Choosing Among Full and Reduced TMDD Models

Why Model Choice Depends on the Question

Target-mediated drug disposition (TMDD) can make drug exposure nonlinear because drug binds to a finite target pool, and bound drug may be internalized. A detailed model can represent these processes separately, but that detail is useful only when the processes matter over the observed time course and the data can inform them. A reduced model is easier to estimate and interpret, but its assumptions may fail during rapid concentration changes, when target abundance changes, or when the data include concentration ranges that reveal binding transients.

The choice is therefore not simply “full is best” or “reduced is simpler.” It depends on the binding and disposition timescales, target turnover, dose range, and what was measured. Here, \(L\) denotes free drug concentration, \(R\) free target concentration, and \(C\) drug–target complex concentration. The rate constants \(k_{\mathrm{on}}\), \(k_{\mathrm{off}}\), and \(k_{\mathrm{int}}\) describe association, dissociation, and complex internalization, respectively. The equations below use a standard one-compartment disposition structure and a target pool with synthesis and loss; additional processes may be needed for a particular biological system.

What Each Formulation Assumes

A representative full model is

Here \(k_{\mathrm{el}}\) is the first-order elimination rate of free drug, and \(k_{\mathrm{syn}}\) and \(k_{\mathrm{deg}}\) represent target synthesis and loss. This version assumes internalization removes complex, drug, and target; different biological mechanisms require corresponding changes. The full model is appropriate when binding and complex dynamics may be resolved by the observations, or when a reduction would discard dynamics relevant to the question.

The QSS relation follows by setting \(dC/dt\) approximately to zero while retaining the association and loss processes:

If total target \(R_{\mathrm{tot}}=R+C\) is used to express the same relation, then \(C\approx R_{\mathrm{tot}}L/(K_{\mathrm{SS}}+L)\). This algebraic expression does not mean the system is at binding equilibrium: internalization contributes to \(K_{\mathrm{SS}}\), and target turnover may make \(R_{\mathrm{tot}}\) change over time. QSS is plausible when the complex relaxes much faster than free drug, target abundance, and the input change on the time interval of interest.

Rapid equilibrium instead uses \(K_D=k_{\mathrm{off}}/k_{\mathrm{on}}\) and \(C\approx R_{\mathrm{tot}}L/(K_D+L)\). It requires binding to equilibrate rapidly and internalization to be negligible relative to the binding processes that establish equilibrium. Thus, RE is not interchangeable with QSS: if \(k_{\mathrm{int}}\) is appreciable, \(K_{\mathrm{SS}}\) and \(K_D\) differ.

An MM disposition form often writes target-mediated loss as \(V_{\max}L/(K_m+L)\), where, under a QSS-based reduction with an approximately constant target pool, \(V_{\max}=k_{\mathrm{int}}R_{\mathrm{tot}}\) and \(K_m=K_{\mathrm{SS}}\). This further simplification is most defensible when binding transients do not need to be represented and drug concentration is a reasonable proxy for the substrate concentration. In particular, if measured drug is total concentration, treating it as free concentration requires bound drug to be a small fraction of total drug. If the target pool changes materially, a constant \(V_{\max}\) may also be inadequate.

Use Timescales to Screen Reductions

A useful first step is to estimate how quickly complex can relax, while being explicit about what is held fixed. If both free drug \(L\) and free target \(R\) are held fixed, the complex equation is a linear differential equation with a constant source:

Its stationary value is \(C^*=k_{\mathrm{on}}LR/(k_{\mathrm{off}}+k_{\mathrm{int}})=LR/K_{\mathrm{SS}}\). Subtracting \(C^*\) from both sides gives \(d(C-C^*)/dt=-(k_{\mathrm{off}}+k_{\mathrm{int}})(C-C^*)\), so the deviation decays exponentially with time constant \(\tau_C=1/(k_{\mathrm{off}}+k_{\mathrm{int}})\). There is no \(k_{\mathrm{on}}L\) term in this fixed-\(L\), fixed-\(R\) calculation.

A different rate applies if \(L\) and total target \(R_{\mathrm{tot}}\) are fixed while free target changes as complex forms. Substituting \(R=R_{\mathrm{tot}}-C\) gives

Under these particular assumptions, the relaxation time is \(1/(k_{\mathrm{on}}L+k_{\mathrm{off}}+k_{\mathrm{int}})\). This distinction matters: a rate derived while free target is fixed cannot be substituted into a calculation that assumes total target is fixed. In a real dose profile, neither \(L\) nor the target pool is necessarily fixed, so these times are screening calculations, not proof that QSS holds.

Compare a candidate complex-relaxation time with the timescales of drug concentration change, dosing or absorption, and target turnover. A useful reduction needs a clear separation: complex relaxation should be substantially faster than the other dynamics being approximated. A modest numerical difference is not a universal pass criterion. It is a reason to inspect predictions over the actual observation window and dose range. Target turnover is especially relevant because an algebraic complex relation may adjust quickly while the amount of available target changes on a timescale that affects exposure.

Distinguishing QSS From Rapid Equilibrium

The comparison between \(K_D\) and \(K_{\mathrm{SS}}\) gives a direct parameter-based check. When \(k_{\mathrm{int}}\) is small relative to \(k_{\mathrm{off}}\), the two constants are close. When it is not, an RE model can misrepresent the occupancy implied by QSS even if both models assume rapid adjustment. The available observations matter too: concentration or occupancy data collected only after transients may not distinguish the formulations, whereas early measurements or direct complex observations may reveal differences.

When an MM Disposition Form Is Adequate

An MM approximation is most useful when the scientific question concerns the overall saturable loss of drug rather than separate estimates of binding and complex dynamics. It generally needs a broad enough concentration range to show the transition from approximately linear loss at \(L\ll K_m\) to near-saturated loss at \(L\gg K_m\). If observations cover only one side of that transition, \(V_{\max}\) and \(K_m\) may not both be well informed. The approximation also becomes questionable when the target pool changes substantially or when total drug differs materially from free drug.

Match the Model to the Observations

Model selection should reflect what the data can resolve, not only what biology could in principle do. Direct measurements of free drug, bound complex, or target can support more mechanistic distinctions than sparse total-drug measurements alone. Dense early sampling can help reveal rapid binding transients; late samples may be more informative about slower disposition and target recovery. If only total drug concentrations are available, different combinations of binding, internalization, and target abundance may produce similar profiles. A more elaborate model is not automatically more informative when its key states are unobserved.

A practical comparison is to fit or simulate plausible candidates over the intended dose range and observation schedule, then examine whether their predictions differ in ways the data could detect. Use residual and predictive diagnostics to look for systematic time- or concentration-dependent discrepancies. A reduced model may show biased early predictions, a dose-dependent pattern in residuals, or consistent errors around the concentration range where saturation changes. Conversely, if the more complex model’s extra processes are unsupported by the observations, its estimates may be unstable or highly correlated. These patterns do not prove which mechanism is true; they identify where the assumptions and available information need closer review.

Common Pitfalls and Limitations

• Using one relaxation rate under different constraints. With fixed free \(L\) and \(R\), the rate is \(k_{\mathrm{off}}+k_{\mathrm{int}}\). With fixed \(L\) and \(R_{\mathrm{tot}}\), it is \(k_{\mathrm{on}}L+k_{\mathrm{off}}+k_{\mathrm{int}}\). State which variables are fixed before calculating a timescale.
• Calling QSS equilibrium. QSS includes internalization in \(K_{\mathrm{SS}}\); RE uses \(K_D\). They coincide only when the contribution of internalization is negligible for the question at hand.
• Assuming a good fit validates the mechanism. Similar exposure predictions can arise from different parameter combinations. A fit alone may not establish that individual binding or disposition processes are identified.
• Applying constant MM parameters when target abundance changes. If the available target pool changes, the effective maximum target-mediated loss may also change, so a constant \(V_{\max}\) can miss time-dependent behavior.
• Using total concentration as if it were free concentration. This can be consequential when bound drug is not small relative to total drug, especially near or below the target concentration scale.

Each reduction has limitations. Timescale comparisons can change across concentrations because association depends on \(L\). A QSS approximation that is adequate late in a profile may fail during a dose transition. An MM model can summarize saturable loss without preserving the mechanistic detail needed to interpret target engagement. The appropriate choice is the simplest formulation that answers the scientific question, remains consistent with the observed dose and time ranges, and does not leave systematic predictive failures.