From CYP2C19 Genotype to Simulated Exposure
A pharmacogenetic virtual cohort is useful only if the path from genotype to predicted concentration is explicit. For a CYP2C19 substrate, that path has several steps: define which alleles are represented, assign each diplotype a functional value, translate that value into an operational phenotype, and use it to alter a model parameter. Exposure differences then arise from the specified pharmacokinetic model, not from the phenotype label by itself.
This workshop uses a deliberately small, hypothetical cohort and a simplified clearance model to make each step visible. CYP2C19 star-allele function and clinical phenotype assignment can depend on the allele set, evidence, and guideline being applied. The functional values and phenotype boundaries below are therefore operational choices for this simulation, not a universal clinical translation table or a substitute for a validated substrate-specific model.
Specify the Alleles and Phenotype Rule
Suppose the virtual population includes four CYP2C19 alleles: *1, treated as normal function; *2 and *3, treated as no function; and *17, treated as increased function. Assign relative allele activity values of 1, 0, 0, and 1.5, respectively. These values are hypothetical model settings chosen for the example. Before applying them to a real compound or population, their suitability would need justification from relevant pharmacogenetic evidence and model calibration.
For alleles \(g_1\) and \(g_2\), define the diplotype activity index \(a\) as their average. The operational categories used here are poor metabolizer (PM) for \(a=0\), intermediate metabolizer (IM) for \(0<a<1\), normal metabolizer (NM) for \(a=1\), rapid metabolizer (RM) for \(1<a<1.5\), and ultrarapid metabolizer (UM) for \(a=1.5\). These boundaries make the example reproducible; they should not be interpreted as a universal clinical standard.
The average makes *1/*1 the reference index \(a=1\). For example, *1/*2 has \(a=(1+0)/2=0.5\), whereas *1/*17 has \(a=(1+1.5)/2=1.25\). This representation assumes that the two allele contributions can be summarized by their average. It does not capture every possible form of gene regulation, allele interaction, or uncertainty in functional evidence.
Worked Example: Assigning Phenotypes to Diplotypes
Assign the index and operational phenotype to *2/*3, *1/*3, *1/*17, and *17/*17.
For *2/*3, \(a=(0+0)/2=0\), so the assigned phenotype is PM.
For *1/*3, \(a=(1+0)/2=0.5\), so the assigned phenotype is IM.
For *1/*17, \(a=(1+1.5)/2=1.25\), so the assigned phenotype is RM.
For *17/*17, \(a=(1.5+1.5)/2=1.5\), so the assigned phenotype is UM. Each assignment follows the stated rule; it does not establish that the same label would be assigned by every clinical guideline.
Construct a Hypothetical Cohort
Assume allele frequencies \(p_{*1}=0.65\), \(p_{*2}=0.25\), \(p_{*3}=0.05\), and \(p_{*17}=0.05\). They sum to 1. For this example, generate each subject's two alleles independently from those frequencies. This is the random-pairing assumption underlying Hardy–Weinberg diplotype probabilities. It may be unsuitable if the target population has relevant substructure, nonrandom mating, or different allele frequencies. The frequencies are invented for teaching and do not describe an actual population.
For two different alleles with frequencies \(p_i\) and \(p_j\), the diplotype probability is \(2p_i p_j\), because either allele can be inherited from either parent. For a homozygous diplotype it is \(p_i^2\). Applying these rules gives the following expected proportions. A finite simulated cohort will have integer counts that fluctuate around these expectations.
| Diplotype group | Expected proportion | Index or indices | Operational phenotype |
|---|---|---|---|
| *2/*2, *2/*3, *3/*3 | 0.0900 | 0 | PM |
| *1/*2, *1/*3 | 0.3900 | 0.5 | IM |
| *2/*17, *3/*17 | 0.0300 | 0.75 | IM |
| *1/*1 | 0.4225 | 1 | NM |
| *1/*17 | 0.0650 | 1.25 | RM |
| *17/*17 | 0.0025 | 1.5 | UM |
Worked Example: Calculating Expected Phenotype Proportions
Calculate the expected proportions of PM, IM, and NM subjects from the hypothetical allele frequencies.
The PM group consists of *2/*2, *2/*3, and *3/*3. Its expected proportion is \(0.25^2+2(0.25)(0.05)+0.05^2=0.0625+0.0250+0.0025=0.0900\).
The IM group includes *1/*2, *1/*3, *2/*17, and *3/*17. Its proportion is \(2(0.65)(0.25)+2(0.65)(0.05)+2(0.25)(0.05)+2(0.05)(0.05)\) \(=0.3250+0.0650+0.0250+0.0050=0.4200\).
The NM group is *1/*1, with proportion \(0.65^2=0.4225\). The remaining increased-function groups have proportions \(2(0.65)(0.05)=0.0650\) for *1/*17 and \(0.05^2=0.0025\) for *17/*17. The categories sum to \(0.0900+0.4200+0.4225+0.0650+0.0025=1.0000\), as required.
For a cohort of 1,000 independently sampled subjects, these proportions imply expected counts of 90 PM, 420 IM, 422.5 NM, 65 RM, and 2.5 UM subjects. Expected counts need not be integers; a realized simulation has integer counts and will not generally match every expectation exactly.
Propagate the Index Into Clearance
To isolate the genotype-to-exposure calculation, suppose a hypothetical CYP2C19 substrate has a CYP2C19-independent clearance component of \(1.00\) L/h and a CYP2C19-mediated clearance component of \(4.00\) L/h at reference activity \(a=1\). Assume total systemic clearance is the sum of those components, with the CYP2C19 component multiplied by the individual's activity index:
This is a linear, additive approximation. It assumes that the substrate is in a range where clearance is not saturated, that the activity index scales the relevant CYP2C19-mediated component proportionally, and that other clearance pathways do not change with genotype. It also treats the stated clearances as systemic clearance terms; a full PBPK model may instead represent enzyme abundance and activity within the liver and calculate hepatic clearance from organ blood flow, binding, and intrinsic clearance.
For a single oral dose, assume bioavailability \(F=0.80\) is constant across genotypes. Under linear pharmacokinetics, the total area under the plasma concentration–time curve is \(AUC=F \times \mathrm{Dose}/CL\). This relation requires the same dose and bioavailability for the subjects being compared. If genotype changes intestinal or hepatic first-pass extraction, then constant \(F\) is not appropriate and the exposure calculation must represent that effect.
Worked Example: Calculating Exposure for Three Diplotypes
Use a hypothetical oral dose of 100 mg and \(F=0.80\). Calculate clearance and AUC for *2/*2, *1/*2, and *1/*17. Assume the linear clearance model above.
For *2/*2, \(a=0\). Thus \(CL=1.00+4.00(0)=1.00\) L/h and \(AUC=(0.80)(100\ \mathrm{mg})/(1.00\ \mathrm{L/h})=80.0\ \mathrm{mg\cdot h/L}\).
For *1/*2, \(a=0.5\). Thus \(CL=1.00+4.00(0.5)=3.00\) L/h and \(AUC=80.0/3.00=26.7\ \mathrm{mg\cdot h/L}\).
For *1/*17, \(a=1.25\). Thus \(CL=1.00+4.00(1.25)=6.00\) L/h and \(AUC=80.0/6.00=13.3\ \mathrm{mg\cdot h/L}\).
In this model and under the stated assumptions, the PM diplotype has the greatest AUC of these three, and the increased-function diplotype has the least. The calculation predicts exposure for this hypothetical dose and model; it does not by itself establish a clinical dose recommendation.
Why Non-CYP Clearance Matters
A useful check on the calculation is the exposure ratio relative to a reference subject with \(a=1\). If dose and bioavailability are unchanged, then \(AUC(a)/AUC(1)=CL(1)/CL(a)\). Substituting the clearance model gives \((1+4)/(1+4a)\). This ratio is not simply \(1/a\), because the non-CYP clearance component remains when CYP2C19 activity changes.
For the PM case, the ratio is \(5/1=5\): the modeled AUC is five times the *1/*1 reference AUC. If the entire clearance had incorrectly been assumed to be CYP2C19-mediated, the predicted ratio would be unbounded at \(a=0\). The non-CYP pathway prevents that result in this model. More generally, the size of a genotype effect depends both on the activity change and on the fraction of clearance attributed to the enzyme.
Summarize Exposure Without Hiding Diplotype Differences
A phenotype category need not correspond to one exact activity index. Here, for example, the IM group includes index values 0.5 and 0.75. Consequently, calculating exposure from a single representative IM value would discard differences between diplotypes. A cohort simulation should retain each subject's diplotype or activity index when calculating clearance and exposure, then summarize the resulting subject-level predictions.
Worked Example: Expected Mean AUC in the Hypothetical Cohort
Using the same 100 mg dose, \(F=0.80\), and clearance model, calculate the expected arithmetic mean AUC implied by the diplotype proportions. The expected mean is the sum of each diplotype group's AUC multiplied by its population proportion.
The PM proportion is 0.0900, with AUC 80.0 mg·h/L, contributing \(0.0900(80.0)=7.20\) mg·h/L to the mean. The *1/*2 and *1/*3 groups together have proportion 0.3900 and index 0.5; their AUC is 26.7 mg·h/L before rounding, contributing \(0.3900(80/3)=10.4\) mg·h/L.
The *2/*17 and *3/*17 groups have proportion 0.0300 and index 0.75. Their clearance is \(1+4(0.75)=4.00\) L/h and AUC is \(80/4=20.0\) mg·h/L, contributing \(0.0300(20.0)=0.600\) mg·h/L. The NM group contributes \(0.4225(80/5)=6.76\) mg·h/L.
The *1/*17 group contributes \(0.0650[80/(1+4(1.25))]=0.0650(13.3333)=0.8667\) mg·h/L. The *17/*17 group contributes \(0.0025[80/(1+4(1.5))]=0.0025(11.4286)=0.02857\) mg·h/L. Adding the contributions gives an expected mean AUC of approximately \(7.20+10.4+0.600+6.76+0.8667+0.02857=25.86\) mg·h/L.
This is a population-weighted mean under the hypothetical frequency and pharmacokinetic assumptions. It is not the AUC of a typical individual, nor does it describe the spread of exposure. A simulation should also inspect the distribution, such as its quantiles and phenotype-stratified summaries, because a mean alone can conceal a high-exposure tail.
Checks, Alternatives, and Limitations
A few checks establish whether the workshop calculation is internally coherent. Allele frequencies must be nonnegative and sum to 1; diplotype probabilities must also sum to 1. Every diplotype should map to exactly one activity index and one operational phenotype. At fixed dose and bioavailability, the specified clearance model should produce lower clearance and higher AUC as the index decreases. If a simulated output violates that direction, first check the implemented parameter mapping and units.
The central alternative to this activity-index approach is direct genotype-specific parameterization: each diplotype can be assigned a separately justified CYP2C19 parameter rather than deriving it from average allele values. That approach can represent known non-additive effects, but requires adequate evidence for each diplotype. Another option is to model enzyme activity as uncertain rather than fixed, which can represent uncertainty in functional effects; it also requires an explicit uncertainty model. A categorical phenotype-only model is simpler to summarize, but can lose within-category differences such as the two activity indices in the IM group here.
The example is deliberately not a complete PBPK model. It omits physiological variability, enzyme abundance variation within a genotype, competing pathways beyond a single lumped clearance term, hepatic extraction and first-pass effects, protein binding, and concentration-dependent metabolism. It also assumes the same dose and bioavailability across subjects. In practice, exposure predictions depend on the substrate, route, model structure, evidence supporting allele effects, and target population. A genotype-to-exposure result is therefore conditional on those assumptions, not a universal property of the allele.
A common mistake is to multiply total clearance by the activity index and then assume exposure is inversely proportional to that index. That reasoning ignores the non-CYP component: in this example, a=0 still gives \(CL=1.00\) L/h, not zero. The correction is to scale only the clearance component explicitly assigned to CYP2C19, add the other clearance component, and then calculate AUC from the resulting total clearance.
Check Your Understanding
Answer each question using the definitions and hypothetical model in this workshop.
- What activity index and operational phenotype are assigned to *2/*17?
- Under the stated allele frequencies and random-pairing assumption, what is the expected proportion of *1/*17 subjects?
- For a subject with \(a=0.75\), calculate clearance and AUC at 100 mg with \(F=0.80\).
- Why can the IM phenotype's exposure not be represented exactly by one value in this cohort?
- Name one assumption that would need reconsideration if CYP2C19 genotype changed first-pass metabolism.
Solutions: (1) \(a=(0+1.5)/2=0.75\), so the operational phenotype is IM. (2) The heterozygote proportion is \(2(0.65)(0.05)=0.0650\), or 6.50%. (3) \(CL=1.00+4.00(0.75)=4.00\) L/h and \(AUC=(0.80)(100)/4.00=20.0\) mg·h/L. (4) IM includes both \(a=0.5\) and \(a=0.75\), which produce different clearances and AUCs under the model. (5) The assumption of genotype-invariant bioavailability \(F\) would need reconsideration; first-pass effects would need to be represented rather than held constant.