| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
,1
* Department of Statistics, University of Florida, Gainesville, Florida 32611
College of Life Sciences, Zhejiang Forestry University, Lin'an, Zhejiang 311300, People's Republic of China
1 Corresponding author: Department of Statistics, 533 McCarty Hall C, University of Florida, Gainesville, FL 32611.
E-mail: rwu{at}stat.ufl.edu
 |
ABSTRACT |
|---|
 TOP ABSTRACT EXPERIMENTAL DESIGNSTATISTICAL METHODRESULTSDISCUSSIONACKNOWLEDGEMENTSLITERATURE CITED |
|---|
A number of quantitative genetic models have been proposed to quantify the contribution of direct genetic effects to the phenotypic variance (LYNCH and WALSH 1998). These models have now been refined to characterize and map individual genetic loci [or quantitative trait loci (QTL)] affecting complex phenotypes, using molecular linkage maps (LANDER and BOTSTEIN 1989; WU et al. 2002; reviewed in JANSEN 2000). While these refined models are instrumental in the genetic mapping of QTL triggering direct effects on a variety of traits, none of them can be effectively used to map QTL of indirect genetic loci because of their failure to take into account gene actions and interaction from different genomes. Many of the current results from the mapping of direct-effect QTL may be insufficient or misleading in the characterization of genetic architecture due to the omission of the potentially important sources of genetic variance contributed by indirect genetic effects (see WOLF et al. 2002).
The presence of indirect genetic effects means that the phenotypic effect of a gene in an individual relies on the genes possessed by its social partners. Perhaps the most important indirect genetic effects occur between parents and their offspring (WOLF 2003). An enormous body of work on parent-offspring conflict suggests that life history traits and other development-related traits are affected by genes expressed in mothers and by maternally and paternally inherited genes expressed in offspring (AGRAWAL et al. 2001; HAGER and JOHNSTONE 2003). Using simple mapping approaches, some researchers have begun to map QTL for indirect effects from maternal interactions (PERIPATO and CHEVERUD 2002; PERIPATO et al. 2002; WOLF et al. 2002). For example, WOLF et al. (2002) identified several direct and maternal-effect QTL for offspring preweaning growth in a cross of LG/J and SM/J inbred mouse strains. They further found that maternal-effect loci contribute much more strongly to the genetic variance in offspring growth than do direct-effect loci. However, results from simple mapping approaches are premature because they were not designed to understand the genetics underlying parent-offspring interactions.
Here we present a novel statistical model for characterizing the genetic effects of the maternal genome on offspring traits. In doing this, we incorporate into our mapping model epistatic interactions between one QTL from the maternal genome and other QTL from the offspring genome. In the QTL mapping study, the direct- and maternal-effect QTL mapped to distinct portions of the genome (WOLF et al. 2002). Epistasis, expressed as the effect of one gene contingent upon the expression of other genes, is thought to play a central role in trait evolution and speciation (DOEBLEY et al. 1995; LARK et al. 1995; WHITLOCK et al. 1995; PHILLIPS 1998; WADE 1998). Despite their considerable evolutionary significance (WOLF 2003), theoretical models of epistatic genetic effects resulting from maternal and offspring genomes are virtually lacking in the literature. To illustrate the idea of our mapping approach, we base our analysis on a simple backcross genetic design for an autogamous plant system. It is possible for our model to be extended to more complex designs, such as the F2, full-sib family, natural populations, and allogamous systems.
|
EXPERIMENTAL DESIGN |
|---|
|
TOPABSTRACTEXPERIMENTAL DESIGNSTATISTICAL METHODRESULTSDISCUSSIONACKNOWLEDGEMENTSLITERATURE CITED |
|---|
Consider two flanking markers, 
(t) and +1(t), derived from sporophytic plants of generation t in the backcross population. The recombination fraction between the two markers is denoted by r. A putative maternal QTL, denoted by 
(t) and located between these two markers (measured by the recombination fraction r1 with or r2 with +1), may affect an embryo trait of interest. Table 1
gives the conditional probabilities (
ij) of backcross individual i (i = 1, ... , n) carrying a maternal QTL genotype j [j = 1 for Qq(t) and 0 for qq(t)], given the four marker genotypes of the backcross population, MmM+1m+1(t), Mmm+1m+1(t), mm M+1m+1(t), and mmm+1m+1(t). This embryo trait is also affected by a QTL on the offspring genome of generation (t + 1), denoted by (t + 1). The offspring QTL may be located on either the same marker interval or a different marker interval, '(t) and '+1(t). Unlike the maternal QTL, the offspring QTL will segregate into three genotypes, QQ(t + 1), Qq(t + 1), and qq(t + 1), through the autogamous pollination of the backcross plants. We derive the conditional probabilities (ij', j' = 2, 1, 0) of these three offspring QTL genotypes given four different marker genotypes of the backcross, M'm' M'+1m'+1(t), M'm'm'+1m'+1(t), m'm'M'+1m'+1(t), and m'm'm'+1m'+1(t) (Table 2)
. The conditional probabilities of joint maternal-offspring QTL genotypes given the marker genotypes of two different marker intervals can be expressed as the product of the corresponding conditional probabilities, i.e.,
 |
|
|
|
(t), and a(t + 1) and d(t + 1) be the additive and dominant effects of the offspring QTL, (t + 1). These two QTL interact to form the additive x additive (iaa) and the additive x dominant epistatic effects (iad), respectively, between the two different genomes. The genotypic means of the six joint QTL are expressed as
 | (1) |
The additive, dominant, and epistatic effects of the two QTL can be estimated from the estimated genotypic means by solving the above regular equations, as can be seen in WU et al. (2002).
|
STATISTICAL METHOD |
|---|
|
TOPABSTRACTEXPERIMENTAL DESIGNSTATISTICAL METHODRESULTSDISCUSSIONACKNOWLEDGEMENTSLITERATURE CITED |
|---|
 | (2) |
= (1, ... , k) are the mixture proportions that are constrained to be nonnegative and sum to unity; 
= (1, ... , k) are the component-specific parameters, with i being specific to component i; and is a parameter (i.e., residual variance) that is common to all components.
Modeling maternal-offspring interactions contains two major tasks: (1) Derive the mixture proportions () expressed as the conditional probabilities of joint maternal-offspring QTL genotypes given marker genotypes (Tables 1–3) and the density functions expressed in terms of six QTL genotypic means and residual variance; (2) provide estimates of unknown parameters 
= (, , ) included in the mixture model using statistical and computational algorithms, such as the maximum-likelihood method (LANDER and BOTSTEIN 1989; WU et al. 2002) and the Bayesian approach (SATAGOPAN et al. 1996; ROBERT and CASELLA 1999). Here, we implement the expectation-maximization (EM) algorithm to obtain the maximum-likelihood estimates (MLEs) of the unknown parameters.
EM algorithm:
On the basis of the mixture model (2), we formulate the likelihood of the marker data and embryo trait phenotypes controlled by the putative QTL as
 | (3) |
contains QTL effects (Equation 1), QTL positions (Tables 1– 3), and residual variance (
2). These parameters are solved using the EM algorithm (DEMPSTER et al. 1977) described below.
In the E step, the conditional probabilities (priors) of the QTL genotypes given the marker genotypes and the normal distribution function are used to calculate
 | (4) |
In the M step, the calculated posterior probabilities were used to solve the unknown parameters
 | (5) |
 | (6) |
Iterations are repeated between Equations 4 and 6 until convergence. The values at convergence are the MLEs. With the MLEs of µj's, the MLEs of the overall mean, the additive, dominant, and epistatic effects of the QTL, as indicated in Equation 1, can be obtained by solving a system of regular equations. The estimation of the QTL positions can be obtained using a grid approach. This approach views the QTL positions as a known parameter in the likelihood function (3) by scanning the QTL over all marker intervals. The positions corresponding to the maximum of the log-likelihood ratio across a linkage group are the MLEs of the QTL positions.
Hypothesis tests:
Our model provides hypothesis tests for the presence of QTL affecting the expression of an offspring trait and their additive, dominant, and maternal-offspring epistatic interactions. The presence of QTL can be tested by using the following hypotheses:
 | (7) |
The test statistic for testing the above hypotheses is calculated as the log-likelihood ratio (LR) of the full model (H1) over the reduced model (H0),
 |
and 
denote the MLEs of the unknown parameters under H0 and H1, respectively. The LR is asymptotically 
2 distributed with 5 d.f. However, for a practical data set with a limited sample size, the critical threshold value for declaring the presence of QTL can be empirically calculated on the basis of permutation tests (CHURCHILL and DOERGE 1994).
We can also test whether the maternal QTL exerts a significant additive (main) effect on the offspring trait by simply formulating
 | (8) |
2 distributed with 1 d.f. The hypothesis about the effect of offspring's own QTL on the offspring trait is tested using
 | (9) |
Increasing evidence has been obtained for the role of maternal-offspring interactions in changing the rate and direction of evolution (WOLF 2003). The significance of maternal-offspring interactions (i.e., iaa and iad in our model) can be tested by formulating the hypotheses
 | (10) |
The likelihood value under H1 of Equation 10 is the same as that under H1 of Equation 7. The likelihood value under H0 of Equation 10 is calculated by plugging the MLEs of µ, a(t), a(t + 1), d(t + 1), and 2 under this null hypothesis into the likelihood function (3). The characterization of critical thresholds for hypotheses 8–10 can be empirically obtained through simulation studies.
|
RESULTS |
|---|
|
TOPABSTRACTEXPERIMENTAL DESIGNSTATISTICAL METHODRESULTSDISCUSSIONACKNOWLEDGEMENTSLITERATURE CITED |
|---|
, 
, 
) are cosegregating in a backcross and cotransmitted into the offspring generation of the backcross.
|
1) or the same marker interval (2). In this simulation, we consider 12 different schemes each representing a combination of three different sample sizes (n = 200, 400, and 800), two different QTL-effect sizes (R2 = 0.1 and 0.4), and two different QTL locations (1 and 2). For 1, the maternal QTL is located at 16 cM from the left marker of the first interval and the offspring QTL is located at 8 cM from the left marker of the third interval (Figure 2)
. For 2, the maternal and offspring markers are located at 8 and 16 cM from the left marker of the first interval, respectively (Figure 3)
. Knowing the locations of these assumed QTL, joint QTL-marker genotypes can be simulated by viewing the QTL as "markers" using the two-stage hierarchical bifurcation strategy described above (Figure 1). The phenotypic values of an assumed offspring trait are simulated as the sum of the QTL genotypic means (determined by both the maternal and offspring QTL) and normally distributed random errors.
|
|
= 0.001 obtained from permutation tests, suggesting that our model has a power of 100% to detect the QTL hidden in our simulated dataset. The locations of both the maternal and offspring QTL can be reasonably estimated (Figures 2 and 3). Second, in general, all parameters can be quite reliably estimated using our model constructed with marker information purely from the maternal population. As expected, the additive genetic effects of the two QTL can be more reliably estimated than the dominant effects. Of different epistatic components, the additive x additive genetic effect can be estimated better than the additive x dominant effect. The overall mean and residual variance can be estimated precisely.
|
|
When both R2 and n are small, the estimates of the QTL parameters display similar precision for the two cases in which QTL are located at different (1; Table 4) or the same marker intervals (2; Table 5). The relative advantage of 1 over 2 for the precision of parameter estimation becomes more remarkable with the increased R2 value and sample size. It seems that the maternal QTL location can be estimated more precisely than the offspring QTL location when they are located at either different (Table 4) or the same marker intervals (Table 5). When R2 is small, the estimation precision of the QTL locations is not very responsive to sample size. But the QTL locations can be more precisely estimated with increased sample size at a large R2 value.
It is interesting to compare the results for genetic mapping of offspring traits from previous models and our model. LANDER and BOTSTEIN's (1989) model associates the marker genotypes with the phenotypes measured at the same generation, whereas WU et al.'s (2002) model associates the marker genotypes at the maternal generation with the phenotypes at the offspring generation. Unlike our model, these two models do not consider possible interactions between the QTL from the maternal and offspring genomes. The simulated data including the maternal-offspring interactions on each of the above 12 schemes were analyzed by the three models, aimed at mapping the QTL for the embryo. As expected, the two previous models cannot provide reasonable estimation of the embryo QTL parameters (results not shown). Also, their power to detect significant QTL was much reduced compared to our model.
We performed the hypothesis test for the significance of the additive x additive (iaa) and additive x dominant epistatic effects (iad) between different genomes. Using the hypothesized values in Tables 4 and 5, iaa and iad are detected to be significant for 80 of 100 simulation replicates at R2 = 0.1 and n = 200. This proportion is increased markedly when the QTL explain a larger proportion of the observed variance and/or when there is a larger sample size (results not shown).
Different gene action modes are suggested to affect the estimates of QTL parameters. An additional simulation study was conducted to investigate the impact of weak main (additive and dominant) effects vs. strong maternal-offspring interaction effects on the precision of parameter estimation. We consider an intermediate sample size (400), two different R2 values, and two different QTL distribution patterns (Table 6) . The simulation results are summarized below. First, both maternal-offspring additive x additive (iaa) and additive x dominant effects (iad) can be be better estimated when these effects are large. Second, large maternal-offspring interaction effects have no marked effects on the estimation of the additive or dominant effects. Third, large maternal-offspring effects would reduce the estimation precision of the two QTL located at the same interval.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTEXPERIMENTAL DESIGNSTATISTICAL METHODRESULTSDISCUSSIONACKNOWLEDGEMENTSLITERATURE CITED |
|---|
Unlike traditional mapping approaches that were all developed to estimate direct genetic effects, our model incorporates both direct and indirect genetic effects into a QTL mapping framework and displays two significant advantages: (1) The results from our model should be closer to biological reality, given the fact that indirect genetic effects account for a great proportion (50% or more) of the genetic variance (WOLF et al. 2002; WOLF 2003), and (2) the model is statistically more powerful because more genetic information (indirect genetic effects) is used. As shown by extensive simulation studies, our model is quite robust in that the QTL parameters can be reliably estimated for a modest sample size (200) and low R2 value (0.1). Increased sample sizes and R2 values can improve the power to detect QTL and the estimation precision of QTL parameters. It is interesting to note that increasing R2 may be more important for precise parameter estimation than increasing sample sizes to the same extent. The increased R2 values may result from more precise phenotyping of a quantitative trait or weaker interaction between the underlying QTL and the environment.
Our strategy for mapping epistatic QTL from the maternal and offspring genomes is based on the marker genotypes derived purely from the maternal genome. Although it is no problem for the maternal markers to infer the maternal QTL, as demonstrated in traditional mapping approaches (LANDER and BOTSTEIN 1989), such a one-stage sampling strategy is limited to predict the offspring QTL. This is because the marker information and offspring QTL are at different generations and undergo Mendelian transmission. In a study of mapping the endosperm of cereals, WU et al. (2002) proposed a two-stage hierarchical genotyping strategy in which both the maternal and offspring genomes are genotyped at the same set of molecular markers. It is likely that the incorporation of Wu et al.'s two-stage hierarchical genotyping strategy can provide more unique information to capture gene transmission.
Our model proposed here is based on a simple backcross design for an autogamous plant system. It is not difficult to extend the model to other reproductive systems, such as allogamous and mixed-pollinated and other mapping populations. More interesting, our model can also be modified to characterize the genes affecting maternal care (PERIPATO and CHEVERUD 2002). The amount of care provided by parents is determined through a complex interaction of offspring signals and responses by parents to those signals (AGRAWAL et al. 2001). Variation in maternal care results from two distinct genetic sources: variation among offspring in their ability to elicit care and variation among parents in their response to offspring signals. By estimating the genetic loci for maternal care derived from these two sources, we can test for one of the most fundamental genetic assumptions in the family conflict theory: i.e., offspring signaling is negatively genetically correlated with maternal provisioning (AGRAWAL et al. 2001; WOLF 2003).
Understanding the maternal and paternal genetic regulation of offspring development and/or maternal care helps to answer many fundamental evolutionary questions in both animals and plants. Our model can be applied immediately to evolutionary genetic studies. But it needs to consider the patterns of gene segregation and transmission in natural plant populations. Parameters characterizing population structure and organization, such as allele frequencies, linkage disequilibrium, and haplotype frequencies, should be incorporated into our mapping model. LOU et al. (2003) have recently derived a closed-form solution for precise and efficient estimates of haplotype frequencies. By incorporating Lou et al.'s model, we will be closer to unraveling the genetic basis of embryogenesis and organ development in natural animal and plant populations.
|
ACKNOWLEDGEMENTS |
|---|
|
TOPABSTRACTEXPERIMENTAL DESIGNSTATISTICAL METHODRESULTSDISCUSSIONACKNOWLEDGEMENTSLITERATURE CITED |
|---|
|
LITERATURE CITED |
|---|
|
TOPABSTRACTEXPERIMENTAL DESIGNSTATISTICAL METHODRESULTSDISCUSSIONACKNOWLEDGEMENTSLITERATURE CITED |
|---|
AGRAWAL, A. F., E. D. BRODIE, III and J. BROWN, 2001 Parent-offspring coadaptation and the dual genetic control of maternal care. Science 292: 1710–1712.
CHEVERUD, J. M., 2000 Detecting epistasis among quantitative trait loci, pp. 58–81 in Epistasis and the Evolutionary Process, edited by J. B. WOLF, E. D. BRODIE, III and M. J. WADE. Oxford University Press, New York.
CHEVERUD, J. M., 2003 Evolution in a genetically heritable social environment. Proc. Natl. Acad. Sci. USA 100: 4357–4359.
CHURCHILL, G. A., and R. W. DOERGE, 1994 Empirical threshold values for quantitative trait mapping. Genetics 138: 963–971.[Abstract]
DEMPSTER, A. P., N. M. LAIRD and D. B. RUBIN, 1977 Maximum likelihood from incomplete data via EM algorithm. J. R. Stat. Soc. Ser. B 39: 1–38.
DOEBLEY, J., A. STEC and C. GUSTUS, 1995 teosinte branched1 and the origin of maize: evidence for epistasis and the evolution of dominance. Genetics 141: 333–346.[Abstract]
FALCONER, D. S., 1965 Maternal effects and selection response, pp. 763–774 in Genetics Today: Proceedings of the 11th International Congress of Genetics, edited by S. J. GEERTS. Pergamon Press, Oxford.
HAGER, R., and R. A. JOHNSTONE, 2003 The genetic basis of family conflict resolution in mice. Nature 421: 533–535.[CrossRef][Medline]
JANSEN, R. C., 2000 Quantitative trait loci in inbred lines, pp. 567–597 in Handbook of Statistical Genetics, edited by D. J. BALDING, M. BISHOP and C. CANNINGS. John Wiley & Sons, New York.
LANDER, E. S., and D. BOTSTEIN, 1989 Mapping Mendelian factors underlying quantitative traits using RFLP linkage maps. Genetics 121: 185–199.
LARK, K. G., K. CHASE, F. ADLER, L. M. MANSUR and J. H. ORF, 1995 Interactions between quantitative trait loci in soybean in which trait variation at one locus is conditional upon a specific allele at another. Proc. Natl. Acad. Sci. USA 92: 4656–4660.
LI, L.-L., E. B. KEVERNE, S. A. APARICIO, F. ISHINO, S. C. BARTON et al., 1999 Regulation of maternal behavior and offspring growth by paternally expressed Peg3. Science 284: 330–333.
LOU, X.-Y., G. CASELLA, R. C. LITTELL, M. K. C. YANG and R. L. WU, 2003 A haplotype-based algorithm for multilocus linkage disequilibrium mapping of quantitative trait loci with epistasis in natural populations. Genetics 163: 1533–1548.
LYNCH, M., and B. WALSH, 1998 Genetics and Analysis of Quantitative Traits. Sinauer, Sunderland, MA.
MOUSSEAU, T. A., and H. DINGLE, 1991 Maternal effects in insect life histories. Annu. Rev. Entomol. 36: 511–534.[CrossRef]
MOUSSEAU, T. A., and C. FOX, 1998 Maternal Effects as Adaptations. Oxford University Press, New York.
PERIPATO, A. C., and J. M. CHEVERUD, 2002 Genetic influences on maternal care. Am. Nat. 160: S173–S185.[CrossRef]
PERIPATO, A. C., R. A. DE BRITOB, T. T. VAUGHNB, L. S. PLETSCHERB, S. R. MATIOLIA et al., 2002 Quantitative trait loci for maternal performance for offspring survival in mice. Genetics 162: 1341–1353.
PHILLIPS, P. C., 1998 The language of gene interaction. Genetics 149: 1167–1171.
REZNICK, D. N., 1991 Maternal effects in fish life histories, pp. 780–793 in The Unity of Evolutionary Biology: Proceedings of the Fourth International Congress of Systematic and Evolutionary Biology, Vol. II, edited by E. C. DUDLEY. Dioscordes Press, Portland, OR.
ROACH, D. A., and R. D. WULFF, 1987 Maternal effects in plants. Annu. Rev. Ecol. Syst. 18: 209–235.[CrossRef]
ROBERT, C. P., and G. CASELLA, 1999 Monte Carlo Statistical Methods. Springer, New York.
SATAGOPAN, J. M., Y. S. YANDELL, M. A. NEWTON and T. C. OSBORN, 1996 A Bayesian approach to detect quantitative trait loci using Markov chain Monte Carlo. Genetics 144: 805–816.[Abstract]
VIELLE-CALZADA, J.-P., R. BASKAR and U. GROSSNIKLAUS, 2000 Delayed activation of the paternal genome during seed development. Nature 404: 91–94.[CrossRef][Medline]
WADE, M. J., 1998 The evolutionary genetics of maternal effects, pp. 5–21 in Maternal Effects as Adaptations, edited by T. MOUSSEAU and C. FOX. Oxford University Press, Oxford.
WEIJERS, D., N. GELDNER, R. OFFRINGA and G. JORGENS, 2001 Early paternal gene activity in Arabidopsis. Nature 414: 709–710.[CrossRef][Medline]
WHITLOCK, M. C., P. C. PHILLIPS, F. B. G. MOORE and S. J. TONSOR, 1995 Multiple fitness peaks and epistasis. Annu. Rev. Ecol. Syst. 26: 601–629.[CrossRef]
WOLF, J. B., 2000 Gene interactions from maternal effects. Evolution 54: 1882–1898.[CrossRef][Medline]
WOLF, J. B., 2003 Genetic architecture and evolutionary constraint when the environment contains genes. Proc. Natl. Acad. Sci. USA 100: 4655–4660.
WOLF, J. B., E. D. BRODIE, III, J. M. CHEVERUD, A. J. MOORE and M. J. WADE, 1998 Evolutionary consequences of indirect genetic effects. Trends Ecol. Evol. 13: 64–69.[CrossRef]
WOLF, J. B., T. T. VAUGHN, L. S. PLETSCHER and J. M. CHEVERUD, 2002 Contribution of maternal effect QTL to genetic architecture of early growth in mice. Heredity 89: 300–310.[CrossRef][Medline]
WU, R. L., C.-X. MA, M. GALLO-MEAGHER, R. C. LITTELL and G. CASELLA, 2002 Statistical methods for dissecting triploid endosperm traits using molecular markers: an autogamous model. Genetics 162: 875–892.
This article has been cited by other articles:
 |
|
 |
|
  Z. Hu, X. Wang, and C. Xu A Method for Identification of the Expression Mode and Mapping of QTL Underlying Embryo-Specific Characters J. Hered., September 1, 2006; 97(5): 473 - 482. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 Y. Cui and R. Wu Mapping genome-genome epistasis: a high-dimensional model Bioinformatics, May 15, 2005; 21(10): 2447 - 2455. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
