Gene Finding in Clinical Trial Populations

Tom Price

SGDP 18th Feb 2009http://sgdp.iop.kcl.ac.uk/tprice/

Translational Value of Pharmacogenetics

Genetic studies can help:• Identify drug targets• Decrease attrition in development of new drugs• Increase safety by predicting adverse events

• Improve treatment by predicting efficacy

Roses AD (2008). Nat Rev Drug Discov. 7(10):807-1.

Problems in Pharmacogenetics

• Questionable value of many diagnostics– Sensitivity/specificity– Clinical relevance

• Misaligned goals of academe / industry– Marketing function of pharma RCTs – Unpublished studies

• Expense

The Good, the Banned, the Ugly

• According to FDA estimates, Vioxx (rofecoxib) caused more than 27,000 excess cases of myocardial infarction and sudden cardiac death before it was withdrawn from the market

Do me-too drugs have a similar risk profile?

Are there subgroups for whom this drug is safe?

COX Isoforms & Prostaglandins

Maree & Fitzgerald Thromb Haemost 2004, 92:1175–81

NSAIDsCoxibsNSAIDs

COX Inhibitors

• NSAIDs (nonsteroidal anti-inflammatories)– e.g. aspirin, ibuprofen– The world's most prescribed drugs– Inhibit both COX-1 and COX-2

• Gastrointestinal side effects

• Coxibs– e.g. rofecoxib (Vioxx), celecoxib– Selectively inhibit COX-2

• Easier on stomach• Elevated cardiovascular risk vs placebo

Side Effects of COX Inhibition

Grosser, Fries & FitzGerald J Clin Invest 2006, 116:4–15

Should We Ban All Coxibs?

• Coxibs have similar biochemical effects – are they all as dangerous as Vioxx?

• Alternatives:– Off-target effects specific to Vioxx– Interindividual variability– Subgroups account for cardiovascular hazard

• Crossover trial of 2 anti-inflammatory drugs

• 50 healthy volunteers received single doses of celecoxib, rofecoxib or placebo in random order with a 7 day washout

• Drug effects measured using in vivo and ex vivo assays of COX-1 and COX-2 enzymatic activity

• 5 individuals underwent this protocol 5 times to study intraindividual variability of response

Biomarkers of Drug Action

PGI2 biosynthesis in vivo2,3-dinor-6keto PGF1αUrine

COX-2 activity ex vivoPGE2Plasma

TxA2 biosynthesis in vivo11-dehydro TxB2Urine

COX-1 activity ex vivoTxB2Serum

MeasureBiomarkerTissue

Drug Responses

Measurements obtained at baseline and 4 hours after dosing

Attained COX-2 Selectivity

Ratio of ex vivo COX-2 to COX-1 inhibition 4h post dose

Variability in Drug Response(COX-2 ex vivo assay)

Within subjectsBetween subjects

Drug Availability

Plasma concentration in blood drawn 4 hours after dosing

Biomarkers of Drug Efficacy

COX inhibition in vivo COX inhibition ex vivo

Conclusions

• Rofecoxib and celecoxib attain similar selectivity for COX-2 ex vivo and in vivo– We have no reason to believe that CV toxicity

is not a class effect of COX-2 inhibitors

• Genetic factors are likely to explain some interindividual variability in drug response– Can we exploit this to predict CV toxicity?

MEDAL Program

• Comparative trial of cardiovascular safety of 2 COX inhibitors – Etoricoxib (Vioxx clone approved in UK)– Diclofenac (NSAID)

• Over 34,000 arthritis patients enrolled

• Average 18 month follow up

Cannon et al. Lancet 2006, 368: 1771-81

MEDAL Results

• No difference between drugs in primary endpoint (heart attack or stroke)

• Some improvement in minor GI symptoms for etoricoxib vs diclofenac

• Blood pressure increase in etoricoxib vs diclofenac

• FDA rejected approval by 19:1

Cannon et al. Lancet 2006, 368: 1771-81

MEDAL Pharmacogenetic Study

• Over 6,000 subjects genotyped on 50K and 38K custom SNP arrays

• Genotyping funded by Rosetta (Merck)

• Intention to investigate gene x drug interactions in blood pressure & gastro side effects

IBC Chip

Brendan Keating

(U Penn)

Stacey Gabriel

(Broad Institute)

• Illumina custom SNP genotyping array for cardiovascular / metabolic / inflammatory disease

• 50K SNPs covering ~2,100 genes

• Dense ‘cosmopolitan’ tagging plus functional variants

• Many resequenced genes

• 200,000 chips manufactured and sold

Merck Chip

• Quickchip V1.5– Illumina custom 38K SNP array– eSNPs (liver, brain, blood, adipose)– All GWAS SNPs– OA/RA genes (WTCCC, franchise nominated)– HTN genes (Current targets)– Network derived (MM)

Sample

Data Analysis

• Genotype data available for 50K chip• Analyzed using PLINK • Genome-wide significance was assessed

as about 10-6 ≈ α of 0.05 after correction for about 50,000 SNPs

• Interesting results followed up by permutation analysis

Preliminary Analysis

• QC removed around 1,600 individuals (mainly non-whites) and 8,000 SNPs (mainly those with low frequency, MAF < 0.2%, in whites).

• The final sample included 4,441 unrelated ethnically white individuals genotyped on 33,661 SNPs, including ancestry informative markers.

QC

Individuals

Total number in PED file 6143

X chromosome genotype data inconsistent with sex

-81

Missing > 1% of genotypes -2

Not ethnic white -1531

Extreme heterozygosity -2

Remove all but one of groups of relatives -57

Remove outliers from EIGENSTRAT -29

Final 6086

SNPs

Total number in PED file 45707

Control SNPs -114

Monomorphic -441

MAF < 0.2% (total sample) -4415

Missing > 1% (MAF > 0.05), Missing > 0.2% (MAF <= 0.05)

-1055

Missing > 1% (ethnic whites) -18

MAF < 0.2% (ethnic whites) -5275

HWE p < 0.001 -435

MAF < 0.2% (after removing EIGENSTRAT outlying individuals)

-277

Final 34216

PhenotypesBaseline Traits• History of hypertension (HBHX)• Antihypertensive treatment at baseline (ANTIHYPE)• Baseline systolic blood pressure (SBP)• Baseline diastolic blood pressure (DBP)• BMI (BMI)• History of diabetes (DMHX)• History of dyslipidaemia (DLHX)

Drug-induced changes in blood pressure• Change in systolic blood pressure over baseline (SC)• Change in diastolic blood pressure over baseline (DC)• Interaction between drug and change in systolic blood pressure over baseline (SCX)• Interaction between drug and change in diastolic blood pressure over baseline (DCX)

Covariates

• First 2 principal components of genotype data from EIGENSTRAT (after removing high LD regions and outlier individuals)

• Age• Sex• Region (US/non US)• RA/OA• Plus (for change in BP over baseline)

– Smoking status– Time since baseline– Square of time since baseline

Eigenstrat

Blue = US

Green = Europe

Eigenstrat

• Interestingly the first 2 PCs of the genotype data, which presumably information on geographic origin within European populations, correlated significantly with baseline systolic and BMI but not baseline diastolic or antihypertensive medication.

History of Hypertension Results• Genome-wide hit for common SNP rs179998 (C−344T)

in promoter region of CYP11B2 (Aldosterone synthase)• LD with rs179998 accounts for subthreshold associations

with other SNPs in CYP11B2

History of Hypertension

CHR SNP UNADJ GC BONF HOLM SIDAK_SS SIDAK_SD FDR_BH FDR_BY 8 rs1799998 1.464e-06 1.464e-06 0.04928 0.04928 0.04809 0.04809 0.04928 0.5422 8 rs3802228 2.96e-05 2.96e-05 0.9963 0.9963 0.6308 0.6307 0.3749 1 8 rs11250163 3.904e-05 3.904e-05 1 1 0.7313 0.7313 0.3749 1 8 rs6433 4.455e-05 4.455e-05 1 1 0.7768 0.7767 0.3749 1 19 rs2230204 9.032e-05 9.032e-05 1 1 0.9522 0.9522 0.5778 1 20 rs6083780 0.0001268 0.0001268 1 1 0.986 0.986 0.5778 1 1 rs1200132 0.0001322 0.0001322 1 1 0.9883 0.9883 0.5778 1 8 rs6410 0.0001373 0.0001373 1 1 0.9902 0.9902 0.5778 1 16 rs7185735 0.0002862 0.0002862 1 1 0.9999 0.9999 0.6259 1

History of Hypertension Results

Genome wide significant hit on rs179998 5’ of CYP11B2

Aldosterone synthase CYP11B2

• Meta-analysis of 19 studies suggests that rs179998 may be associated with essential hypertension– CC homozygotes had a 17% lower risk than TT homozygotes

under a fixed effects model (OR 0.83; CI 0.76–0.91; p < 0.001) but not under a random effects model (OR 0.89; CI 0.76–1.04; p = .13)

– Heterogeneity between studies would suggest a random effects model is appropriate

– HTN defined as SBP>140 or DBP>90 or antihypertensive Rx

– No effect of rs179998 on SBP or DBP in untreated patients– Sookoian et al. J. Hypertens. 2007, 25:5-13

– Staessen et al. J. Hypertens. 2007, 25:37-39

Antihypertensive Use Results• Antihypertensive use at baseline correlates highly with

history of hypertension• There is some evidence of association with rs179998 but

below the threshold for genome-wide significance

Antihypertensive use at baseline

CHR SNP UNADJ GC BONF HOLM SIDAK_SS SIDAK_SD FDR_BH FDR_BY 21 rs2073362 3.989e-05 3.989e-05 1 1 0.7389 0.7389 0.4605 1 21 rs4986956 4.033e-05 4.033e-05 1 1 0.7427 0.7427 0.4605 1 8 rs1799998 4.104e-05 4.104e-05 1 1 0.7488 0.7488 0.4605 1 5 rs1498928 0.0001537 0.0001537 1 1 0.9943 0.9943 1 1 13 rs532625 0.0003381 0.0003381 1 1 1 1 1 1 1 rs4072431 0.0003654 0.0003654 1 1 1 1 1 1 19 rs2230204 0.0004178 0.0004178 1 1 1 1 1 1 14 rs12896130 0.0004234 0.0004234 1 1 1 1 1 1 2 rs2059693 0.0004714 0.0004714 1 1 1 1 1 1

Blood Pressure Results

Nothing much came up for any of the blood pressure phenotypes

Blood Pressure Results

• Nothing much came up for any of the measured blood pressure phenotypes - possibly because antihypertensive use was included as a covariate, so genetic influences on liability to antihypertensive use were already excluded

• Among the nonsignificant top hits were FTO on chromosome 16 with baseline systolic BP

Baseline Systolic

CHR SNP UNADJ GC BONF HOLM SIDAK_SS SIDAK_SD FDR_BH FDR_BY 16 rs12324955 1.991e-05 1.991e-05 0.6703 0.6703 0.4885 0.4885 0.6703 1 3 rs3774061 5.446e-05 5.446e-05 1 1 0.8401 0.8401 0.7331 1 12 rs11172124 6.948e-05 6.948e-05 1 1 0.9035 0.9035 0.7331 1 16 rs6499656 8.712e-05 8.712e-05 1 1 0.9467 0.9467 0.7331 1 1 rs300267 0.0001123 0.0001123 1 1 0.9772 0.9772 0.7562 1 2 rs4675278 0.0001524 0.0001524 1 1 0.9941 0.9941 0.7891 1 2 rs10200844 0.0002173 0.0002173 1 1 0.9993 0.9993 0.7891 1 6 rs2295591 0.0002566 0.0002566 1 1 0.9998 0.9998 0.7891 1 4 rs2069763 0.0003063 0.0003063 1 1 1 1 0.7891 1

BMI Results• A low frequency (1%) SNP rs3781637 in intron 1 of MTNR1B melatonin receptor

1B was associated with BMI at genome-wide significant level• Statistical significance was confirmed by permutation analysis (p = 0.03)• There is also a cluster of SNPs associated in the range p=10-4 - 10-5 in the NRG1

neuregulin1 gene locus on chromosome 8. There was also a hit on NRG1 with a similar p value for history of hypertension.

• FTO and MCR4, the most consistently replicated associations with BMI, do not feature in the top hits

BMI

CHR SNP UNADJ* GC* BONF* HOLM* SIDAK_SS* SIDAK_SD* FDR_BH* FDR_BY* 11 rs3781637 2.242e-08 2.242e-08 0.0007548 0.0007548 0.0007545 0.0007545 0.0007548 0.008304 8 rs12675298 2.151e-05 2.151e-05 0.7239 0.7239 0.5151 0.5151 0.2076 1 8 rs2881544 3.524e-05 3.524e-05 1 1 0.6946 0.6946 0.2076 1 10 rs196335 3.697e-05 3.697e-05 1 1 0.7119 0.7119 0.2076 1 8 rs989465 4.09e-05 4.09e-05 1 1 0.7476 0.7475 0.2076 1 8 rs1383961 4.436e-05 4.436e-05 1 1 0.7753 0.7753 0.2076 1 6 rs3734681 4.637e-05 4.637e-05 1 1 0.7901 0.79 0.2076 1 8 rs1979565 4.935e-05 4.935e-05 1 1 0.8101 0.81 0.2076 1 12 rs3213900 5.868e-05 5.868e-05 1 1 0.8613 0.8612 0.2195 1

*Test statistics are approximate since BMI is not normally distributed

BMI Results

Genome wide significant hit on Chromosome 11

MTNR1B

• Meta analyses have not previously identified SNPs in this gene to be associated with BMI.

• Murine KOs for the homologous gene GPR50 are resistant to diet-induced obesity (PMID: 17957037).

• The MAGIC consortium has reported that MTNR1B genotype is associated with fasting glucose levels and diabetes susceptibility (Prokopenko et al., poster AHSG 2008).

Diabetes Results

• MTNR1B has been associated with type II diabetes, but no association was found with history of diabetes in this sample

• The top 2 hits were in GLUT5 and CETP

History of diabetes

CHR SNP UNADJ GC BONF HOLM SIDAK_SS SIDAK_SD FDR_BH FDR_BY 2 rs2018414 2.015e-05 2.015e-05 0.6784 0.6784 0.4926 0.4926 0.3436 1 16 rs9923854 2.042e-05 2.042e-05 0.6873 0.6872 0.4971 0.497 0.3436 1 1 rs12145292 3.372e-05 3.372e-05 1 1 0.6786 0.6786 0.3783 1 17 rs9892909 6.299e-05 6.299e-05 1 1 0.88 0.88 0.5301 1 5 rs2895795 0.0001022 0.0001022 1 1 0.9679 0.9679 0.5809 1 10 rs7903146 0.0001046 0.0001046 1 1 0.9704 0.9704 0.5809 1 5 rs1042718 0.0001656 0.0001656 1 1 0.9962 0.9962 0.5809 1 2 rs315921 0.0001699 0.0001699 1 1 0.9967 0.9967 0.5809 1 6 rs1057293 0.0002449 0.0002449 1 1 0.9997 0.9997 0.5809 1

Further possibilities

• Gene x drug effects on GI side-effects– Summer 2009

• Case-only estimation of epistasis in osteo-/rheumatoid arthritis

Case Control Study

G- G+

E- a b

E+ c d

Cases Controls

G- G+

E- e f

E+ g h

ORGXE = a d f g / b c e h

Var (ln ORGXE) = 1/a + 1/b + 1/c + 1/d + 1/e + 1/f + 1/g + 1/h

Case Only Study

G- G+

E- a b

E+ c d

Cases

ORGXE = a d / b c

Var (ln ORGXE) = 1/a + 1/b + 1/c + 1/d

Assuming that G and E are uncorrelated

Case Only Study nested in RCT

G- G+

Drug A

a b

Drug B

c dSide effect cases

Drug exposure is randomized →

Drug exposure and genotype are uncorrelated

Conclusions

• Genetic epidemiology studies can make use of RCT populations

• RCT samples of convenience can have disadvantages e.g.– Inappropriate size – Unrepresentative populations– Nonrandom recruitment into PhGx / dropout

• Some innovative research designs are possible