association signal was obtained for PC2 with this SNP. Most recently, Daley et al [Daley, et al. 2009] reported an association between IFNGR2 gene and atopic asthma using four asthma population samples from Canada and Australia. Finally, data from the International HapMap consortium suggest that, 6 out 10 snps of IFNGR2 gene in the EGEA data are within a large block of linkage disequilibrium, capturing most of the variation in the IFNGR2 gene.
D21S1252 microsatellite is located within one of the markers belonging to the Claudin 14 (CLDN14) gene. [Tsukita and Furuse 2002] reported that primary and secondary dysfunction of the claudin-based barrier in various organs can cause various pathological conditions. The facilitation of antigen presentation via a paracellular pathway is hypothesised to trigger or aggravate chronic inflammation. In asthma it has been proposed that various peptidases contained in allergens, such as house dust mites and fungus, cleave the extracellular loops of claudins/occludin to affect the TJ barrier [Wan, et al. 1999], [Robinson, et al. 2001]. Along the same lines, it is possible that dysfunction of the claudin-based barrier in the epidermis is involved in the pathogenesis of epidermal inflammation such as atopic dermatitis [Leung 1995]. Recently, Rodriguez et al [Rodriguez, et al. 2009] carried out a meta analysis to show strong associations of fillagrin(FLG) mutations with eczema and asthma. This may be another example of a disrupted barrier which may support the hypothesis with the CLDN14 gene.
In conclusion, the present study shows that both use of quantitative phenotypes of %FEV1 and SPTQ, and using two different multivariate approaches, indicated support for a pleiotropic effect at two different positions on chromosome 21. An association analysis further indicated a moderate association signal for PC2 with two loci: D21S1252 and rs2834213 belonging to IFNGR2. Multi-marker analysis with these two loci strengthened this signal, especially when considering jointly allele 11 of D21S1252 and rs2834213. Although the p values obtained are generally low, these are considered nominal uncorrected p values and hence these results need to be followed up with a confirmatory study and these results bode well for a replication study. Future functional analysis and further genotyping may help to identify the causal variants in the IFNGR2 gene for %FEV1 and allergen polysensitization.
Data Availability
Data and quality management: Inserm ex-U155 (Egea1): J Hochez; Inserm U 780, Villejuif: N Le Moual, C Ravault; Inserm U 794: N Chateigner; Grenoble: J Ferran.
EGEA cooperative group
Coordination
F Kauffmann; F Demenais (genetics); I Pin (clinical aspects).
Respiratory epidemiology
Inserm U 700, Paris M Korobaeff (Egea1), F Neukirch (Egea1); Inserm U 707, Paris: I Annesi-Maesano; Inserm U 780, Villejuif: F Kauffmann, N Le Moual, R Nadif, MP Oryszczyn; Inserm U 823, Grenoble: V Siroux.
Genetics
Inserm U 393, Paris: J Feingold; Inserm U 535, Villejuif: MH Dizier; Inserm U 794, Evry: E Bouzigon, F Demenais; CNG, Evry: I Gut, M Lathrop.
Clinical centers
Grenoble: I Pin, C Pison; Lyon: D Ecochard (Egea1), F Gormand, Y Pacheco; Marseille: D Charpin (Egea1), D Vervloet; Montpellier: J Bousquet; Paris Cochin: A Lockhart (Egea1), R Matran (now in Lille); Paris Necker: E Paty, P Scheinmann; Paris-Trousseau: A Grimfeld, J Just.
Data and quality management
Inserm ex-U155 (Egea1): J Hochez; Inserm U 780, Villejuif: N Le Moual, C Ravault; Inserm U 794: N Chateigner; Grenoble: J Ferran.
This study was funded by INSERM, the French Ministry of Higher Education and Research, University of Evry, Fondation pour la Recherche Médicale (ACE20061209064) and the French National Agency for Research (ANR 05-SEST-020-02/05-9-97) and ANR 06-CEBS-029-2).
ACKNOWLEDGMENTS
Appendix
1. Asymptotic Distribution of the bivariate VC test
The bivariate variance component model leads to a non-regular likelihood with a nuisance parameter present only under the alternative. Indeed, the correlation parameter of the variance-covariance matrix of the linked QTL disappears when variances are equal to 0. This type of non-regular likelihood was studied by Davies (1977, Biometrika 64: 247-254; 1987, Biometrika 74: 33-43). However, since the likelihood function is regular when the correlation ρ is known, the asymptotic law of the LRT for the bivariate VC model can be studied as the supremum of the asymptotic law of the likelihood ratio test given ρ.
Let lN (θ, τ, ρ) denote the log likelihood function of the bivariate VC model where N is the number of independent families, ρ is the QTL additive genetic correlation, andτ are the other unknown parameters. Assuming that the true value of τ does not lie on the boundary of the parameter space and that ρ is known, we can restrict our attention to θ. Following Self and Liang (1987), let UN (θ ; ρ) and −I N (θ ; ρ) denote the first and second derivatives of the log likelihood function given. In the neighborhood of the null hypothesisθ0 = (0,0)′, the likelihood ratio test given ρ is asymptotically equal to Where
And we found an asymptotic probability function that is a mixture , and . However, when the bivariate variance component model is unconstraint, the correlation parameter is estimated under the alternative hypothesis so we get
Therefore, the LRT of the bivariate variance component model is asymptotically equal to which is the supremum of a process.
For the upper bound of the distribution the results obtained by simulation in Amos et al 2001 is used. Their simulations study showed that the quantiles are conservative for the unconstraint bivariate VC test, therefore .
Footnotes
Literature updated.