Human genetic evidence supports approved drug indications Matt Nelson October 12, 2015 Acknowledgements • GSK – Hannah Tipney – Judong Shen – Jeff Painter – Mark Hurle – Pankaj Agarwal – John Whittaker – Philippe Sanseau • GWASdb – Hong Kong University – Mulin Jun Li – Pak Chung Sham – Junwen Wang • Pointer variant-gene mapping – Columbia University – Paola Nicoletti – Yufeng Shen – Aris Floratos • ENCODE DHS correlations – University of Washington – John Stamatoyannopoulos • Manual review of top GWAS diseases – McGill University – Rui Li – Vincent Forgetta – Mark Lathrop – Brent Richards 2 Failure due to lack of efficacy is a major challenge in drug development Drug will progress to next phase Overall probability of success Drug in current phase will be approved × × ~$1.5B in R&D costs per drug Hay et al. (2014) Nat. Biotech. 32:40-51 Arrowsmith & Miller (2013) Nat. Rev. Drug. Disc. 12:569 3 Genetics provides natural human experiments that can help guide target selection and quality assessment Preclinical Studies Experimental Medicine Phenotypic Screens Clinical Validation Molecule Target Disease Mechanism Cellular Models Rare Diseases Animal Models GWAS & PheWAS Literature Mining Human Epidemiology Regulatory Networks 4 Genome-wide association studies identify genetic factors that affect human health ~20,000 subjects QQ (PP) Plot 5 Kathiresan et al. (2009) Nat. Genet. 41:56-65 Genome-wide association studies have identified thousands of variants that influence human traits Associations P ≤ 5×10-8 Publications New Unique Traits Source: NHGRI/EBI GWAS Catalog, July 10, 2015 6 Next-generation sequencing is uncovering genetic causes of rare monogenic disorders 10 discovery probands, 7 validation, followed up in 53 total families 7 Ng et al (2010) Nat Genet 42:790–3; Bamshad et al (2011) Nat Rev Genet 12:745–55 Next-generation sequencing is uncovering genetic causes of rare monogenic disorders at a rapid rate (~3/week) Chong et al. (2015) Am. Jour. Hum. Genet 97:199-215 % LDL-C reduction LDL-C –log10(p) Can human genetics improve drug discovery, development and repositioning decisions? Kathiresan et al. (2009) Nat. Genet. 41:56-65 Rossi S, ed. Australian Medicines Handbook. Adelaide, 2010 Questions Is genetic evidence enriched within targets for successful drugs? How often are successful drug mechanisms supported by a genetic association? Does this vary among diseases and therapeutic areas? STOPGAP Project: Systematic Target Opportunity Assessment by Genetics Association Predictions 9 STOPGAP project integrated three sources of data to assess the genetic evidence for successful drug targets 10 Genetic variants are mapped to genes through LD, gene expression (eQTL), and regulatory elements Disease Kawasaki disease Variant PubMed Pvalue rs1600249 22446962 2.40E-10 Rheumatoid arthritis rs1600249 21225715 5.00E-06 Physical position Linkage disequilibrium Regulatory control Prior development efforts in liver fibrosis, AD, and cancer DHS Correlation 11 GeneScore: quantifying the causal connection between variant and gene Association source GWAS GWAS Other Suppl. Catalog OMIM <0.75 ≥0.75 ≥ 0.9 -log10(p-value) LD with reported variant 6060 –log10 (p-value) Variant function DHS eQTL or eQTL & Other Missense Rdb 3 DHS (2) DHS 4040 Number of independent associations <3 ≥3 ≥5 ≥ 10 2020 Gene Score Contribution 0 1 Low 2 3 4 5 6 Gene Score Medium 7 8 9 10 11 High 12 Genetic variants around targets of approved drugs are more likely to affect human traits than other genes *Protein sequence variation *Protein sequence variation Odds Ratio (log scale) Enrichment of target genes with any genetic evidence *RVIS: Residual variation intolerance score from Petrovski et al. PLoS Genet. 9:1003709 13 Using the “is a” relationships among MeSH→UMLS terms to assess similarity for analysis Applied both Lin and Resnik “path + information content” methods, normalized, and averaged From Pedersen et al. IHI 2012 Tutorial http://www.comp.hkbu.edu.hk/ihi2011/Tutorials%20-%20IHI2012.htm Using the relationships among MeSH→UMLS terms to assess trait and indication similarity for analysis Example similarity measures Cardiovascular Diseases Heart Diseases Neoplasms by Histologic Type Neoplasms by Site Mycardial Ischemia Glandular and Epithelial Neoplasms Neoplasms, Genitourinary Carcinoma Neoplasms, Urologic Squamous cell carcinoma Neoplasms, Kidney Coronary Disease Coronary Stenosis Neoplasms Coronary Artery Disease Relative similarity = 0.79 Relative similarity = 0.37 15 Most approved drug indications have 1+ similar traits that have been investigated genetically 60 Count 40 20 0 0.00 0.25 0.50 0.75 1.00 Relative Similarity 16 Direct genetic evidence for approved target–indications varies significantly among disease areas Test of heterogeneity among disease categories: p = 10-12 17 The percentage of targets with genetic evidence increases with later stages of clinical development 18 Both Mendelian and complex trait genetics correlate with successful drug targets 19 Genetic effect size is not related to approved drug status Effect size does not predict likelihood of success Target for approved drug with genetic evidence In collaboration with Brent Richards, McGill University 20 The relative advantage of genetic association in predicting success is hampered by a lack of failure data >127,000 drug trials Trial Outcome Recorded Phase 2 and/or 3 Failed Due to Efficacy Disease Area Count (%) Neoplasms 152 (68%) Digestive System 17 (8%) Musculoskeletal 15 (7%) Hemic Lymphatic 10 (4%) Nutritional and Metabolic 5 (2%) Only 15% of approved targetindication for Neoplasms 224 Target-indication failures due to efficacy 21 Estimating the relative advantage of targets with genetic support Assuming the overall probability of a drug in phase II progressing to approval is 27% we estimate the impact of genetic information on probability of success – P(Success | Genetic Assn) ≈ 0.52 – P(Success | No Genetic Assn) ≈ 0.26 – Relative value of genetic information (ratio): 2.0 (95% CI = {1.6, 2.4}) Genetic evidence increases the probability that a drug will progress to approval 23 Increasing the proportion of drug targets with genetic evidence can lead to higher clinical success rates Assume 15% of historic portfolio with genetic evidence 30% with 95% confidence intervals (blue) 2-fold 24 Uncovering the genetic contributions to human health will have measurable contributions to discovering new medicines 60 Count 40 20 Associations P ≤ 5×10-8 0 0.00 0.25 0.50 0.75 1.00 Relative Similarity Publications New Unique Traits Source: NHGRI/EBI GWAS Catalog, July 10, 2015 25 Summary • Targets of successful drugs are more likely to be coded by genes that influence human health • About 8% of successful drug mechanisms have direct genetic evidence for the approved indication – This is a ~3.5 fold enrichment over target–indications that did not progress beyond phase 1 – As genetic knowledge continues to accrue, this proportion should increase • Based on historical drug development data and existing GWAS and OMIM data, direct genetic evidence supporting target role with the indication doubles the probability of success over drugs without it • There is ample opportunity and need for genetic studies of disease that lack adequate therapies and further elucidation of the genetic risk factors that influence them 26 Important provisos • This study does not mean that genetic evidence is a panacea – Not all drugs with genetically supported targets and indications will succeed – Most successful drugs do not currently have genetic evidence • All other things being equal genetics evidence increases probability of success • Genetic evidence provides guidance for choice of indications – Currently, most genetic studies deal with relatively crude disease outcomes in broadly defined populations • Increasing targets with genetic evidence should reduce attrition due to lack of efficacy 27 Thank you Permutation algorithm Permute this relationship Variant-Trait Drug-Indication Gene-Trait Gene-Indication Permute by creating new trait-trait mappings to retain relationships among variants/genes and new traits within each region Original Trait Trait Permutation HDL-C T2D LDL-C Allopecia Total-C Kawasaki Disease Etc. Maintains structure at gene-trait level, but does break structure among correlated traits Merge on bestmatch trait-indication for each gene Gene-Trait-Indication Compute overlap at several relative similarity and pvalue thresholds Permutation test demonstrates enrichment of genetic associations among approved drugs is highly significant Result of 1000 permutations Observed overlap Applying human genetics to target validation Criteria for gene–drug pairs in drug discovery • The gene harbors a causal variant that is unequivocally associated with a medical trait of interest • The biological function of the causal gene and causal variant are known • The gene harbors multiple causal variants of known biological function, thereby enabling the generation of genotype–phenotype dose–response curves • The gene harbors a loss-of-function allele that protects against disease, or a gain-of-function allele that increases the risk of disease • The genetic trait is related to the clinical indication targeted for treatment • The causal variant is associated with an intermediate phenotype that can be used as a biomarker • The gene target is druggable • The causal variant is not associated with other adverse event phenotypes • Corroborating biological data support genetic findings 31 Genome-wide association studies identify genetic factors that affect human health ~37,000 schizophrenia cases ~113,000 controls 32 Schizophrenia Working group of the Psychiatric Genomics Consortium (2014) Nature 511:421-7