43 b 32 o 59 55 48 51 54 po 45 45 po 32 39 + + breast/ovarian family 22 † 57 † 49
TRANSCRIPT
43 43
B 32B 32O 59O 59
55 55 484851515454
POPO 4545
POPO 4545
3232 39 39
++ ++
Breast/Ovarian Family
22
† 57 † 49
Inherited predisposition
More BRCA-like genes
Rare, moderately strong variants
Common genetic variation
Role of normal genetic variation in determining individual risk.
How useful is this information in selection for screening and prevention?
How do we find the genes?
Breast cancer as an example
Evidence that genetic variation affects risk
Measure of variation = familial clustering
Risk in close blood relative compared to risk in population as a whole
= roughly 2-fold.
Is family clustering genetic?
Incidence % per yearMZ twin 1.31
DZ twin 0.5Mother/sister 0.36
Patient’s contralateral breast 0.66
(Peto & Mack, Nat Genet 26, 411 (2000))
How much genetic predisposition is there?How is it distributed?
Determines potential for discriminating individual risks
risk
43 43
B 32B 32O 59O 59
55 55 484851515454
POPO 4545
POPO 4545
3232 39 39
++ ++
Breast/Ovarian Family
22
† 57 † 49
Population
BRCA1/2mutation
OBS EXP Excess
177 106 71
13 1.47 11.5
Fraction of excess familial clustering attributable to BRCA1/2 = 15-20%
Familial clustering of breast cancer
Familial clustering of breast cancer
1
2
Excess familial risk
Roughly 15-20%due to BRCA1/2
ATMChk-2Ha-rasPTEN
Risk to1o relativeof case
What sort of genes may account for familial risk apart from BRCA1/2?
Common low-penetrant genes
BRCA3 etc BRCA1, 2
Allele freq. XsFRR Number Allele freq. XsFRR Number 1% .25 350 0.2% 16 5 10% 2.3 35 30% 5.3 16
1.5 10 Relative risk
Patterns of breast cancer in families
1500 cases, population basedBRCA1/2 excluded
What model fits best?
Best fit = combined result of several factors, individually of small effect
= log-normal distribution of risk
in population.
0.010
0.020
0.030
0.040
0.01 0.10 1.00 10.00 100.00
Relative risk
SD = 1.2
CasesPopulation
Distribution of genotypes inpopulation and cases by
genotype risk
0.000
Proportion of population and cases above specified risk: SD =
1.2P
ropo
rtio
n ab
ove
give
n ris
k (x
)
Risk of breast cancer by age 70
0%
50%
100%
0% 20% 40% 60% 80%
CasesPopulation
88%
10%
12%
46%
3%
Effects of normal genetic variation on breast cancer
risks
Population10% 50%
46% 12%
Cancers
Individual risk by age 70 > 1 : 8 < 1 : 30
Proportion of population and cases above specified risk: SD =
0.8
0%
50%
100%
0% 20% 40% 60% 80%
CasesPopulation
Pro
port
ion
abov
e gi
ven
risk
(x)
Risk of breast cancer by age 70
80%
31%
10%
11%4%
Proportion of population and cases above specified risk: SD =
0.3
0%
25%
50%
75%
100%
0% 20% 40% 60% 80%
CasesPopulation
Pro
port
ion
abov
e gi
ven
risk
(x)
Risk of breast cancer by age 70
Gail model of breast cancer risk Nurses Health Study Analysis
Excellent prediction of breast cancer incidence in specified population.
Poor prediction of risk to individual.
2.8-fold between upper and lower deciles
cut-off for tamoxifen use defined 33% of population with 44% of cases.
(Rockhill, JNCI 93, 358 (2001))
- find genes- interactions- validation
40x
risk
1/5 1/5
QuickTime™ and aPhoto - JPEG decompressor
are needed to see this picture.
How to find the genes?
Association studiesarg cys
directindirect
linkage disequilibrium
C T
V
Problems: recombination origins different time
multiple origins
Common variant : common disease Rare variants
MarkerDisease allele
Candidate genes
Estrogen synthesis and degradation; ER
Cell cycle checkpointsDNA repairTGF pathwayIGF pathwayCarcinogen metabolism
Sample setsInitial : 2000 cases, 2000 controlsConfirmatory : 2000 cases, 2000 controls
Cases - Population based, East Anglia simple epidemiology data, survival;
paraffin blocksControls - EPIC cohort, East Anglia
extensive epidemiological data, follow-up, serum, mammography, bone density, etc
(Antoniou & Easton, submitted)
Percentage polygenic variance explained.
0
1000
2000
3000
4000
5000
6000
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Allele Frequency
Sam
ple
size 1%
2%
5%
10%
90% power p = 10-4
multiplicative
Power
Provisional positive associations : breast cancer
98 snps 47 candidate genes
Risk Br Ca Fraction to age 70 of excess
Freq OR PAF (5.7%) RR
TGF 14% 1.25 2.9% 6.8% 0.2%BRCA27% 1.31 2.1% 7.4% 0.3%XRCC3 15% 1.34 4.4% 7.4% 0.5%ER 20% 1.27 5% 6.8% 0.5%
Chk2 0.5% 2.4 0.6% 16% 0.5%
~2.0%
0.1 1 10
Joint NN
Joint NH
Joint HH
UK set 1 HH
UK set 2 HH
UK set 3 HH
HDB HH
Finns HH
p=0.02
BRCA2 N372H association with breast cancer risk
0.1 1 10 100
Tee et al. In prep.
Fiegelson et al. 2001
Haiman et al. 1999
Mitrunen et al. 2000
Kristensen et al. 1999
Spurdle et al. 2000
Miyoshi et al. 2000
Kuligina et al. 2000
Hamajima et al. 2000
Huang et al. 1999
Helzlouler et al. 1998
Weston et al. 1998
Weston et al. 1998
Young et al. 1999
Bergman-Jungestrom et al. 1999
OR breast cancerCYP17 t -34 c (cc Vs. tt)
Conclusion:This SNP has no main effecton breast cancer risk!
Ye & Parry, 2002 Mutagenesis 17:119-126 N
226
230
3133
310
744
1081
Why a p value of p = 0.01 is not persuasive
Prior probability of result (snp causing 1% of FRR, 100,000 snps in genome) 1/1000999/1000
Probability given result has p = 0.01 99/100 1/100
99/100,000 999/100,000
Assuming random choice of ‘candidate’ gene only ~ 10% results at p = 0.01 are true
(~50%, at p = 0.001)
True Falseassociation association
SNP
0.001
0.01
0.10
1.000 10 20 30 40 50 60 70 80 90 100
p-v
alu
e
0.05
observed
chance
Summary of results 96 snps, 47 genes~2000 cases, 2000 controls
p = 0.01/0.0004 for comparison of distributions
0.5 1 1.3 2 relative risk
% of excess FRR explained
Some reasons why human association studies may be
difficultInappropriate genetic models eg rare/multiple alleles
Regulatory vs coding polymorphisms
Numbers : inadequate statistical power
Genetic background effects; interactions weak ‘main effect’, high-order interactions ‘null’ result = balance of susceptible and
resistant on different BG
Phenotypic heterogeneity eg ER+/ER-; histology
Cancer/no cancer endpoint lacks power
Intermediate phenotypes
P homogeneity = 0.0005P trend <0.0001
Serum estradiol and CYP19Exon 10 t>c 3’UTR
10
12
14
16
18
20
tt tc cc
Serum SHBG and SHBGExon 8 g>a or D356N
20
30
40
50
60
gg ga aa
P homogeneity = 0.006P trend = 0.006
(Ponder, Dowsett labs; EPIC; unpublished)
Implications for breast cancer risk
2 fold increase in estradiol 30% increase in risk of breast cancer
tt genotype of CYP19 c>t associated with 14% increase in estradiol: equivalent to 1.04 fold increase in breast cancer risk
Where next?
Empirical vs candidate approaches
Snp genotyping now ~17c/genotype : ? screen 600 “enriched”
cases/600 controls vs 1150 coding snps
~$240,000
Candidate gene approaches
Candidates from cell biology
Epidemiology
Regulatory variants
Quantitative phenotypes
Leads from mouse models
Mouse/human collaborations
1. Candidate susceptibility genes/regions
mapped in susceptible/resistant crossesrefined by amplicons/deletions in tumoursallele-specific differences in expression/somatic change (easier in mouse because extended haplotypes)
loci involved in control of gene regulation
loci influencing intermediate phenotypesset up large cross and score multiple phenotypes
How tightly should the region be defined?
Say 5 genesFirst pass = find all coding region snps at >5%Construct haplotypes, select minimum snp set = ? 30 snps
Genotype 30 snps in 2000 cases/2000 controls = 120,000 genotypes
Genotyping cost ~$20,000 @ 17c/genotype
BUT : currently requires ~1000 snps at a time
300 kb
Mouse/human collaborations
2. Interactions
Identification of interacting loci potentially approachable in
mouse
Develop and evaluate programmes to search for higher order
interactions;? applicability to man
Mouse/human collaborations
3. Stages of cancer development
? Distinguish loci that influencemultiplicitylatency; progressioninvasionmetastasis and resistance to
these
? Loci that affect treatment response
Mouse/human collaborations
4. “End game” - which is the active gene, snp?
strain comparisons of variantsdissection of complex QTLs
transgenic models
“‘Risk factor’ analysis will facilitate environmental modification, screening and therapeutic management of people before they develop symptoms”
(Bell, BMJ 1998)
“Differences in social structure, lifestyle and environment account for much larger proportions of disease than genetic differences …… Those who make medical and scientific policies ….. would do well to see beyond the hype”
(Holtzman & Marteau, NEJM 2000)
A new horizon in medicine?
Strangeways Research Laboratories - University of Cambridge
Bruce Ponder Doug EastonPaul Pharoah Antonis Antoniou UCSFAlison Dunning Mitul Shah Allan BalmainFabienne Lesueur Julian Lipscombe Mandy TolandBettina Kuschel Joe GrayAnnika Auranen Nick Day; EPIC Mark SternlichtKatie Healey NCICraig Luccarini Kent HunterJenny He Louise Tee Biochemistry, CambridgeGary Dew Jim Metcalfe
Cancer Research UK; MRC
TGF
t/c Pro/Leu
-509 10
t P
c P
c L
0.25
0.11
0.60
PP vs LL OR 1.25 (1.1 - 1.4) p = 0.01
tt vs cc OR 1.30 (1.1 - 1.5) p = 0.01
Which SNP is the functional variant?
0.1
cc LeuLeu
cc LeuPro
ct LeuPro
cc ProPro
ct ProPro
tt ProPro
1.0 10Odds Ratio
Pro10 homozygoteshave increased riskregardless of c-509tgenotype
TGF in vitro secretion
0
1
2
3
4
TGF1ng/ml
untransfected
cells
CM
V-E
+ßgalC
MV
-LC
MV
-L+CM
V-E
CM
V-L+ßgal
CM
V-P
CM
V-P
+CM
V-E
CM
V-P
+ßgal
181260
hours
Time CourseEnd Point
Leu10
Pro10Ratio P:L
(Metcalfe, Ponder labs, 2002)
Funnel Plot For TGF L10P
0.1 1 10
Frei
Ziv et al.
Hishido et al.
Finn
HDB
ABC
OR (PP Vs. LL)
N
238
3075
404
939
875
4517
** Cohort study
146 cases 2929 controls