1. Detecting selection using genome scans
Roger Butlin
University of Sheffield

2. Nielsen R (2005) Molecular signatures of natural selection. Annu. Rev
Nielsen R (2005) Molecular signatures of natural selection. Annu. Rev. Genet. 39, 197–218.
What signatures does selection leave in the genome?
Population differentiation – today’s focus!
Frequency spectrum, e.g. Tajima’s D
Selective sweeps
Haplotype structure (linkage disequilibrium)
MacDonald-Kreitman tests (or PAML over long time-scales)

3. Frequency distribution:
From Nielsen (2005): frequency of derived allele in a sample of 20 alleles.
Tajima’s D = (π-S)/sd, summarises excess of rare variants

4. Selective sweep:

5. Extended haplotype homozygosity (Sabeti et al. 2002)

6. MacDonald-Kreitman and related tests
dN = replacement changes per replacement site
dS = silent changes per silent site
dN/dS = neutral
dN/dS < conserved (purifying selection)
dN/dS > adaptive evolution (positive selection)

7. Selection on phenotypic traits:
QTL
Association analysis
Candidate genes

8. Genome scans
(aka ‘Outlier analysis’)

9. Littorina saxatilis – locally adapted morphs
What signatures of selection might we look for?
‘H’
NB divergent selection (sort of equivalent to RI)
‘M’
Thornwick Bay

10. Signatures of selection:
Departure from HWE
Low diversity (selective sweep)
Frequency spectrum tests
High divergence
Elevated proportion of non-synonymous substitutions LD

11. Neutral loci

12. Stabilizing selection

13. Local adaptation

14. Charlesworth et al. 1997 (from Nosil et al. 2009)

15. A concrete example: adaptation to altitude in Rana temporaria (Bonin et al. 2006)
High – 2000m
Intermediate – 1000m
Low – 400m
190 individuals
392 AFLP bands

16. Generating the expected distribution
DetSel – Vitalis et al. 2001 N0 N1 N2 t μ to
F1,2 – measure of divergence of population 1,2 from population 2,1
Dfdist – Beaumont & Nichols 1996 N m
FST – symmetrical population differentiation, as a function of heterozygosity
Separation of timescales argument…
Does the structure/history matter?

17. DetSel
Dfdist
95% CI
95%
50% 5%
‘Low 1’ vs ‘High 1’

18. DetSel
Dfdist
Both
Interpretation
Monomorphic in one population 35
N/A
Unreliable outliers
Significant in one comparison 14 29
False positives
Significant in comparisons involving one population 3 11
Local effects
Significant in at least 2 comparisons 2 1
Adaptation to altitude
Significant in global comparison across altitudes 6
(2 at 99%)
392 AFLPs, 12 pairwise comparisons across altitude or 3 altitude categories, 95% cut off

19. 343 loci
8 loci

20. Outlier AFLP in homologous set*
Outliers and selected traits
Rogers and Bernatchez (2007):
Dwarf x Normal cross  both backcrosses
Measure ‘adaptive’ traits (9)
QTL map (>400 AFLP plus microsatellites)
Homologous AFLP in 4 natural sympatric population pairs
Outlier analysis (forward simulation based on Winkle)
Coregonus clupeaformis (lake whitefish)
Homologous AFLP
Outlier AFLP in homologous set*
Outlier within QTL (based on 1.5 LOD support)
Hybrid x Dwarf
180 19 9
(3.6 expected, P=0.0015)
Hybrid x Normal
131 8 4
(0.5 expected, P=0.0002)
Expectation based on overall proportion of AFLP associated with QTL.
*Only 3 outliers shared between lakes

21. Roger Butlin - Genome scans

22. Nosil et al. 2009 review of 14 studies:
0.5 – 26% outliers, most studies 5-10%
1 - 5% outliers replicated in pair-wise comparisons
% of outliers specific to habitat comparisons
No consistent pattern for EST-associated loci
LD among outliers typically low
But many methodological differences between studies
Population sampling
Marker type
Analysis type and options
Statistical cut-offs

23. Environmental correlations SAM – Joost et al. 2007
IBA – Nosil et al. 2007
FST for each locus correlated with ‘adaptive distance’, controlling for geographic distance (partial Mantel test)

24. Methodological improvements – Bayesian approaches
BayesFst – Beaumont & Balding 2004
Bayescan – Foll & Gaggiotti 2008
For each locus i and population j we have an FST measure, relative to the ‘ancestral’ population, Fij
Then decompose into locus and population components,
Log(Fij/(1-Fij) = αi + βj
αi is the locus-effect – 0 neutral, +ve divergence selection, -ve balancing selection
βj is the population effect
Assuming Dirichlet distribution of allele frequencies among subpopulations, can estimate αi + βj by MCMC
In Bayescan, also explicitly test αi = 0
Ancestral

25. Apparently much greater power to detect balancing selection than FDIST
Lower false positive rate
Wider applicability

26. Methodological improvements – hierarchical structure
Arlequin – Excoffier et al. 2009

27. Circles – simulated STR data, grey – null distribution

29. Multiple analyses? Candidate vs control? E.g. Shimada et al. 2010
Bayenv – Coop et al. 2010
Estimates variance-covariance matrix of allele frequencies then tests for correlations with environmental variables (or categories).
Software available at:
Multiple analyses? Candidate vs control? E.g. Shimada et al. 2010

31. Hohenlohe et al. 2010

32. Mäkinen et al 2008 7 populations 3 marine, 4 freshwater 103 STR loci
Analysed by BayesFst
(and LnRH)
5 under directional selection
(3 in Eda locus)
15 under balancing selection
Used as a test case by Excoffier et al
2 directional
3 balancing

33. Can we replicate these results?
Bayescan
Stickleback_allele.txt – input file
Output_fst.txt – view with R routine plot_Bayescan
Arlequin
Stickleback_data_standard.arp – IAM
Stickleback_data_repeat.arp – SMM
Run using Arlequin3.5
Try hierarchical and island models, maybe different hierarchies

35. Sympatric speciation?
FST distribution as evidence of speciation with gene flow
Savolainen et al (2006)
Cf. Gavrilets and Vose (2007)
few loci underlying key traits
intermediate selection
initial environmental effect on phenology
Howea - palms