1. Identification of Paralogs in RADseq data
Garrett McKinney, Ryan Waples, Lisa Seeb, Jim Seeb
University of Washington
Seattle

2. Paralogs in Salmon
Salmon whole genome duplication has resulted in a high proportion of paralogs
Up to 25% of the genome experienced delayed diploidization
Up to 10% of the genome is still inherited tetrasomically
Include citations?
Lien et al. 2016

3. Paralogs in Salmon
Paralogs are concentrated on the distal ends of eight pairs of homeologous chromosome arms (Brieuc et al. 2014, Kodama et al. 2015, Waples et al. 2015, Larson et al , McKinney et al. 2016)
Distal portions of these homeologs can still undergo tetrasomic inheritance
Waples et al. 2015

4. Allelic Ratios and Genotypes in Heterozygous Individuals
Paralogs in Salmon
Three broad locus classes result from the rediploidization process
Locus type*
Inheritance
Allelic Ratios and Genotypes in Heterozygous Individuals
Singleton
Disomic
1:1 AB
Duplicate
Tetrasomic
3:1
2:2
1:3
AA | AB
AB | AB
AA | BB
AB | BB
Diverged duplicates
AA || AB
AA || BB
* Based on nomenclature and definitions given in Allendorfand Thorgaard (1984) and Limborg et al. (2016)

5. Issues With Paralogs It is difficult to accurately genotype paralogs
AAAB vs AABB vs ABBB
Paralogs can’t always be analyzed using standard population genetic methods
Many models assume diploid system
Paralogs are often removed from analyses due to these considerations
Must first be identified

6. Methods to Identify paralogs
Genome/transcriptome alignments
Genomes not available or incomplete for many species
Gene annotation
Most RAD loci outside of coding regions
Haploids/doubled haploids
Sequence depth
Heterozygosity/Hardy-Weinberg deviations
Number of alleles/polymorphic sites

7. How to better identify paralogs
Paralogs differ from singleton loci in two important characteristics
Heterozygosity
Allele ratios within heterozygotes
Can be combined to accurately identify paralogs in population datasets
HDplot

8. Heterozygosity
Duplicate loci have more heterozygotes than singleton loci
Increase in heterozygosity occurs at every allele frequency but is maximized when p=q=0.5

9. Difference is maximized when allele frequencies are even

10. Allele Ratio
Allele ratio is discrete within individuals but continuous at the population level
Singleton heterozygotes (AB) always 1:1
Duplicates have three heterozygous classes
AAAB, AABB, ABBB (3:1, 1:1, 1:3)
Proportion of each heterozygous class depends on allele frequencies
Stochastic variation in RADseq can mask true ratio in individuals
Reads for each allele can be summed over heterozygous individuals to increase power

13. Difference is maximized when one allele is rare

14. A statistical test can be done to see if the observed allelic ratio differs from 1:1

15. Chinook salmon data Larson et al. 2014 McKinney et al. 2016
Population Dataset
266 individuals
5 populations
McKinney et al. 2016
Linkage map: 14,620 loci
Use copy-status from mapped loci to verify accuracy of HDplot

16. Data simulation Simulated data based on Chinook salmon dataset
20,000 loci (17,000 singleton, 1,500 duplicate, 1,500 diverged duplicate)
Allele frequency for each locus randomly generated
2,000 genotypes created based on allele frequency
250 individuals randomly sampled from genotypes
Read depth per locus and read depth per allele modeled based on Chinook salmon dataset

17. Data simulation Calculating H and D H is the heterozygosity per locus
D (read ratio deviation) is a z-score for each locus
Measure deviation of observed allelic ratio from expected 1:1

18. Data Simulation Results
Singletons and Duplicates are distinguished both by heterozygosity and read ratio deviation
Duplicates and Diverged duplicates are distinguished primarily by heterozygosity

19. Real Data Total Loci Singleton Duplicate 19,299 16,019 (83%)
3,280 (17%)

20. Singleton Loci Duplicate Loci Diverged Duplicate Loci

21. Accuracy of Other Methods
Haploid Mapping
Accurate
97% agreement with HDplot for singletons
95% agreement with HDplot for duplicates
Limited to identifying duplicates that are polymorphic in parents of haploids
Could only assign copy-status to ~50% of the population loci

22. Accuracy of Other Methods
Read Depth
~30% of duplicate loci misidentified

23. Accuracy of Other Methods
Heterozygosity
Used FIS threshold of <-0.5 to classify loci as duplicate
Subset of duplicates identified (~30%)
No singletons misidentified

24. Accuracy of Other Methods
Number of Polymorphic Sites/Alleles
Duplicate loci do have more alleles and polymorphic sites but overlap in distribution is too great to distinguish
Singleton Duplicate

25. How to use knowledge of copy-status
Filter duplicates
Has been default approach, removes regions of the genome which may be important
Incorporate into population analyses
Estimate population allele frequencies (polyfreqs)
Assign allele dosage

26. Genotyping Paralogs: RADseq
AAAB
AABB
ABBB
1:3
1:1
3:1
1:3
1:1
3:1
Sequence reads too variable to assign allele dosage
Dosage may be assigned with few ambiguities

27. Genotyping Paralogs: RADseq
AAAB
AABB
ABBB
28 reads per individual
119 reads per individual
*Difference is average read depth

28. Genotyping Paralogs: Amplicon
Sample
Haplotype
Reads
Dosage 43
AAT
468 2
TGG
407
Sample
Haplotype
Reads
Dosage 48
AAT
246 1
TAT
242
TGG
430 2

29. Acknowledgements Seeb Lab Funding Wes Larson Morten Limborg
Carita Pascal
Carolyn Tarpey
Funding