s11538 019 00651 8

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 24

Bulletin of Mathematical Biology

https://doi.org/10.1007/s11538-019-00651-8

SPECIAL ISSUE: MATHEMATICS TO SUPPORT DRUG


DISCOVERY AND DEVELOPMENT

Analytical Solution and Exposure Analysis of a


Pharmacokinetic Model with Simultaneous Elimination
Pathways and Endogenous Production: The Case of
Multiple Dosing Administration

Xiaotian Wu1,2 · Fahima Nekka2,3 · Jun Li2,3

Received: 18 July 2018 / Accepted: 15 July 2019


© Society for Mathematical Biology 2019

Abstract
In this paper, a typical pharmacokinetic (PK) model is studied for the case of multiple
intravenous bolus-dose administration. This model, of one-compartment structure,
not only exhibits simultaneous first-order and Michaelis–Menten elimination, but also
involves a constant endogenous production. For the PK characterization of the model,
we have established the closed-form solution of concentrations over time, the existence
and local stability of the steady state. Using analytical approaches and the concept
of corrected concentration, we have shown that the area under the curve (AUCcorr ss,τ )
at steady state is higher compared to that at the single dose (AUCcorr 0−∞ ). Moreover,
by splitting the dose and dosing interval into halves, we have revealed that it can
result in a significant decrease in the steady-state average concentration. These model-
based findings, which contrast with the current knowledge for linear PK, confirm the
necessity to revisit drugs exhibiting nonlinear PK and to suggest a rational way of
using mathematical analysis for the dosing regimen design.

Keywords Pharmacokinetic model · Simultaneous first-order and Michaelis–Menten


elimination · Endogenous production · Area under the curve at steady state · Average
concentration at steady state

B Jun Li
jun.li.2@umontreal.ca
Fahima Nekka
fahima.nekka@umontreal.ca

1 Department of Mathematics, Shanghai Maritime University, Shanghai 201306,


People’s Republic of China
2 Faculté de pharmacie, Université de Montréal, Montreal, QC H3C 3J7, Canada
3 Centre de Recherches Mathématiques, Université de Montréal, Montreal, QC H3C 3J7, Canada

123
X. Wu et al.

Mathematics Subject Classification 92C45 · 92C50 · 34N05 · 37N25

1 Introduction

Pharmacokinetics (PK) is a scientific branch of clinical pharmacology that informs


about the temporal disposition of a drug within a living organism. Based on the
measured plasma concentrations of a drug, mathematical models are widely used to
explain the behavior of these observations, predict the PK process and help in making
scientific-based decisions of potential drug candidates. In drug research and develop-
ment, compartment modeling is the most used approach as it provides a convenient
mathematical framework in which the drug kinetics can be approximately described
with usual ordinary differential equations or algebraic functions. This approach consid-
ers the body as a system of compartments, without necessarily an explicit physiological
or anatomical meaning, and generally assumes that the rates of transfer between com-
partments and the rates of drug elimination from compartments follow linear kinetics
(Gibaldi and Perrier 2007; Boroujerdi 2015). This empirical compartment modeling
approach has been well accepted in the drug research community, designed as a stan-
dard procedure and implemented in many commercial software packages. At the same
time, many PK surrogates derived from these linear compartment models are proposed
to predict PK behavior and used for the evaluation of therapeutic effect.
In the last two decades, biologics have gained a widespread attention from drug
research companies and regulatory agencies (Keizer et al. 2010; Leader et al. 2008; Shi
2014). Biological products, which may be directly produced from living organisms
or obtained through biotechnological methods, have a long history in medical use.
Typical ones are blood and blood components, vaccines, modified human hormones,
monoclonal antibodies (mAbs), etc. (Shi 2014). However, the disposition of these
drugs could differ markedly compared to what is already known for traditional chem-
ical drugs. As their names suggest, many of these biologics have their counterparts in
the human body and are endogenously produced to fulfill specific biochemical func-
tions. Another peculiarity of biologics pertains to their elimination, which, for many
of them, does not only follow the general linear elimination pathway, but also involves
saturate target-mediated mechanisms. Examples cited here can be mAbs (Dirks and
Meibohm 2010), growth hormones (Klitgaard et al. 2009), blood components as gran-
ulocyte colony-stimulating factor (G-CSF) (Craig et al. 2015, 2016), erythropoietin
(Frymoyer et al. 2017) and thrombopoietin (Jin and Krzyzanski 2004). Compared
to small molecules, biologics have different physicochemical properties and mani-
fest complex PK characteristics with different mechanisms underlying the process of
absorption, distribution, metabolism and excretion (ADME) (Shi 2014). Therefore,
PK knowledge based on linear compartment modeling should be revisited to assess its
validity in the drug area of biologics, so that it can be adapted to the current research
reality of new biologics.
Mathematical models of different levels of complexity coexist in the PK literature
of biologics, going from simple linear compartment models to compartment models
with nonlinear components, target-mediated drug disposition models (TMDD) and the
more complete physiologically based PK models (PBPK) (Keizer et al. 2010; Zhao

123
Analytical Solution and Exposure Analysis of a…

et al. 2012). The development of these models is to meet various needs of pharmaceu-
tical research in different contexts. Specific PK properties of biologics are discussed
(Kloft et al. 2004; Kuester et al. 2008) or included in model structures (Craig et al.
2015; Quartino et al. 2014). However, a deep mathematical analysis is needed since
most of these studies only consider these properties in their models’ formulation or
use them to perform simulations. It is thus worth mathematically investigating the
PK models for biologics, to either confirm the validity of former methodologies, or
propose new ways for their analysis.
Studies for analytical solutions of PK models have proved to be a great advan-
tage and remained interest for a better understanding of drug qualitative PK behavior
(van der Graaf et al. 2016). For instance, Yu and Cao (2017) recently have studied PK
parameters through the analytical relationship between drug amount (or concentration)
and time using the principle of calculus; Schnell and Mendoza (1997), Tang and Xiao
(2007) have been expressed the closed-form solutions of one-compartment nonlinear
PK models with Michaelis–Menten elimination with multiple modes of administra-
tion using Lambert W function (Corless et al. 1996). Motivated by these works and
biologics such as aforementioned G-CSF, we have introduced a new transcendent
function to express the analytical solution of a model for drugs exhibiting simulta-
neous first-order and Michaelis–Menten elimination (Wu et al. 2015). Moreover, the
study for such drugs has allowed us to update the pharmacological knowledge of their
elimination half-lives (t1/2 ) and steady-state volumes of distribution (Vdss ) (Wu et al.
2015, 2016).
By taking into account a constant endogenous production, we have studied one-
compartment PK models having simultaneous first-order and Michaelis–Menten
elimination for the case of a single intravenous (IV) bolus administration (Wu et al.
2018). Using our introduced transcendent function, we were able to provide the closed-
form solution of C(t) and explicit expressions of the total and partial areas under the
concentration–time curve (AUC). We also have elucidated the role of endogenous
production in PK with nonlinear elimination and its impact on the estimation of AUC.
The case of multiple-dose administrations of such models is also an important issue,
particularly in the context of therapeutic drug monitoring (Boroujerdi 2015). In fact,
for linear compartment models, the principle of superposition is valid and implies the
equality between the AUC at steady state and total AUC for a single dose (Gibaldi
and Perrier 2007). In terms of drug exposure, this indicates that the resulting average
concentration will not be affected by changes in dosing regimen given the same total
daily dose. However, with the involvement of nonlinear elimination and endogenous
production, how the average concentration at steady state is affected by drug regimens
is not clear in the literature.
In this paper, using one-compartment PK model with simultaneous first-order and
Michaelis–Menten elimination and a constant endogenous production (Wu et al. 2018),
we investigate its PK properties in the case of multiple dosing administration. In
Sect. 2, we provide the closed-form solution of the concentration–time course, the
existence, uniqueness and local stability of the solution at steady state and discuss
the pharmacological meanings of the newly introduced parameters. In Sect. 3, the
explicit expression of the steady-state AUCcorr ss,τ and its relationship to the single-
dose AUCcorr0−∞ are given. As well, the impact of the administered dose amount and

123
X. Wu et al.

endogenous production are discussed. More importantly, we show how the average
concentration at steady state is influenced by different dosing regimens in this section.
Perspectives of the current work are discussed in Sect. 4.

2 Analytical Solutions of Concentration–Time Course for Multiple


Intravenous Bolus Administrations

2.1 Description of the Pharmacokinetic Model

In addition to the usual exogenous administration of biologics drug substances (IV


bolus doses here), the endogenous production of these substances can be an impor-
tant source of input to the system. For the sake of simplicity, a constant endogenous
production rate rprod (unit of concentration/time) is considered here as suggested in
Craig et al. (2015), Foley and Mackey (2009) and Quartino et al. (2014).
Two simultaneous elimination pathways are used to represent the drug elimination:
one is of linear first-order and the other has nonlinear Michaelis–Menten saturate
kinetics. For the former, kel is used to denote the first-order elimination rate constant,
while Vmax (unit of mass/time) and K m (unit of drug concentration) are required
for the latter, representing, respectively, the maximum velocity of Michaelis–Menten
kinetics and the concentration value at which 50% of the maximum velocity is reached.
Moreover, a one-compartment structure is considered for drug disposition, where
Vd denotes the apparent volume of distribution of the compartment. The schematic
diagram is displayed in Fig. 1.
For the multiple administrations, a same dose amount D is repeated for each fixed
time interval τ ; then, the drug disposition can be described as:

d
dt C(t) = rprod − kel C(t) − Vd V(Kmax C(t)
m +C(t))
, t = nτ,
(1)
C(nτ + ) = C(nτ ) + Vd , t = nτ, n = 0, 1, 2, . . .
D

where C(nτ + ) refers to the drug concentration immediately after a dose D admin-
istered at time nτ and C(0) is the drug concentration immediately prior to the very
Exogenous
IV bolus (D)

Linear elimination (kel)

Endogenous
C(t), Vd
production (rprod) Michaelis-Menten
elimination (Vmax, Km)

Fig. 1 (Color figure online) Schematic diagram of the one-compartment PK model representing drug
substances with two sources of input (endogenous production and exogenous IV bolus administration) and
two elimination pathways (linear and Michaelis–Menten kinetics). C(t) is the plasma concentration at time
t; Vd the apparent volume of distribution; D the dose amount; rprod the endogenous production rate; kel the
elimination rate constant; Vmax the maximum velocity of Michaelis–Menten kinetics; K m the concentration
value at which 50% of Vmax is reached

123
Analytical Solution and Exposure Analysis of a…

first dose administration in the multiple dosing scheme which may vary for different
patients. In the following, we will use (D, τ ) to denote the associated multiple dosing
regimen.

2.2 Closed-form Solution of Concentration–Time Course

To give the closed-form solution of the PK model (1), a transcendent function, named
X function (Wu et al. 2015), and its unique real branch in the first quadrant in R2
are needed. More details are given in “Appendix A.” In the rest part of the paper, X
represents the real branch of X function in the first quadrant.
Definition 1 For given p, q ∈ R+ , the X function is the multivalued inverse of the
function f (z) = z p (z + 1)q , i.e.,
(X (z, p, q)) p (X (z, p, q) + 1)q = z,
where z is a complex number.
Theorem 1 For the PK model (1), the closed-form solution of concentration–time
course for t ∈ ((n − 1)τ, nτ ] (n = 1, 2, . . .) is given by
⎛⎛ ⎞p
C((n − 1)τ ) + VD − Chs
C(t) = Chs + (Chs + Cβen ) · X ⎝⎝ d ⎠
Chs + Cβen
⎛ ⎞q ⎞
C((n − 1)τ ) + VD + Cβen
⎝ d ⎠ e−(t−(n−1)τ ) , p, q ⎠ , (2)
Chs + Cβen

and
D
C((n − 1)τ + ) = C((n − 1)τ ) + , (3)
Vd
where p, q and Cβen are positive and Cβen > K m (Wu et al. 2018). Meanwhile, we have
 
1 2
Chs = C L,hs − Cβ + C L,hs − Cβ + 4C L,hs K m , Cβen = Chs − C L,hs + Cβ ,
2
rprod Vmax
C L,hs = , Cβ = + Km,
kel kel Vd
1 Cβ − K m
en
1 Chs + K m
p= , q = .
kel Chs + Cβen kel Chs + Cβen
(4)

Proof For t > 0, the first equation of model (1) can be mathematically rearranged
into

p q
+ dC(t) = − dt (5)
C(t) − Chs C(t) + Cβen

when C(t) = Chs , and Chs , Cβen , p and q are defined in Eq. 4.

123
X. Wu et al.

Integration of Eq. 5 with respect to time from (n − 1)τ + to t leads to


q
(C(t) − Chs ) p C(t) + Cβen = C((n − 1)τ + ) − Chs
p

q
C((n − 1)τ + ) + Cβen e−(t−(n−1)τ ) ,

which can be further rewritten as


 p  q  p
C(t) − Chs C(t) − Chs C((n − 1)τ + ) − Chs
+1 =
Chs + Cβen Chs + Cβen Chs + Cβen
  q
C((n − 1)τ + ) + Cβen
e−(t−(n−1)τ )
Chs + Cβen

by dividing both sides by (Chs + Cβen ) p+q .


Following the definition of X function, C(t)−C hs
Chs +C en
in the last equation can be expressed
β
as
 p
C(t) − Chs C((n − 1)τ + ) − Chs
=X
Chs + Cβen Chs + Cβen
 q 
C((n − 1)τ + ) + Cβen −(t−(n−1)τ )
e , p, q . (6)
Chs + Cβen

Substituting C((n − 1)τ + ) by C((n − 1)τ ) + D/Vd and using a direct transformation
lead to the closed-form solution of C(t) (Eq. 2). 

Remark 1 It is noteworthy that, with the X function implemented in mathematical
software, C(t) can be iteratively evaluated from one period to the next. Moreover, its
dependence on the initial condition C(0) is shown in Fig. 2.
For multiple IV bolus administrations, a stable drug PK is important. This requires
the assessment of the highest (peak) and lowest (trough) drug concentrations, in terms
of their existence, uniqueness and periodicity.
For simplicity, Css (t) and Csstr (D, τ ) are used to denote the periodic and trough

concentrations, respectively, at steady state of the PK model (1).


Theorem 2 For the PK model (1), we have
(1) Trough concentration at steady state, Css tr (D, τ ), uniquely exists.

(2) The steady-state solution, Css (t), is periodic with period τ and given by
 tr (D, τ ) +
p
Vd − C hs
D
Css
Css (t) = Chs + (Chs + Cβen ) · X
Chs + Cβen
 tr (D, τ ) +
q 
Vd + C β
D en
Css
e−(t−(n−1)τ ) , p, q , (7)
Chs + Cβen

123
Analytical Solution and Exposure Analysis of a…

for t ∈ ((n − 1)τ, nτ ], n = 1, 2, . . .; Chs , Cβen , p and q are as defined in Eq. 4.


In particular, when C(0) = Css tr (D, τ ), the solution of model (Eq. 1) is genuinely

periodic as given by Eq. 7.


(3) Css (t) is locally asymptotically stable with respect to a perturbation of the initial
condition.
Proof (1) Denote Cn = C(nτ ) be trough concentration at each dosing interval. It
follows from Eq. 2 that we have the following difference equation as
 p
Cn + D
Vd − Chs
Cn+1 = Chs + (Chs + Cβen ) · X
Chs + Cβen
 q 
Cn + D
+ Cβen
e−τ , p, q ,
Vd
(8)
Chs + Cβen

which implies that there is a positive periodic solution of model (1) if model (8) has
at least positive steady state.
tr (D, τ ) satisfies
In fact, the steady-state trough concentration Css
 tr (D, τ ) +
p
Vd − C hs
D
Css
tr
Css (D, τ ) = Chs + (Chs + Cβen ) · X
Chs + Cβen
 tr (D, τ ) +
q 
Vd + C β
D en
Css −τ
e , p, q .
Chs + Cβen

In terms of the definition of X function and re-arrangement of the above equation, it


leads to
 tr (D, τ ) − C 1−s  tr (D, τ ) + C en
s
Css Css β
= e−kel τ , (9)
hs
tr (D, τ ) − C + D/V
Css C tr (D, τ ) + C en + D/V
hs d ss β d

Cβen −K m
where s = Chs +Cβen
∈ (0, 1) since it can be proved Cβen > K m . Following “Appendix
A” in Wu et al. (2015), we know that Css tr (D, τ ) uniquely exists.

(2) Since the trough concentration uniquely exists, the periodic solution at steady-
state Css (t) can be immediately obtained by replacing C((n − 1)τ ) by Css tr (D, τ ) in

Eq. 2, which results in Eq. 7.


(3) Concerning the local stability of Css (t), we will show any solution of model
(Eq. 1) with positive initial value will asymptotically approach the periodic solution
(Eq. 7). In fact, it is equivalent to the stability of peak concentration at steady-state
max which is right differentiable. To do that, let us define P : C(nτ + ) ∈ (0, ∞)  →
Css
C((n + 1)τ + ) ∈ (0, ∞) at steady state in the sense that

C((n + 1)τ + ) = C((n + 1)τ ) + D/Vd



= Chs + Chs + Cβen X z(C(nτ + )), p, q , (10)

123
X. Wu et al.

where
 p  q
+ C(nτ + ) − Chs C(nτ + ) + Cβen
z(C(nτ )) = e−τ . (11)
Chs + Cβen Chs + Cβen

Obviously, Css max is an equilibrium of the difference equation (Eq. 10), and it is asymp-

totically stable if the following condition holds (Agarwal 2000):

∂P(C(nτ + )) 
 < 1. (12)
∂C(nτ + ) C(nτ + )=Cssmax

Applying the chain rule to Eq. 10, we have


∂ P (C(nτ + ))  
 ∂z 

+  = (Chs + Cβen )X z (z, p, q) · + 
∂C(nτ ) C(nτ )=Css
+ max C(nτ )=Css ∂C(nτ ) C(nτ + )=Css
+ max max

(13)

where X z (z, p, q) is the derivative of the real branch in the first quadrant of X function
with respect to the variable z . The derivative of X function with respect to variable z is

  −1
 1 p q
X z (z, p, q) = max )
+ .
C(nτ + )=Css max z(Css X (z(Css
max ), p, q) 1 + X (z(Css
max ), p, q)

(14)

Since at steady state,


tr (D, τ ) − C
Css C max − Chs − D/Vd
hs
X (z(Css
max
), p, q) = = ss , (15)
Chs + Cβ en Chs + Cβen

Eq. 14 turns into



 1 1
X z (z, p, q) = ·
max ) C + C en
·
C(nτ + )=Css
max z(Css hs β
 −1
p q
+ max . (16)
Css − Chs − D/Vd
max Css + Cβ − D/Vd )
en

From Eq. 11, we have


 
∂z 
 p q
+  = + max max
z(Css ). (17)
∂C(nτ ) C(nτ + )=Cssmax Css − Chs
max Css + Cβen

Substituting Eqs. 16 and 17 into Eq. 13, we easily have

∂ P (C(nτ + )) 
max − C ) + q/(C max + C en )
p/(Css hs ss β
+  = <1
∂C(nτ ) C(nτ )=Css
+ max p/(Css − Chs − D/Vd ) + q/(Css
max max + C en − D/V )
β d

123
Analytical Solution and Exposure Analysis of a…

250
C(0) = 20 mIU/ml
C(0) = 200 mIU/ml
Concentration (mIU/ml) 200

150

100

50

0
0 50 100 150 200 250
Time (h)

Fig. 2 (Color figure online) An illustration to show that any solution of the studied model with positive initial
value will asymptotically approach the periodic solution. Here two initial conditions C(0) = 20 mIU/ml and
200 mIU/ml are considered; other parameters are rprod = 10 mIU/ml/h, Vd = 61.18 ml/kg, kel = 0.02/h,
K m = 67.28 mIU/ml, Vmax = 1500 mIU/h/kg, D = 2700 mIU, τ = 24 h

since the numerator is smaller than the denominator.


An illustration of the local stability of the periodic solution with respect to the
perturbation of initial conditions is presented in Fig. 2. 


2.3 Interpretation of the Parameters

In terms of pharmacology, the parameters defined in Eq. 4 can be explained as follows:


(i) C L,hs is the baseline concentration of a linear model (the first differential equation
of model (1) with the absence of Michaelis–Menten kinetics).
(ii) Chs is the baseline concentration of the studied model (1).
(iii) Consider two linear PK models without endogenous production:

dC1 (t)
a) = −kel C1 (t)
dt  
dC2 (t) Vmax
b) = − kel + C2 (t),
dt K m Vd

dC(t) 
and let dt C(t)=c denote the value of derivative of concentration with respect
to time t when concentration value is equal to c. Then Cβ is the concentration in
Model (a) that gives the same rate of change as in Model (b) for concentration
K m , that is,
 
dC1 (t)  dC2 (t) 
= .
dt C1 (t)=Cβ dt C2 =K m

123
X. Wu et al.

(iv) Consider two linear PK models with constant endogenous productions:

dC3 (t)
c) = kel Chs − kel C3 (t)
dt  
dC4 (t) Vmax
d) = kel C L,hs − kel + C4 (t),
dt K m Vd

then Cβen is the concentration in Model (c) that gives the same rate of change as
in Model (d) for concentration K m , that is,
 
dC3 (t)  dC4 (t) 
= .
dt C3 (t)=C en dt C4 =K m
β

It is noteworthy that Cβen is the extension of Cβ when the constant endogenous


production is taken into account.
(v) Since 1/kel is the average time (Boroujerdi 2015) that drug molecules for the
case of an IV bolus are assumed to be eliminated through the linear elimination
pathway alone (Model (c)), the pair p and q give a partition of 1/kel modulated
Chs +K m Cβen −K m
by two multipliers Chs +Cβen
and Chs +Cβen
, respectively.

We point out that the parameters described in this section will be equally explained
hereafter.

3 Drug Exposure at Steady State for Multiple Intravenous Bolus


administrations

3.1 Steady-State AUCcorr corr


ss, and Single-Dose AUC0−∞

In the presence of endogenous production, the corrected plasma concentration calcu-


lated by subtracting the baseline from the observed concentration was recommended
(European Medicines Agency 2010; FDA Guidance 2014; Health Canada 2018).
Namely, within a dosing interval τ , drug exposure at steady state is defined as
 τ
AUCcorr
ss,τ = (Css (t) − Chs ) dt. (18)
0

The single-dose AUCcorr


0−∞ has been discussed in Wu et al. (2018), and we recall
the result as follows:
Lemma 1 Wu et al. (2018) For drugs following kinetics described by Eq. 1, AUCcorr
0−∞
after a single IV bolus is
 
D Cβen − K m D/Vd
AUCcorr
0−∞ = − ln 1 + . (19)
kel Vd kel Chs + Cβen

123
Analytical Solution and Exposure Analysis of a…

For the case of multiple dosing administration, the results are as follows:
Theorem 3 For the PK model described by Eq. 1, we have
(1) Steady-state AUCcorr
ss,τ is
 
D Cβen − K m D/Vd
AUCcorr
ss,τ = − ln 1 + tr , (20)
kel Vd kel Css (D, τ ) + Cβen

tr (D, τ ) is the trough concentration at steady state satisfying Eq. 9.


where Css
(2) Single-dose AUCcorr corr
0−∞ is strictly less than steady-state AUCss,τ , and both are
upper bounded by AUC of the drugs being the PK model with linear elimination
alone for a IV bolus input, i.e.,

D
0−∞ < AUCss,τ <
AUCcorr .
corr
(21)
kel Vd

Proof Equation 5 is still valid for the model (1) at steady state; thus, we have

p q
+ dCss (t) = − dt,
Css (t) − Chs Css (t) + Cβen

which can be rearranged as



Chs + Cβen
(Css (t) − Chs ) dt = − p + q − q dCss (t). (22)
Css (t) + Cβen

Integrating Eq. 22 within a dosing interval [0, τ ] yields


 τ
AUCcorr
ss,τ = (Css (t) − Chs ) dt
0
 
D Cβen − K m D/Vd
= − ln 1 + tr , (23)
kel Vd kel Css (D, τ ) + Cβen

which corresponds to Eq. 20.


It is clear that the trough concentration at steady state is always greater than the
tr (D, τ ) > C , and C en > K , we further have
baseline concentration, i.e., Css hs β m

 
D Cβen − K m D/Vd
AUCcorr
ss,τ > − ln 1 + .
kel Vd kel Chs + Cβen

Using Lemma 1, we have

ss,τ > AUC0−∞ .


AUCcorr corr

123
X. Wu et al.

D
ss,τ <
AUCcorr directly results from Cβen > K m . 

kel Vd
Remark 2 In the absence of Michaelis–Menten kinetics, even for the linear PK model
with a constant endogenous input, the inequalities (Eq. 21) turn into an equality, which
can easily be seen by taking Vmax = 0, which yields Cβen = K m in Theorem 3. This
is a known fact for linear PK models that we simply summarize below.
Corollary 1 For the PK model (1) while having only the first-order elimination and
constant endogenous production, we have

D
0−∞ = AUCss,τ =
AUCcorr .
corr
(24)
kel Vd

3.2 Impact of Administered Dose Amount and Endogenous Production

As expected, a higher dose amount always induces a higher AUC, which is true for
both single- and multiple-dose situations. In the case of linear PK, the increase in
AUC is positively proportional to the increase in the administered dose amount. Par-
ticularly, the principle of the equality between steady-state and single-dose AUC is
currently used to guide dose adaptation. However, for the nonlinear PK model as the
one discussed in this paper, both single- dose AUC and steady-state AUC are no more
linearly proportional to the administered dose amount. Moreover, AUCcorr ss,τ is always
larger than AUCcorr
0−∞ . Thus, the way how these two differ for different dose amounts
becomes an important issue when performing dose adjustment.
For this purpose, we simulated the two curves of AUCcorr corr
ss,τ and AUC0−∞ in function
of dose (Fig. 3a). It clearly indicates the nonlinear trends of AUC versus dose with a
faster increase of AUCcorr
ss,τ . This feature should be given a careful consideration for a
safe dose adaptation, which can only be reached using a model-based approach instead
of applying a simple algebraic formula.

700 1100
(a) 1000
(b)
Drug exposure (mIU× h/ml)
Drug exposure (mIU× h/ml)

600
900
500
800

400 700

600
300
500
200
400
AUCcorr
0-∞
AUCcorr
0-∞
100 300
AUCcorr
ss,τ
AUCcorr
ss,τ

0 200
0 2000 4000 6000 8000 10000 12000 0 50 100 150 200 250 300 350 400

Dose (mIU/kg) rprod (mIU/h/ml)

Fig. 3 (Color figure online) Drug exposure of the PK model (1) in terms of dose amount and endogenous
production rate. a AUC versus D, with rprod = 10 mIU/ml/h. b AUC versus rprod , with D = 13500 mIU/kg.
Other parameters are Vd = 61.18 ml/kg, kel = 0.21/h, K m = 67.28 mIU/ml, Vmax = 6000 mIU/h/kg,
τ =2 h

123
Analytical Solution and Exposure Analysis of a…

AUC corr
ss,τ
-AUCcorr
0-∞

180

160
50

140
Dose amount D (mIU/kg)

120
100

100

150 80

60

200
40

20
250

10 20 30 40 50 60 70 80
rprod (mIU/h/ml)

Fig. 4 (Color figure online) Heatmap of AUCcorr corr


ss,τ −AUC0−∞ in terms of administered dose amount D and
endogenous production rate rprod , where D ∈ [100, 13500 mIU/kg] and rprod ∈ [0,400 mIU/h/ml]. Other
parameters are Vd = 61.18 ml/kg, kel = 0.21/h, K m = 67.28 mIU/ml, Vmax = 6000 mIU/h/kg, τ = 2 h

Endogenous production may be another issue for the evaluation of drug exposure.
Though the baseline concentration generated by endogenous production rprod has been
subtracted prior to the calculation of AUCcorr corr
ss,τ and AUC0−∞ , the dependence on rprod
is still observed (Fig. 3b). Increasing rprod leads to an increase in both AUCcorr
ss,τ and
AUCcorr0−∞ . The gap between these two becomes, however, smaller, and eventually
both converge to a same value of D/(Vd kel ). Importantly, the smaller the endogenous
production rate rprod , the larger the difference between AUCcorr corr
ss,τ and AUC0−∞ . Partic-
ularly, rprod = 0 gives the largest difference. In other words, for small values of rprod ,
the nonlinear feature of Michaelis–Menten elimination is important, whereas for large
rprod , the linear elimination becomes dominant. The difference between AUCcorr ss,τ and
AUCcorr0−∞ can be further seen within a heatmap in terms of administered dose amount D
and endogenous production rate rprod (Fig. 4). One may observe that for high doses and
low endogenous production levels, the difference in drug exposure can be important.

3.3 Average Concentration at Steady State

Average concentration at steady-state Css,av is a useful indicator for drug exposure as it


can be easily calculated through samplings. In fact, it can be derived from AUCcorrss,τ by
taking the ratio of the latter over the dosing interval τ . For a linear PK process, Css,av
remains the same either by IV infusion or IV bolus administrations and not affected
by changes in dosing frequency given the same total daily dose as the resulting AUC
at steady state is the same. Several principles are widely used in clinical practice. For

123
X. Wu et al.

example, Css,av doubles when the dose amount is doubled, or Css,av is the same for the
same daily dose amount regardless of the dosing regimens such as once daily (QD)
and twice daily (BID).
However, with the involvement of nonlinear kinetics as for the PK model considered
here (Eq. 1), these principles have to be revisited. The following theorem summarizes
our major findings.

Theorem 4 For the PK model (1) of multiple dosing administration, we have

(1) If the dose amount D and dosing interval τ are split into two equal halves, namely
multiple dosing drug regimen (D, τ ) is changed into ( D2 , τ2 ); thus, the resulting
average concentration at steady state decreases, i.e.,

Css,av ( D2 , τ2 ) < Css,av (D, τ ). (25)

(2) For an IV infusion of constant rate D/τ , the concentration at steady-state Css,inf
gives the minimum value of all Css,av produced by the regimens obtained by suc-
cessively splitting dose and dosing time interval into two equal halves.

To prove Theorem 4, we need the following lemma for trough concentrations and
AUCcorr
ss,τ at steady state.

Lemma 2 For the PK model (1), if the dose amount D and dosing interval τ are split
into two equal halves, then the resulting trough concentration increases, i.e.,
 
D τ
tr
Css (D, τ ) < tr
Css , . (26)
2 2

However, the corresponding drug exposure at steady state during the dosing interval
τ decreases, i.e.,

2AUCcorrτ D
2 ss,τ (D).
< AUCcorr (27)
ss, 2

Proof First we prove Eq. 26. It follows from Theorem 2 and Eq. 9 that the trough
tr (D, τ ) and C tr ( D , τ ) uniquely exist and satisfy
concentrations at steady-state Css ss 2 2

 1−s  s
D/Vd D/Vd
1 + tr 1+ = ekel τ (28)
Css (D, τ ) − Chs tr (D, τ ) + C en
Css β

and
 1−s  s
1 1
2 D/Vd 2 D/Vd τ
1+ 1+ = ekel 2 . (29)
tr ( D , τ ) − C
Css tr ( D , τ ) + C en
Css
2 2 hs 2 2 β

123
Analytical Solution and Exposure Analysis of a…

By raising both sides of Eq. 29 to the power of two and comparing with Eq. 28, we
obtain
 1−s  s
D/Vd D/Vd
1 + tr 1 + tr
Css (D, τ ) − Chs Css (D, τ ) + Cβen
⎡ 2 ⎤1−s ⎡ 2 ⎤s (30)
1 1
D/V D/V
=⎣ 1+ ⎦ ⎣ 1+ ⎦ .
2 d 2 d
tr ( D , τ ) − C
Css tr ( D , τ ) + C en
Css
2 2 hs 2 2 β

Let us consider Csstr (D, τ ) ≥ C tr ( D , τ ), then the right side can be seen larger than
ss 2 2
the left side in Eq. 30 if the two squared terms on the right are expanded. Hence, we
have Csstr (D, τ ) < C tr ( D , τ ) as required by Eq. 30.
ss 2 2
Now we prove Eq. 27. By applying Eq. 20, we have
 
D Cβen − K m D/Vd
ss,τ (D)
AUCcorr = − ln 1 + tr (31)
kel Vd kel Css (D, τ ) + Cβen
⎡ 2 ⎤
C en − K 1
D β m D/V
ln ⎣ 1 + ⎦
d
2AUCcorrτ ( D2 ) = − 2
(32)
ss, 2 kel Vd kel tr ( D , τ ) + C en
Css 2 2 β

for dosing regimens (D, τ ) and ( D2 , τ2 ), respectively.


If we write
D/Vd D/Vd
A= D τ
and B = ,
Css ( 2 , 2 ) + Cβen
tr Css (D, τ ) + Cβen
tr

then we have A < B by Eq. 26. From Eqs. 31–32, the statement of Eq. 27 is equivalent
to the inequality

(1 + 21 A)2 > 1 + B (33)

as Cβen > K m for the model (Eq. 1).


To prove the inequality, we can rewrite Eq. 30 as
 2  −1
1  − s
2 D/Vd D/Vd 2
(1 + B)−1
1−s
1+ 1 + tr = 1 + 21 A .
tr ( D , τ ) − C
Css Css (D, τ ) − Chs
2 2 hs
(34)

However, we can further prove (see “Appendix B”) that


 2  −1
1
2 D/Vd D/Vd 2
1+ 1 + tr < 1 + 21 A (1 + B)−1 .
tr ( D , τ ) − C
Css Css (D, τ ) − Chs
2 2 hs
(35)

123
X. Wu et al.

Following Eqs. 34–35, we have


 2
− s
2
(1 + B)−1 (1 + B)−1 ,
1−s
1 + 21 A < 1 + 21 A

which is equivalent to

 2
 1
(1 + B)−1
1−s
1 + 21 A > 1.

Since s ∈ (0, 1), we have

2
1 + 21 A (1 + B)−1 > 1,

which confirms Eq. 33. 



From Lemma 2, we can obtain an interesting corollary for nonlinear PK, which is
in contrast to the superposition principle for linear PK.
Corollary 2 For the PK. model (1), if we double the dosing amount for each dosing
interval τ , we have

ss,τ (2D) > 2AUCss,τ (D).


AUCcorr corr
(36)

Hence, for drugs satisfying the considered nonlinear PK model, we can expect there
is a larger than proportional increase in AUC for each dose increment at steady state.
Here we go back to the proof of Theorem 4.
Proof For dosing regimens (D, τ ) and ( D2 , τ2 ), the average concentrations at steady
state (Gibaldi and Perrier 2007; Boroujerdi 2015) are

2AUCcorrτ ( D2 )
ss,τ (D)
AUCcorr ss, 2
Css,av (D, τ ) = + Chs and Css,av ( D2 , τ2 ) = + Chs ,
τ τ
(37)

respectively. From Lemma 2, we have Css,av (D, τ ) > Css,av ( D2 , τ2 ).


By repetitively making equal divisions of the dose and time interval, we have

D τ D τ D τ
Css,av 2, 2 > Css,av 4, 4 > Css,av 8, 8 > ···

PK of constant rate IV infusion for a one-compartment model can be regarded as


infinite times (n) of IV bolus injection with an equal dose and zero interval (Li and
Nekka 2007; Yu and Cao 2017). We can prove the concentration at steady state induced
by constant infusion (Css,inf ) is the limit of Css,av ( 2Dn , 2τn ) when n tends to infinity,
that is,

Css,av (D, τ ) > Css,av ( D2 , τ2 ) > · · · > lim Css,av ( 2Dn , 2τn ) = Css,inf .
n→∞

123
Analytical Solution and Exposure Analysis of a…

(a) (b) (c)


250 250 250
C(t) by (D,τ) C(t) by (D/2,τ/2) C(t) by (D/4,τ/4)
Css,av Css,av Css,av
IV infusion with R0 =D/τ IV infusion with R0 =D/τ IV infusion with R0 =D/τ
200 200 200
Concentration (mIU/ml)

Concentration (mIU/ml)

Concentration (mIU/ml)
150 150 150

100 100 100

50 50 50

0 0 0
4 6 8 10 4 6 8 10 4 6 8 10
1 1
D D Time (h) 2
D 12 D 12 D Time (h) 4
D 14 D 14 D Time (h)

Fig. 5 Plasma concentrations of drug substance versus time (solid blue lines) and the corresponding steady-
state average concentrations (dashed blue lines) for different multiple IV regimens, and all are in contrast
to that of IV infusion with rate R0 = D/τ (solid red lines). a (D, τ ), b (D/2, τ/2), c (D/4, τ/4). In all
simulations, rprod = 10 mIU/ml/h, Vd = 61.18 ml/kg, kel = 0.21 /h, K m = 67.28 mIU/ml, Vmax = 19930
mIU/h/kg, D = 13500 mIU/kg, τ = 2 h

This indicates that, given a dose amount D in a fixed time interval τ , IV infusion
gives the minimum concentration at steady state compared to all average steady-state
concentrations induced by above multiple IV bolus-dose administrations. 

Numerical simulations are performed and displayed in Fig. 5. It shows that, with
the increase in dosing frequency, the trough concentrations Css tr increase and average

concentrations at steady-state Css,av decrease, whereas the steady-state concentration


produced by IV infusion of rate D/τ , Css,inf , gives the lowest concentration level
to all these Css,av . Conversely, reducing dosing frequency will increase the average
concentration at steady state. All these facts could influence the drug therapeutic effect
and need more consideration in clinical practice. In addition to the equal division, we
also simulate some cases of unequal divisions of drug regimen (D, τ ), which indicate
that Css,inf always gives the lowest average steady-state concentration level (Fig. 6).

4 Discussion and Conclusion

In this paper, we aim to investigate the multiple dosing behavior of a PK model


bearing a specific mathematical structure that is common to many biologics. This is
a one-compartment model with two simultaneous first-order and Michaelis–Menten
elimination pathways and also involves a constant endogenous production additional
to the exogenous IV bolus administration. In fact, we have explored this model in a
previous paper for the case of a single IV bolus-dose administration (Wu et al. 2018).
Despite the existence of the nonlinear component in elimination and the endogenous
input, we were able to provide a closed-form solution of the concentration–time course
using the X function that we introduced in Wu et al. (2015). Compared to the existing

123
X. Wu et al.

(a) (b) (c)


250 250 250
C(t), D=2/3D+1/3D,τ=1/2τ+1/2τ C(t),D=1/2D+1/2D,τ=2/3τ+1/3τ C(t), D=2/5D+3/5D,τ=1/5τ+4/5τ
Css,av Css,av Css,av
IV infusion with R =D/τ IV infusion with R =D/τ IV infusion with R =D/τ
0 0 0
200 200 200
Concentration (mIU/ml)

Concentration (mIU/ml)

Concentration (mIU/ml)
150 150 150

100 100 100

50 50 50

0 0 0
4 6 8 10 4 6 8 10 4 6 8 10
2
D 13 D 23 D 13 D Time (h) 1
2
D 1
D 12 D 1
2
D Time (h) 2
5
D 35 D 25 D 35 D Time (h)
3 2

Fig. 6 Plasma concentrations of drug substance versus time (solid blue lines) and the corresponding steady-
state average concentrations (dashed blue lines) for multiple IV regimens obtained by unequal division of
dose and time interval of the regimen (D, τ ), and all are in contrast to that of IV infusion with rate
R0 = D/τ (solid red lines). a D is split into 2/3D and 1/3D, which are alternately administered at
the beginning of each τ/2 dosing interval, b τ is split into two dosing intervals of 2/3τ and 1/3τ , D/2
are repeatedly administered at the beginning of each dosing interval, c D is split into 2/5D and 3/5D,
which are administered at the beginning of dosing intervals 1/5τ and 4/5τ , respectively. In all simulations,
rprod = 10 mIU/ml/h, Vd = 61.18 ml/kg, kel = 0.21 /h, K m = 67.28 mIU/ml, Vmax = 19930 mIU/h/kg,
D = 13500 mIU/kg, τ = 2 h

results for models having either only one type or two parallel types of elimination
with/without endogenous production (Schnell and Mendoza 1997; Tang and Xiao
2007; Wu et al. 2015, 2018; Yu and Cao 2017), we have been able to provide the non-
trivial generalization by extending their expressions to simultaneous elimination and
endogenous production. Moreover, many interesting results around the drug exposure
of a single IV bolus dose that differ from the linear case are explored therein. In the
current work, we have successfully established the closed-form solution of this PK
model for multiple IV bolus at transient and steady states. The analytical solution has
been made possible through the use of X function, which had formerly been key to the
closed-form solution in the simpler case when endogenous production is absent (Wu
et al. 2015). By exploring the PK properties of the current model, we found that, in
the presence of Michaelis–Menten elimination, a much higher steady-state AUCcorr ss,τ
than the single-dose AUCcorr
0−∞ can be reached, thus contrasting with the well-known
superposition principle for the linear case. In other words, these drug exposure indices
are no more proportional to the prescribed dose amounts. Moreover, by assuming a
same total dose amount D during a time interval τ , we have shown that the average
concentration at steady-state Css,av decreases with increasing dosing frequency and
eventually reaches downward to Css,av in the limit case of IV infusion administration.
Additionally, in the presence of endogenous production, we found that, even when the
corrected concentration is used for the estimation of steady-state AUCcorr
ss,τ , we cannot
get rid of the effect of endogenous production. Driven by the nonlinear component
and the endogenous production part and their mutual interaction, these drug exposure

123
Analytical Solution and Exposure Analysis of a…

indices heavily depend on the routes of administration or the regimens used. In this
regard, we advocate that, for safety and efficacy purposes, the estimation methods
established for the classical linear PK have to be used carefully if the drug exhibits
one of the features as shown in the current model.
The classical PK analysis is based on linear compartment modeling. The choice
of this linear approach is natural and can be attributed to several reasons. One is
that, at the early period of drug discovery and research, limited by analytical and
experimental techniques, there was a large tolerance margin for precision and accuracy
when describing drug kinetics. Thus, satisfaction with the results brought by data
fitting using linear compartment models was the rule. Another reason was the reliance
of linear compartment modeling on system theory which was well developed and has
brought a great success to other fields such as physics, mechanical engineering and
system control. Many principles have been discovered and set up, for example the so-
called linear time variance property. The adaptation of this approach to the PK analysis
allowed to make a rapid progress in the field. The corresponding rules are relatively
simple and can be easily applied in daily clinical practice, particularly in the context
of therapeutic drug monitoring. However, with the new drug candidates such as mAbs,
which often have a narrow therapeutic range, along with the development of cutting-
edge measurement tools, the applicability of these principles is to be challenged. For
these drugs, deviation from linear models can lead to surprising results. For instance,
in the case of linear PK, a steady-state level is assumed to be achieved after five
half-lives in clinical practice since 97% of the steady-state level is close enough to be
considered 100% (Boroujerdi 2015). However, if we estimate the half-life by wrongly
assuming linear elimination for the nonlinear model, the time to reach the steady state
will be largely overestimated. In fact, two or three of such half-lives are enough for
the nonlinear PK considered here.
The PK model discussed here is far from complete compared to many PK models
proposed for biologics in the recent literature. The most representative is the so-called
target-mediated drug disposition (TMDD), in which saturate, high-affinity binding
of the drug to its pharmacological target is mathematically modeled to represent the
observed nonlinear PK behavior (Mager and Jusko 2001; Mager 2006; Woo et al.
2007). Another progressively popular approach is the whole-body physiologically
based PK model (PBPK), which uses multiple compartments representing various
tissues and organs, connected by circulating blood to reach a mechanistic description
of drug distribution by the incorporation of physiological concepts (Wong and Chow
2017). However, since the mathematical maneuverability is inversely proportional
to models’ complexity, we do believe that considering simpler models that never-
theless pertain to the main features (nonlinearity and endogenous production here)
helps in understanding the raised critical issues. With further integrating more realis-
tic characteristics and mechanisms, such as the negative and/or positive feedback of
endogenous production (Quartino et al. 2014) and circadian influence, we may, step by
step, mathematically approach and keep up with the most up-to-date pharmaceutical,
physiological and biological knowledge.
From the mathematical point of view, this study can have many perspectives. The
first concerns the X function, which is a transcendent function that we introduced and
used to express the closed-form solution of the PK model of simultaneous first-order

123
X. Wu et al.

and Michaelis–Menten elimination. X function turns out to be as well useful for the
more complex PK model as considered here. This mathematical analysis is the first
to propose an analytical treatment contrasting with the usual recourse to numerical
methods to find solutions for PK models deviating from linearity. However, numerical
approaches raise several mathematical questions such as the existence, uniqueness
or stability of the solution. When the proposed transcendent function is repetitively
used and well accepted as for the PK model here, it is important to mathematically
investigate its properties, such as the X function. This is what we intended to do here by
exploring this function, which generalizes the well-known Lambert W function. The
second point worth of mentioning concerns model parameters. With the modification
of model components such as the case here, the introduced parameters should be
revisited and updated. For instance, Cβ is given for the model without endogenous
input, while Cβen is its updated version when a constant endogenous production is added
to the model. The harmony of these parameters through model transition is crucial to
ensure the modeling integrity. Our last comment concerns the need for mathematical
proofs. As shown here, we took the special case of equally dividing the dose and dosing
interval such that a decreasing trend of drug exposure can be mathematically found
at steady state, reaching the minimum at the limit case of IV infusion. Though this
strategy is quite common in clinical situations, various combinations of subdivisions
should be mathematically investigated in the future to verify our conjecture concerning
the validity of this result for other drug regimens.

Acknowledgements This research is supported by NSERC-Industrial Chair in Pharmacometrics—


Novartis, Pfizer and Inventiv Health Clinical and FRQNT Projet d’équipe led by F. Nekka as well as
NSERC and FRQNT (F.N. and J.L.). FRQNT Fellowship and NSFC (No. 11501358) hold by X.W. are also
acknowledged.

Appendix A: Introduction of X Function and Unique Real Branch in the


First Quadrant in R2

The aim of introduction of X function is to express the closed-form solution of a


one-compartment PK open model exhibiting simultaneous first-order and Michaelis–
Menten elimination (Wu et al. 2015). Specifically, the model is described as

 dC(t) Vmax C(t)


dt = −kel C(t) − Vd (K m +C(t)) , t >0
(A.1)
C(0+ ) = D/Vd ,

where D is an intravenous bolus dose, kel is the rate constant of the first-order elimina-
tion, Vmax and K m are the corresponding parameters of Michaelis–Menten elimination
representing the maximum velocity of its kinetics and the concentration value at which
50% of Vmax is reached, and Vd is the apparent volume of distribution.
Integrating Eq. A.1 gives rise to the following algebraic equation
  p1   p2   p1   p2
C(t) C(t) C0 C0
+1 = +1 e−t (A.2)
Cβ Cβ Cβ Cβ

123
Analytical Solution and Exposure Analysis of a…

where

Vmax 1 Km 1 Cβ − K m
Cβ = K m + > K m , p1 = > 0, p2 = > 0.
kel Vd kel Cβ kel Cβ

Using the definition of X function (see Definition 1), we are able to express C(t)
as
  p1   p2 
D/Vd D/Vd −t
C(t) = Cβ · X +1 e , p1 , p2 , (A.3)
Cβ Cβ

For more details, refer to Wu et al. (2015).


Next, we show the proof of unique real branch of X function in the first quadrant
in R2 , where p > 0 and q > 0.
From the definition of X function, we denote

F(z, X ) = (X (z, p, q)) p (X (z, p, q) + 1)q − z. (A.4)

In the scope of real number in the first quadrant in R2 , namely for each z, X > 0, we
can have
 
p q
Fz (z, X ) = −1, FX (z, X ) = z + = 0.
X X +1

By implicit function theorem, there is a unique real solution X (z, p, q) > 0 such that
the equation X (z, p, q) p (X (z, p, q) + 1)q = z is valid. Moreover, the derivative of
X (z, p, q) with respect to z is given by

 −1
dX (z, p, q) Fz (x, X ) 1 p q
=− = + > 0,
dz FX (z, X ) z X X +1

which implies the function X (z, p, q) is increasing with the variable z in the first
quadrant.

Appendix B: Proof of the Relationship of Eq. 35

To prove Eq. 35, we define a continuous function as

 2  −1
1
de f 2 D/Vd D/Vd
h(x) = 1+ 1 + tr ,
tr ( D , τ ) + C en
Css −x Css (D, τ ) + Cβen − x
2 2 β
 
x ∈ 0, Cβen + Chs , (C.1)

123
X. Wu et al.

where Cβen and Chs are concentrations defined in Sect. 2.3, D is dose amount, Vd is
tr (D, τ ) and C tr ( D , τ ) are respective steady-
the apparent volume of distribution, Css ss 2 2
state trough concentrations which are greater than Chs .
Taking derivative of h(x) with respect to the variable x yields

  −2
1
D 2 D/Vd D/Vd
h (x) = 1+ 1 + tr · g,
Vd tr ( D , τ ) + C en
Css −x Css (D, τ ) + Cβen − x
2 2 β
(C.2)

where g = g1 − g2 with

 
1 D/Vd
g1 = 1 + tr
tr ( D , τ ) + C en − x
Css
2 Css (D, τ ) + Cβen − x
2 2 β

and

 
1 D/Vd
g2 = 1+ .
tr (D, τ ) + C en − x 2 tr ( D , τ ) + C en − x
Css
Css β 2 2 β

Considering the right hand of h (x), the left portion before g is positive; thus, the
sign of h (x) depends on that of g alone. Furthermore, g can be rewritten as

1
g= (w1 − w2 )
tr ( D , τ ) + C en
Css − x (Css
2
tr (D, τ ) + C en − x)2
2 2 β β

with

 
w1 = D
Vd + Css
tr
(D, τ ) + Cβen − x tr
Css (D, τ ) + Cβen − x

and

 
tr D τ tr D τ
w2 = D
Vd + Css ( 2 , 2 ) + Cβen − x Css ( 2 , 2 ) + Cβen − x .

tr ( D , τ ) > C tr (D, τ ) > C and x ∈ [0, C en + C ], we have 0 < w < w ,


Since Css 2 2 ss hs β hs 1 2
and thus, g is negative. Therefore, h(x) is a decreasing function with respect to its
variable x ∈ [0, Cβen + Chs ]. Subsequently, we have

123
Analytical Solution and Exposure Analysis of a…

 2
1
2 D/Vd
h(Cβen + Chs ) = 1 + τ
tr ( D ,
2 ) + C β − (C β + Chs )
Css en en
2
 −1
D/Vd
1+
tr (D, τ ) + C en − (C en + C )
Css β β hs
 2  −1
1
2 D/Vd D/Vd
= 1+ 1 +
tr ( D , τ ) − C
Css tr (D, τ ) − C
Css hs
2 2 hs
2
< h(0) = 1 + 21 A (1 + B)−1 .

References
Agarwal Ravil P (2000) Difference equations and inequalities. Theory, methods, and applications, 2nd edn.,
rev. and expanded. New York
Corless RM, Gonnet GH, Hare DEG, Jeffrey DJ, Knuth DE (1996) On the Lambert W function. Adv
Comput Math 5:329–359
Craig M, Humphries AR, Nekka F, Bélair J, Mackey MC LJ (2015) Neutrophil dynamics during concur-
rent chemotherapy and G-CSF administration: mathematical modelling guides dose optimisation to
minimise neutropenia. J Theor Biol 385:77–89
Craig M, Humphries AR, Mackey MC (2016) A mathematical model of granulopoiesis incorporating the
negative feedback dynamics and kinetics of G-CSF/neutrophil binding and internalization. Bull Math
Biol 78(12):2304–2357
Dirks NL, Meibohm B (2010) Population pharmacokinetics of therapeutic monoclonal antibodies. Clin
Pharmacokinet 49(10):633–659
European Medicines Agency (2010) https://www.ema.europa.eu/en/documents/scientific-guideline/
guideline-investigation-bioequivalence-rev1_en.pdf. Accessed 11 Aug 2019
FDA Guidance (2014) Guidance for industry. Bioavailability and bioequivalence studies submitted in
NDAs or INDs—general considerations. https://www.fda.gov/media/88254/download. Accessed 11
Aug 2019
Foley C, Mackey MC (2009) Mathematical model for G-CSF administration after chemotherapy. J Theor
Biol 257:27–44
Frymoyer A, Juul SE, Massaro AN, Bammler TK, Wu YW (2017) High-dose erythropoietin population
pharmacokinetics in neonates with hypoxic-ischemic encephalopathy receiving hypothermia. Pediatr
Res 81(6):865–872
Gibaldi M, Perrier D (2007) Pharmacokinetics. Informa Healthcare USA Inc, New York
Health Canada (2018) Guidance document: conduct and analysis of comparative bioavailability
studies. https://www.canada.ca/content/dam/hc-sc/documents/services/drugs-health-products/drug-
products/applications-submissions/guidance-documents/bioavailability-bioequivalence/conduct-
analysis-comparative.pdf. Accessed 11 Aug 2019
Jin F, Krzyzanski W (2004) Pharmacokinetic model of target-mediated disposition of thrombopoietin.
AAPS Pharm Sci 6(1):E9
Keizer RJ, Huitema AD, Schellens JH, Beijnen JH (2010) Clinical pharmacokinetics of therapeutic mono-
clonal antibodies. Clin Pharmacokinet 49(8):493–507
Klitgaard T, Nielsen JN, Skettrup MP, Harper A, Lange M (2009) Population pharmacokinetic model for
human growth hormone in adult patients in chronic dialysis compared with healthy subjects. Growth
Horm IGF Res 19(6):463–470
Kloft C, Graefe EU, Tanswell P, Scott AM, Hofheinz R, Amelsberg A, Karlsson MO (2004) Population
pharmacokinetics of sibrotuzumab, a novel therapeutic monoclonal antibody, in cancer patients. Invest
New Drugs 22(1):39–52
Kuester K, Kovar A, Lüpfert C, Brockhaus B, Kloft C (2008) Population pharmacokinetic data analysis of
three phase I studies of matuzumab, a humanised anti-EGFR monoclonal antibody in clinical cancer
development. Br J Cancer 98(5):900–906

123
X. Wu et al.

Leader B, Baca QJ, Golan DE (2008) Protein therapeutics: a summary and pharmacological classification.
Nat Rev Drug Discov 7:21–39
Li J, Nekka F (2007) A pharmacokinetic formalism explicitly integrating the patient drug compliance. J
Pharmacokinet Pharmacodyn 34(1):115–139
Mager DE, Jusko WJ (2001) General pharmacokinetic model for drugs exhibiting target-mediated drug
disposition. J Pharmacokinet Pharmacodyn 28(6):507–532
Mager DE (2006) Target-mediated drug disposition and dynamics. Biochem Pharmacol 72(1):1–10
Mehdi B (2015) Pharmacokinetics and toxicokinetics. CRC Press, Boca Raton
Quartino AL, Karlsson MO, Lindman H, Friberg LE (2014) Characterization of endogenous G-CSF and
the inverse correlation to chemotherapy-induced neutropenia in patients with breast cancer using
population modeling. Pharm Res 31(12):3390–3403
Schnell S, Mendoza C (1997) Closed form solution for time dependent enzyme kinetics. J Theor Biol
187:207–212
Shi S (2014) Biologics: an update and challenge of their pharmacokinetics. Curr Drug Metab 15(3):271–290
Tang S, Xiao Y (2007) One-compartment model with Michaelis–Menten elimination kinetics and thera-
peutic window: an analytical approach. J Pharmacokinet Pharmacodyn 34:807–827
van der Graaf PH, Benson N, Peletier LA (2016) Topics in mathematical pharmacology. J Dyn Diff Equ
28:1337–1356
Wong H, Chow TW (2017) Physiologically based pharmacokinetic modeling of therapeutic proteins. J
Pharm Sci 106(9):2270–2275
Woo S, Krzyzanski W, Jusko WJ (2007) Target-mediated pharmacokinetic and pharmacodynamic model
of recombinant human erythropoietin (rHuEPO). J Pharmacokinet Pharmacodyn 34(6):849–868
Wu X, Li J, Nekka F (2015) Closed form solutions and dominant elimination pathways of simultaneous
first-order and Michaelis–Menten kinetics. J Pharmacokinet Pharmacodyn 42:151–161
Wu X, Nekka F, Li J (2016) Steady-state volume of distribution of two-compartment models with simulta-
neous linear and saturated elimination. J Pharmacokinet Pharmacodyn 43(4):447–459
Wu X, Nekka F, Li J (2018) Mathematical analysis and drug exposure evaluation of pharmacokinetic models
with endogenous production and simultaneous first-order and Michaelis–Menten elimination: the case
of single dose. J Pharmacokinet Pharmacodyn 45(5):693–705
Yu RH, Cao YX (2017) A method to determine pharmacokinetic parameters based on andante constant-rate
intravenous infusion. Sci Rep 7(1):13279
Zhao L, Shang EY, Sahajwalla CG (2012) Application of pharmacokinetics–pharmacodynamics/clinical
response modeling and simulation for biologics drug development. J Pharm Sci 101(12):4367–4382

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.

123

You might also like