Why Implementation Details Matter
A target-mediated drug disposition model can be conceptually sound yet produce incorrect predictions because a complex is counted incorrectly, a rate constant is expressed in incompatible units, or a model is initialized with a total concentration where it expects a free concentration. These mistakes can change apparent exposure, target occupancy, and clearance without producing an obvious software error. The remedy is to audit the model as a system of reactions and balances before interpreting its output.
The checks here apply to a simple drug–target binding system and to its quasi-steady-state (QSS) reduction. They do not replace checking whether the biological model is appropriate. They help establish that the implemented equations, units, and initial conditions faithfully represent the stated model.
Begin With Reaction Stoichiometry
For a monovalent drug \(A\) binding one target \(R\) to form complex \(C\), the binding reactions are \(A+R \rightleftharpoons C\). Let \(k_{\mathrm{on}}\) be the association rate constant, \(k_{\mathrm{off}}\) the dissociation rate constant, and \(k_{\mathrm{int}}\) the rate constant for internalization of complex. In a single well-mixed compartment, the binding and internalization contributions are:
Here \(k_{\mathrm{el}}\) is elimination of free drug, and target synthesis and degradation are represented by \(k_{\mathrm{syn}}\) and \(k_{\mathrm{deg}}\). All concentrations must refer to compatible volumes. Binding removes one molecule of free drug and one molecule of free target for every complex formed; dissociation returns one of each. Internalization removes one complex. These coefficients follow from that stoichiometry.
With synthesis, degradation, elimination, or internalization switched off, binding and dissociation alone must conserve \(A+C\) and \(R+C\). This gives a direct way to detect sign or coefficient errors. For example, a term \(-2k_{\mathrm{on}}AR\) in \(dA/dt\) would incorrectly remove two drug molecules per one-to-one complex formed.
Worked Example: Auditing a Multistep Complex Network
Suppose a bivalent antibody can bind one or two identical target molecules, forming \(C_1\) and \(C_2\), respectively. Write the reaction rates and check whether the proposed derivatives conserve antibody and target during binding and dissociation. Let the forward and reverse rates be \(v_{1f}=k_1AR\), \(v_{1r}=k_{-1}C_1\), \(v_{2f}=k_2C_1R\), and \(v_{2r}=k_{-2}C_2\).
The reaction-rate contributions are:
For the illustrative values \(A=10\), \(R=4\), \(C_1=2\), and \(C_2=1\) nM, and \(k_1=0.1\), \(k_{-1}=0.2\), \(k_2=0.05\), and \(k_{-2}=0.1\) in compatible nM and hour units, the rates are \(v_{1f}=4\), \(v_{1r}=0.4\), \(v_{2f}=0.4\), and \(v_{2r}=0.1\) nM/h. Thus \(dA/dt=-3.6\), \(dR/dt=-3.9\), \(dC_1/dt=3.3\), and \(dC_2/dt=0.3\) nM/h.
One antibody is present in \(A\), \(C_1\), and \(C_2\); two target molecules are present in each \(C_2\). Therefore, antibody total is \(A+C_1+C_2\), whose rate of change is \(-3.6+3.3+0.3=0\) nM/h. Target total is \(R+C_1+2C_2\), whose rate of change is \(-3.9+3.3+2(0.3)=0\) nM/h. Both totals are conserved by binding and dissociation. The example is a stoichiometric check, not a claim that every bivalent antibody should be represented with only these states.
Units and Concentration Bases
A reaction rate must have the units of concentration per time. If \(A\) and \(R\) are in nM, then \(k_{\mathrm{on}}AR\) has units nM/h only when \(k_{\mathrm{on}}\) is in nM\(^{-1}\)h\(^{-1}\). A first-order rate constant such as \(k_{\mathrm{off}}\) must be in h\(^{-1}\) if time is measured in hours. Using the numerical value of a rate constant in a different unit system without conversion can create errors by many orders of magnitude.
Worked Example: Converting an Association Rate Constant
Suppose a reported association rate constant is \(2.00\times10^5\) M\(^{-1}\)s\(^{-1}\). Convert it to nM\(^{-1}\)h\(^{-1}\), then calculate the association rate at \(A=6\) nM and \(R=2\) nM.
Since \(1\) nM is \(10^{-9}\) M and one hour is \(3600\) seconds, the conversion is:
The association rate is \(k_{\mathrm{on}}AR=(0.720)(6)(2)=8.64\) nM/h. Inserting \(2.00\times10^5\) directly into an equation using nM and hours would not give a valid rate: its units would not match the units of the concentration terms. Check all rate constants, time units, concentration units, compartment volumes, and dose units together.
A related error is mixing amounts and concentrations. If the model state is amount, the reaction equations must use amount-based rates or explicitly convert amounts to concentrations using compartment volume. For example, an amount of 3 nmol in 0.5 L corresponds to \(3/0.5=6\) nmol/L, or 6 nM. A reaction term using an amount as though it were a concentration changes with the implied compartment volume and is generally incorrect.
Initial Conditions: Free and Total Are Not Interchangeable
A free concentration counts unbound molecules. A total concentration includes the specified bound forms. For a one-to-one system, \(A_{\mathrm{tot}}=A+C\) and \(R_{\mathrm{tot}}=R+C\). A dose or assay value must be translated into the state variables the model actually uses; setting free drug equal to total drug while also assigning a nonzero initial complex double-counts drug. Conversely, setting free drug equal to a reported total while setting complex to zero may misrepresent a system that is already at binding equilibrium.
Under the QSS approximation for the simple binding system, setting \(dC/dt\) approximately to zero gives \(C=AR/K_{\mathrm{SS}}\), where \(K_{\mathrm{SS}}=(k_{\mathrm{off}}+k_{\mathrm{int}})/k_{\mathrm{on}}\). In terms of total target, this becomes \(C=R_{\mathrm{tot}}A/(K_{\mathrm{SS}}+A)\), provided \(R_{\mathrm{tot}}=R+C\) is the target total used in this one-to-one model.
Worked Example: Initializing From a Total Drug Concentration
Suppose the initial free drug concentration is 3.72 nM, total target is 4.00 nM, and \(K_{\mathrm{SS}}=0.28\) nM. Under the stated QSS relationship, calculate complex, free target, and total drug.
The complex concentration is \(C=R_{\mathrm{tot}}A/(K_{\mathrm{SS}}+A)=(4.00)(3.72)/(0.28+3.72)=14.88/4.00=3.72\) nM. Free target is \(R=R_{\mathrm{tot}}-C=4.00-3.72=0.28\) nM. Total drug is \(A+C=3.72+3.72=7.44\) nM. Thus entering 7.44 nM as the free drug state would be wrong if the intended initial condition was 3.72 nM free drug.
The reverse conversion is also important. If the specified total drug is 3.72 nM, the free drug is not 3.72 nM. Using \(A_{\mathrm{tot}}=A+R_{\mathrm{tot}}A/(K_{\mathrm{SS}}+A)\), the positive solution is:
Then \(C=A_{\mathrm{tot}}-A=3.72-0.7783=2.9417\) nM. These values satisfy the stated total and QSS relation. This conversion assumes the QSS algebraic relationship applies at initialization; a full kinetic model instead needs initial free and complex states specified or determined from an explicitly justified initial equilibrium calculation.
When the QSS Approximation Is a Mistake
QSS replaces the dynamic complex equation with an algebraic condition, \(dC/dt\approx0\). It is an approximation, not a declaration that binding is instantaneous or that the system is necessarily at thermodynamic equilibrium. For the simple model, \(K_{\mathrm{SS}}\) includes complex internalization as well as dissociation. Substituting an equilibrium dissociation constant for \(K_{\mathrm{SS}}\) is generally unjustified when internalization is appreciable.
The approximation is most plausible when complex formation and relaxation are fast relative to the changes in free drug and target that drive the system. A useful first check is the complex relaxation time at fixed free concentrations, \(1/(k_{\mathrm{off}}+k_{\mathrm{int}})\). This is not a complete timescale analysis of a coupled model: binding also couples the state variables, so the relevant relaxation rates can depend on concentrations and the full system. But if even this basic complex timescale is not short relative to the timescale of drug or target change, assuming QSS requires strong additional justification.
Worked Example: Checking a QSS Timescale
Consider a model with \(k_{\mathrm{off}}=0.03\) h\(^{-1}\) and \(k_{\mathrm{int}}=0.02\) h\(^{-1}\), while free drug changes substantially on a timescale of about 4 h. The fixed-free-concentration complex relaxation time is \(1/(0.03+0.02)=20\) h. Since 20 h is not short compared with 4 h, the complex may not track the changing free concentrations closely. A QSS reduction is not supported by this comparison alone; the full dynamic complex model is the safer starting point unless further analysis or data justify the approximation.
Other warning signs include large early-time changes after a dose, predicted abrupt changes in complex concentration, or poor agreement between full and reduced models in the concentration range of interest. Compare predicted trajectories, not just a single steady-state value. Also check whether the timescale separation holds across dose levels and target abundance: binding may relax quickly in one regime and slowly in another.
A Practical Audit and Its Limits
Use this sequence when reviewing an implementation. A failed check identifies a problem to investigate; passing these checks does not establish that the model captures all relevant biology.
For each reaction, state how many molecules of each species are consumed or produced. Make sure each complex has a clear molecular composition.
Confirm that each reaction-rate term has units of state per time, that amounts and concentrations are not mixed, and that all volumes and time units are consistent.
With loss and production terms removed, verify conservation of each molecular component. Record whether each input is free or total and initialize states accordingly.
Compare complex relaxation with drug and target change, and compare full and reduced predictions where feasible. Do not use an algebraic QSS relation merely because it is available.
A full TMDD model retains dynamic complex formation and can represent transient binding, but it has more states and may require informative data to estimate its parameters. QSS can reduce model complexity when its assumptions are reasonable, but it can miss transient behavior if complex adjustment is not fast enough. A quasi-equilibrium approximation is a distinct assumption and should not be substituted for QSS without checking the kinetic conditions. Neither a full nor a reduced model repairs wrong stoichiometry, incompatible units, or inconsistent initialization.
Common diagnostics include negative concentrations, totals that drift when only binding reactions should act, unexpectedly large or small rates, and strong dependence on the chosen numerical step or solver tolerances. Such behavior can also arise from numerical or model-structure issues, so it is a prompt to inspect equations, parameter units, and solver settings rather than proof of one specific error. Finally, reductions and balance checks rely on their stated species definitions and compartment assumptions; transport between compartments, target synthesis, and internalization must be represented explicitly when included in the biological question.
Check Your Understanding
Answer each question, then compare with the solutions below.
- For one-to-one binding, which totals are conserved by binding and dissociation alone?
- A model uses concentrations in nM and time in hours. What units must \(k_{\mathrm{on}}\) and \(k_{\mathrm{off}}\) have?
- In a one-to-one model, \(A=2.0\) nM and \(C=0.5\) nM. What is total drug?
- If \(k_{\mathrm{off}}+k_{\mathrm{int}}=0.10\) h\(^{-1}\), what is the fixed-free-concentration complex relaxation time? What would you conclude if free drug changes on a 2 h timescale?
Solutions. (1) \(A+C\) and \(R+C\), because each binding event removes one free molecule of each and adds one complex containing one of each. (2) \(k_{\mathrm{on}}\) must be nM\(^{-1}\)h\(^{-1}\), and \(k_{\mathrm{off}}\) must be h\(^{-1}\), so each derivative has concentration-per-hour units. (3) Total drug is \(A+C=2.0+0.5=2.5\) nM. (4) The relaxation time is \(1/0.10=10\) h. Since 10 h is not short relative to 2 h, this comparison does not support QSS; inspect the coupled timescales or retain dynamic complex states.