When Pooling Tissues Preserves Predictions

Why Pooling Can Preserve Some Predictions but Not Others

A whole-body PBPK model may represent several tissues with one pooled compartment to reduce the number of states. Whether that simplification is appropriate depends on the prediction target. If the target is the tissues’ total amount or their volume-weighted mean concentration, a pooled state can sometimes reproduce the result exactly. It need not reproduce the concentration in each constituent organ. The key is not merely that the tissues receive blood from the same circulation: their concentration responses to that shared blood input must also satisfy a matching condition.

Consider perfusion-limited tissues indexed by \(i\). Let \(V_i\) be tissue volume, \(Q_i\) its blood flow, \(C_i(t)\) its tissue concentration, \(C_a(t)\) the shared arterial concentration, and \(K_i\) the tissue-to-blood partition coefficient. In this model, the venous concentration leaving tissue \(i\) is \(C_i/K_i\), and the exchange equation is

\frac{dC_i}{dt} = \frac{Q_i}{V_i}\left(C_a-\frac{C_i}{K_i}\right).

This equation assumes constant \(V_i\), \(Q_i\), and \(K_i\), a well-mixed tissue compartment, and the same arterial input \(C_a(t)\) for all the tissues being pooled. It does not include tissue-specific metabolism or other fluxes. Those processes would need to be included explicitly and could change whether aggregation is exact.

Definition: For tissues with total volume \(V=\sum_i V_i\), their volume-weighted mean concentration is \(\bar C=\sum_i V_iC_i/V\). The corresponding total tissue amount is \(A=\sum_i V_iC_i=V\bar C\). Pooling preserves the aggregate prediction when the pooled model reproduces this mean concentration, and therefore this amount, over time.

The Matching Condition and Its Consequence

Rewrite each tissue equation to distinguish its input coefficient from its response rate:

\frac{dC_i}{dt} = \frac{Q_i}{V_i}C_a-\lambda_i C_i, \qquad \lambda_i=\frac{Q_i}{V_iK_i}.

The matching condition for exact aggregation is that the tissues have the same \(\lambda_i\), denoted by \(\lambda\). The partition coefficients do not have to be equal. Because \(Q_i/K_i=\lambda V_i\), summing the equations after weighting each by its volume gives

\frac{d\bar C}{dt} = \frac{\sum_i Q_i}{V}C_a-\lambda\bar C.

This is a single-compartment equation. Write \(Q=\sum_i Q_i\), and define the pooled partition coefficient \(K_{\mathrm{pool}}=Q/(V\lambda)\). Then the pooled equation takes the usual perfusion-limited form:

\frac{d\bar C}{dt} = \frac{Q}{V}\left(C_a-\frac{\bar C}{K_{\mathrm{pool}}}\right), \qquad K_{\mathrm{pool}} = \frac{Q}{V\lambda} = \frac{\sum_i V_iK_i}{V}.

The last equality follows from the matching condition: \(Q_i=\lambda V_iK_i\), so \(Q=\lambda\sum_i V_iK_i\). Thus the effective partition coefficient is the volume-weighted mean of the individual partition coefficients. This pooled equation holds for any shared time-varying arterial input and any initial tissue concentrations; the pooled initial concentration must be set to their volume-weighted mean.

The same condition preserves the aggregate exchange with blood. If net uptake from blood into tissue \(i\) is \(J_i=Q_i(C_a-C_i/K_i)\), then

\sum_i J_i = Q C_a-\lambda\sum_iV_iC_i = Q\left(C_a-\frac{\bar C}{K_{\mathrm{pool}}}\right).

Consequently, if a plasma or central-compartment equation uses these tissues only through their total exchange flux, replacing them with the exact pooled compartment preserves that flux. With the same initial central state and all other model components unchanged, the central prediction is preserved as well. This conclusion depends on the coupling actually being through the summed flux; it does not imply that any arbitrary plasma-linked process is unaffected by pooling.

Worked Example: Equal Partitioning and Equal Response Rates

Suppose two tissues have volumes \(V_1=1\ \mathrm{L}\) and \(V_2=3\ \mathrm{L}\), partition coefficients \(K_1=K_2=2\), and flows \(Q_1=0.20\ \mathrm{L/h}\) and \(Q_2=0.60\ \mathrm{L/h}\). The shared arterial concentration is held at \(C_a=10\ \mathrm{mg/L}\). Initially, \(C_1(0)=4\ \mathrm{mg/L}\) and \(C_2(0)=8\ \mathrm{mg/L}\).

Both have \(Q_i/V_i=0.20\ \mathrm{h}^{-1}\), so \(\lambda_i=Q_i/(V_iK_i)=0.10\ \mathrm{h}^{-1}\). They meet the matching condition. The initial pooled concentration is

\bar C(0) = \frac{(1)(4)+(3)(8)}{1+3} = 7\ \mathrm{mg/L}.

For a constant arterial concentration, solving the tissue equation gives \(C_i(t)=K_iC_a+[C_i(0)-K_iC_a]e^{-\lambda_i t}\). At \(t=2\ \mathrm{h}\), the exponential factor is \(e^{-(0.10)(2)}=e^{-0.20}\approx0.8187\), not \(e^{-0.40}\). Therefore,

C_1(2)=20+(4-20)e^{-0.20}\approx 6.900\ \mathrm{mg/L}, \qquad C_2(2)=20+(8-20)e^{-0.20}\approx 10.18\ \mathrm{mg/L}.

Their volume-weighted mean is \(\{(1)[20-16e^{-0.20}]+(3)[20-12e^{-0.20}]\}/4\approx9.357\ \mathrm{mg/L}\). The pooled model has \(V=4\ \mathrm{L}\), \(Q=0.80\ \mathrm{L/h}\), and \(K_{\mathrm{pool}}=2\), so its rate is \(Q/(VK_{\mathrm{pool}})=0.10\ \mathrm{h}^{-1}\). Starting from \(7\ \mathrm{mg/L}\), it predicts \(20+(7-20)e^{-0.20}\approx9.357\ \mathrm{mg/L}\), agreeing with the weighted mean. The organ concentrations are not equal, even though the aggregate prediction is exact.

Different Partition Coefficients Can Still Aggregate

A common misconception is that pooling requires identical partition coefficients. The derivation shows that identical response rates \(\lambda_i\), not identical \(K_i\), are the relevant condition. With different \(K_i\), the tissue concentrations may approach different equilibrium values under the same arterial concentration, while their weighted mean still follows one exact pooled equation.

Worked Example: Different Partition Coefficients, Matched Response Rates

Take \(V_1=2\ \mathrm{L}\), \(V_2=1\ \mathrm{L}\), \(K_1=1\), and \(K_2=3\). Set both response rates to \(\lambda=0.10\ \mathrm{h}^{-1}\), which requires \(Q_i=\lambda V_iK_i\): \(Q_1=0.20\ \mathrm{L/h}\) and \(Q_2=0.30\ \mathrm{L/h}\). Let \(C_a=8\ \mathrm{mg/L}\), \(C_1(0)=0\ \mathrm{mg/L}\), and \(C_2(0)=6\ \mathrm{mg/L}\).

The pooled initial concentration is \(\bar C(0)=\{(2)(0)+(1)(6)\}/3=2\ \mathrm{mg/L}\). The effective partition coefficient is the volume-weighted value \(K_{\mathrm{pool}}=\{(2)(1)+(1)(3)\}/3=5/3\). At \(t=2\ \mathrm{h}\), \(e^{-0.20}\approx0.8187\). The organ concentrations are

C_1(2)=8+(0-8)e^{-0.20}\approx1.450\ \mathrm{mg/L}, \qquad C_2(2)=24+(6-24)e^{-0.20}\approx9.263\ \mathrm{mg/L}.

Thus \(\bar C(2)=\{(2)(1.450)+(1)(9.263)\}/3\approx4.054\ \mathrm{mg/L}\). The pooled calculation gives the same result:

\bar C(2) = \frac{5}{3}(8) + \left(2-\frac{5}{3}(8)\right)e^{-0.20} \approx4.054\ \mathrm{mg/L}.

The aggregate prediction is preserved, but a single pooled concentration cannot report both \(1.450\ \mathrm{mg/L}\) and \(9.263\ \mathrm{mg/L}\). If either organ concentration is itself the prediction target, the pooled state is insufficient.

Why Mismatched Response Rates Break Exact Pooling

If the \(\lambda_i\) differ, the weighted sum of the tissue equations contains the term \(\sum_i V_i\lambda_iC_i\). In general, this cannot be calculated from \(\bar C\) alone: knowing the weighted mean does not reveal how concentration is distributed among tissues with different rates. The aggregate trajectory can then depend on organ-specific states that the pooled model has discarded.

Worked Example: The Same Pooled State, Different Aggregate Changes

Consider two equal-volume tissues, each with \(K_i=2\), and flows such that \(Q_1/V_1=0.20\ \mathrm{h}^{-1}\) and \(Q_2/V_2=0.80\ \mathrm{h}^{-1}\). Their response rates are \(\lambda_1=0.10\ \mathrm{h}^{-1}\) and \(\lambda_2=0.40\ \mathrm{h}^{-1}\), so the matching condition fails. At an instant when \(C_a=10\ \mathrm{mg/L}\), compare two possible pairs of tissue concentrations.

In allocation A, let \((C_1,C_2)=(2,10)\ \mathrm{mg/L}\); in allocation B, let \((C_1,C_2)=(10,2)\ \mathrm{mg/L}\). Both have the same pooled value, \(\bar C=6\ \mathrm{mg/L}\). But their rates of change differ. For allocation A,

\frac{d\bar C}{dt} = \frac{1}{2}\left[0.20\left(10-\frac{2}{2}\right) +0.80\left(10-\frac{10}{2}\right)\right] = 2.90\ \mathrm{mg/(L\,h)}.

For allocation B,

\frac{d\bar C}{dt} = \frac{1}{2}\left[0.20\left(10-\frac{10}{2}\right) +0.80\left(10-\frac{2}{2}\right)\right] = 4.10\ \mathrm{mg/(L\,h)}.

The same pooled concentration and arterial input therefore produce different aggregate rates of change. No single-valued one-state equation based only on \(\bar C\) and \(C_a\) can reproduce both cases. A pooled model may still be a useful approximation for a particular target, but it is not an exact representation under these conditions.

Checking Whether Pooling Is Appropriate

Before pooling, calculate or inspect each tissue’s response rate \(Q_i/(V_iK_i)\), verify that the tissues share the same arterial input, and specify whether the target is an aggregate or organ-specific prediction. Exact aggregation also relies on the stated model structure: constant parameters, the perfusion-limited exchange equation, and no unrepresented tissue-specific fluxes. A shared blood input alone is not enough.

Useful diagnostics include comparing tissue concentration-time profiles and checking whether the aggregate rate of change is determined by the weighted mean and current arterial concentration. If two allocations with the same weighted mean imply different aggregate derivatives, as in the example, a one-state pool cannot be exact. Disagreement between pooled and organ-resolved simulations can also reveal a failure of the matching assumptions, although agreement in one scenario does not prove exactness for other inputs or initial conditions.

The main alternative is to retain separate tissue states. This preserves organ-specific predictions and allows different response rates, at the cost of more states and parameters. A single pooled state is preferable when its exactness conditions hold and the target is aggregate; it is a deliberate approximation otherwise. An exact aggregate concentration is not evidence that every organ concentration is accurate. Nor does preserving an aggregate under the stated exchange model guarantee preservation when the model includes nonlinear, tissue-specific metabolism or other processes that do not aggregate in the same way.

Common pitfall: “The tissues have the same arterial input, so their average concentration follows one pooled compartment.” A shared input does not make the aggregate equation closed. The correction is to check the response rates \(\lambda_i=Q_i/(V_iK_i)\); if they are not matched, the aggregate change can depend on how concentration is allocated across tissues.

Key takeaway: Matched response rates and a shared arterial input make the volume-weighted tissue concentration follow an exact pooled equation, with \(K_{\mathrm{pool}}=\sum_iV_iK_i/\sum_iV_i\). This can preserve aggregate tissue amount and summed blood exchange, but generally cannot preserve organ-specific concentrations.

Check Your Understanding

Answer each question, then compare with the solutions.

What response rate must be shared for the exact pooled equation derived here?
Under the matching condition, what is the pooled partition coefficient?
Why can a pooled concentration fail to determine the aggregate rate of change when response rates differ?
If exact pooling preserves total tissue amount, does it necessarily preserve each organ concentration? Explain.

Solutions. (1) Each tissue must have the same \(\lambda_i=Q_i/(V_iK_i)\). (2) \(K_{\mathrm{pool}}=\sum_iV_iK_i/\sum_iV_i\), the volume-weighted mean partition coefficient. (3) The derivative contains \(\sum_iV_i\lambda_iC_i\); with unequal rates, that term is not determined by \(\sum_iV_iC_i\) alone. (4) No. Pooling retains only the weighted mean, not the individual concentrations; different organ concentrations can have the same aggregate value.