Recovering Microconstants From a Biexponential Curve

From a Biexponential Curve Back to Rate Constants

A two-compartment concentration curve is often summarized by two exponential rates and their amplitudes. These macroparameters describe the observed curve, but they are not themselves the rates of elimination and exchange between compartments. Recovering those microconstants requires using both the curve’s shape and its initial condition.

There is an important qualification: the two macro rates and the initial concentration do not by themselves determine all three microconstants. The initial concentration is the sum of the two amplitudes, so recovering the rates also requires the amplitude split, or equivalent information such as the initial slope. This distinction matters when a curve has been reported only as a pair of exponents and a concentration at time zero.

Definition: For an IV bolus into the central compartment, let \(k_{10}\) be the first-order elimination rate constant from the central compartment, \(k_{12}\) the rate constant from central to peripheral, and \(k_{21}\) the rate constant from peripheral to central. A biexponential concentration curve is written \(C(t)=A e^{-\alpha t}+B e^{-\beta t}\), where \(A\) and \(B\) are concentration amplitudes, and \(\alpha>\beta>0\) are the macro rates. Under the stated model, \(C(0)=A+B\).

Assume a linear, time-invariant two-compartment model with positive first-order rate constants, a bolus initially in the central compartment, no drug initially in the peripheral compartment, and no further input. Concentration is measured in the central compartment. The amplitudes and \(C(0)\) have concentration units; \(\alpha,\beta,k_{10},k_{12},k_{21}\) have units of inverse time.

Deriving the Inverse Conversion

The amount-balance equations for this model give the characteristic polynomial whose roots are the two macro rates. Equivalently, the earlier mapping from microconstants to macroparameters gives their sum and product:

\alpha+\beta=k_{10}+k_{12}+k_{21}, \qquad \alpha\beta=k_{10}k_{21}.

These equations provide two constraints on three microconstants. The initial slope supplies the third. At the instant after the bolus, the peripheral amount is zero, so the only loss from the central compartment is elimination. Thus \(C'(0)=-k_{10}C(0)\). Differentiating the fitted biexponential curve also gives \(C'(0)=-\alpha A-\beta B\). Equating the expressions yields the recovery formulas.

\begin{aligned} C(0)&=A+B,\\ k_{10}&=\frac{\alpha A+\beta B}{A+B},\\ k_{21}&=\frac{\alpha\beta}{k_{10}},\\ k_{12}&=\alpha+\beta-k_{10}-k_{21}. \end{aligned}

The first recovered rate is a weighted average of the macro rates, with weights given by the amplitude fractions. The second follows from the product relation, and the last from the sum relation. In particular, knowing \(C(0)\) without knowing how it is divided between \(A\) and \(B\) does not give \(k_{10}\).

Admissibility check: For positive amplitudes and \(\alpha>\beta>0\), \(k_{10}\) lies strictly between \(\beta\) and \(\alpha\). The recovered outward rate can be written \(k_{12}=\frac{(\alpha-k_{10})(k_{10}-\beta)}{k_{10}}\), so it is positive under those conditions. A valid conversion should also give \(k_{21}>0\). Nonpositive recovered rates indicate that the inputs do not meet the assumptions of this positive-rate model, or that estimation and rounding have produced an incompatible combination.

Worked Conversions and Checks

Worked Example: Recovering Three Positive Rate Constants

Suppose a hypothetical fitted curve is \(C(t)=8e^{-1.20t}+2e^{-0.120t}\) mg/L, with time measured in hours. Then \(A=8\) mg/L, \(B=2\) mg/L, \(\alpha=1.20\) h\(^{-1}\), and \(\beta=0.120\) h\(^{-1}\). The initial concentration is \(C(0)=8+2=10\) mg/L.

First calculate central elimination using the initial slope:

k_{10}= \frac{(1.20\ {\rm h}^{-1})(8\ {\rm mg/L})+ (0.120\ {\rm h}^{-1})(2\ {\rm mg/L})} {10\ {\rm mg/L}} =\frac{9.84\ {\rm mg/(L\cdot h)}}{10\ {\rm mg/L}} =0.984\ {\rm h}^{-1}.

Next, use the macro-rate product:

k_{21}=\frac{(1.20)(0.120)\ {\rm h}^{-2}}{0.984\ {\rm h}^{-1}} =0.1463\ {\rm h}^{-1}.

Finally, use the macro-rate sum:

k_{12}=1.20+0.120-0.984-0.1463 =0.1897\ {\rm h}^{-1}.

All three rates are positive, and \(0.120<0.984<1.20\) h\(^{-1}\). The amplitude-weighted initial slope and the sum and product checks are consistent with a positive-rate two-compartment model under the assumptions stated above.

Worked Example: Same Initial Concentration, Different Microconstants

Consider two hypothetical curves with the same macro rates, \(\alpha=1.00\) h\(^{-1}\) and \(\beta=0.200\) h\(^{-1}\), and the same initial concentration, \(C(0)=10\) mg/L. First let \(A=B=5\) mg/L. Then

k_{10}=\frac{(1.00)(5)+(0.200)(5)}{10} =0.600\ {\rm h}^{-1},\qquad k_{21}=\frac{(1.00)(0.200)}{0.600} =0.3333\ {\rm h}^{-1},

k_{12}=1.00+0.200-0.600-0.3333 =0.2667\ {\rm h}^{-1}.

Now let \(A=8\) mg/L and \(B=2\) mg/L, which still sum to 10 mg/L:

k_{10}=\frac{(1.00)(8)+(0.200)(2)}{10} =0.840\ {\rm h}^{-1},\qquad k_{21}=\frac{0.200}{0.840} =0.2381\ {\rm h}^{-1},

k_{12}=1.20-0.840-0.2381 =0.1219\ {\rm h}^{-1}.

Both sets of rates are positive and reproduce the same macro rates and initial concentration. They differ because the amplitude split, which determines the initial slope, differs. Therefore \(\alpha,\beta\), and \(C(0)\) alone are insufficient to identify the three microconstants.

Worked Example: Detecting an Inadmissible Amplitude Pattern

Suppose a hypothetical fitted curve has \(A=12\) mg/L, \(B=-2\) mg/L, \(\alpha=1.00\) h\(^{-1}\), and \(\beta=0.200\) h\(^{-1}\). Although \(C(0)=10\) mg/L, the amplitude-weighted initial slope implies

k_{10}=\frac{(1.00)(12)+(0.200)(-2)}{10} =\frac{11.6}{10}=1.16\ {\rm h}^{-1}.

Then

k_{21}=\frac{0.200}{1.16}=0.1724\ {\rm h}^{-1}, \qquad k_{12}=1.20-1.16-0.1724=-0.1324\ {\rm h}^{-1}.

A negative \(k_{12}\) is not an admissible outward transfer rate in this model. The fitted components may describe a curve over the observed time range, but this parameter combination cannot represent the assumed positive-rate central bolus model. The appropriate conclusion is not to replace the negative rate with zero without further analysis; instead, check the fit, assumptions, and numerical precision.

Why the Admissibility Check Works

The weighted-average formula gives a direct justification. If \(A>0\), \(B>0\), and \(A+B>0\), then

k_{10} =\frac{A}{A+B}\alpha+\frac{B}{A+B}\beta, \qquad \frac{A}{A+B}+\frac{B}{A+B}=1.

Both weights are positive, so \(\beta<k_{10}<\alpha\). Substituting \(k_{21}=\alpha\beta/k_{10}\) into the sum formula gives

\begin{aligned} k_{12} &=\alpha+\beta-k_{10}-\frac{\alpha\beta}{k_{10}}\\ &=\frac{\alpha k_{10}+\beta k_{10}-k_{10}^{2}-\alpha\beta}{k_{10}}\\ &=\frac{(\alpha-k_{10})(k_{10}-\beta)}{k_{10}}. \end{aligned}

Since \(k_{10}>\beta>0\) and \(\alpha-k_{10}>0\), the numerator and denominator are positive; hence \(k_{12}>0\). Also \(k_{21}=\alpha\beta/k_{10}>0\). This establishes why a positive amplitude split produces positive recovered rates when the macro rates are ordered and positive. It does not establish that every curve fit has positive amplitudes or that the model is appropriate for every data set.

Diagnostics, Alternatives, and Limitations

A practical check is to calculate each rate before rounding, then verify the defining relationships:

\(k_{10}\) should equal the amplitude-weighted macro-rate average.
\(k_{10}+k_{12}+k_{21}\) should equal \(\alpha+\beta\).
\(k_{10}k_{21}\) should equal \(\alpha\beta\).
All three rate constants should be positive under the assumed model.

A failure can reflect a sign or unit error, rounded or uncertain parameter estimates, a poor biexponential fit, or a mismatch between the curve and the assumed IV bolus initial condition. Very small amplitudes deserve particular caution: although the algebra is defined, uncertainty in a weakly identified component can make the recovered rates highly sensitive to small changes in the fitted values. A recovered negative rate is a warning about compatibility, not proof of which assumption failed.

One approach is to fit the biexponential curve and transform its amplitudes and rates using these equations. This makes the curve’s macroparameters explicit, but uncertainty in the fitted quantities must be carried through the transformation. Another approach is to fit the microconstants directly within the compartment model, often constraining them to be positive. That can enforce physical admissibility during fitting, but it does not guarantee that the model is scientifically appropriate or that the data identify the rates precisely. Neither approach removes the need to inspect the curve, the initial-condition assumptions, and parameter uncertainty.

The conversion is exact for the specified linear two-compartment IV bolus model. It should not be applied unchanged when there is drug initially in the peripheral compartment, a continuing input, nonlinear kinetics, or a different observation model. In those settings, the initial slope or observed concentration curve need not obey the relationships used in the derivation.

Common pitfall: Treating \(C(0)=A+B\) as enough information to recover \(k_{10}\). It is only the sum of the amplitudes. The slope depends on their separate contributions, \(-\alpha A-\beta B\); without the amplitude split or an equivalent initial-slope measurement, the inverse conversion is underdetermined.

Check Your Understanding

Use the formulas and assumptions in this tutorial. Each solution is included so you can check both the arithmetic and the model interpretation.

A curve has \(A=6\) mg/L, \(B=4\) mg/L, \(\alpha=0.80\) h\(^{-1}\), and \(\beta=0.10\) h\(^{-1}\). Calculate \(k_{10}\), \(k_{21}\), and \(k_{12}\), and check positivity.
Why can two curves with the same \(\alpha,\beta\), and \(C(0)\) imply different microconstants?
For positive amplitudes and \(\alpha>\beta>0\), explain why \(k_{10}\) lies between the macro rates.
A conversion gives \(k_{12}<0\). What conclusion is warranted, and what conclusion is not warranted?

Solutions. (1) \(C(0)=10\) mg/L and \(k_{10}=[(0.80)(6)+(0.10)(4)]/10=0.52\) h\(^{-1}\). Then \(k_{21}=(0.80)(0.10)/0.52=0.1538\) h\(^{-1}\), and \(k_{12}=0.90-0.52-0.1538=0.2262\) h\(^{-1}\). All are positive, with \(0.10<0.52<0.80\) h\(^{-1}\). (2) The initial concentration fixes only \(A+B\); different amplitude splits give different initial slopes and therefore different \(k_{10}\). (3) \(k_{10}\) is a weighted average of \(\alpha\) and \(\beta\), with positive weights \(A/(A+B)\) and \(B/(A+B)\) that sum to one. (4) A negative \(k_{12}\) is incompatible with the assumed positive-rate model for those inputs. It does not, by itself, identify whether the cause is fitting error, violated assumptions, or numerical imprecision.