Why Mechanistic Kp Methods Give Different Predictions

Why the Same Drug Can Have Different Predicted Kp Values

A tissue-to-plasma partition coefficient, \(K_p\), summarizes the ratio of drug concentrations in tissue and plasma at equilibrium, on a specified concentration basis. Mechanistic prediction methods estimate that ratio by combining compound properties with tissue and plasma composition. They can therefore give different values for the same drug without any arithmetic error: they may represent the phases differently, treat ionization differently, or account for binding in different ways.

The goal of comparing methods is not to select the largest or most elaborate prediction automatically. It is to identify which assumptions drive the difference, check that the inputs and concentration basis are comparable, and decide whether those assumptions suit the compound and tissue. The earlier material on Kp definitions and reference phases is essential here: two ratios are not directly comparable if one uses total tissue concentration and another uses a different reference or concentration basis.

What the Four Approaches Represent

All four approaches connect composition to partitioning, but they do not encode the same model. The Poulin-Theil approach is principally a neutral-compound method: it uses tissue and plasma phase contributions to estimate how a neutral drug distributes. Applying it to an ionizable drug as though only the neutral form mattered is a useful baseline in some comparisons, but it does not represent the full ionization-dependent distribution.

Rodgers-Rowland (RR) methods extend phase-based prediction to ionizable compounds. They use the compound’s ionization properties and the pH of relevant aqueous environments, alongside partitioning into tissue components such as phospholipids. Consequently, they can represent differences in neutral and ionized species and pH-dependent accumulation. Their result depends on the phase terms, ionization treatment, and input assumptions used in the particular implementation.

The Schmitt method also combines compound properties with tissue composition, but its formulation accounts for a broader set of tissue and plasma interactions, including aqueous, lipid, and binding-related contributions. This additional structure does not guarantee a more accurate prediction: it makes the result sensitive to the definitions and values assigned to those contributions.

Berezhkovskiy is most usefully understood here as a correction framework linking total partitioning to unbound partitioning and unbound fractions. It is not simply another independent tissue-phase ledger. The identity behind the correction clarifies how binding can change total \(K_p\) even when the ratio of unbound concentrations is held fixed.

The table is a comparison of modelling emphasis, not a claim that every software implementation uses identical equations or inputs. Before interpreting a numerical difference, identify the actual implementation and the definitions it uses.

Common Bookkeeping Behind the Predictions

A useful way to organize a phase-based prediction is to think in terms of a drug’s effective capacity in tissue and in plasma. If the contributions from the represented phases are \(C_{t,j}\) for tissue and \(C_{p,j}\) for plasma, a schematic expression is

This is bookkeeping, not a single formula that defines all four methods. The phase terms, their calculation, and which terms are present depend on the method. For example, an ionizable-drug method may distinguish the contribution of the neutral and ionized forms in aqueous phases, while a neutral-only calculation does not. A binding-related term may affect the tissue sum, the plasma sum, or both.

The schematic also shows why changing only the tissue numerator does not tell the whole story. If one method predicts a larger tissue capacity but also represents the plasma reference differently, the ratio can increase, decrease, or stay similar. A comparison that reports numerator differences but ignores the denominator can therefore misattribute the source of a Kp discrepancy.

Ionization and pH Can Separate Predictions

For a basic compound, the neutral fraction in a monoprotic aqueous environment follows from the Henderson-Hasselbalch relationship. With \(pK_a\) defined for the conjugate acid, the ratio of ionized to neutral concentration is \(10^{pK_a-pH}\), so the total-to-neutral concentration factor is \(1+10^{pK_a-pH}\). If only the neutral species partitions freely across a reference boundary, changing pH changes the total concentration associated with a given neutral concentration.

This factor illustrates why tissue and plasma pH can matter for an ionizable drug. It is not, by itself, a complete Kp prediction: real methods may also include partitioning or binding of ionized species and multiple tissue phases. In a method comparison, therefore, a pH-related difference should be traced through the method’s full phase and species accounting rather than added as an isolated multiplier without justification.

Why the Berezhkovskiy Correction Changes Total Kp

Let \(C_{u,t}\) and \(C_{u,p}\) be unbound concentrations in tissue and plasma, and let \(C_{t,\mathrm{tot}}\) and \(C_{p,\mathrm{tot}}\) be the corresponding total concentrations on a consistent basis. Define \(f_{u,t}=C_{u,t}/C_{t,\mathrm{tot}}\) and \(f_{u,p}=C_{u,p}/C_{p,\mathrm{tot}}\). The unbound partition coefficient is \(K_{p,uu}=C_{u,t}/C_{u,p}\). Rearranging these definitions gives the correction identity.

The identity assumes that the unbound fractions and concentrations refer to compatible conditions, tissue and plasma bases, and equilibrium definitions. It shows why equal unbound concentrations do not imply equal total concentrations: tissue binding can make \(f_{u,t}\) small, increasing total tissue concentration relative to plasma.

Tracing a Discrepancy in Practice

A disciplined comparison proceeds from shared definitions to model-specific assumptions. First confirm that the outputs refer to the same tissue, species, plasma reference, total or unbound basis, and equilibrium concept. Then compare input values: ionization constants, lipophilicity-related properties, pH, phase fractions, and binding inputs. Only after these checks should the remaining difference be attributed to method structure.

Useful diagnostics include checking that phase fractions and capacities are nonnegative, that ratios use consistent units, and that a neutral-compound comparison does not accidentally invoke pH-dependent terms. For an ionizable drug, inspect whether the supplied pH values are tissue- and plasma-specific and whether the ionization convention matches the compound. For a binding correction, verify that \(f_{u,p}\) and \(f_{u,t}\) are on compatible bases and that binding has not already been counted in the underlying prediction.

Failure can be subtle. A result can be numerically plausible while combining a tissue total concentration with an unbound plasma reference. A large difference can also arise from a mistaken pKa convention or an inappropriate pH input rather than from a meaningful mechanistic disagreement. A useful diagnostic is to recompute a neutral-drug case or temporarily remove one disputed contribution, then confirm that the predicted change has the expected direction.

Choosing What a Comparison Can Establish

These methods are alternatives in the sense that they encode different ways to estimate tissue partitioning from composition and compound properties. They are not interchangeable calculations that should necessarily converge to one number. Poulin-Theil provides a useful neutral-compound phase-based estimate; RR explicitly addresses ionization and pH-dependent contributions; Schmitt uses a broader composition and interaction structure; and the Berezhkovskiy relationship makes the role of unbound fractions explicit.

A comparison can establish which assumptions account for a numerical difference under the chosen inputs. It cannot, by itself, establish which prediction is closest to reality. That requires appropriate measured data and attention to whether the measurements represent equilibrium and the same quantity. Nor does adding more terms automatically improve prediction: extra terms can add important mechanisms, but they can also introduce uncertain inputs or double-count an effect.

The central limitation is that each prediction remains conditional on its representation of the drug, phases, and binding. Treat a method’s output as a model-based estimate, not a direct observation. When methods disagree, preserve the disagreement as information about structural assumptions, rather than averaging the values without a reason.