1. Causes of regulatory variation in the human genome
Manolis Dermitzakis
The Wellcome Trust Sanger Institute
Wellcome Trust Genome Campus
Cambridge, UK

2. Human Genome:
~25,000 genes
1-1.5% of the human DNA is coding
Is the remaining 98.5% “junk”

3. Gene expression as a phenotype
Altered patterns of gene expression  disease.
e.g., Type 1 diabetes, Burkitt’s lymphomas.
Widespread intraspecific variation.
Heritable genetic variation for transcript levels.
Familial aggregation of expression profiles (Cheung et al. 2003).
In humans, ~30% of surveyed loci exhibited a genetic component for expression differences (Monks et al. 2004; Schadt et al. 2003).
Much of the influential variation is located cis- to the coding locus.
In humans, mouse, and maize, 35%-50% of the genetic basis for intraspecific differences in transcription level are cis- to the coding locus (e.g. Morley et al. 2004; Schadt et al. 2003; Stranger et al. 2005; Cheung et al. 2005, etc.).
As an introduction, I’d like to give you a couple of quick facts about gene expression:
In general, normal cell function and many aspects of development are highly dependent on having the right genes transcribed at the right time and place…and certainly some diseases are associated with altered patterns of gene expression. Extreme effects or subtle changes in expression.
However, in many species there is also quite a lot of variation among individuals with respect to gene expression patterns.
Much of this variation has a genetic component, for example in humans, nearly 30% of surveyed loci exhibited a genetic component for expression differences.
And more studies are showing that much of the genetic component influencing expression is located cis- to the coding locus, for example a survey of humans mouse and maize estimates that approx 30-50% is attributable to cis-located variants
Stranger and Dermitzakis 2006

4. Why study gene expression
Describe and dissect regulatory variation
Annotate regulatory elements in the human genome
Support disease studies to interpret statistical signals
Distribution of molecular effects in the genome
Natural selection

5. Outline Gene expression variation – recent studies
Analysis of gene expression with HapMap phase II SNPs
Update on CNV-expression associations
Natural selection and cis regulatory effects

6. Nature of regulatory variation
DNA
REG
GENE
i) Pre-mRNA
ii) mRNA
iii) Protein
iv) DNA
Expression
Stranger and Dermitzakis, Human Genomics 2005

7. Effects of Copy Number Variation on
gene expression
Copy number variation
kb to Mb size variable DNA copy number
contribute to disease
heritable
common in humans/recently appreciated

8. Gene expression association mappingAA AG GG
Quantitative phenotype
Stranger et al. PLoS Genet 2005

9. Whole-genome gene expression
~48,000 transcripts
24,000 RefSeq
24,000 other transcripts
270 HapMap individuals:
CEU: 30 trios, 90 total
CHB: 45 unrelated
JPT: 45 unrelated
YRI: 30 trios, 90 total
2 IVTs each person
2 replicate hybridizations each IVT
Quantile normalization of all replicates of each individual.
Median normalization across all individuals of a population.
illumina Human 6 x 2 gene GEX arrays
Cell line
RNA
IVT1
IVT2
rep1
rep2
rep3
rep4

10. HapMap SNPs 60 CEU 45 CHB 44 JPT 60 YRI 14,072 genes
Phase I HapMap; MAF > 0.05
CEU: 762,447 SNPs
CHB: 695,601
JPT: 689,295
YRI: 799,242
~1/5kb
The number of expression phenotypes is not a direct correlation to the number of genes in these regions because there were 2 probes per gene

11. Copy Number Variation dataset
Genome Structural Variation Consortium
Redon et al. Nature Nov 22, 2006
Array-CGH using a whole genome
tile path array
Median clone size ~170 kb
All 270 HapMap individuals
Quantitative values (log2 ratios) representing
diploid genome copy number, not genotypes.
1117 CNVs called from log2 ratios
Calls based on standard deviation of log2 ratios
Many CNVs experimentally verified
26,563 clones
93.7% euchromatic genome

12. Linear regression for SNPs CNV and expression
Clone signal (log2 ratio)

13. SNP cis-analysis: SNPs within 1Mb of probe midpoint
2Mb window
probe
gene
SNPs

14. CNV cis-analysis: clone midpoint within 2Mb of probe midpoint
4Mb window
probe
gene
clones

15. Permutation GENOTYPES GENE EXPRESSION g11 g12 g13 g14 … g1n
gi1 gi2 gi3 gi4 … gin
Exp1
Exp2
Exp3 …
Expi
permute
- 10,000 permutations – each time keep lowest p-value
- Null distribution of 10,000 extreme p-values
- Compare observed p-values to the tails of the null
Doerge and Churchill 1996

16. CNV vs. SNP associations
Stranger et al. Science 2007

18. CNVs and SNPs mostly capture different effects
Relative impact on gene expression: 82% SNPs
18% CNVs
Only 13% of genes with CNV association also had a SNP association in the same population
biased toward large effect size.
CNV and SNP variation are highly correlated (p-value 0.001).

19. 2 batches of 60 CEU individuals
Custom vs. Genome-wide [Stranger et al PLoS Genet and Stranger et al Science]
2 batches of 60 CEU individuals
grown independently at two different labs
RNA extraction and labelling by different labs and people
Run in custom and gw illumina arrays
97% of associations at the 0.05 permutation threshold from the custom array analysis were also detected in gw analysis

20. HapMap phase II analysis
~ 4 million SNP genotypes made publicly available for the 270 HapMap individuals.
Density: 1 SNP/ 700 bps
Includes ~50% of expected common SNPs in these populations.
2.2 million SNPs analyzed (MAF>0.05)

21. Phase I vs. Phase II cis- significant genes (0.001) phase I HapMap
both
phase II HapMap
CEU
286
258
299
CHB
317
269
318
JPT
337
297
341
YRI
356
310
394
90%
86%
85%
85%
87%
87%
87%
79%

22. Phase I vs. Phase II

23. Population sharing of cis- associations

24. Associated SNP position relative to TSS

25. Distribution of regulatory elements around the TSS
ENCODE Nature 2007

26. Direction of allelic effect same SNP-gene combination across populations
AGREEMENT
log2 expression
log2 expression
OPPOSITE
log2 expression
log2 expression

27. Direction of allelic effect

28. Pooling populations
Spurious associations
Pop1
Pop2
Pop1
Pop2

29. Conditional permutations
Permute data within each pop separately then perform test
X 4

30. Multi-population analysis

31. Figure 2A
Proportion of single pop cis associated genes detected in multi-population analysis
Number of populations sharing association in cis: single population analysis

32. SGPP2

33. Trans- phase II HapMap association
Biological hypotheses: functional categories
Regulatory SNPs identified from cis- analysis (52%)
Non-synonymous SNPs (39%)
Splice site SNPs (7%)
miRNA SNPs (1%)
DNA
REG
GENE
rSNPs
nsSNPs
spliceSNPs
miRNA SNPs
Genome-wide associations
Network analysis
GENE
~ 25,000 SNPs per population x 14,072 genes

34. Trans- associations 10-3 threshold
correction at genes estimated false positives
FDR = 33%-39%
correction at genes estimated false positives
FDR = 60%-75%
14,072 genes
tested

35. Enrichment of regulatory SNPs and deficit of nsSNPs in trans- associations
regulatory SNPs (cis 0.001)
ns SNPs
splice SNPs
miRNA SNPs
ratio
p-value
Ratio
CEU
6.05
3.23E-24
0.15
1.22E-21
0.49
0.07 1
CHB
3.69
7.90E-10
0.24
1.91E-09
0.76
0.71
JPT
3.15
2.06E-07
0.31
8.82E-07
0.55 !
3-6x more likely that a cis regulatory effect explains a trans regulatory effect

36. Multi-pop CNV analysis
Combined 4 populations: 193 genes at (48 overlap with the 99 from single population analysis)
Combined 3 populations: 173 genes at (42 overlap with the 99 from single population analysis)

37. CNV trans effects Variable expression Biological pathway
Copy number variation
kb to Mb size variable DNA copy number
contribute to disease
heritable
common in humans/recently appreciated
Biological pathway

38. Trans-position Copy number variation
kb to Mb size variable DNA copy number
contribute to disease
heritable
common in humans/recently appreciated

39. Trans effects - CEU

40. Trans effects - YRI

41. Gene expression and natural selection
-logpval
TSS
TSS
With Sridhar Kudaravalli and Jonathan Pritchard (unpublished)

42. Gene expression and natural selection
With Sridhar Kudaravalli and Jonathan Pritchard (unpublished)

43. Co-segregating regulatory variants can drive differential isoform expression

44. SUMMARY
Cis- and trans- acting genetic variation influencing mRNA levels.
CNV effects detected are largely not captured by SNPs
Structural variation (copy number polymorphism) influences transcript level variation.
Many detected associations are shared across human populations – replication of effects
Signal concentrated within 100 Kb from the promoter symmetrically
Trans-acting effects of CNVs - interpretation
Primary effects of trans associations are largely cis regulatory effects
Cis regulatory effects under positive selection

45. Acknowledgements Cambridge University Stanford illumina
Mark Dunning
Natalie Thorne
Simon Tavaré
Barbara Stranger
Alexandra Nica
Antigone Dimas
Christine Bird
Matthew Forrest
Catherine Ingle
Claude Beazley
Panos Deloukas
Matt Hurles
Stanford
Daphne Koller
illumina
Jill Orwick
Mark Gibbs
Genome Structural Variation Consortium
Richard Redon, Nigel Carter, Charles Lee, Chris Tyler-Smith, Stephen Scherer,
The HapMap
Consortium
Wellcome Trust for funding