Lecture 11: Selection on the Environmental...
Transcript of Lecture 11: Selection on the Environmental...
1
Lecture 11:Selection on the
Environmental varianceBruce Walsh lecture notes
Synbreed courseversion 4 July 2013
2
Scales of Environmentalsensitivity
• Clones (pure lines) typically showdifferential (relative) performance acrossmacroenvironments,– evidence of G x E
• Genotypes can also show variation in theirsensitivity to microenvironmentaldifferences,– Fluctuating asymmetry
– Developmental noise
3
Environmental vs. genetic canalization
• Waddington (1942) suggested thedevelopment systems become bufferedover time to be somewhat robust toenvironmental and/or genetic noise
• Environmental canalization– A genotype has robust performance over a set
of microenvironments (G x E)
• Genetic canalization– A particular genotype (one or several loci) has
robust performance when randomized overgenetic backgrounds (epistasis)
4
Fraser & Schadt’s experiment
• Using variation within and between a series ofinbred lines, one can distinguish between geneticand environmental robustness (canalization)
• Fraser & Schadt examined mRNA levels atthousands of genes for their traits over a seriesof inbred lines of mice
• Environmental robustness means that the traitwill have reduced within-line variation
• Genetic robustness implies that the mean traitvalue over lines (the between-line variance) isreduced
5
Their design: eQTLswere mapped and thenexamined both within(down a row) and acrosslines (across rows)
Environmental robustness(ER) QTLs: within-linevariation, no between-line
Genetic robustness (GR) QTLs: Between-linevariation, no within-line
6
•Figure 2. Genetic robustness QTLs in mouse.
Fraser HB, Schadt EE (2010) The Quantitative Genetics of Phenotypic Robustness. PLoS ONE 5(1): e8635.
GR QTLs tend to bemainly trans-acting,modest sex-specificoverlap.When present, due toCis-acting factors
7
•Figure 3. Environmental robustness QTLs in mouse.
Fraser HB, Schadt EE (2010) The Quantitative Genetics of Phenotypic Robustness. PLoS ONE 5(1): e8635.
ER QTLs trans-acting, No sex-specific overlap.
No overlap between ER, GR QTLs,Hence, appears to bedifferent pathways forgeneticvs environmentalrobustness
8
Evidence for geneticvariation in ER
• Inbred lines differ in Var(E)
• Sire differences (in cattle) for variation intheir offspring– Could also simply be due to major genes
segregating in that sire
• Specific QTL mapping– Early QTL experiments showed marker effects
on variance
– Direct mapping of such vQTLs in more recentstudies
9
Selection on the environmentalvariance
• Genotypes can differ in their environmentalvariances– Example: Different inbred lines show different
variances
– Different QTL genotypes can show different traitvariances
• Can get response to selection on the environmentalvariance– Response for increased cannalization (greater left-right
symmetry in traits like bristle number in flies)
• Hence, selection on a trait can also potentiallychange its environmental variance
10
Modeling breeding values ofVar(E)
• Assuming that the environmental variationis like any other complex trait, it has abreeding value (passed from parent ofoffspring) and residual values that are not
• Several different ways to model thegenetic structure of the environmentalvariance
11
Models of heritable environmentalvariation
• Base model is P = G + E, where E now variesover genotypes and hence we can speak ofa breeding value for E
• Gavrilets-Hastings multiplicative model– E = !i*e, where e ~ (0, "e
2)
– A genotype-specific effect ! (sensitivity) +noise (e)
– Hence, Var(E | G, !) = !2 * "e2
– Simplest approach ! = Av, the BV for variance• Hence, Var(E | G, Av) = Av
2* "e2
12
Multiplicative Model (cont)
• Total environmental variance (forpopulation)– "E
2 = (µ!2 + "!2)*"e
2
• Selecting µ! to zero minimizes popenvironmental variance
• When ! = Av + Dv, (dominance inenvironmental sensitivity), then– "E
2 = (µAv2 + "Av
2)*"e2 + "Dv
2*"e2
– selection can decrease mean of Av and also itsvariance (by generating negative LD), but can’timpact the dominance term
13
The exponential model
• E = exp(Av/2)*e– Where e ~ N(0, "e
2), Av ~ N(µAv, "Av2),
• Also called the log-additive model, as– Ln ("2(E | Av) = ln("e
2) + Av
• Environmental variation for the population– "E
2 = "e2 exp( µAv + "2
Av/2)
– Under this model, decreasing the mean (tonegative values) continues to decrease thevariance
14
The additive model
• While the multiplicative and exponentialmodels ensure a non-zero variance, theycan be hard to work with
• The additive model is easier to use butdoes not ensure a positive environmentalvariance
– Here, "E2 = "e
2 + µAv
15
16
Under the additive model, we can define a heritability in Var(E) as
Under normalityassumptions,
Average heritability is small (0.038), but potential forresponse is significant, its evolability CV(A) = 0.41. Thus,while the actual response can be small, the percentageincrease in the trait is not
here, CV(A)2 = 0.16 is the expected change in the traitmean given a standard unit of selection
Breeding values for trait mean, trait Var(E)tends to be negatively correlated (r ~ -0.24)
17
18
Selection response
• Assume the additive model for Var(E) andconsider truncation selection with theupper p saved. Hill & Zhang (2004) showedthat the probability P(a,b) that a genotypewith mean value µ + a and variance effect"2 + b is selected is
• P(a,b)/p ~ 1 + a (i/"z) + (b/2) i x[1-p] / "z2
• Where i is the selection intensity and x[1-p]
is defined by P(unit normal > x[1-p] ) = p.
• Hence, direction selection favors higherVar(E)
19
20
Selection response (additive model)
Hence strong directional selection (by selecting outliers)can result in a significant increase in Var(E), reducing h2.
21
22
Artificial selection
• An increase in the environmental varianceis often seen following several generationsof directional selection
• As the models show, this is expected whenthere is heritable variation in Var(E) and isexpected to be more dramatic thestronger the directional selection
23
Stabilizing selection
• Selection to minimize deviations fromsome optimal value can occur throughthree pathways
– Change the mean to the optimal value
– Reduce the additive genetic variation(generation of negative LD)
– Reduce the environmental variance
24
Stabilizing selection (cont)Assume Multiplicative model
25
Direct selection on Var(E)using repeated records
• While direct and stabilizing selection applyindirect selection on Var(E), direct selection canbe applied for traits with repeated records
• Construct an index I for each individual whichamounts to their sample variance in the records,then selection on I directly
26
Assuming the exponential model for the variance