Why Method Selection Matters
A predicted tissue-to-plasma partition coefficient is not a direct measurement of tissue exposure. It is an estimate conditional on a model’s representation of tissue composition, drug ionization, and binding. The same compound can therefore receive different predictions from different methods, even when each method is applied correctly. Choosing a method is not simply a matter of selecting the most familiar name: it requires matching the compound’s properties and the target tissue to the method’s assumptions and available inputs.
This tutorial gives a practical selection framework for the Poulin-Theil, Rodgers-Rowland, Schmitt, and Berezhkovskiy approaches. The earlier material on mechanistic differences among Kp methods and their high- and low-lipophilicity limits is useful context: no method should be treated as universally reliable across all compounds and tissues.
The methods considered here are related but not interchangeable. Poulin-Theil emphasizes partitioning into tissue water and lipid compartments. Rodgers-Rowland represents ionization and pH effects and includes interactions with acidic phospholipids, particularly relevant for basic drugs. Schmitt uses a broader tissue-composition framework, with multiple tissue components and compound properties represented in the prediction. Berezhkovskiy is especially useful as a binding correction when suitable unbound fractions are available; it should not be mistaken for an independent substitute for every tissue-partition mechanism.
A Practical Selection Framework
Start with the requested tissue and the quality of the input data, rather than with a preferred model. Record the compound’s lipophilicity and ionization properties, the tissue-specific information required by the candidate methods, and whether relevant unbound fractions are available. Then ask which mechanisms are likely to matter. A neutral compound with no important pH-dependent behavior does not need an ionization term merely because a more elaborate method can provide one. Conversely, a basic compound may be poorly represented if the selected approach omits ionization or acidic-phospholipid interactions.
| Method | Most informative use case | Inputs and assumptions to check | Important caution |
|---|---|---|---|
| Poulin-Theil | Neutral compounds for which water and lipid partitioning are the main mechanisms of interest | Compound lipophilicity and tissue lipid and water composition; check that ionization and specific binding are not dominant | A simple lipid-based representation may miss pH-dependent accumulation or important binding interactions |
| Rodgers-Rowland | Ionizable compounds, especially when pH differences and acidic-phospholipid interactions may affect partitioning | pKa, lipophilicity, relevant pH values, and tissue composition, including acidic phospholipids where needed | Predictions depend on the quality and relevance of pH and interaction inputs; a computed equilibrium ratio does not establish the rate of tissue entry |
| Schmitt | Cases requiring a broader tissue-composition model and explicit representation of multiple tissue components | Compound properties and suitable tissue-specific composition and binding information, as required by the implementation | Additional input detail does not guarantee accuracy; missing or poorly matched inputs can undermine the apparent mechanistic detail |
| Berezhkovskiy correction | Cases where a baseline unbound partition estimate can be adjusted using relevant plasma and tissue unbound fractions | A defensible unbound partition ratio, \(f_{u,p}\), and tissue-specific \(f_{u,t}\) | Uncertain or mismatched unbound fractions can dominate the corrected result |
This is a selection guide, not a ranking from best to worst. If two methods are both plausible, calculate both and investigate the reasons for any substantial difference. Agreement can support robustness to the choice between those models, but it does not establish that their shared assumptions are true. Disagreement is useful diagnostic information: it may point to a sensitive input, a mechanism represented differently, or a limitation in the data.
Binding Correction: A Short Derivation
The Berezhkovskiy correction can be understood by separating total concentration from unbound concentration. Let \(f_{u,p}\) and \(f_{u,t}\) denote the unbound fractions in plasma and tissue, respectively, and let \(K_{pu,t}=C_{u,t}/C_{u,p}\) be the unbound tissue-to-plasma ratio. Since \(C_{u,p}=f_{u,p}C_{p,\mathrm{total}}\) and \(C_{u,t}=f_{u,t}C_{t,\mathrm{total}}\), rearranging gives:
This relation is an identity given the definitions; the modeling assumption enters when \(K_{pu,t}\) is specified or approximated. At equilibrium, for a compound whose unbound concentration equilibrates between plasma and tissue, \(K_{pu,t}=1\), so \(K_{p,t}=f_{u,p}/f_{u,t}\). If an underlying model instead supplies a non-unit unbound ratio, retain that ratio in the correction. Do not silently replace it with one.
The assumptions require attention. The fractions must refer to the relevant species, concentration range, and biological matrices; the tissue fraction must correspond to the target tissue. The equilibrium relationship is not a claim that total concentrations are equal. Nor does it account automatically for active transport, pH gradients, intracellular sequestration, or other departures from the assumed unbound equilibration. The earlier material on unbound fractions as Kp inputs provides further context for interpreting those quantities.
Worked Method Choices
Worked Example: A Neutral Compound in Liver
Hypothetical case. A neutral compound has measured logP \(=2.1\). The intended output is an equilibrium liver \(K_p\). The available tissue-composition inputs support both Poulin-Theil and Schmitt calculations. An approved implementation gives hypothetical predictions of 3.80 and 4.10, respectively. There is no evidence in the supplied data that ionization or a measured unbound correction is needed.
First check whether the neutral-compound assumption is reasonable: the compound is specified as neutral, so a pH-dependent ionization mechanism is not the primary basis for selecting the method. Both candidate methods are within scope, but the water-and-lipid partitioning representation of Poulin-Theil directly matches the stated mechanism and requires fewer additional mechanistic inputs in this case.
For a consistency check, suppose the implementation’s Poulin-Theil result corresponds to a hypothetical liver concentration of \(19.0\ \mathrm{\mu mol/L}\) and plasma concentration of \(5.00\ \mathrm{\mu mol/L}\) at equilibrium. Then
The units cancel, as required for a partition coefficient. Select Poulin-Theil as the primary estimate because its assumptions fit the stated neutral, lipid-partitioning case and its output is consistent with the specified concentration ratio. Retain the Schmitt result as a useful comparison, not as evidence that the true value lies between 3.80 and 4.10. These hypothetical method outputs are illustrative; they are not a claim that the numbers follow from logP alone.
Worked Example: An Ionizable Base in Tissue
Hypothetical case. A basic compound has pKa \(=8.4\). Consider a simplified comparison of the fraction present as neutral base at plasma pH 7.4 and tissue pH 7.0. For a base \(B\), the neutral fraction is \(1/(1+10^{\mathrm{p}K_a-\mathrm{pH}})\). This calculation is a diagnostic for whether ionization may matter; it is not a complete Kp prediction.
At pH 7.4, the neutral fraction is
At pH 7.0, it is
The neutral fraction differs across the two pH values, so an approach that can represent ionization and tissue pH is more defensible than relying only on neutral-lipid partitioning. Rodgers-Rowland is a reasonable candidate because it includes ionization and acidic-phospholipid interactions; Schmitt is also worth comparing if its required tissue-specific inputs are available. The choice is not based on the neutral fractions alone: verify the pH values, pKa, and implementation inputs, and assess whether the tissue is adequately represented. For a brain application, this equilibrium estimate must not be interpreted as evidence that a compound crosses the blood-brain barrier rapidly or freely.
Worked Example: Applying a Binding Correction
Hypothetical case. For a specified tissue, suppose the unbound partition ratio is taken to be \(K_{pu,t}=1.00\) under an equilibrium-unbound assumption. The hypothetical plasma unbound fraction is \(f_{u,p}=0.0800\), and the tissue unbound fraction is \(f_{u,t}=0.00400\). Both fractions are assumed to apply to the same compound and relevant conditions.
Apply the correction:
The corrected total tissue-to-plasma ratio is therefore \(20.0\), dimensionless. The result is driven by the twenty-fold ratio of the unbound fractions, not by a twenty-fold difference in unbound tissue and plasma concentrations; those are equal under the assumed \(K_{pu,t}=1\). If \(K_{pu,t}\) were instead \(0.500\), the same fractions would give \(K_{p,t}=0.500\times20.0=10.0\). Thus the correction must preserve the stated unbound ratio, and its result is only as reliable as the tissue-specific fractions and equilibrium assumption.
Diagnostics, Alternatives, and Limitations
Before adopting a prediction, check whether it behaves plausibly and whether the calculation actually used the intended inputs. Confirm that the target tissue composition corresponds to the species and tissue definition of interest; that lipophilicity and pKa values are appropriate for the modeled compound; and, for pH-dependent methods, that the pH values are relevant to the compartments represented. When a Berezhkovskiy correction is used, verify that the unbound fractions are tissue-specific and compatible with the underlying partition estimate. The earlier material on sensitivity diagnostics and measured-predicted Kp discrepancies is relevant when these checks reveal an unstable or unexpected result.
Warning signs include a prediction that changes sharply after a small, plausible input change; a large disagreement between methods that cannot be traced to a mechanism or input; or a high predicted total \(K_p\) being interpreted as rapid entry into a tissue. A very high or very low value is not automatically an error, particularly at the lipophilicity extremes, but it calls for a check of model scope and assumptions. Also verify that the comparison concerns equilibrium data: a concentration ratio sampled before distribution has equilibrated may not be comparable to an equilibrium prediction.
The main alternatives serve different purposes. Poulin-Theil is a relatively direct choice when neutral lipid and water partitioning are the relevant features. Rodgers-Rowland offers a more explicit treatment of ionization and acidic-phospholipid interactions. Schmitt offers a broader composition-based representation when the necessary inputs are available and justified. A Berezhkovskiy correction is most informative when defensible unbound fractions can adjust an unbound partition estimate. Using a more complex method is not automatically preferable: extra parameters can introduce uncertainty, and a simpler method may be more transparent when its assumptions fit.
No method eliminates uncertainty in tissue composition, compound properties, or the biological meaning of the target ratio. Predicted values should be documented as model-dependent estimates, with the method, input sources, tissue and species, equilibrium basis, and important assumptions recorded. When multiple methods are plausible, report the comparison and explain the primary choice rather than averaging predictions without a defined rationale.
Common Pitfall
Check Your Understanding
Answer each question, then compare your reasoning with the solutions below.
- A neutral compound is modeled for a tissue where water and lipid partitioning are considered the main mechanisms. Which method is a reasonable starting choice, and what should still be checked?
- Why might Rodgers-Rowland be preferable to a lipid-only approach for a basic compound?
- If \(K_{pu,t}=1.00\), \(f_{u,p}=0.0500\), and \(f_{u,t}=0.0100\), calculate \(K_{p,t}\) using the binding correction.
- Two suitable methods predict different tissue \(K_p\) values. Why is their arithmetic mean not automatically the best estimate?
Solutions. (1) Poulin-Theil is a reasonable starting choice because the compound is neutral and the stated mechanism is water and lipid partitioning; check tissue composition, compound lipophilicity, and whether other mechanisms are important. (2) A basic compound may show pH-dependent ionization and interactions with acidic phospholipids, which Rodgers-Rowland can represent. (3) \(K_{p,t}=1.00\times(0.0500/0.0100)=5.00\), dimensionless. (4) The methods may encode different mechanisms and assumptions, so averaging can conceal meaningful disagreement and has no automatic statistical justification; investigate the inputs and report model uncertainty.