Pharmacogenomics of GPCR Drug Targets

Abstract

Natural genetic variation in the human genome is a cause of individual differences in responses to medications and is an underappreciated burden on public health. Although 108 G-protein-coupled receptors (GPCRs) are the targets of 475 (∼34%) Food and Drug Administration (FDA)-approved drugs and account for a global sales volume of over 180 billion US dollars annually, the prevalence of genetic variation among GPCRs targeted by drugs is unknown. By analyzing data from 68,496 individuals, we find that GPCRs targeted by drugs show genetic variation within functional regions such as drug- and effector-binding sites in the human population. We experimentally show that certain variants of μ-opioid and Cholecystokinin-A receptors could lead to altered or adverse drug response. By analyzing UK National Health Service drug prescription and sales data, we suggest that characterizing GPCR variants could increase prescription precision, improving patients' quality of life, and relieve the economic and societal burden due to variable drug responsiveness. VIDEO ABSTRACT.

Keywords: FDA approved drugs; GPCR; clinical trial; drug response; economic burden; natural variation; opioid receptor; personalized medicine; pharmacogenomics; polymorphism.

Graphical abstract

Figure 1

Pharmacogenomic Landscape of GPCR Drug Targets Schematic highlighting the different types of data analyzed in this study, ranging from data on drug targets, variants, functional effects, sequences, structures to information on prescription, and sales of drugs in the UK.

Figure 2

Distribution of Individuals Harboring Missense Variation in GPCR Drug Targets (A and B) Estimates based on genotype data from 2,504 individual genomes was made per individual on (A) number of missense variants in GPCR drug targets (left) and the number of GPCR drug targets harboring a missense variation (right) and (B) number of clinically known variants that alter efficacy of drug response or toxicity in GPCR drug targets (left) and the number of affected GPCR drug targets with clinically known mutations (right). (C) Allele frequencies of variants, known to alter drug response in 2,504 individuals (number of homozygous/heterozygous carriers) (Table S2). (D) Analysis of 1,762 studied trios (father-mother-offspring) revealed a total of 9 de novo missense mutations in 6 GPCR drug targets.

Figure 3

Genetic Variation Landscape of GPCR Drug Targets (A–C) Scatterplots of (A) missense variation (red), (B) loss-of-function mutations (blue), and (C) copy-number variation (purple) for GPCR drug targets. Each mutation type shows the number of observed variants (separated into deletions and duplications for CNVs) for a given GPCR drug target. Missense variation density was obtained by normalizing number of missense mutations to the receptor sequence length. Loss-of-function mutations are presented as the minimum percentage of individuals harboring at least one copy of a protein-truncating variant (STAR Methods). Correlations and mean values (μ) are shown for MVs and LoFs. Mean values (μ) for the distributions are provided. Genetic variation landscapes of GPCR drug targets that are in clinical trials are provided in Table S3 and S4). Lower half of the figure shows the distribution of top 10 (upper panels) and bottom 10 ranking GPCR drug targets (lower panels). See also Figures S1, S2, and S3.

Figure S1

Frequency of Genetic Variants in GPCR Drug Targets, Related to Figure 3 (A) The allele frequency spectrum (ExAC data) of the 108 GPCR targets of approved drugs shows that most genetic variants are rare (single observations or allele frequency ≤ 0.01%). Common variants (> 0.01%) exist for 358 sites. The coloring shows missense variations with an amino acid property change (‘changed’) and missense variations, where the mutant amino acid substitution is within the same class of amino acid properties (‘similar’). (B) To assess the most polymorphic GPCR drug targets (rows) across different categories including for MVs, LoFs and CNVs, Z-scores were calculated within categories (columns) and grouped by hierarchical clustering. Receptors with high genetic variation are highlighted in red (Table S3).

Figure S2

Loss-of-Function and Copy-Number Variations in GPCR Drug Target, Related to Figure 3 (A–C) Distribution of human GPCRs by the number of FDA approved drugs that target them (y axis, logarithmic scale) and (A) the number of missense variants (x axis) along with the fraction of MVs by receptor length (red color scale), (B) the number of loss of function variants (x axis) along with a conservative estimate of the minimum population frequency (blue color scale) and (C) the number of observed deletions (x axis) and duplications (purple color scale). GPCRs that are frequently targeted by drugs (i.e., many FDA-approved drugs) are highly polymorphic in terms of MVs, LoF variants and CNVs.

Figure 4

Missense Variations in GPCR Drug Targets and Their Possible Functional Impact (A) Variants predicted to have impact by SIFT or PolyPhen (dark green). MVs can affect different functional sites (light green), which were defined as ligand binding (left), post-translational-modification site (bottom), and micro-switches including allosteric sodium ion binding pocket and G-protein/arrestin interaction interface (right). The displayed structures show missense variants within 5 Å (red) of an approved drug (left and right) or MVs within 5 Å distance to the G protein or arrestin. PDB IDs are provided in the bottom of each structure sub-panel. (B) Disease ontology (left), FDA-approved drug (middle), and variant (right) known (i.e., statistical association in clinical-genetics studies) to alter drug response or efficacy or lead to adverse drug response (Table S2). In some cases, the drug and disease are linked to reflect the clinical study design and are not drugs given to treat those diseases. See also Figure S4.

Figure S3

All Available X-Ray Crystal Structures of GPCRs in Complex with FDA-Approved Drugs, Related to Figure 4 FDA approved drugs bound to their respective receptors (n = 15) are shown in green. Missense variations from 60,706 individuals within 5Å of the co-crystallized drug are highlighted in red.

Figure S4

Missense Variants in Post-translational Modification Sites and Structural Segments, Related to Figure 4 (A) Missense variants were mapped onto experimentally verified post-translational modification sites of GPCR drug targets (n = 846). Number of missense variants per post-translational modification site type. (B) Structural segments were assigned for each receptor. Each segment was then aggregated into higher-order groups: C terminus, extracellular loops, transmembrane region, intracellular loops, helix 8 and N terminus (top). Cartoon representation of the β₂AR (PDB: 2RH1). (C) Missense variants were projected onto each structural segment. Variants that are predicted to have a functional impact map (green) significantly more often into the transmembrane region and loops (Wilcoxon rank sum test; EC-Loops: p < 8.0x10⁻⁷, IC-Loops: p < 1.6x10⁻³, TM: p < 2.2x10⁻¹⁶) variants of unknown functional impact (gray). (D) Missense variants were projected onto each structural segment and normalized by segment length. (Wilcoxon rank sum test; EC-Loops: p < 2.8x10⁻⁰⁷, IC-Loops: p < 2.3x10⁻⁷, TM: p < 2.2x10⁻¹⁶).

Figure 5

Effects of Natural Mutations on Drug Activity and G-Protein Coupling (A) Positions of selected missense variations of the μ-opioid receptor (MOR) near the ligand-binding pocket. (B) Schema of the BRET assay for real-time monitoring of G-protein activation. Activating μ-opioid receptors by agonist leads to the dissociation of inactive heterotrimeric G proteins into active GTP-bound Gα and Venus-Gβγ subunits. The free Venus-Gβγ then interacts with the Gβγ effector mimetic masGRK3ct-Nluc to increase the BRET signal. (C) Ligand/drug-induced maximum BRET amplitude (R_Max) and activation rate constants (k_ON) by wild-type μ-opioid receptor (mean ± SEM, n = 6 wells). (D) Real-time monitoring of ligand/drug actions on μ-opioid receptor mutants (mean response trace, n = 3 or 6). (E) Quantification of stimulus bias (R_MAX, left and k_ON, right) of μ-opioid receptor mutants. The values of agonist-induced responses were normalized to the reference, wild-type μ-opioid receptor (black line). The values of naloxone-induced responses were normalized to the K235N^3.36x36 mutant (thickness represents the SEM over n = 3). (F) Positions of selected missense variations of the Cholecystokinin receptor type A. (G) Schema of the BRET assay for real-time monitoring of G-protein activity for CCKAR experiments. (H) Agonist-induced maximum BRET amplitude (R_Max) and activation rate constants (k_ON) by wild-type CCKAR (mean ± SEM, n = 6 wells). (I) Real-time monitoring of G-protein activation by CCKAR mutants stimulated with caerulein (30 μM, applied at 5 s, n = 3) normalized to the maximum amplitude of the wild-type receptor. (J) Quantification of G-protein-coupling bias (R_MAX, left and k_ON, right) of CCKAR mutants normalized to wild-type CCKAR (black line, thickness represents the SEM over n = 3). See also Figure S5.

Figure S5

Effects of Mutations on R_Max and k_ON for μ-Opioid Receptor and CCKAR and Allele-Specific Expression of GPCR Drug Targets, Related to Figure 5 (A) Chemical structures of opioid receptor ligands. (B) The bar graphs quantitate the relative R_Max and k_ON of μ-opioid receptor mutants to μ-opioid receptor wild-type. (C) Chemical structure of Cholecystokinin receptor ligand caerulein. (D) The bar graphs quantitate the relative R_Max and k_ON of CCKAR mutants to CCKAR wild-type. For both panels, results are expressed as the mean ± SEM. One-way ANOVA with Dunnett post hoc multiple comparison test relative to “wild type receptor,” ^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗p < 0.001, n = 3 independent experiments. (E) Enrichment of GPCR drug targets among genes with mono-allelic expression. The Venn diagram shows the overlap between GPCR drug targets and the genes with allelic expression data. Enrichment was tested with permutation tests by performing 100,000 iterations. The random expectation (gray histogram) and the actual observation (green arrow) of GPCRs with mono-allelic expression are shown on the right.

Figure S6

Resource and Tools for the Analysis of Variation Data of GPCR Drug Targets within GPCRdb, Related to Figure 6 Datasets for natural genetic variations comprising of 60,706 individuals have been integrated into the GPCRdb. (A) Sortable variant table is provided for every receptor with more detailed information on the type and nature of each amino acid substitution, information about allele counts and frequency, predicted functional impact scores (SIFT and PolyPhen) and functional site annotation. (B) Genetic variation density (red intensity levels) on a GPCR classification tree (item ‘statistics’). (C) Data points can be visualized for every selected non-olfactory GPCR (missense variants in red shown for adrb2_human) on a snake-like diagram (top) or a helix plot (bottom) with additional information shown on mouse-over (allele count, allele frequency, amino acid change, number of homozygotes, predicted effects by SIFT and PolyPhen). (D) Missense variants can also be visualized on a consensus snake-like diagram for single families or ligand-types of GPCRs (gradient red for all Class A peptide angiotensin receptors). Each position then gives the number of observed mutations and the list of observed amino acid changes. (E) National Health Service spending data from 2011 to 2017 for 279 FDA-approved drugs. This is shown for Buprenorphine. The natural variation dataset can be downloaded per receptor (www.gpcrdb.org/mutational_landscape/) or accessed programmatically via an extensive API.

Figure 6

Drugs, GPCR Functional Site Variability, and Associated Economic Impact (A) Number of FDA-approved drugs (y axis, log-scale) against their missense mutation density within known functional sites (x axis) for GPCR drug targets. Color represents the total number of missense variants for each receptor within known functional sites as seen in the ExAC dataset. (B) Number of prescribed items by the National Health Service each month (y axis, log-scale) against the maximum number of missense variations in known and putative functional sites of its therapeutic GPCR target for each FDA-approved drug. UK sales volume is shown in million GBP per month. (C) Estimated economic burden on the National Health Service per year due to ineffective drug prescription (STAR Methods). Please see Table S7 for each drug. See also Figures S6 and S7.

Figure S7

Possible Economic Burden Due to Loss-of-Function and Copy-Number Variation Observed for GPCR Drug Targets, Related to Figure 6 (A–C) For each FDA-approved drug, a pot of the number of prescribed items by the National Health Service each month (y axis, logarithmic scale in thousands) against (B) the percentage of individuals with loss of function mutations and (C) maximum number of copy number variations in its therapeutic GPCR target is shown. Sales volume is shown in million GBP per month.

Figure 7

GPCR Variants Can Contribute to Altered Drug Response (A) Receptor response outcome upon binding to an endogenous ligand binding (E) and upon drug treatment (D). A receptor variant may induce phenotype-altering (e.g., disease) perturbations (endogenous ligand) and/or altered drug response (and combinations thereof). The position of the missense variation (top right) renders its effect. Some mutations may have no effect on the binding of the endogenous ligand or the drug and are entirely neutral. Mutations in/near the ligand-binding interface might affect endogenous ligand signaling, drug response, or both. Mutations in the effector-binding interface (G protein/arrestin) most likely affect both endogenous ligand signaling and drug response. (B) Differences in drug response due to different mutations between individuals in a population. The drug target variation spectrum may differently affect individual drug responses by potentially altering ligand potencies and efficacies, receptor conformation, surface expression, and/or biasing signaling. Personalized target sequencing could facilitate prognosis of a patient’s drug response. Additionally, pharmacological characterization of genetic variants that have been cataloged in humans could foster precision prescription. See also Figure S6.