Genotype to phenotype map and the G-matrix

Question

Suppose I have a genotype-phenotype map defined by the matrix $\mathbf{Z}$:

       

The scalars $G,P$ represent the number of genotypes and traits, respectively. The map assigns a set of phenotypic traits (the columns) to a set of genotypes (the rows), where $z_{ij}$ is the quantitative measure of an individual of genotype $i$ with respect to a trait $j$. I want to know how the average value of each of the phenotypic traits changes over time, suggesting I should use quantitative genetics theory and formulate the equation

$\Delta \bar{\boldsymbol{z}} = \mathbf{G}\mathbf{P}^{-1}\mathbf{s}$,

where, $\mathbf{G}$ is the additive genetic variance covariance matrix, $\mathbf{P}$ is the phenotypic variance covariance matrix, and $\mathbf{s}$ is a vector of covariances of each trait with fitness. Assume that I know the frequencies of the different genotypes in the population.

Question 1 Can I formulate the additive genetic variance-covariance matrix from the genotype-phenotype map alone (and pop freq)? Note, I don't know anything about the underlying genetic model - number of loci representing each genotype, dominance structure, epistatic interactions, etc.

Question 2 If I assume the genotypes represent haploid individuals, does this change things? Is the additive genetic variance always equal to the phenotypic variance (assuming no environmental effects) in a haploid model? In other words does $\mathbf{G}\mathbf{P}^{-1} = \mathbf{I}$? What I'm thinking is that epistatic variance might affect things, but it shouldn't affect heritability in a haploid model, which is what $\mathbf{G}\mathbf{P}^{-1}$ is a measure of.

Answer 1

From the pheno-geno map and the genotypes frequencies, you have the whole distribution of phenotypes in your population. The mean of the phenoype $n$, $P_n$ is the

$$\bar P_n = \sum f_{G_i} P_{G_i}$$

, where $f_{G_i}$ is the proportions of individuals having genotype $G_i$ and $P_{G_i}$ is the phenotype of the individuals with genotype $G_i$. Therefore, the variance of $P_n$ is

$$var(P_n) = \sum f_{G_i} (P_{G_i} - \bar P_n)^2$$.

The covariance between the phenotypic traits $n$ and $m$ is

$$cov(P_n,P_m) = \sum f_{G_i} (P_{G_i} - \bar P_n)(P_{G_i} - \bar P_m)$$

The only assumptions you need is that there is no environmental variance. I don't think the above calculations required any other assumption. $G$ can be defined at the haploid or diploid stage, it doesn't change anything.

The above calculations however give you the genetic variance and covariance but not the additive genetic variance covariance. I think (but I might be mistaken) that the additive genetic variance would be equal to

$$\sigma_D^2 = \sum_i f_i^2(1-f_i)^2(2 \cdot P_{n,i,12} - P_{n,i,11} - P_{n,i,22})^2$$

$$\sigma_A^2 = \sum_i 2f_i(1-f_i)(f_iP_{n,i,11}+(1-2f_i)P_{n,i,12} - (1-f_i)P_{n,i,22})^2$$

, where $f_i$ is the frequency of the allele $A$ at the locus $i$ (you should get those data by reformatting your geno-pheno map matrix) and $P_{n,i,11}$, $P_{n,i,12}$ and

$$P_{n,i,22} =\frac{1}{l-1} \sum_j P_{n,i,j,22}$$

is the $n^{th}$ phenotypic trait for the genotype $22$ and locus $i$ averaged over all other genotypes. There are $l$ genotypes in total. Of course, this works only for biallelic loci. I don't know what it would look like for more alleles.