CS 563

Lecture 5
Oct. 11, 2018

Assigned Reading: Falconer & Mackay, Chapters 6 and 7

Quantitative Genetics Model

We discussed previously the factors affecting the relationship between the genotype
and the range of phenotypic values that can be expressed by the given genotype under
environmental influence. We have discussed the following concepts into detail:
1. That quantitative genetics models are essentially general extensions of
Mendelian genetics, such that alleles that segregate in a Mendelian fashion are
extreme of the spectrum of allelic effects that are associated with quantitative
traits. For example, we saw that homozygous (a) and heterozygous (d) effects at
one locus affecting a trait can vary in magnitude. Variants with large homozygous
effects, such as the mh mutation affecting double muscling in whippets (Mosher
et al., 2007, PLoS Genetics 3: 379) segregate as Mendelian alleles, while
variants with smaller homozygous effects, such as the difference in obesity-
related phenotypes associated with the FTO locus (Scutari et al., 2007, PLoS
Genetics 3: e115) are associated with much smaller differences between
genotypes.

2. That homozygous and heterozygous effects at a single locus affecting a

quantitative trait are not necessarily constant, but can vary depending on
interactions with other loci affecting variation in the trait. Such inter-locus
interactions can be additive or non-additive (epistatic).

3. Homozygous and heterozygous effects are also highly sensitive to environmental

conditions, such that there will be a range of phenotypes associated with a given
genotype even if the environment is strictly controlled. This environmental
sensitivity further blurs the distinction between genotypes, especially when the
homozygous effect is small.

4. Finally, different genotypes can vary in the magnitude and/or direction of their
response to a particular environmental change, giving rise to genotype by
environment or genotype by sex interactions.

1
The ultimate goal of quantitative genetics research is to understand population variation
for a trait in sufficient detail to answer the following questions:

1. What loci affect trait variation?

2. How are these loci organized on the chromosomes (linkage)?
3. What are the homozygous (additive) and heterozygous (dominance) effects of
each locus?
4. How do the loci combine/interact (in the haplotype) to give the trait phenotype
(epistasis)?
5. What are the effects of the loci on other traits, including reproductive fitness
(pleiotropy)?
6. How do answers to questions 1-5 depend on the environment?
7. What determines allelic variation for quantitative traits at the molecular level?
8. What are the allele frequencies?

It is important to note that we can quantify the magnitude and determine the direction of
the net effects of all loci and the effect of environment that give rise to population
variation for a quantitative trait without knowing the exact underlying genetic details, that
is, the DNA sequence.The strategy to do this, that is determine magnitude and direction
of trait variation in a population is
1. To quantify the degree to which related individuals resemble each other for a
trait measurement,
2. To use the degree of resemblance to predict how statistical properties of the trait
(mean, variance, correlation with other traits) will change in the populations
under the influence of forces of evolution.

To achieve the above, we develop a theoretical model which relates the genetic
properties (additivity, dominance, epistasis, pleiotropy) of individual loci
affecting the trait and environmental effects of the expression of alleles at these loci
to statistical properties of the trait. Having developed the theoretical framework we are
then ready to incorporate the genetic properties and environmental effects of loci
(Quantitative Trait Loci (QTLs) into tools for manipulating populations for trait
improvement.

First, we note that the population distribution of phenotypes of quantitative traits often
approximates a statistical normal distribution. Meristic (involving a change in the
number or arrangement of body parts or segments) traits, like Drosophila bristle number
and mouse litter size, are strictly not continuous (but discrete) as their phenotypes are
counted, not measured. If the number of classes is not too small, however, meristic
traits can be treated as if they are continuous. Threshold traits, such as occur in disease
2
(affected or not affected) fall into one or a few classes. They can be treated as having
an underlying continuous liability to contract the disease, with a threshold value above
which individuals are affected and below which they are not. Threshold traits require
different methods of analysis than continuously varying or meristic traits.
Drosophila showing bristles and compound eye. Drosophila
has four pairs of chromosomes: three autosomes, and one
sex chromosome. The number of bristles in individuals in a
population may vary.

The frequency distributions of most metric characters approximate more or less close
to normal curves. This is seen in Fig. 5.1 where the smooth curves drawn through the
histograms are normal curves having means and variances calculated from the data. In
the study of metric characters, it is possible to make use of the properties of normal
distribution and to apply the appropriate statistical techniques. Sometimes for certain
data, the scale of measurement must be modified by transformation in order to
obtain a distribution approximating to normal. In Fig. 5.1d for example, the distribution
will be skewed if measurements were taken and plotted simply as the number of facets
in Drosophila. But when a transformation is performed on the data, the distribution can
approximate a normal curve and becomes symmetrical. Some examples of
transformations are y , arcsin y , and log y. It must be noted here that the choice
of a transformation to use is difficult, but can be chosen by trial and error.

The first step toward developing a quantitative theoretical genetic model is to determine
how Mendelian segregation of alleles affecting a quantitative trait gives rise to the
typically observed continuous distribution of quantitative traits in populations.

3
Fig. 5.1. Frequency distributions of four metric characters, with normal curves
superimposed. The means are indicated by arrows. The characters are as follows:
(a) Mouse: growth from 3 to 6 weeks of age.
(b) Mouse: litter size (number of live young in first litter).
(c) Drosophila melanogaster : number of bristles on ventral surface of 4th and 5th
abdominal segments, together.
(d) Drosophila melanogaster : number of facets (a lens segment in the compound eye of
an insect or arthropod) in the eye of the mutant ‘’Bar”. Source: Falconer and
Mackay, 1996.

The quantitative genetic model assumes that continuous distribution of phenotypes

results from multiple genetic factors (QTLs), as well as environmental variation. The loci
affecting quantitative traits can be mapped on chromosomes and such loci identified this
way are called Quantitative Trait Loci (QTLs). Drosophila has variable numbers of
sensory bristles on the thorax and abdomen, called sternopleural and abdominal
bristles, respectively. At least 17 QTLs affecting sternopleural bristle number and 5
affecting abdominal bristle number have been mapped to the third chromosome.
Similarly, at least 6 QTLs affecting fruit mass have been found to differentiate the
domestic tomato (Lycopersicon esculentum) from a related wild species (L.
pimpinellifolium).

To illustrate how increasing the number of loci affecting a trait leads to a continuous
distribution of phenotypes, let us compare the phenotypic distributions when 1, 2 or 5
loci, each with two alleles, affect the trait. Assume for simplicity:
 strict additivity; d = 0 for each locus
4
  no epistasis; the different loci combine together additively to affect the phenotype
 genotypic effects are the same for all loci
 allele frequencies = 0.5 at all loci, i.e., p = q
 the range of phenotypes stays the same regardless of the number of loci; thus,
as the number of loci increases, the effect of each decreases
 no environmental variation.

For one locus the genotypes and phenotypes

0.6
correspond exactly:
0.5

Genotypic 0.4

Frequency
Genotype Value Frequency 0.3

A1A1 a 5 p2 0.2

A1A2 d 0 2 pq 0.1

A2A2 −a −5 q2 0.0
-8 -6 -4 -2 0 2 4 6 8

Value

For two loci, there are 16 genotypes and 5 phenotypes, and the genotype: phenotype
correspondence is no longer exact:
Genotypic
0.4
Genotype Value Frequency
A1A1B1B1 5 p4 0.3
Frequency

A1A2B1B1, A1A1B1B2 2.5 4 p3q

0.2

A1A1B2B2, 0.1

A1A2B1B2, 0 6 p2q2
0.0
A2A2B1B1 -8 -6 -4 -2 0 2 4 6 8

Value
A1A2B2B2, A2A2B1B2 −2.5 4 pq3
A2A2B2B2 −5 q4

For five loci, the genotypic values and frequencies are:

Genotypic Value Frequency

0.30
5 p10
0.25
4 10 p9q
3 45 p8q2 0.20
Frequency

2 120 p7q3 0.15

1 210 p6q4 0.10

0 252 p5q5 0.05

0.00
-8 -6 -4 -2 0 2 4 6 8 5
Value
−1 210 p4q6
−2 120 p3q7
−3 45 p2q8
−4 10 pq9
−5 q10

As the number of loci increases, there is no longer a 1:1 correspondence between

genotype and phenotype; multiple genotypes can give rise to the same phenotype. The
distinction between genotypic classes decreases as the number of loci increases, and
hence the effect of each becomes smaller. In addition, we assume any one genotype
can take on a range of phenotypic values, depending on the environment
(environmental sensitivity). Superimposed environmental variation is truly continuous
and further blurs the distinction between genotypes. It is also possible that the
environmental sensitivities of different genotypes are not the same, and there is
genotype by environment interaction.

For these reasons, we cannot infer the effects of segregation at individual loci
affecting variation for quantitative traits from analysis of phenotypes in pedigrees
or crosses, as we can for alleles with large qualitative effects. Rather, we must infer
genetic properties of the entire constellation of segregating alleles affecting the trait, as
well as the effect of the environment, from measurements of the trait on populations of
individuals. To do this, we need to merge our understanding of the effects of alleles at
multiple loci on the trait (homozygous and heterozygous effects, interactions between
loci, and environmental sensitivities and interactions) with our understanding of variation
of genes in populations. The properties of a population that we can observe in
connection with a metric character are means, variances, and covariances. We can
then relate summary statistics such as the population mean and variance, which we can
estimate from measurements of individuals in a population, to genotypic effects, allele
frequencies, and environmental effects.

6
Population Mean
What is the expected mean value of a quantitative trait under the quantitative genetic
model? To infer this, we assume an ‘ideal’ randomly mating population, in which the
genotypes at all loci controlling the trait are in HW equilibrium, and that no evolutionary
forces are acting on the population (i.e., no natural or artificial selection, no migration or
inbreeding, and no mutation). We also assume that we can measure the phenotypic
value, P, for each individual in the population (e.g. 18 bristles, 5'9", or whatever).

According to the quantitative genetic model, P has two primary components: P = G + E.

G, the genotypic value is due to the individual's unique genotype at all loci controlling
the trait, or the sum of the additive, dominance and epistatic effects of all loci affecting
the trait. E, the environmental deviation, arises from environmental influences on the
genotype. Thus, according to the quantitative genetic model, the phenotype of any
individual is caused by the genes it carries and the environment in which it was raised.

We make the further assumption that in the population as a whole, the environmental
fluctuations cancel out, so the mean phenotypic value (P) is equal to the mean
genotypic value (G). The mean genotypic value of a genotype can be measured if the
genotype can be replicated, and many individuals of the same genotype are reared in
the normal range of environments. As we will see later, this can be done for fully inbred
genotypes. In random breeding populations, however, the genotypic value of an
individual is not directly measurable. However, for any one locus affecting the trait, we
can compute the average of the phenotypic measurements of each genotype, which are
the mean genotypic values of the genotypes under our assumption that the mean of the
environmental deviations is equal to zero. We saw in the last lecture that we could
convert these genotypic values at a single locus to additive and dominance effects:
A2A2 A1A2 A1A1

–a 0 d +a

Now we include the assumption that the genotypes are segregating on the population
according to HW proportions. All we need to do to determine the mean genotypic value
(and the mean phenotypic value) of the trait attributable to this locus is to multiply the
values of each genotype by their expected frequency, and add up the products of
frequencies and values over all genotypes. Note that when we calculate the
population mean in this manner, it is expressed as a deviation from the mid-homozygote
value (0).

7
Genotype Frequency Value Frequency × Value
A1A1 p 2 a p2 a
A1A2 2pq d 2pqd
A2A2 q 2 −a −q2a
∑ = a(p2− q2) + 2pqd
= a(p−q)(p+q) + 2pqd
G = P = a(p−q) + 2pqd

Because many loci affect the trait, the simplest model for the population mean is to
assume each locus contributes additively to the trait (i.e., no epistasis): G = P =
∑a(p− q) + 2∑pqd.

Therefore, we might expect the mean of a quantitative trait to be different between

populations, for two possible reasons. First, the population mean depends on gene
frequency, so the contribution of any locus to the mean can range from a high of G =
+a if the population is monomorphic for the A1A1 genotype (p = 1, when A1 is fixed), to a
low of G = −a if the population is monomorphic for the A2A2 genotype (q = 1, when A2
is fixed) (if there is no overdominance). Second, the values of a and d are likely to be
context dependent, and vary if the environment changes. If populations experience
difference in environments, the means could be different even if the allele frequencies of
loci affecting the trait are the same.

Example:

Recall the mh mutation that affects body size of whippets. The mean phenotypic
(genotypic) values of neck girth of whippets of each of the three genotypes are:

Genotype mh/mh mh/+ +/+

Mean Neck girth (cm) 38.48 30.63 29.26

We estimated additive and dominance effects as:

the additive effect, a = (A1A1 – A2A2)/2.
a = (38.48 – 29.26)/2 = 4.61 cm
the dominance effect, d = A1A2 – (A1A1 + A2A2)/2.
d = 30.63 – (38.48 + 29.26)/2 = –3.24 cm

Compute the mean neck girth in a population of whippets in which

(a) The frequency (p) of the mh allele is 0.8, and
(b) The frequency of the mh allele is 0.2.
8
Solution:
(a) G = P = a(p−q) + 2pqd = 4.61(0.8–0.2) + 2(0.8)(0.2)(–3.24) = 2.766 – 1.0368 =
1.7292 cm
(b) G = P = a(p−q) + 2pqd = 4.61(0.2–0.8) + 2(0.8)(0.2)(–3.24) = –2.766 – 1.0368 =
–3.8028 cm
These means are relative to the average genotypic value of the two homozygous
genotypes being set to 0. We can convert the means back to the original measurement
scale by adding 33.87 cm, the midparent, or average of the homozygote genotypic
values {(38.48 + 29.26)/2}. Then the mean of population (a) is 33.87 cm + 1.7292 cm =
35.5992 cm, and of population (b) is 33.87 cm + (-3.8028) = 30.0672 cm.

The population mean M =a(p-q) + 2pqd = mean genotypic value.

There are two terms
(1) a(p-q) is attributable to the homozygotes and
(2) 2pqd is attributable to the heterozygotes

If there is no dominance, d=0 so

M=a(p-q),
but p=1-q so M becomes M=a(1-q-q).
M=a(1-2q)

For complete dominance, d=a, so

M=a(q-p) + 2pqd becomes
M=a(1-2q)+2pqa
= a-2qa+2pqa
= a(1-2q-2pq) but p=1-q
= a(1-2q+2(1-q)q)
=a(1-2q+2q-2q2)=M=a(1-2q2)

Random Mating Populations

The population mean does not describe the breeding properties of the population,
because alleles and not entire genotypes are transmitted from generation to generation.
We therefore have to derive a measure of the value of an individual allele (in terms of
its effect on the phenotype), and express the value of each genotype in terms of the
sum of the values of the alleles it carries. We call the value of an allele the average
effect of the allele, and the value of each genotype expressed in terms of the sums of
the average effects of alleles the breeding value of a genotype. Note that for a diploid

9
organism, there are two alleles at each locus. A multilocus trait would therefore have
sum of the alleles per locus added across all loci to give breeding value of the genotype.
These are two key concepts in quantitative genetics, and also two of the most difficult
concepts to understand.

We define the average effect of an allele as “the mean value of all progeny that received
that allele from a parent, the other allele having come at random from the population,
expressed as a deviation from the overall population mean”. As found earlier,
mean=frequency × value

Values and Mean Value

Type of Frequencies of of Genotypes Average Effect, i
Gamete Genotypes Produced Produced of Allele
A1A1 A1A2 A2A2
a d −a
A1 p q pa + qd 1 = pa + qd − [a(p−q) +
2dpq] = q [a + d(q−p)]
A2 p q −qa + pd 2 = −qa + pd −[a(p−q) +
2dpq] = −p [a + d(q−p)]

Like the population mean, the average effect of an allele is not an inherent property of
the allele, but depends on the particular population under consideration (the
environment, the genotype) such that the values of a and d can change depending on
the particular constellation of environments that individuals of the population are
exposed to. Like the overall mean, the average effect of an allele depends on allele
frequency. When a large number of individuals possess a desirable allele, the larger the
population mean. By definition, the average effect is the effect of an allele in
combination with a random sample of alleles from the population.

Defined in terms of average effects, the breeding value of a genotype is the sum of the
average effects of the genes (loci) it carries. Because breeding values are expressed
as deviations from the population mean, we expect the breeding values of some
genotypes to be positive, and some to be negative.

Genotype Breeding Value

A1A1 2α1
A1A2 α1 + α2
A2A2 2α2

Example:
10
Breeding values for neck girth at the mh locus in whippets.
Compute the breeding values of the three genotypes in a population of whippets in
which the frequency (p) of the mh allele is (a) 0.8 and (b) 0.2. From the previous
examples, recall that:
a = 4.61 cm
d = –3.24 cm
(a) G = P = 1.7292 cm
(b) G = P = −3.8028

Average effects:
(a) p(mh) = 0.8; q(+) = 0.2
α1 = q [a + d(q−p)] = 0.2 [4.61 – 3.24 (0.2 − 0.8)] = 1.3108
α2 = −p [a + d(q−p)] = −0.8 [4.61 – 3.24 (0.2 − 0.8)] = −5.2432

(b) p(mh) = 0.2; q(+) = 0.8

α1 = q [a + d(q−p)] = 0.8 [4.61 – 3.24 (0.8 − 0.2)] = 2.1328
α2 = −p [a + d(q−p)] = −0.2 [4.61 – 3.24 (0.8 − 0.2)] = −0.5332

Genotype Breeding Value (Pop. a) Breeding Value (Pop.b)

mh/mh 2α1 = 2.6216 2α1 = 4.2656
mh/+ α1 + α2 = −3.9324 α1 + α2 = 1.5996
+/+ 2α2 = −10.4864 2α2 = −1.0664

The relationships between breeding values, genotypic values and the population mean
can be visualized by regressing the genotypic value (y axis) on the number of mh alleles
(x axis), and weighting the regression by the frequency of each genotype:

12

10
G
A
(a) p(mh) = 0.8 At high allele frequency, p, the
8 X Population mean population mean is high, close to
6
genotypic value of the
4

2
homozygous genotype; genotypic
Value

X
0 value for all genotypes is greater
-2 than breeding value (the
-4
remaining coming from
-6
dominance and epistasis);
-8

-10
genotypic value of heterozygote
-12 almost equals breeding value
+/+ +/mh mh/mh

Number of mh alleles

11
 12
A (b) p(mh) = 0.2 At low allele frequency, p,
10 G

8 X Population mean
population mean is low, close to
genotypic value of the
6
homozygous recessive genotype;
Value

4
genotypic value for all genotypes
2
is lower than breeding value;
0
genotypic value of homozygote
-2
mh almost equals breeding value
X
-4

-6
+/+ +/mh mh/mh

Number of mh alleles

The breeding values and the population mean lie on the regression line. The exact
regression depends on allele frequencies and effects (additive, dominance, etc.), and
differs between populations (a) and (b) because they have different frequencies of the
mh allele. Because the breeding value of the heterozygote is always intermediate
between the breeding values of the two homozygotes, breeding value is also called the
“additive genetic value”, and is symbolized A. The difference between genotypic value
and breeding values is due to dominance. G –A = D, the dominance deviation. If D =
0, the breeding values are the same as the genotypic values. D represents the effect of
putting different alleles together in pairs to make genotypes.

In both populations, the mh/mh homozygotes have genotypic values that are greater
than their breeding values. This is because some of their progeny will be mh/+ and thus
lighter than the parents. In population (a), the breeding value of the +/+ homozygous
genotype is substantially less than their genotypic value. How can this be when the
lowest genotypic value is –4.61 cm? It is a consequence of breeding value being
defined as a deviation from the population mean. All progeny of +/+ genotypes will be
+/+ or mh/+, with genotypic values of –4.61 cm and –3.24 cm, respectively. The
population mean is 1.7292 in population (a) and thus, all the progeny of the +/+
genotypes are below the mean in this population. In contrast, in population (b) the
breeding value of +/+ homozygotes is much better than the genotypic value. This is
because the mean of this population is −3.8028, and thus the progeny of mh/+
individuals, with genotypic values of –3.24 cm, will be above the mean.

For this example, we can fully partition the effects due to the mh locus. All effects are
expressed as deviations from the population mean:
______________________________________________________________________

12
 mh/mh mh/+ +/+
(a) p(mh)= 0.8
G a−µ d−µ −a − µ
4.61−1.7292 −3.24−1.7292 −4.61−1.7292
2.8808 −4.9692 −6.3392
A 2.6216 −3.9324 −10.4864
D G−A G−A G−A
2.8808−2.6216 −4.9692−(−3.9324) −6.3392−(−10.4864)
0.2592 −1.0368 4.1472

(b) p(mh) = 0.2

G a−µ d−µ −a − µ
4.61−(−3.8028) −3.24−(−3.8028) −4.61−(−3.8028)
8.4128 0.5628 −0.8072
A 4.2656 1.5996 −1.0664
D G−A G−A G−A
8.4128−1.0664 0.5628−(−1.5996) −0.8072−(−4.2656)
4.1472 −1.0368 0.2592
______________________________________________________________________

In general, the values of genotypes at a locus with two alleles, expressed as deviations
from the population mean are:
______________________________________________________________________

Genotype A1A1 A1A2 A2A2

Frequency p2 2pq q2
Value a d −a
Deviations from population mean:
G 2q(a−pd) a(q−p)+d(1−2pq) −2p(a+qd)
A 2q[a + d(q−p)] (q−p) [a + d(q−p)] −2p[a + d(q−p)]
D −2q d2 2pqd −2p2d
______________________________________________________________________

Breeding value is a central concept in quantitative genetics, because it gives us a way

to define the value of an individual in the absence of knowledge of the genetic details
underlying the inheritance of the trait. We define breeding value empirically as A = the
value of an individual judged by the mean value of its progeny. A defined in this
way can be can be determined experimentally:

Individual i Random individuals

13
Genotypic value = Gi × from population
Breeding value = Ai
↓ ↓ ↓
Progeny 1 Progeny 2 ..... Progeny n
Mean genotypic value of progeny = Ai/2

We assume the expected mean environmental deviation of the progeny is zero. The
breeding value of individual i, is 2× the difference in the mean phenotype of its progeny
from the population mean (2× because a parent only provides ½ the genes in the
progeny, the other ½ coming at random from the population). A represents the part of
the genotypic value that is transmitted to progeny.

14