Unbalances in Tree Breeding

1
Unbalances in Tree Breeding

• Dag Lindgren, SLU, Sweden

Lets have the discussion on the net instead of
oral, this show and a discussion site is
available at http//www-genfys.slu.se/staff/dagl/
Komi/DagActivityKomi.htm An effort was done to
discuss at a workplace but that is now
abandoned https//arbetsplats.genfys.slu.se/TreeBr
eedingBook/
2
Genetic unbalances are

• The basis for evolution
• Natural balance is extremely unnatural
• Unavoidable
• Essence of breeding.

3
Genetic contributions varies. Natural selection
favors some and disfavors others, (survival of
the fittest), thus some contributions will
increase, other decrease
4

Selection means always unbalance, not selected
components get no contribution
Selected
Not selected
Contribution of component
I talk about more sophisticated unbalance
5
Different types of unbalances

• Unbalance in genetic components (parents)
• Unbalance in resources
• Structure of breeding population (e.g. mating
  probability, PAM)

Often they come together, thus Nucleus
(elite-main) has all three components of unbalance
6
Why unbalance?

• Breeding has to consider
• Gain
• Gene diversity
• Cost
• Time
• Interaction breeding ? seed orchards.
• Unbalances may make the breeding system more
  optimal and efficient.
• Unbalances offer more degrees of freedom for
  optimization (balance is a form of simplification
  fundamentalism)

7
Reasons against unbalance?

• Unbalance may just make things worse if not done
  wisely and skillfully
• It may overshoot and be too much unbalance!
• Unbalance requires competence!
• Unbalance is more demanding on management skills
• Often the tools for handling unbalance are badly
  developed!
• Seldom transparent
• Sometimes the advantage is small, usually limited
  (3-8) and seldom drastic (20)
• Advantages are often calculated for an ideal
  situation, and are usually somewhat less in the
  real world!
• Historically unbalances were difficult to manage,
  thus all traditional wisdom is against, now
  computers can do everything!!!?
• It requires calculations to be done.

8
Unbalance is a black box! Requires lots of
competence!Risky
9
Why balance?

• Simpler
• More transparent
• Less demanding on competence
• Less demanding on skilful management
• More fail-safe.

10
More reasons for unbalance

• Even a limited extra gain (e.g. 5 increase in
  gain) means enormous economic returns for some
  extra thinking
• Although all extra gain predicted may not be
  reached, it is unlikely it will not give an extra
  gain (using some common sense)
• Unbalance may offer fast gain. That is more worth
  than options some centuries ahead
• Competence can be increased by education and
  research!
• Complete balance is an extreme alternative, that
  makes it unlikely to be optimal
• Complete balance is practically unrealistic!
  Unbalance must anyway be managed, so why not do
  it efficient!

11
Quantitative evaluation of unbalance often
overestimates the practical benefit!

• Genetic parameters (genetic correlations) change
  over time and environment
• Environment changes
• Unreliable parameter estimations
• Breeding goals change and are not exactly
  predictable
• Planned unbalances are influenced by unplanned
• This is likely to lead to overestimates of the
  practical benefits of unbalances and that optimum
  is missed.

12
SuggestionApply unbalance, but with moderation!

• I suggest to often apply unbalances
• But do it with moderation and not too drastic
• It might often be a good idea to try compromising
  between balance and the predicted optimal
  unbalance
• After gaining experience of unbalance, a larger
  share of predicted advantages may be utilized.

13
Unbalances in production population

• Simplest case, only unbalances in different
  contributions (e.g. clones, parents) matter.

14
Equal (balanced) contribution of clones!
Balance
A
B
C
D
Breeding value of clone
E
Contribution of clone
More clones with different contributions can
result in both more gain and more diversity!
15
Linear deployment is optimal for establishment!

• Relate contribution linearly to breeding value
• No other deployment combines higher gain with
  higher effective number.

16
(No Transcript)
17
At thinning ramets cannot be added, just
withdrawn. Thus there is a highest number of
ramets!
Linear deployment works with constraints also
18
(No Transcript)
19
The Swedish model

• In the following, many of presented figures
  intend to be relevant for Sweden or the Swedish
  breeding strategy
• Heading for a number of long term breeding
  populations, each of size 50
• Heading for balance Within family selection
  Each parent get two full sib families One
  selection per family.
• Start with tested plus trees (typical 200 per
  breeding population)
• Test recruitment population (clone-testing or
  progeny-testing)
• Genetic parameters, costs and time estimates
  should be relevant.

20
Unbalanced contributions at the initiation of a
tree improvement program
Closing the breeding population is irreciprocal
and can not be undone! Argument to play on the
safe side!
21
Generalizations from Wei PhD thesis (Wei 1995)

• Method developed for optimal selection in a
  population with family structure (can be
  visualized as unrelated full sibs with plus tree
  parents)
• Optimal selection among individuals with a family
  structure is close to linear deployment from
  parent offspring
• Seems reasonable to start with crosses from about
  150 plus trees to start up a breeding population
• Differences from current Swedish program no
  testing (phenotypic selection), thus
  heritability not high no initial knowledge of
  plus tree breeding values.

22
Unbalances in setting up the first recruitment
population

• Generalized from Ruotsalainen (2002) PhD thesis
• An approximation to linear deployment (3,2,1).

23

• The same resources, the same resulting gene
  diversity.

Unbalanced (60 founders) Unbalanced (60 founders) Balanced (50 founders) Balanced (50 founders) Balanced (50 founders)
Rank of plus tree Progenies Rank of plus tree Progenies  
1-10 3 1-50 2  
11-30 2 -  
31-60 1 -  
61-200 0 51-200 0  
Gain selection intensity 1.368 1.271  
Eight percent more gain with unbalance in the F1
recruitment population!
24
Unbalance, eight percent more gain than balance! Unbalance, eight percent more gain than balance!
Share of represented founders (tested plus trees) Progenies per founder
Best 1/6 3
Medium 1/3 2
Bottom 1/2 1
25
Result Andersson PhD thesis 1999
Unbalanced selection is superior to balanced in
the initiation of a breeding program
26
Unbalance by refreshing in F1Inspired from
Andersson PhD thesis (1999)

• Unbalance could perhaps be introduced in F1 by
  refreshing
• In model-calculations it was favorable to replace
  5-10 of founders at F1 with new plus trees
• That indicates that it may sometimes be
  beneficial to replace one or a few of bottom
  ranking BP members with new founders in the
  Swedish breeding
• The introduced founders may have slightly lower
  BV, but the Group merit of the BP could increase.
  That would mean that a few F1 BP would be crossed
  with new founders to form the next BP generation
• The bottom ranking selected founders are only
  slightly superior to the best non-selected
  candidates.

27
Unbalances in long term breeding
28
Unbalances in long term breeding

• Wei (1995) demonstrated the possible disastrous
  effects to use the strongly unbalanced selection
  resulting from maximizing breeding value in each
  generation
• When gene diversity is exhausted, genetic gain
  drops
• Sanchez (2000) studied the effect of a slight
  unbalance with quantitative simulation and small
  populations. It was noted that a slight unbalance
  often was more favorable in breeding than
  complete balance.

29
Results generalized from PhD thesis Rosvall (1999)

• Used POPSIM (tree improvement simulator) to study
  different aspects of long term breeding with
  simulation of a program similar to Swedish Norway
  spruce breeding.
• The capacity of the breeding population to
  support a seed orchard was used as a criterion.

30
Genetic gain
Breeding population size 48, SPM , progeny size
50, GMS selection, high heritability, after
five generations
80
70
60
50
40
30
20
10
Gene diversity (status number) in the breeding
population
based on Rosvall 1999
0
0
2
4
6
8
10
12
14
16
31
Unbalances in long term breeding?

• Some advantage of unbalance is found, but so
  marginal and uncertain (Rosvall 1999) that it
  seems doubtful applying unbalance in the Swedish
  long term breeding.

32
Some reasons Rosvall (1999) found little
advantage of unbalance

• The benefit of the breeding is measured as its
  ability of supporting seed orchards
• Less sophisticated selection for advanced
  generation seed orchards than will be used in
  practice
• Testing of the recruitment population (clonal or
  progeny), thus high heritability (Swedish pine
  breeding may turn to phenotypic selection next
  cycle, when unbalance may appear more favorable)
  
• Not exactly optimal unbalance, constraints in the
  simulator makes it hard to use optimum
• Intensively selected breeding populations (Bulmer
  effect)
• Distinct generations, that will not be so!
• Breeding population size is small, that makes
  conservation of diversity relatively important
• Mainly a closed breeding population.

I guess the advantage of unbalance is slightly
greater in real world, and in particular in
initiation!
33
Gain at a given diversity. h20.25 and P0.1
Modified From Lindgren and Wei 1993
Combined indexestimated BV (maximizes gain)
Between family(exhausts diversity)
Gain
Within family(preserves diversity)
Relative diversity
Within family selection does not look efficient.
Information from sibs was used for estimating
breeding values (selection index). Infinite
normal populations were assumed.
34
Distinct generations rolling synchronously will
not work! Rolling front breeding is more
operational, and must be unbalanced!

• Of the Swedish breeding populations, which
  reached F1 in field, 75 are not synchronized in
  time. In spite of time lost in efforts!
• Mates can be selected in several ways. Trees in
  field trials can probably only function as pollen
  parents, while grafts may first be available as
  seed parents. Optimal use of such factors will
  force unbalances.
• It will be found optimal to utilize genotypes
  technically in different generations
• Some materials will be remeasured at higher age,
  some not
• The management of rolling front will be
  unbalanced anyway, so that balance is simple will
  be irrelevant.

35
Not only contributions but also resource
allocation matters!
More resources for improving larger components
may result in higher average gain!
More resources for better
Equal resources
Genetic value of component
A
A
B
B
Contribution of component
36
More attention on the better may improve
efficiency

• If the predicted best contributions get more
  attention (larger test families, more mating
  partners etc) the best contributions benefit more
  from breeding.

37
Results interpreted from PhD thesis Lstiburek
(2005). POPSIM simulations. Linear deployment of
family sizes related to their breeding value
combined with PAM and within family selection
boosts ability to select for high gain seed
orchards!
Strong unbalance gives 20 more gain
Family size
Intermediate unbalance, 10 more gain
After Mullin et al 2005
Low Family breeding value High
38
Why not stronger unbalance?

• It cannot be good breeding economy to spend lots
  of resources to produce a family (including their
  parents) and when make the family size very small
  for some families
• Test environments and optimal test criteria for
  optimizing family size is different from there
  families are deployed, thus the advantage will be
  reduced
• The accepted conventional wisdom is the same
  family size, safer not to make too extreme
  changes, while experience and considerations
  accumulate successively stronger unbalance may be
  applied.

39
Average selection intensity
The selection gain by within family selection
drops if the same total testing effort is
unequally distributed among families, but
marginally little if the unbalance is moderate.
  Balanced   Balanced   Moderate unbalance   Moderate unbalance   Strong unbalance   Strong unbalance  
  Size i Size i Size i
Large family 5 1.163 6 1.267 9 1.485
Small family 5 1.163 4 1.029 1 0
Average selection intensity   1.163   1.148   0.742
of balanced   100   98.7   63.8
40
Population structuringPh-thesis Rosvall (1999),
Lstiburek (2005)

• To structure a breeding population in elite and
  main offers advantages, but more advantages of
  the same type can be achieved by proper
  management of a single population.

41
Stratified subliningFrom Ruotsalainen PhThesis
(2001)

• Extends the PAM concept to several generations
• A better alternative to elite main is to use
  many strata.

42
Regeneration
Stratified sublining
F1 families
HIGHEST
Index breeding value
60(100)
Breeding population
LOWEST
Taken from Finnish breeding strategy (Haapanen
2004)
43
Stratified sublines

• stratification allows prioritising testing and
  breeding efforts on the sublines which are most
  likely contribute trees to seed orchards
• Complete control of inbreeding
• enables enough unrelated selections to be
  deployed in seed orchards
• Flexible
• sublines can be merged or entirely abandoned if
  desired

Taken from Finnish breeding strategy (Haapanen
2004)
44
Slight overestimate of advantages

• Parents are paired for PAM or allocated to
  stratified sublines based on certain optimal
  indices, but the optimal indices will be
  different in the offspring (different
  environment, different genetic parameters,
  different desires), thus the positive effect of
  PAM and stratified sublines may be slightly
  overestimated.

45
End