Cucurbit Genetics Cooperative Report 19:13-16 (article 5) 1996
F.C. Serquen and J.E. Staub
Vegetable Crops Research, USDA/ARS, Department of Horticulture, University of Wisconsin-Madison, WI 53706
Introduction. Our laboratory has been interested in using random amplified polymorphic DNAs (RAPDs) as a tool for marker-assisted selection (MAS) in cucumber (Cucumis sativus L.). Phenotype selection based on traits which are conditioned by additive allelic effects can produce economically important changes in breeding populations. MAS provides a potential for increasing selection efficiency by allowing for earlier selection and reducing plant population size used during selection. Selection for multiple loci or quantitative trait loci (QTL) using genetic markers can be effective if a significant association is found between a quantitative trait and markers (1,2,3). Nevertheless, the phenotypic variation that marker loci define is often non-additive, and is a function of genetic linkage, pleiotropy and environment (5). Thus, the efficiency of application of marker loci as predictors of phenotypic variation is dependent upon many factors, and predictions of response to selection (R) are often difficult.
The utility of MAS is further exacerbated when dominant markers are used for characterizing QTLs because of their inability to completely classified all possible phenotypes (i.e., aa, Aa, and AA). In this report, we use sex expression data (see the previous companion article in this issue) and a hypothetical data set in a simulation experiment to provide evidence for the decreased ability of RAPD markers to detect QTLs when compared to codominant markers.
Materials and Methods. There are typically two analytical methods used for the analysis and mapping of QTLs: one-way analysis of variance (ANOVA) and maximum likelihood estimation. While the one-way analysis approach involves test squares estimation and produces an alpha value as a test of significance, maximum likelihood is designed to choose values for variables which maximizea defined function and linkage is tested by the logarithm of the ratio (LOD) of two likelihood; i.e., the null hypothesis that there is not a QTL at a given site [interval mapping, (4)]. In many analyses a significance level of LOD> 3.0 is appropriate as an acceptance level of linkage between two loci. When many loci are considered in QTL mapping, a multilocus analysis involves either maximum likelihood or multi-regression analysis.
Comparison by one-way analysis of variance. Phenotypes in a codominant model are expressed as 1 (AA), 0.5 (Aa) and 0 (aa). In a dominant model phenotypes are expressed as 1 (AA, Aa) and 0 (aa). We have assumed two types of overdominance in our investigations. One type is where the phenotypic value of the heterozygote is higher than either parent and the parents themselves are similar (designated here as type-1), and the other type is where the phenotypic value of one of the parents is higher than the other (designated here as type-2, which may be most realistic) (Table 1.).
In the simulations presented here, overdominance models along with the additive and the dominant models are compared to define the efficiency of marker types for their ability to detect QTLs. A one-way ANOVA was performed on a data set which contained hypothetical values for dominant and codominant markers and a trait which was given a range of different types of gene action (Table 1). The mean square error resulting from each ANOVA was used as a sensitivity value for comparisons because it is directly related to the F-test performed in each case. Theoretically, one might expect that while in the case of additive or overdominance, codominant markers would have greater sensitivity in detecting QTLs than dominant markers, the sensitivity of codominant and dominant markers would be similar where the trait was inherited as a dominant.
Comparison by multi-locus analysis. The efficiency of QTL detection was also examined in a multi-locus analysis using backward elimination and forward selection of markers (x1-n ). Multi-locus analysis can be used to examine the association between a marker and a QTL, and estimate the contribution which a QTL makes to the expression of a trait’s phenotype. Theoretically, the ability of dominant and codominant markers to detect QTLs should differ because of the inability of dominant marker systems to classify all phenotypes. We tested this hypothesis by using a data set containing sex expression phenotypes in F3 families (see companion article, this issue). Depending upon the grouping of sex phenotypes, the expression of the F locus could be considered codominant (FF, F_, ff) or dominant (F_, ff). We also observed that a random amplified polymorphic DNA (RAPD), OP_AO7, segregated as a codominant marker in this population. Thus, a unique opportunity was provided which allowed for a trait and an independent RAPD marker to be considered as either having codominant or dominant inheritance in any combination in the analysis.
The regression model used attempted to define the predicted dependent variable (y), gynoecious sex expression (conditioned by the F locus) using morphological and marker phenotypes. Two or more x’s (e.g., markers) are used to give information about y by means of multiple regression on the x’s (y = Y + b1 x1 + b2 x2 + …bn xn) where b is the regression coefficient (the phenotypic effect associated with each marker) of a given x and Y represents the mean phenotypic distribution of the population). In the case of the analysis we modeled y as y = y + B1 x1 + b2 x2 (where x1 = the RAPD marker OP_AO7). This regression model was used in backward elimination and forward selection as follows:
Case 1 y = x1 and x2 (both codominant)
Case 2 y = x1 and x2 (both dominant)
Case 3 y = x1 and x2 (dominant and codominant)
Results and Discussion. It was determined that both through simulation and by the analysis of a data set composed of sex and associated RAPD marker phenotypes that QTL effects are underestimated by dominant molecular markers. This observation was made by comparing the sensitivity of dominant and codominant marker systems.
Comparison by one-way analysis of variance. The mean square error (MSE) was larger in each case of gene action examined for a dominant marker when compared to a codominant marker (Table 2). The increase in MSE in the case of dominant marker systems is due to the inability to distinguish the AA nd Aa as genotype classes.
This increase in sensitivity (lower MSEs) in codominant marker systems results in the potential detection of a higher frequency of QTLs when compared to dominant marker systems. Thus, dominant marker systems tend to provide a conservative estimate of the number of QTLs that affect a trait’s expression. Because of this fact, the misidentification of spurious association between dominant markers and QTLs is decreased. Moreover, the QTLs detected when individual dominant marker probability values (P<0.01) are combined in the analyses (P<0.05) result in accurate, but conservative, reflections of reality.
Comparison by multi-locus analysis. The results of multi-locus regression analysis was as follows (x1 = F and x2 = OP_AO7):
Forward selection |
Marker |
R2 |
P>4 |
| Case 1 Both codominant | 1 | 0.621 | 0.0001 |
| 2 | 0.015 | 0.0509 | |
| Case 2 Both dominant | 1 | 0.289 | 0.0002 |
| 2 | 0.027 | 0.0591 | |
| Case 3 Codominant & dominant | 1 | 0.621 | 0.0001 |
Backward elimination |
|||
| Case 1 Both codominant | 1 & 2 | 0.637* | 0.0001 |
| 0.0509 | |||
| Case 2 Both dominant | 1 & 2 | 0.637 | 0.0001 |
| 0.0509 | |||
| Case 3 Codominant & dominant | 1 | Kept co-dominant marker | |
| 2 | Eliminated dominant marker | ||
* combined R2 for both equations.
From these analyses, two conclusions may be drawn: 1) forward selection underestimates the contribution of dominant markers (case 2 above) to the phenotypic variance; and 2) in a combination codominant-dominant marker (case 3 above), the dominant marker tends to be eliminated from the model (both in forward selection and backward elimination). Markers with small effects tend to be eliminated from the model, thus supporting the contention of the conservative nature of dominant markers for the detection of QTLs.
Table 1. Raw data used for simulation to examine the detection of quantitative trait loci (QTL) using dominant and codominant markers by one-way ANOVA, for traits which are conditioned by different types of gene action.
Trait |
Marker |
|||||
Overdominance |
||||||
Individual |
Additive |
Dominance |
Type-1 |
Type-2 |
Dominant |
Codominant |
| 1 | 0.4 | 0.4 | 0.4 | 0.4 | 0 | 0 |
| 2 | 0.5 | 0.5 | 0.5 | 0.5 | 0 | 0 |
| 3 | 0.5 | 0.5 | 0.5 | 0.5 | 0 | 0 |
| 4 | 0.9 | 0.9 | 1.1 | 1.5 | 1 | 0.5 |
| 5 | 1 | 1 | 1 | 1.5 | 1 | 0.5 |
| 6 | 1 | 1 | 0.9 | 1.6 | 1 | 0.5 |
| 7 | 1 | 1 | 1 | 1.5 | 1 | 0.5 |
| 8 | 1 | 1 | 1.1 | 1.5 | 1 | 0.5 |
| 9 | 1.1 | 0.9 | 0.9 | 1.6 | 1 | 0.5 |
| 10 | 1.5 | 1 | 0.5 | 1 | 1 | 1 |
| 11 | 1.5 | 1 | 0.5 | 1 | 1 | 1 |
| 12 | 1.6 | 1 | 0.6 | 1 | 1 | 1 |
Table 2. One-way analysis of variance of dominant and codominant markers for traits which are conditioned by different types of gene action.
Gene Action |
|||||||||
Type of Marker |
Additivity |
Dominance |
Overdominance-1 |
Overdominance-2 |
|||||
df |
MS Error |
Pr>F |
MS Error |
Pr>F |
MS Error |
Pr>F |
Ms Error |
Pr>F |
|
| Dominant | 10 | 0.6022 | 0.0015 | 0.0222 | 0.0001 | 0.4889 | 0.03 | 2.0267 | 0.03 |
| Codominant | 9 | 0.3333 | 0.0001 | 0.0200 | 0.0001 | 0.0533 | 0.0001 | 0.0267 | 0.0001 |
Literature Cited
- Edwards, M amd L:. Johnston. 1994. RFLPs for rapid recurrent selection.Proc. Symposium on Analysis of Molecular Marker Data, p. 33-40. August, 1994. Corvallis, Oregon.
- Edwards, M.D. and N.J. Page.1994. Evaluation of marker-assisted selection through computer simulation. Theor. Appl. Genet. 88:376-382.
- Lande, R. and R. Thompson. 1990. Efficiency of marker-assisted selection in the improvement of quantitative traits. Genetics 124:743-756.
- Lander, E.S. and D. Botstein. 1989. Mapping Mendelian factors underlying quantitative traits using RFLP linkage maps. Genetics 12:185-199.
- 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. Proc. Natl. Acad. Sci. 92:4656-4660.