1. The Ashkenazi Genome Project
Shai Carmi
Pe’er lab, Columbia University
Joint Group Meeting
November 2012

2. Recent History of Ashkenazi Jews
Mediterranean origin (?)
Ca. 1000: Small communities in N. France, Rhineland
Migration east
Expansion
~10M today, mostly in US and Israel
Relative isolation

3. Ashkenazi Jewish Genetics
Recently, AJ shown to be a genetically distinct group
Close to Middle-Eastern & South-European populations
300 Jewish individuals; SNP arrays
Jewish non-AJ
Europeans AJ
Middle-Eastern
Price et al., PLoS Genetics 2008.
Olshen et al., BMC Genetics 2008.
Need et al., Genome Biology 2009.
Kopelman et al., BMC Genetics, 2009.
Behar et al., Nature 2010.
Bray et al., PNAS 2010.
Guha et al., Genome Biology 2012.
Atzmon et al., AJHG 2010

4. Recent Demography & IBD
In small populations, common ancestors are likely recent.
Generation 𝑔 1 2 3 A B

5. Recent Demography & IBD
In small populations, common ancestors are likely recent.
Generation 𝑔
For g-generation ancestor, chances of IBD ~ 4 −𝑔 , but length ~ 1 𝑔 (M).
IBD is highly informative on recent history! A B A B
A shared segment
⟹ Many long haplotypes identical-by-descent

6. Formal Inference Using IBD
Assume a population of historical size 𝑁 𝑡 = 𝑁 0 𝜆 𝑡 .
Total shared segments of length ℓ 1 <ℓ< ℓ 2 :
0 ∞ 𝑒 − 0 𝑡 𝑑𝑡′ 𝜆( 𝑡 ′ ) 𝜆(𝑡) 𝑒 −2 ℓ 1 𝑁 0 𝑡 1+2 ℓ 1 𝑁 0 𝑡 − 𝑒 −2 ℓ 2 𝑁 0 𝑡 1+2 ℓ 2 𝑁 0 𝑡 𝑑𝑡
Detect IBD in sample ⟹ Infer history 𝑁 𝑡 . A B
Palamara et al., AJHG 2012
IBD sharing abundant in AJ
Atzmon et al., AJHG 2012
Gusev et al., MBE 2011 A B
A shared segment

7. Power of imputation by IBD
AJ Genetic History
2,300 N t
Effective size
45,000
270
4,300,000
Years ago
800
Present
High potential for genetic studies! AJ UK
Power of imputation by IBD
Palamara et al., AJHG 2012
Expansion rate ≈34% per generation

8. The Ashkenazi Genome Consortium
10 labs from NY area and Israel.
Goals:
Sequence to high coverage hundreds of healthy AJ
Use as a reference panel for
Association studies
Imputation
Clinical interpretation
Understand AJ population history
Understand AJ functional genetic variation (negative/positive selection)

9. The Ashkenazi Genome Consortium
Phase I:
144 AJ personal genomes
~60yo, healthy controls
Unrelated, PCA-validated AJ
Selected to maximize sharing with rest of cohort
Technology: Complete Genomics
Sequenced so far: ~100 genomes
Data presented: 58 genomes
Phase II:
Hundreds of genomes (2013?)
More collaborators

10. Quality Measures Property Genome (exome) Coverage ~55x Fraction called
96.5±0.003% (98%)
Fraction with coverage > 20x
92.4±0.018% (94.9%)
Concordance with SNP array
99.87±0.1%
Ti/Tv ratio
2.14±0.003 (3.05)
Ti/Tv

11. Multi-nucleotide variant count
Variant statistics
Statistic
Per genome (exome)
Total SNPs
3.4M (22k)
Novel SNPs
3.7% (4%)
Het/hom ratio
1.64 (1.67)
Insertions count
223k (246)
Deletions count
237k (218)
Multi-nucleotide variant count
83k (374)
Synonymous SNPs
10525
Non-synonymous SNPs
9695
Nonsense SNPs 71
Other disrupting
241
CNV count
336
SV count
1486
MEI count
3475

12. Comparison to Europeans
Extrapolated to 100% genome
TAGC
Flemish
(M)
Similar results in 13 CG European public genomes.
(k)

13. Het/Hom Ratio Significant in comparison to both Flemish and HapMap EU.
Was observed in SNP arrays (Need et al., Genome Biology 2009).
Did I not just say that AJ have more IBD?

14. Het/Hom Ratio
Years ago t AJ EU
IBD observed
Present

15. Data Flow Pipeline Backup 3x CGA tools testVariants VCF Fix Plink/SeqQC
Compress, index
Plink
Phase
Distribute

16. Quality Control
False positive rate assessment by runs of homozygosity:
Assume hets in high confidence roh are FP.
hets
Paternal
Maternal
High confidence rohs only (>7.5MB, no gaps).
7 segments in 7 individuals (total 72MB).
Count het SNPs in original files.
Genome wide extrapolation: ~20,000 per genome.
~3-5% FP rate for indels.

17. Quality Control Remove: ⇒FP after QC: ~5,000 per genome.
Indels and MNPs
Low-quality SNPs
⇒FP after QC: ~5,000 per genome.
Multi-allelic SNPs
Half-calls
SNPs with high no-call rate
SNPs not in HWE
Monomorphic reference SNPs
Inbred individual

18. Applicability to Clinical Genomics
Variants of unknown significance
Technical false positives
True variants without health impact
Novel variants per sample
Not in TAGC

19. Distance between phased hets
Phasing
Sequencing is in mate-pairs
Haplotype information available for ~30-35% of hets.
BEAGLE error rate: 3-4%.
Seqphase: new phasing tool
Based on SHAPEIT
Incorporates reads
18 hours on chromosome 1.
Distance between phased hets
100
300
500
Frequency

20. Variant Discovery Number of non-reference variants.
Extrapolation using Gravel et al., PNAS 2011.

21. Variant Discovery
Number of segregating sites Sn(t), heterozygosity H(t).
Zivkovic and Stephan, Theor. Pop. Biol
𝑆 𝑛 𝑡 =𝜃 𝑘=1 𝑛/2 (4𝑘−1) 𝑛 2𝑘 𝑛+2𝑘−1 2𝑘 −∞ 𝑡 exp⁡ − 2𝑘 2 𝑠 𝑡 𝑑𝑢 𝜌(𝑢) 𝑑𝑠
𝐻 𝑡 =𝜃 −∞ 𝑡 exp⁡ − 𝑠 𝑡 𝑑𝑢 𝜌(𝑢) 𝑑𝑠
N(t): # diploids at time t; N=N(t=0); ρ(t)=N(t)/N; n: # diploid samples
t: #generations/2N; θ=4Nμ; μ: mutation rate per generation
Use double expansion model of Palamara et al., AJHG 2012.
Define t=0 at the start of the first expansion.
Match H(t).

22. Variant Discovery ?

23. Allele Frequency Spectrum
Counts
Fractions
All
Pop.-specific

24. Demographic Inference
Folded allele frequency spectrum + coalescent simulations.
Double expansion model + ancient AJ foundation bottleneck.
Find maximum likelihood solution (Gutenkunst et al., PLoS Genet. 2009)
Average over simulations to obtain expected spectrum.
Assume mutation frequency is drawn according to expected spectrum.
Multinomial probability approximated as Poisson.
100 10 1
0.1
%sites

25. Demographic Inferencet
Years ago
3,000
Similar to Palamara et al., with somewhat larger population sizes.
To do: Gene flow from EU; better inference tools.
5000
90,000
875
500
Present
7,500,000 N
Effective size

26. Ongoing Analysis Exome analysis Mobile elements insertion
Genes w/ AJ-specific high mutation load
Mobile elements insertion
Common insertions frequencies correlated with 1KG
AJ disease genes (Ostrer & Skorecki, Human Genetics 2012)
Some carriers detected
276 non-synonymous mutations, >65 known
60 loss-of-function

27. Summary
AJ bottleneck and expansion reveal potential for genetics studies.
High quality genomes sequenced by TAGC indicate utility in clinical setting.
Complete variant discovery improves demographic inference; subtle differences from Europeans.
Future directions:
Imputation power using TAGC vs. 1000Genomes
Local ancestry inference
Effect of natural selection

28. Thank you! TAGC consortium members: Funding:
Columbia University Computer Science:
Itsik Pe’er, Pier Francesco Palamara Undergrads: Fillan Grady, Ethan Kochav, James Xue IT: Shlomo Hershkop
Long-Island Jewish Medical Center:
Todd Lencz, Semanti Mukherjee, Saurav Guha
Columbia University Medical Center:
Lorraine Clark, Xinmin Liu
Albert Einstein College of Medicine:
Gil Atzmon, Harry Ostrer
Mount Sinai School of Medicine:
Inga Peter, Laurie Ozelius
Memorial Sloan Kettering Cancer Center:
Ken Offit, Vijai Joseph
Yale School of Medicine:
Judy Cho, Ken Hui, Monica Bowen
The Hebrew University of Jerusalem:
Ariel Darvasi
VIB, Gent, Belgium
Herwig Van Marck, Stephane Plaisance
Complete Genomics Jason Laramie
Funding:
Human Frontiers Science program.