 Methodology article
 Open Access
A multilocus likelihood approach to joint modeling of linkage, parental diplotype and gene order in a fullsib family
 Qing Lu^{1},
 Yuehua Cui^{1} and
 Rongling Wu^{1, 2}Email author
https://doi.org/10.1186/14712156520
© Lu et al; licensee BioMed Central Ltd. 2004
 Received: 10 March 2004
 Accepted: 26 July 2004
 Published: 26 July 2004
Abstract
Background
Unlike a pedigree initiated with two inbred lines, a fullsib family derived from two outbred parents frequently has many different segregation types of markers whose linkage phases are not known prior to linkage analysis.
Results
We formulate a general model of simultaneously estimating linkage, parental diplotype and gene order through multipoint analysis in a fullsib family. Our model is based on a multinomial mixture model taking into account different diplotypes and gene orders, weighted by their corresponding occurring probabilities. The EM algorithm is implemented to provide the maximum likelihood estimates of the linkage, parental diplotype and gene order over any type of markers.
Conclusions
Through simulation studies, this model is found to be more computationally efficient compared with existing models for linkage mapping. We discuss the extension of the model and its implications for genome mapping in outcrossing species.
Keywords
 Linkage Analysis
 Gene Order
 Recombination Fraction
 Linkage Phase
 Repulsion Phase
Background
The construction of genetic linkage maps based on molecular markers has become a routine tool for comparative studies of genome structure and organization and the identification of loci affecting complex traits in different organisms [1]. Statistical methods for linkage analysis and map construction have been well developed in inbred line crosses [2] and implemented in the computer packages MAPMAKER [3], CRIMAP [4], JOINMAP [5] and MULTIMAP [6]. Increasing efforts have been made to develop robust tools for analyzing marker data in outcrossing organisms [7–12], in which inbred lines are not available due to the heterozygous nature of these organisms and/or longgeneration intervals.
Genetic analyses and statistical methods in outcrossing species are far more complicated than in species that can be selfed to produce inbred lines. There are two reasons for this. First, the number of marker alleles and the segregation pattern of marker genotypes may vary from locus to locus in outcrossing species, whereas an inbred lineinitiated segregating population, such as an F_{2} or backcross, always has two alleles and a consistent segregation ratio across different markers. Second, linkage phases among different markers are not known a priori for outbred parents and, therefore, an algorithm should be developed to characterize a most likely linkage phase for linkage analysis.
To overcome these problems of linkage analysis in outcrossoing species, Grattapaglia and Sederoff [13] proposed a twoway pseudotestcross mapping stratety in which one parent is heterozygous whereas the other is null for all markers. Using this strategy, two parentspecific linkage maps will be constructed. The limitation of the pseudotestcross strategy is that it can only make use of a portion of molecular markers. Ritter et al. [7] and Ritter and Salamini [9] proposed statistical methods for estimating the recombination fractions between different segregation types of markers. Using both analytical and simulation approaches, Maliepaard et al. [10] discussed the power and precision of the estimation of the pairwise recombination fractions between markers. Wu et al. [11] formulated a multilocus likelihood approach to simultaneously estimate the linkage and linkage phases of the crossed parents over multiple markers. Ling [14] proposed a threestep analytical procedure for linkage analysis in outcrossing populations, which includes (1) determining the parental haplotypes for all of the markers in a linkage group, (2) estimating the recombination fractions, and (3) choosing a most likely marker order based on optimization analysis. This procedure was used to analyze segregating data in an outcrossing forest tree [15]. Currently, none of these models for linkage analysis in outcrossing species can provide a onestep analysis for the linkage, parental linkage phase and marker order from segregating marker data.
Estimation from twopoint analysis of the recombination fraction ( ± SD) and the parental diplotype probability of parent P ( ) and Q ( ) for five markers in a fullsib family of n = 100
Parental diplotype  r = 0.05  r = 0.20  

Marker  P^{ a }  ×  Q^{ a } 





 
        
 a  b  c  d  
        0.530 ± 0.0183  0.2097 ± 0.0328  
 a  b  a  b  0.9960  0.9972  0.9882  0.9878  
        0.0464 ± 0.0303  0.2103 ± 0.0848  
 a  o  ×  o  a  1 (0^{ b })  0(1^{ b })  1 (0^{ b })  0(1^{ b })  
        0.0463 ± 0.0371  0.1952 ± 0.0777  
 a  b  b  b  1  1/0^{ c }  1  1/0^{ c }  
        0.0503 ± 0.0231  0.2002 ± 0.0414  
 a  b  c  d  1  1/0^{ c }  1  1/0^{ c } 
Twolocus analysis
A general framework
Suppose there is a fullsib family of size n derived from two outcrossed parents P and Q. Two sets of chromosomes are coded as 1 and 2 for parent P and 3 and 4 for parent Q. Consider two marker loci and , whose genotypes are denoted as 12/12 and 34/34 for parent P and Q, respectively, where we use / to separate the two markers. When the two parents are crossed, we have four different progeny genotypes at each marker, i.e., 13, 14, 23 and 24, in the fullsib family. Let r be the recombination fraction between the two markers.
In general, the genotypes of the two markers for the two parents can be observed in a molecular experiment, but the allelic arrangement of the two markers in the two homologous chromosomes of each parent (i.e., linkage phase) is not known. In the current genetic literatuire, a linear arrangement of nonalleles from different markers on the same chromosomal region is called the haplotype. The observable twomarker genotype of parent P is 12/12, but it may be derived from one of two possible combinations of maternally and paternallyderived haplotypes, i.e., [11] [22] or [12] [21], where we use [] to define a haplotype. The combination of two haplotypes is called the diplotype. Diplotype [11] [22] (denoted by 1) is generated due to the combination of twomarker haplotypes [11] and [22], whereas diplotype [12] [21] (denoted by ) is generated due to the combination of twomarker haplotypes [12] and [21]. If the probability of forming diplotype [11] [22] is p, then the probability of forming diplotype [12] [21] is 1  p. The genotype of parent Q and its possible diplotypes [33] [44] and [34] [43] can be defined analogously; the formation probabilities of the two diplotypes are q and 1  q, respectively.
for [11] [22] × [33] [44],
for [11] [22] × [34] [43],
for [12] [21] × [33] [44] and
Let n = (n_{j 1j 2})_{4 × 4} denote the matrix for the observations of progeny where j_{1},j_{2} = 1 for 13, 2 for 14, 3 for 23, or 4 for 34 for the progeny genotypes at these two markers. Under each parental diplotype combination, n_{j 1j 2}follows a multinomial distribution. The likelihoods for the four diplotype combinations are expressed as
where N_{1} = n_{11} + n_{22} + n_{33} + n_{44}, N_{2} = n_{14} + n_{23} + n_{32} + n_{41}, N_{3} = n_{12} + n_{21} + n_{34} + n_{43}, and N_{4} = n_{13} + n_{31} + n_{24} + n_{42}. It can be seen that the maximum likeihood estimate (MLE) of r ( ) under the first diplotype combination is equal to one minus under the fourth combination, and the same relation holds between the second and third diplotype combinations. Although there are identical plugin likelihood values between the first and fourth combinatins as well as between the second and third combinations, one can still choose an appropriate from these two pairs because one of them leads to greater than 0.5. Traditional approaches for estimating the linkage and parental diplotypes are to estimate the recombination fractions and likelihood values under each of the four combinations and choose one legitimate estimate of r with a higher likelihood.
In this study, we incorporate the four parental diplotype combinations into the observed data likelihood, expressed as
where Θ = (r, p, q) is an unknown parameter vector, which can be estimated by differentiating the likelihood with respect to each unknown parameter, setting the derivatives equal to zero and solving the likelihood equations. This estimation procedure can be implemented with the EM algorithm [2, 11, 16]. Let H be a mixture matrix of the genotype frequencies under the four parental diplotype combinations weighted by the occurring probabilities of the diplotype combinations, expressed as
where
Similar to the expression of the genotype frequencies as a mixture of the four diplotype combinations, the expected number of recombination events contained within each twomarker progeny genotype is the mixture of the four different diplotype combinations, i.e.,
where the expected number of recombination events for each combination are expressed as
Define
The general procedure underlying the {τ + 1}th EM step is given as follows:
E Step: At step τ, using the matrix H based on the current estimate r^{{τ}}, calculate the expected number of recombination events between two markers for each progeny genotype and ,
where d_{j 1j 2}, h_{j 1j 2}, p_{j 1j 2}and q_{j 1j 2}are the (j_{1}j_{2})th element of matrix D, H, P and Q, respectively.
M Step: Calculate r^{{τ+1}}using the equation,
The E step and M step among Eqs. (4) – (7) are repeated until r converges to a value with satisfied precision. The converged values are regarded as the MLEs of Θ.
Model for partially informative markers
Unlike an inbred line cross, a fullsib family may have many different marker segregation types. We symbolize observed marker alleles in a fullsib family by A_{1}, A_{2}, A_{3} and A_{4}, which are codominant to each other but dominant to the null allele, symbolized by O. Wu et al. [11] listed a total of 28 segregation types, which are classified into 7 groups based on the amount of information for linkage analysis:
A. Loci that are heterozygous in both parents and segregate in a 1:1:1:1 ratio, involving either four alleles A_{1}A_{2} × A_{3}A_{4}, three nonnull alleles A_{1}A_{2} × A_{1}A_{3}, three nonnull alleles and a null allele A_{1}A_{2} × A_{3}O, or two null alleles and two nonnull alleles A_{1}O × A_{2}O;
B. Loci that are heterozygous in both parents and segregate in a 1:2:1 ratio, which include three groups:
B_{1}. One parent has two different dominant alleles and the other has one dominant allele and one null allele, e.g., A_{1}A_{2} × A_{1}O;
B_{2}. The reciprocal of B_{1};
B_{3}. Both parents have the same genotype of two codominant alleles, i.e., A_{1}A_{2} × A_{1}A_{2};
C. Loci that are heterozygous in both parents and segregate in a 3:1 ratio, i.e., A_{1}O × A_{1}O;
D. Loci that are in the testcross configuration between the parents and segregate in a 1:1 ratio, which include two groups:
D_{1}. Heterozygous in one parent and homozygous in the other, including three alleles A_{1}A_{2} × A_{3}A_{3}, two alleles A_{1}A_{2} × A_{1}A_{1}, A_{1}A_{2} × OO and A_{2}O × A_{1}A_{1}, and one allele (with three null alleles) A_{1}O × OO;
D_{2}. The reciprocals of D_{1}.
The marker group A is regarded as containing fully informative markers because of the complete distinction of the four progeny genotypes. The other six groups all contain the partially informative markers since some progeny genotype cannot be phenotypically separated from other genotypes. This incomplete distinction leads to the segregation ratios 1:2:1 (B), 3:1 (C) and 1:1 (D). Note that marker group D can be viewed as fully informative if we are only interested in the heterozygous parent.
In the preceding section, we defined a (4 × 4)matrix H for joint genotype frequencies between two fully informative markers. But for partially informative markers, only the joint phenotypes can be observed and, thus, the joint genotype frequencies, as shown in H, will be collapsed according to the same phenotype. Wu et al. [11] designed specific incidence matrices (I) relating the genotype frequencies to the phenotype frequencies for different types of markers. Here, we use the notation for a (b_{1} × b_{2}) matrix of the phenotype frequencies between two partially informative markers, where b_{1} and b_{2} are the numbers of distinguishable phenotypes for markers and , respectively. Correspondingly, we have . The EM algorithm can then be developed to estimate the recombination fraction between any two partial informative markers.
E Step: At step τ, based on the matrix (DH)' derived from the current estimate r^{{τ}}, calculate the expected number of recombination events between the two markers for a given progeny genotype and :
M Step: Calculate r^{{τ+1}}using the equation,
The E and M steps between Eqs. (8) – (11) are repeated until the estimate converges to a stable value.
Threelocus analysis
A general framework
Let r_{12} denote the recombination fraction between markers and , with r_{23} and r_{13}defined similarly. These recombination fractions are associated with the probabilities with which a crossover occurs between markers and and between markers and . The event that a crossover or no crossover occurs in each interval is denoted by D_{11} and D_{00}, respectively, whereas the events that a crossover occurs only in the first interval or in the second interval is denoted by D_{10} and D_{01}, respectively. The probabilities of these events are denoted by d_{00}, d_{01}, d_{10}and d_{11}, respectively, whose sum equals 1. According to the definition of recombination fraction as the probability of a crossover between a pair of loci, we have r_{12} = d_{10} + d_{11}, r_{23} = d_{01} + d_{11} and r_{13} = d_{01} + d_{10}. These relationships have been used by Haldane [17] to derive the map function that converts the recombination fraction to the corresponsding genetic distance.
Estimation from threepoint analysis of the recombination fraction ( ± SD) and the parental diplotype probabilities of parent P ( ) and Q ( ) for five markers in a fullsib family of n = 100
Parental diplotype 

 

Marker  P  ×  Q  Case 1  Case 2 

 Case 1  Case 2 

 
Recombination fraction = 0.05  
        
 a  b  c  d  
        0.0511 ± 0.0175  
 a  b  a  b  0.1008 ± 0.0298  0.9978  0.9986  
        0.0578 ± 0.0269  0.0557 ± 0.0312  
 a  o  ×  o  a  0.9977  0  0.0988 ± 0.0277  1  0  
        0.0512 ± 0.0307  0.0476 ± 0.0280  1  1/0  
 a  b  b  b  0.0932 ± 0.0301  1  1/0  1  1/0  
        0.0514 ± 0.0229  
 a  b  c  d  1  1  
        
Recombination fraction = 0.20  
        
 a  b  c  d  
        0.2026 ± 0.0348  
 a  b  a  b  0.3282 ± 0.0482  0.9918  0.9916  
        0.2240 ± 0.0758  0.2408 ± 0.0939  
 a  o  ×  o  a  0.9944  0  0.3241 ± 0.0488  1  0  
        0.1927 ± 0.0613  0.1824 ± 0.0614  
 a  b  b  b  0.3161 ± 0.0502  1  1/0  1  1/0  
        0.2017 ± 0.0393  
 a  b  c  d  1  1  
       
The joint genotype frequencies of the three markers can be viewed as a mixture of 16 diplotype combinations and three orders, weighted by their occurring probabilities, and is expressed as
Similarly, the expected number of recombination events contained within a progeny genotype is the mixture of the different diplotype and order combinations, expressed as:
Also define
The occurring probabilities of the three marker orders are the mixture of all diplotype combinations, expressed, in matrix notation, as
We implement the EM algorithm to estimate the MLEs of the recombination fractions between the three markers. The general equations formulating the iteration of the {τ + 1}th EM step are given as follows:
E Step: As step τ, calculate the expected number of recombination events associated with D_{00}(α), D_{01} (β), D_{10}(γ), D_{11}(δ) for the (j_{1}j_{2}j_{3})th progeny genotype (where j_{1}, j_{2} and j_{3} denote the progeny genotypes of the three individual markers, respectively):
where n_{j 1j 2j 3}denote the number of progeny with a particular threemarker genotype, h_{j 1j 2j 3}, , , , , p_{1(j 1j 2j 3)}, p_{2(j 1j 2j 3)}, q_{1(j 1j 2j 3)}and q_{2(j 1j 2j 3)}are the (j_{1}j_{2}j_{3})th element of matrices H, D_{00}, D_{01}, D_{10}, D_{11}, P_{1}, P_{2}, Q_{1} and Q_{2}, respectively.
M Step: Calculate , , and using the equations,
The E and M steps are repeated among Eqs. (19) – (32) until d_{00}, d_{01}, d_{10} and d_{11} converge to values with satisfied precision. From the MLEs of the g's, the MLEs of recombination fractions r_{12}, r_{13} and r_{23} can be obtained according to the invariance property of the MLEs.
Model for partial informative markers
Consider three partially informative markers with the numbers of distinguishable phenotypes denoted by b_{1}, b_{2} and b_{3}, respectively. Define is a (b_{1}b_{2} × b_{3}) matrix of genotype frequencies for three partially informative markers. Similarly, we define , and .
Using the procedure described in Section (2.2), we implement the EM algorithm to estimate the MLEs of the recombination fractions among the three partially informative markers.
mpoint analysis
Threepoint analysis considering the dependence of recombination events among different marker intervals can be extended to perform the linkage analysis of an arbitrary number of markers. Suppose there are m ordered markers on a linkage group. The joint genotype probabilities of the m markers form a (4^{m1}× 4)dimensional matrix. There are 2^{m1}× 2^{m1}such probability matrices each corresponding to a different parental diplotype combination. The reasonable estimates of the recombination fractions rely upon the characterization of a most likely parental diplotype combination based on the multilocus likelihood values calculated.
The mmarker joint genotype probabilities can be expressed as a function of the probability of whether or not there is a crossover occurring between two adjacent markers, where l_{1}, l_{2}, ..., l_{m1}are the indicator variables denoting the crossover event between markers and , markers and , ..., and markers and , respectively. An indicator is defined as 1 if there is a crossover and 0 otherwise. Because each indicator can be taken as one or zero, there are a total of 2^{m1}D's.
Monte Carlo simulation
Simulation studies are performed to investigate the statistical properties of our model for simultaneously estimating linkage, parental diplotype and gene order in a fullsib family derived from two outbred parents. Suppose there are five markers of a known order on a chromosome. These five markers are segregating differently in order, 1:1:1:1, 1:2:1, 3:1, 1:1 and 1:1:1:1. The diplotypes of the two parents for the five markers are given in Table 1 and using these two parents a segregating fullsib family is generated. In order to examine the effects of parameter space on the estimation of linkage, parental diplotype and gene order, the fullsib family is simulated with different degrees of linkage (r = 0.05 vs. 0.20) and different sample sizes (n = 100 vs. 200).
As expected, the estimation precision of the recombination fraction depends on the marker type, the degree of linkage and sample size. More informative markers, more tightly linked markers and larger sample sizes display greater estimation precision of linkage than less informative markers, less tightly linked markers and smaller sample sizes (Tables 1 and 2). To save space, we do not give the results about the effects of sample size in the tables. Our model can provide an excellent estimation of parental linkage phases, i.e., parental diplotype, in twopoint analysis. For example, the MLE of the probability (p or q) of parental diplotype is close to 1 or 0 (Table 1), suggesting that we can always accurately estimate parental diplotypes. But for two symmetrical markers (e.g., markers and in this example), two sets of MLEs, = 1, = 0 and = 0, = 1, give an identical likelihood ratio test statistic. Thus, twopoint analysis cannot specify parental diplotypes for symmetrical markers even when the two parents have different diplotypes.
The estimation precision of linkage can be increased when a threepoint analysis is performed (Table 2), but this depends on different marker types and different degrees of linkage. Advantage of threepoint analysis over twopoint analysis is more pronounced for partially than fully informative markers, and for less tightly than more tightly linked markers. For example, the sampling error of the MLE of the recombination fraction (assuming r = 0.20) between markers and from twopoint analysis is 0.0848, whereas this value from a threepoint analysis decreases to 0.0758 when combining fully informative marker but increases to 0.0939 when combining partially informative marker . The threepoint analysis can clearly determine the diplotypes of different parents as long as one of the three markers is asymmetrical. In our example, using either asymmetrical marker or , the diplotypes of the two parents for two symmetrical markers ( and ) can be determined. Our model for threepoint analysis can determine a most likely gene order. In the threepoint analyses combining markers , markers and marker , the MLEs of the probabilities of gene order are all almost equal to 1, suggesting that the estimated gene order is consistent with the order hypothesized.
To demonstrate how our linkage analysis model is more advantageous over the existing models for a fullsib family population, we carry out a simulation study for linked dominant markers. In twopoint analysis, two different parental diplotype combinations are assumed: (1) [aa] [oo] × [aa] [oo] (cis × cis) and (2) [ao] [oa] × [ao] [oa] (trans × trans). The MLE of the linkage under combination (2), in which two dominant alleles are in a repulsion phase, is not as precise as that under combination (1), in which two dominant nonalleles are in a coupling phase [12]. For a given data set with unknown linkage phase, the traditional procedure for estimating the recombination fraction is to calculate the likelihood values under all possible linkage phase combinations (i.e., cis × cis, cis × trans, trans × cis and trans × trans). The combinations, cis × cis and trans × trans, have the same likelihood value, with the MLE of one combination being equal to the subtraction of the MLE of the second combination from 1. The same relationship is true for cis × trans and trans × cis. A most likely phase combination is chosen corresponding to the largest likelihood and a legitimate MLE of the recombination fraction (r ≤ 0.5) [10].
Comparison of the estimation of the linkage and parental diplotype between two dominant markers in a fullsib family of n = 100 from the traditional and our model
Traditional model  Our model  

cis × cis  cis × trans  trans × cis  trans × trans  
Data simulated from cis × cis  
Correct diplotype combination  Correct  Incorrect  Incorrect  Incorrect  
Loglikelihood^{ a }  46.2  92.3  92.3  46.2  
under each diplotype combination  0.1981 ± 0.0446  0.5000 ± 0.0000  0.5000 ± 0.0000  0.8018 ± 0.0446  
Estimated diplotype combination  Selected  
under correct diplotype combination  0.1981 ± 0.0446  0.1982 ± 0.0446  
Diplotype probability for parent P ( )  1.0000 ± 0.0000  
Diplotype probability for parent Q ( )  1.0000 ± 0.0000  
Data simulated from trans × trans  
Correct diplotype combination  Incorrect  Incorrect  Incorrect  Correct  
Loglikelihood^{ a }  89.6  89.6  89.6  89.6  
under each diplotype combination  0.8573 ± 0.1253  0.0393 ± 0.0419  0.0393 ± 0.0419  0.1426 ± 0.1253  
Estimated diplotype combination  Selected  Selected  
under correct diplotype combination  0.1426 ± 0.1253  0.1428 ± 0.1253  
Diplotype probability for parent P ( )  0.0000 ± 0.0000  
Diplotype probability for parent Q ( )  0.0000 ± 0.0000 
Comparison of the estimation of the linkage and gene order between three dominant markers in a fullsib family of n = 100 from the traditional and our model
MLE  Traditional model  Our model  



 
Data stimulated from [aaa] [ooo] × [aaa] [ooo]  
Correct gene order  Correct  Incorrect  Incorrect  
Estimated best gene order (%^{ a })  100  0  0  
 0.2047 ± 0.0422  0.2048 ± 0.0422  
 0.1980 ± 0.0436  0.1985 ± 0.0434  
 0.3245 ± 0.0619  0.3235 ± 0.0618  
 0.9860 ± 0.0105  
 0.0060 ± 0.0071  
 0.0080 ± 0.0079  
Data simulated from [aao] [ooa] × [aao] [ooa]  
Correct gene order  Correct  Incorrect  Incorrect  
Estimated best gene order (%^{ a })  80  11  9  
 0.1991 ± 0.0456  0.8165 ± 0.1003  0.9284 ± 0.0724  0.2104 ± 0.0447 
 0.1697 ± 0.0907  0.8220 ± 0.0338  0.1636 ± 0.0608  0.2073 ± 0.0754 
 0.3218 ± 0.0755  0.2703 ± 0.0586  0.7821 ± 0.0459  0.2944 ± 0.0929 
 0.9952 ± 0.0058  
 0.0045 ± 0.0058  
 0.0003 ± 0.0015 
Our model is further used to perform joint analyses including more than three markers. When the number of markers increases, the number of parameters to be estimated will be exponentially increased. For fourpoint analysis, the speed of convergence was slow and the accuracy and precision of parameter estimation have been affected for a sample size of 200 (data not shown). According to our simulation experience, the improvement of morethanthreepoint analysis can be made possible by increasing sample size or by using the estimates from two or threepoint analysis as initial values.
A worked example
We use an example from published literature [18] to demonstrate our unifying model for simultaneous estimation of linkage, parental diplotype and gene order. A cross was made between two triple heterozygotes with genotype AaVvXx for markers , and . Because these three markers are dominant, the cross generates 8 distinguishable genotypes, with observations of 28 for A_{}/V_{}/X_{}, 4 for A_{}/V_{}/xx, 12 for A_{}/vv/X_{}, 3 for A_{}/vv/xx, 1 for aa/V_{}/X_{}, 8 for aa/V_{}/xx, 2 for aa/vv/X_{} and 2 for aa/vv/xx. We first use twopoint analysis to estimate the recombination fractions and parental diplotypes between all possible pairs of the three markers. The recombination fraction between markers and is , whose the estimated parental diplotypes are [Av] [aV] × [AV] [av] or [AV] [av] × [Av] [aV]. The other two recombination fractions and the corresponding parental displotypes are estimated as , [Vx] [vX] × [VX] [vx] or [VX] [vx] × [Vx] [vX] and , [AX] [ax] × [AX] [ax], respectively. From the twopoint analysis, one of the two parents have dominant alleles from markers and are repulsed with the dominant alleles from marker .
Discussion
Several statistical methods and software packages have been developed for linkage analysis and map construction in experimental crosses and wellstructured pedigrees [2–6], but these methods need unambiguous linkage phases over a set of markers in a linkage group. For outcrossing species, such as forest trees, it is not possible to know exact linkage phases for any of two parents that are crossed to generate a fullsib family prior to linkage analysis. This uncertainty about linkage phases makes linkage mapping in outcrossing populations much more difficult than that in phaseknown pedigrees [7, 9].
In this article we present a unifying model for simultaneously estimating the linkage, parental diplotype and gene order in a fullsib family derived from two outbred parents. As demonstrated by simulation studies, our model is robust to different parameter space. Compared to the traditional approaches that calculate the likelihood values separately under all possible linkage phases or orders [9, 10, 18], our approach is more advantageous in three aspects. First, it provides a onestep analysis of estimating the linkage, parental diplotype and gene order, thus facilitating the implementation of a general method for analyzing any segregating type of markers for outcrossing populations in a package of computer program. For some shortgenerationinterval outcrossing species, we can obtain marker information from grandparents, parents and progeny. The model presented here allow for the use of marker genotypes of the grandparents to derive the diplotype of the parents. Second, our model for the first time incorporates gene ordering into a unified linkage analysis framework, whereas most earlier studies only emphasized on the characterization of linkage phases through a multilocus likelihood analysis [11, 14, 15]. Instead of a comparative analysis of different orders, we proposed to determine a most likely gene order by estimating the order probabilities.
Third, and most importantly, our unifying approach can significantly improve the estimation precision of the linkage for dominant markers whose alleles are in repulsion phase. Previous analyses have indicated that the estimate of the linkage between dominant markers in a repulsion phase is biased and imprecise, especially when the linkage is not strong and when sample size is small [12]. There are two reasons for this: (1) the linkage phase cannot be correctly determined, and/or (2) there is a fairly high possibility (20%) of detecting a wrong gene order. Our approach provides more precise estimates of the recombination fraction because correct parental diplotypes and a correct gene order can be determined.
Our approach will be broadly useful in genetic mapping of outcrossing species. In practice, a twopoint analysis can first be performed to obtain the pairwise estimates of the recombination fractions and using this pairwise information markers are grouped based on the criteria of a maximum recombination fraction and minimum likelihood ratio test statistic [2]. The parental diplotypes of markers in individual groups are constructed using a threepoint analysis. With a limited sample size available in practice, we do not recommend morethanthreepoint analysis because this would bring too many more unknown parameters to be precisely estimated. If such an analysis is desirable, however, one may use the results from these lowerpoint analyses as initial values to improve the convergence rate and possibly the precision of parameter estimation.
In any case, our two and threepoint analysis has built a key stepping stone for map construction through two approaches. One is the leastsquares method, as originally developed by Stam [5], that can integrate the pairwise recombination fractions into reconstruction of multilocus linkage map. The second is to use the hidden Markov chain (HMC) model, first proposed by Lander and Green [2], to construct genetic linkage maps by treating map construction as a combinatorial optimization problem. The simulated annealing algorithm [19] for searching for optima of the multilocus likelihood function need to be implemented for the HMC model. A userfriendly package of software that is being written by the senior author will implement two and threepoint analyses as well as the algorithm for map construction based on the estimates of pairwise recombination fractions. This software will be online available to the public.
Our maximum likelihoodbased approach is implemented with the EM algorithm. We also incorporate the Gibbs sampler [20] into the estimation procedure of the mixture model for the linkage characterizing different parental diplotypes and gene orders of different markers. The results from the Gibbs sampler are broadly consistent with those from the EM algorithm, but the Gibbs sampler is computationally more efficient for a complicated problem than the EM algorithm. Therefore, the Gibbs sampler may be particularly useful when our model is extended to consider multiple fullsib families in which the parents may be selected from a natural population. For such a multifamily design, some population genetic parameters describing the genetic structure of the original population, such as allele frequencies and linkage disequilibrium, should be incorporated and estimated in the model for linkage analysis. It can be anticipated that the Gibbs sampler will play an important role in estimating these parameters simultaneously along with the linkage, linkage phases, and gene order.
Declarations
Acknowledgements
We thank two anonymous referees for their constructive comments on the manuscript. This work is partially supported by a University of Florida Research Opportunity Fund (02050259) and a University of South Florida Biodefense Grant (722206112) to R. W. The publication of this manuscript is approved as Journal Series No. R10073 by the Florida Agricultural Experiment Station.
Authors’ Affiliations
References
 Flint J, Mott R: Finding the molecular basis of quantitative traits: Successes and pitfalls. Nat Rev Genet. 2001, 2: 437445. 10.1038/35076585.View ArticlePubMedGoogle Scholar
 Lander ES, Green P: Construction of multilocus genetic linkage maps in humans. Proc Natl Acd Sci USA. 1987, 84: 23632367.View ArticleGoogle Scholar
 Lander ES, Green P, Abrahamson J, Barlow A, Daly MJ, Lincoln SE, Newburg L: MAPMAKER: an interactive computer package for constructing primary genetic linkage maps of experimental and natural populations. Genomics. 1987, 1: 174181.View ArticlePubMedGoogle Scholar
 Green P, Falls K, Crooks S: Documentation for CRIMAP, version 2.4. Washington Univ. School of Medicine, St. Louis, MO. 1990Google Scholar
 Stam P: Construction of integrated genetic linkage maps by means of a new computer package: JOINMAP. Plant J. 1993, 3: 739744. 10.1046/j.1365313X.1993.03050739.x.View ArticleGoogle Scholar
 Matise TC, Perlin M, Chakravarti A: Automated constrcution of genetic linkage maps using an expert system (MULTIMAP): a human genome linkage map. Nat Genet. 1994, 6: 384390.View ArticlePubMedGoogle Scholar
 Hitter E, Gebhardt C, Salamini F: Estimation of recombination frequencies and construction of RFLP linkage maps in plants from crosses between heterozygous parents. Genetics. 1990, 125: 645654.Google Scholar
 Arus P, Olarte C, Romero M, Vargas F: Linkage analysis of 10 isozyme genes in Fl segregating almond progenies. J Am Soc Hort Sci. 1994, 119: 339344.Google Scholar
 Ritter E, Salamini F: The calculation of recombination frequencies in crosses of allogamous plant species with applications to linkage mapping. Genet Res. 1996, 67: 5565.View ArticleGoogle Scholar
 Maliepaard C, Jansen J, van Ooijen JW: Linkage analysis in a fullsib family of an outbreeding plant species: overview and consequences for applications. Genet Res. 1997, 70: 237250. 10.1017/S0016672397003005.View ArticleGoogle Scholar
 Wu RL, Ma CM, Painter I, Zeng ZB: Simultaneous maximum likelihood estimation of linkage and linkage phases in outcrossing populations. Theor Pop Biol. 2002, 61: 349363. 10.1006/tpbi.2002.1577.View ArticleGoogle Scholar
 Wu RL, Ma CM, Wu SS, Zeng ZB: Linkage mapping of sexspecific differences. Genet Res. 2002, 79: 8596. 10.1017/S0016672301005389.View ArticlePubMedGoogle Scholar
 Grattapaglia D, R Sederoff: Genetic linkage maps of Eucalyptus grandis and Eucalyptus urophylla using a pseudotestcross: mapping strategy and RAPD markers. Genetics. 1994, 137: 11211137.PubMed CentralPubMedGoogle Scholar
 Ling S: Constructing genetic maps for outbred experimental crosses. Ph.D. thesis, University of California, Berkeley, CA. 1999Google Scholar
 Butcher PA, Williams ER, Whitaker D, Ling S, Speed TP, Moran CF: Improving linkage analysis in outcrossed forest trees – an example from. Acacia mangium. Theor Appl Genet. 2002, 104: 11851191. 10.1007/s0012200108201.View ArticlePubMedGoogle Scholar
 Dempster AP, Laird NM, Rubin DB: Maximum likelihood from incomplete data via EM algorithm. J Roy Stat Soc Ser B. 1977, 39: 138.Google Scholar
 Haldane JBS: The combination of linkage values and the calculation of distance between the loci of linked factors. J Genet. 1919, 8: 299309.View ArticleGoogle Scholar
 Ridout MS, Tong S, Vowden CJ, Tobutt KR: Threepoint linkage analysis in crosses of allogamous plant species. Genet Res. 1998, 72: 111121. 10.1017/S0016672398003371.View ArticleGoogle Scholar
 van Laarhoven PJM, Aarts EHL: Simulated Annealing: Theory and Application. 1987, D. Reide Publishing Co., Dordrecht, The NetherlandsView ArticleGoogle Scholar
 Casella G: Empirical Bayes Gibbs sampling. Biostatistics. 2001, 2: 485500. 10.1093/biostatistics/2.4.485.View ArticlePubMedGoogle Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.