Journal of Pharmacokinetics and Pharmacodynamics (2021) 48:861–871

https://doi.org/10.1007/s10928-021-09776-7 (0123456789().,-volV)(0123456789().
,- volV)

ORIGINAL PAPER

Minimal brain PBPK model to support the preclinical and clinical

development of antibody therapeutics for CNS diseases
Peter Bloomingdale1 • Suruchi Bakshi2 • Christian Maass3 • Eline van Maanen4 • Cesar Pichardo-Almarza5 •

Daniela Bumbaca Yadav1 • Piet van der Graaf5 • Nitin Mehrotra1

Received: 8 April 2021 / Accepted: 29 July 2021 / Published online: 10 August 2021
Ó The Author(s) 2021

Abstract
There are several antibody therapeutics in preclinical and clinical development, industry-wide, for the treatment of central
nervous system (CNS) disorders. Due to the limited permeability of antibodies across brain barriers, the quantitative
understanding of antibody exposure in the CNS is important for the design of antibody drug characteristics and determining
appropriate dosing regimens. We have developed a minimal physiologically-based pharmacokinetic (mPBPK) model of
the brain for antibody therapeutics, which was reduced from an existing multi-species platform brain PBPK model. All
non-brain compartments were combined into a single tissue compartment and cerebral spinal fluid (CSF) compartments
were combined into a single CSF compartment. The mPBPK model contains 16 differential equations, compared to 100 in
the original PBPK model, and improved simulation speed approximately 11-fold. Area under the curve ratios for minimal
versus full PBPK models were close to 1 across species for both brain and plasma compartments, which indicates the
reduced model simulations are similar to those of the original model. The minimal model retained detailed physiological
processes of the brain while not significantly affecting model predictability, which supports the law of parsimony in the
context of balancing model complexity with added predictive power. The minimal model has a variety of applications for
supporting the preclinical development of antibody therapeutics and can be expanded to include target information for
evaluating target engagement to inform clinical dose selection.

Keywords PBPK  Antibody  Neuroscience  Pharmacokinetics  Drug development

Introduction 79 that have been approved by the FDA [2]. The majority
of therapeutic antibodies have been developed for the
Over the past few decades, there has been a surge of treatment of cancer and immune-related diseases. A few
antibody therapeutics that have made their way into clinical antibodies have been FDA approved for the treatment of
practice [1]. The first FDA approved antibody therapeutic, neurological and CNS disorders, such as multiple sclerosis,
approved in 1986, was muromonab, which is an anti-CD3 migraine, and neuromyelitis optica [3–5]. However, the site
antibody used for organ transplantation to prevent graft- of action for most of these therapeutic antibodies is
versus-host disease. As of December 2019, there were at peripheral and they do not need to cross brain barrier for
least 570 antibody therapeutics clinically investigated and pharmacological effects.
There has been increased interest and significant
investments made to develop therapeutic antibodies as a
& Peter Bloomingdale
passive immunotherapy strategy for the treatment of neu-
[email protected]
rodegenerative diseases [6]. However, to date, there has not
1
Pharmacokinetics, Pharmacodynamics, and Drug been any clinical success and no antibody therapeutic has
Metabolism, Merck & Co. Inc., Boston, MA, USA been fully approved for the treatment of a neurodegener-
2
Certara QSP, Oss, The Netherlands ative disease. On June 7th of 2021, the FDA granted
3
Present Address: esqLABS GmbH, Berlin, Germany accelerated approval of aducanumab, a beta-amyloid (Ab)
4 antibody for the treatment of Alzheimer’s disease. Dis-
Certara IDD, Oss, The Netherlands
5
rupted proteostasis is a hallmark of neurodegeneration as
Certara QSP, Canterbury, UK

123
862 Journal of Pharmacokinetics and Pharmacodynamics (2021) 48:861–871

there exists an underlying protein aggregation and clear- demonstrate the efficacy of aducanumab from prior trials
ance problem associated with neurodegenerative disease. has sparked interest in the AD community for conducting
Alzheimer’s disease (AD) is the most prevalent neurode- another clinical trial using a high dose [14].
generative diseases affecting over 46 million people Pharmacokinetic modeling is used for understanding the
worldwide [7]. There are several antibody therapeutics time course of drug exposure through mathematically
under clinical development targeting Ab and tau for the describing kinetic processes that govern drug absorption,
treatment of AD [8]. The hypothesis behind this proposed distribution, and elimination. Over the past couple of
treatment strategy is that antibodies will bind to extracel- decades there has been a shift from classical compart-
lular forms of the pathological proteins, which could mental pharmacokinetic models towards more mechanistic
facilitate clearance and prevent protein aggregation, neu- models that include detailed physiological processes,
ron-to-neuron transmission, and neuronal damage. compartments that represent specific organs, and realistic
Numerous phase 2/3 clinical trials investigating anti-Ab parameter values that are constrained by known physiol-
antibodies for the treatment of AD were negative, as ogy. These types of models are known as physiologically-
clinical endpoints of cognition (e.g. ADAS-Cog and CDR- based pharmacokinetic (PBPK) models. PBPK models date
SB) were not improved [9]. Tau-targeting antibodies are in back to the 1930s, but have become popularized due to the
early clinical development (phase 1/2). In addition to AD, applications in drug discovery and development [15].
there are several other neurodegenerative diseases where There has been an increased interest in reducing PBPK
therapeutic antibodies are being investigated, such as models to less complex structures, known as minimal
Parkinson’s disease, tauopathies, amyotrophic lateral scle- PBPK (mPBPK) models, in order to increases transparency
rosis, and Huntington’s disease. and enhance the ease of application, whilst retaining the
One of the main challenges facing the potential utility of key mechanistic features and behaviors [16].
biologics for the treatment of CNS diseases is achieving As far as we are aware, a mPBPK approach has not been
brain exposures that is above a therapeutic threshold [10]. reported yet for antibodies targeting the CNS. Therefore,
The physicochemical properties of biologics, primarily we have developed a mPBPK model of the brain for
their molecular size, limits their ability to cross brain antibody therapeutics, based on an existing multi-species
barriers, which results in low CNS exposure. Physical platform brain PBPK model [17].
barriers, such as the blood–brain-barrier (BBB) and blood–
cerebral-spinal-fluid-barrier (BCSFB), are designed by
nature to be highly regulated gateways in the body in order Methods
to protect the brain from toxins and pathogens. The BBB is
the barrier interfacing systemic circulation and brain par- Structural model reduction
enchyma. The components that make up the BBB, known and reparameterization
as the neurovascular unit, are vascular endothelial cells
forming tight junctions, basal lamina, pericytes, astrocytes, A mPBPK model of the brain was constructed through the
microglia, and neurons [11]. The BCSFB is the barrier that stepwise reduction of a previously developed PBPK model,
interfaces systemic circulation with ventricular CSF. The further referred to as the ‘‘full’’ model [17]. The reduction
components that make up the BCSFB are vascular approach was implemented to reduce model complexity,
endothelial cells, basement membrane, and epithelial cells while conserving physiological details and model pre-
forming tight junctions [12]. Once in the ventricular sys- dictability. Fourteen compartments representing all non-
tem, antibodies need to cross the ventricular barrier, con- brain organs were combined into a single tissue compart-
sisting of an ependymal cell layer, to reach the brain ment, which is divided into three subcompartments (vas-
parenchyma. The relative leakiness of the BCSFB com- cular, endosomal, and interstitium). The four CSF
pared to the BBB is an attractive hypothesis how antibodies compartments (lateral ventricle, third-fourth ventricle,
enter the brain. However, the routes of antibody disposition cisterna magna, and subarachnoid space) were combined
into the brain and their relative contributions remains into a single CSF compartment. Brain vascular, endosomal,
unclear. and interstitial spaces remained the same. Model parame-
Ability to predict the level of antibody exposure in the ters are provided in Table 1.
human brain required for adequate target engagement is Due to the structural reduction, a reparameterization of
critical to clinical drug development programs. This was certain volumes, flow rates, endosomal uptake rates, and
recently exemplified when aducanumab was discontinued tissue reflection coefficients was required. Parameter val-
following a phase III futility analysis, but subsequent data ues of the minimal model were either conserved or com-
from patients exposed to a higher dose for a longer period bined from the full model. Three parameters for tissue
of time suggested potential efficacy [13]. The failure to volumes, tissue vascular volume ðV T V Þ, tissue endosomal

123
Journal of Pharmacokinetics and Pharmacodynamics (2021) 48:861–871 863

Table 1 Minimal brain PBPK

Parameter Units Mouse Rat Monkey Human
model parameters
FcRn M 4.98E-05 4.98E-05 4.98E-05 4.98E-05
kdeg 1/h 26.6 26.6 26.6 26.6
FR – 0.715 0.715 0.715 0.715
FRB – 0.715 0.715 0.715 0.715
konFcRn 1/M/h 8.06E?07 8.00E?08 7.92E?08 5.59E?08
koffFcRn 1/h 6.55 144 46.8 23.9
VP L 9.44E-04 0.00906 0.187 3.13
VTv L 8.06E-04 0.00789 0.151 1.68
VTe L 1.28E-04 0.00132 0.0286 0.335
VTi L 0.00482 0.0483 0.976 11.1
VBv L 1.07E-05 5.02E-05 0.00207 0.0319
VBE_BBB L 2.09E-06 9.82E-06 4.27E-04 0.00659
VBE_BCSFB L 3.37E-07 1.58E-06 4.27E-05 6.59E-04
VBi L 8.73E-05 4.10E-04 0.0169 0.261
VCSF L 1.93E-05 2.97E-04 0.00926 0.143
VL L 1.13E-04 0.00115 0.0251 0.274
QT L/h 0.361 2.88 20.9 160.5
QB L/h 0.0118 0.0653 1.51 21.5
LT L/h 7.23E-04 0.00577 0.0419 0.321
LB L/h 2.16E-05 1.62E-04 0.00369 0.0345
QB_ECF L/h 1.80E-06 3.00E-05 0.00123 0.0105
QB_CSF L/h 1.98E-05 1.32E-04 0.00246 0.0240
kCLUPT 1/h 0.550 0.550 0.550 0.550
kCLUPB 1/h 0.0195 0.0195 0.0195 0.0195
CLUPT L/h 7.06E-05 7.26E-04 0.0157 0.184
CLUPB L/h 4.74E-08 2.23E-07 9.19E-06 1.42E-04
CLUPBBB L/h 4.08E-08 1.92E-07 8.35E-06 1.29E-04
CLUPBCSFB L/h 6.58E-09 3.09E-08 8.35E-07 1.29E-05
rBBB – 1 1 1 1
rBCSFB – 0.9973 0.9973 0.9973 0.9973
rTv – 0.9172 0.9212 0.9239 0.9233
rTL – 0.2 0.2 0.2 0.2
rISF – 0.2 0.2 0.2 0.2
rCSF – 0.2 0.2 0.2 0.2
SABBB m2 0.0155 0.0155 17 17
SABCSFB m2 0.0025 0.0025 1.7 1.7
fBBB – 0.861 0.861 0.909 0.909
All model parameter values were obtained from the original brain PBPK model described in Chang et al.
[17]. Neonatal Fc receptor (FcRn), antibody endosomal degradation rate (kdeg), fraction of antibody
recycled (FRB), fraction of antibody recycled in brain (FRB), antibody-FcRn association rate (konFcRn),
antibody-FcRn complex dissociation rate (koffFcRn), plasma volume (VP), tissue vascular volume (VTv),
tissue endosomal volume (VTe), tissue interstitial volume (VTi), brain vascular volume (VBv), blood–brain-
barrier endosomal volume (VBE_BBB), blood–CSF-barrier endosomal volume (VBE_BCSFB), brain interstitial
volume (VBi), brain CSF volume (VCSF), tissue blood flow (QT), brain blood flow (QB), tissue lymphatic
flow (LT), brain lymphatic flow (LB), brain ECF flow (QB_ECF), brain CSF flow (QB_CSF), tissue uptake
clearance rate (kCLUPT), brain uptake clearance rate (kCLUPB), tissue uptake clearance (CLUPT), brain
uptake clearance (CLUPB), blood–brain barrier uptake clearance (CLUPBBB), blood–CSF-barrier uptake
clearance (CLUPBCSFB), blood–brain-barrier reflection coefficient (rBBB), blood–CSF-barrier reflection
coefficient (rBCSFB), tissue vascular reflection coefficient (rTv), tissue lymph reflection coefficient (rTL),
brain ISF reflection coefficient (rISF), brain CSF reflection coefficient (rCSF), blood–brain-barrier surface
area (SABBB), blood–CSF-barrier surface area (SABCSFB)

123
864 Journal of Pharmacokinetics and Pharmacodynamics (2021) 48:861–871

volume ðV T E Þ; and tissue interstitial volume ðV T I Þ; were clearance (CLUPT Þ was calculated as the product of tissue
calculated based upon the sum of the respective individual uptake clearance rate ðkCLUPT Þ and total tissue endosomal
tissue compartments of the full model. For example, the volume ðV T E Þ:
calculation for V T V is:
CLUPT ¼ kCLUPT  V T E ð6Þ
V T V ¼ V LungV þ V HeartV þ V KidneyV þ V MuscleV þ V SkinV
þ V AdiposeV þ V ThymusV þ V SIntestineV þ V LIntestineV The same calculation was used for brain uptake clear-
þ V SpleenV þ V PancreasV þ V LiverV þ V BoneV þ V OtherV ance ðCLUPB Þ, the product of brain uptake clearance rate
ðkCLUPB Þ and brain endosomal volume ðV BE Þ.
ð1Þ
CLUPB ¼ kCLUPB  V BE ð7Þ
The same calculation was performed for V T E and V T I .
Brain vascular ðV BV Þ, endosomal ðV BE Þ, and interstitial Brain uptake clearance is split between antibody dis-
ðV BI Þ volumes were unchanged. Four brain CSF volumes, position across the BBB ðCLUPBBB Þ and BCSFB
lateral ventricle ðV LV Þ, third-fourth ventricle ðV TFV Þ, cis- ðCLUPBCSFB Þ, which is scaled based upon their relative sur-
terna magna ðV CM Þ, and subarachnoid space ðV SAS Þ, were face areas (SA).
combined into a single CSF volume: CLUPB ¼ CLUPBBB þ CLUPBCSFB ð8Þ
V CSF ¼ V LV þ V TFV þ V CM þ V SAS ð2Þ
SABBB
f BBB ¼ ð9Þ
The plasma flow rate to tissue ðQT Þ was calculated SABBB þ SABCSFB
based on the difference between lung ðQL Þ and brain ðQB Þ CLUPBBB ¼ kCLUPB  V BE  f BBB ð10Þ
plasma flow:
CLUPBCSFB ¼ kCLUPB  V BE  ð1  f BBB Þ ð11Þ
QT ¼ QL  QB ð3Þ
Note that the total plasma flow is equivalent to the lung where f BBB is the fraction of drug disposition across the
plasma flow. Tissue lymph flow ðLT Þ was calculated as BBB, SABBB is the surface area of the BBB, and SABCSFB is
0.2% of tissue plasma flow, equivalent to the original the surface area of the BCSFB. The endosomal volume for
model: the blood–brain-barrier (V BEBBB Þ is equal to the product of
V BE and f BBB . The endosomal volume for the blood-CSF-
LT ¼ 0:002  QT ð4Þ
barrier (V BEBCSFB Þ is equal to the product of V BE and
The original model used three different values for tissue ð1  f BBB Þ.
reflection coefficients based upon the leakiness of the organ The proportionality constant called LNLF, originally
vasculature, categorized as loose ðrLoose ¼ 0:85Þ, medium described by Shah and Betts in 2012 [18], was used in the
ðrMedium ¼ 0:90Þ, and tight ðrTight ¼ 0:95Þ. In order to full model to calculate lymph flow. This parameter was
conserve this feature, we obtained a single tissue reflection removed from the model because it was found to be
coefficient ðrT V Þ based upon the weighted average reflec- insensitive, as there were no changes in pharmacokinetic
tion coefficient for each organ: profiles upon changing this value. Total lymph flow was
recalculated as the sum of tissue lymph flow ðLT Þ and brain
rT V ¼ rLung  V LungV þ rHeart  V HeartV þ rKidney  V KidneyV
lymph flow ðLB Þ. Brain lymph flow was calculated as the
þ rMuscle  V MuscleV þ rSkin  V SkinV þ rAdipose  V AdiposeV sum of brain ISF and CSF flow:
þ rThymus  V ThymusV þ rSIntestine  V SIntestineV þ rLIntestine
 V LIntestineV þ rSpleen  V SpleenV þ rPancreas  V PancreasV LB ¼ QBECF þ QBCF ð12Þ
þ rLiver
 V LiverV þ rBone  V BoneV þ rOther Hence, physiological lymph flow rates were used instead
 V OtherV =V TissueV of an empirical proportionality constant.
ð5Þ
Comparison of minimal and full PBPK model
A single value was used for the lymph reflection coef-
simulations
ficients in all tissues in the full model, which was the same
value used in the minimal model ðrT L ¼ 0:2Þ.
Simulations were performed to assess the accuracy of the
All clearances and rates were conserved between the full
minimal model for capturing the behavior of the full
and minimal models. However, in an attempt to avoid
model. Single ascending dose (1, 3, 10, 30, 100 mg/kg)
confusing nomenclature going forward, we have redefined
simulations were performed for each species (mouse, rat,
the symbol for uptake clearance rates from CLUP to kCLUP ,
monkey, and human) up to 1000 h, using an output step
since this value is a rate constant with units of inverse time
size of 0.01 h. Doses were administered directly into the
and not units of flow, volume per time. Tissue uptake

123
Journal of Pharmacokinetics and Pharmacodynamics (2021) 48:861–871 865

plasma compartment to emulate intravenous (IV) dosing. Results

Predictions using the full and minimal models were over-
laid for a direct comparison. Additionally, area under the Brain minimal PBPK model
concentration time curve (AUC) for plasma and CNS
compartments were calculated for each simulated profile The brain mPBPK model contains 16 differential equations
using the trapz function in MATLAB. The AUC ratio as compared to 100 in the original model. Brain mPBPK
between the minimal and full model simulations were model diagram and equations are shown in Figs. 1 and 2,
calculated: respectively. Antibody concentrations in plasma go to the
AUC Minimal brain and non-brain tissues at flow rates of QB and QT,
AUC Ratio ¼ ð13Þ while returning flow rates are lower by a difference of
AUC Full
lymph flow (QB–LB and QT–LT). In the tissue vascular
An AUC ratio of 1 would indicate that predicted anti- space, antibody can traverse blood vessel endothelium
body exposure by the minimal model is identical to the full through paracellular and transcellular routes. Mathemati-
model. An AUC ratio of[ 1 or\ 1 indicates that predicted cally, paracellular transport is described as a function of the
antibody exposure by the minimal model is greater than or tissue lymph flow and tissue vascular reflection coefficient,
less than the full model, respectively. LT  ð1  rT V Þ. Transcellular transport via pinocytosis is
To determine the computational speed for executing the described as the tissue uptake clearance (CLUPT Þ from the
minimal and full PBPK models, the tic–toc function in tissue vasculature into the endosomal space. In the endo-
MATLAB was implemented before (tic) and after (toc) somal space, antibody is able to bind to FcRn to form an
calling the model function. antibody-FcRn complex and is either taken up into the
tissue interstitial space or recycled back to the tissue vas-
Sensitivity analysis culature, where FR is the fraction of FcRn-bound antibody
recycled to the tissue vascular space. Unbound antibody in
A one-at-a-time sensitivity analysis method was utilized to the endosome is subject to lysosomal degradation at a rate
assess the sensitivity of antibody exposure when modu- of kdeg. Antibody then leaves the tissue interstitial space by
lating parameter values. Parameters were sampled from a the lymphatic system, which is a function of tissue lymph
log normal distribution (l = 0, r = 0.25), using the logn- flow and tissue lymph reflection coefficient: LT  ð1  rT L Þ
rnd function in MATLAB. 1000 and 100 parameter sets or by uptake clearance (CLUPT Þ back into endosomal
were generated for 27 and 25 parameters in the minimal compartment.
and full model, respectively. Simulations were performed Antibody in the brain vasculature space can cross the
for each parameter set for a 70 kg human administered two brain barriers, BBB and BCSFB. Antibodies trans-
10 mg/kg over 1000 h, using a step size of 1 h. ported across the BBB enter the brain ISF, whereas anti-
AUC was determined, using the trapz function in bodies that move across the BCSFB enter the CSF. The
MATLAB, for three compartments: (1) plasma, (2) CSF, paracellular transport across the BBB and BCSFB is gov-
and (3) ISF. A metric of sensitivity (S), representing the erned by brain vascular reflection coefficients rBBB and
percentage change in antibody exposure, was calculated as rBCSFB . Mathematically, antibody flow across the BBB and
follows: BCSFB via paracellular transport is defined by the func-
 
AUC Perturb tions: QBECF  ð1  rBBB Þ and QBCSF  ð1  rBCSFB Þ. Where,
Sx ð % Þ ¼  1  100% ð14Þ
AUC Base QBECF and QBCSF are the brain extracellular fluid and cere-
bral spinal fluid (CSF) flow rates. Due to the tight nature of
where Sx represents the percentage change in antibody
the neurovascular unit, rBBB was fixed to one, as originally
exposure by modulation of parameter x. AUCBase and
described, to represent the absence of paracellular transport
AUCPertrub represent the area under the concentration–time
across the BBB under normal conditions. The paracellular
curve before and after perturbation of parameter x,
transport of antibodies across the BCSFB is thought to be
respectively. Figures were created in R, using ggplot2.
possible, although quite limited, which is supported by an
Total and tissue blood flow were excluded from the
estimated value of 0.9974 for rBCSFB [17]. Transcellular
analysis for the full model, due to issues where simulations
transport across the brain via pinocytosis is described by
would take a long time and not complete. The lung was
two uptake clearances, CLUPBBB and CLUPBCSFB , and the
selected as the tissue to represent the sensitivity of the
fraction of antibody recycled to the brain vascular space
tissue vascular ðV T V Þ, tissue endosomal volume ðV T E Þ, and
(FRB). Antibodies in the brain endosomal spaces that are
tissue interstitial volume ðV T I Þ for the full PBPK model.
not bound to FcRn are degraded at a rate of kdeg. As
described by the model, antibodies in the brain ISF and

123
866 Journal of Pharmacokinetics and Pharmacodynamics (2021) 48:861–871

Fig. 1 Brain mPBPK model structure. The model contains 16 crosses two brain barriers, BBB and BCSFB. Antibodies that cross the
compartments and three regions: plasma (red), brain (blue) and BBB and BCSFB enter the brain ISF and CSF, respectively.
non-brain tissues (green). Antibodies in plasma flow between brain Paracellular transport across the BBB and BCSFB is governed by
and non-brain tissues. In tissue vascular, antibody travels transcellu- brain vascular reflection coefficients that represent the leakiness of the
larly through the endosomal space or paracellularly by directly vasculature space. Transcellular transport across the brain via
entering tissue interstitium. In the endosome, antibody binds FcRn to pinocytosis is described by uptake clearance processes. Antibody in
form an antibody-FcRn complex, which either is taken up into the brain endosomal spaces is either recycled via FcRn or eliminated via
tissue interstitial space or recycled back to the tissue vasculature. lysosomal degradation. Antibody in brain ISF and CSF can be cleared
Unbound antibody in the endosome is cleared through lysosomal back to systemic circulation via the glymphatic system. Diagram was
degradation. Antibody in tissue interstitial space leaves via lymphatic drawn using Inkscape (Color figure online)
flow as well as via endosomal uptake. Antibody in brain vasculature

CSF undergo three kinetic processes: (1) uptake into brain performed (1, 3, 10, 30, 100 mg/kg) for human (Fig. 3).
endosomal space, (2) flow between brain ISF and CSF Minimal (dotted lines) and full (solid lines) PBPK models
compartments, and (3) glymphatic clearance. The flow rate predictions nicely overlay across all doses for plasma, ISF,
between the ISF and CSF compartments was set to the and CSF compartments. CSF PK predictions from the
brain ECF flow rate QBECF . Glymphatic clearance from mPBPK model were compared against the full PBPK
brain ISF and CSF compartments are described by the model for two different CSF compartments, lateral ven-
following functions:QBECF  ð1  rBISF Þ and tricle (Fig. 3c) and subarachnoid space (Fig. 3d). Minimal
QBCSF  ð1  rBCSF Þ. Lastly, flow from the lymph compart- and full PBPK model predictions also overlay precisely
ment circulates back into systemic circulation ðLB þ LT Þ. across different preclinical species, mouse (Fig. S1), rat
(Fig. S2), and non-human primates (NHP) (Fig. S3).
Minimal PBPK model performance AUCs were determined for minimal and full PBPK
model simulations across all doses and species for plasma,
Model performance was assessed by comparing mPBPK ISF, and CSF compartments. AUC ratios, minimal divided
model predictions against simulated data from the full by full, were calculated and are reported in Table 2. AUC
PBPK model. A single ascending dose simulation was ratios range between 0.94 and 1.22, but generally are close

123
Journal of Pharmacokinetics and Pharmacodynamics (2021) 48:861–871 867

Fig. 2 Brain mPBPK model

equations. Antibody
concentration in (1) plasma, (2)
tissue vascular, (3) tissue
endosome (unbound), (4) tissue
endosome (FcRn-bound), (6)
tissue interstitium, (7) brain
vascular, (8) BBB endosome
(unbound), (9) BBB endosome
(FcRn-bound), (11) brain
interstitium, (12) BCSFB
endosome (unbound), (13)
BCSFB endosome (FcRn-
bound), (15) brain CSF, (16)
lymph. FcRn concentration in
(5) tissue endosome, (10) BBB
endosome, and (14) BCSFB
endosome. Equations written
using Mathcha

to 1, which indicates that PK predictions are comparable Sensitivity analysis

between models and the reduction methodology did not
significantly impact model predictability. The reduced The sensitivity of minimal PBPK model parameters for
model seemed to better retain model predictability for rats altering plasma (Fig. 5a), brain ISF (Fig. 5b), and brain
relative to other species as evidenced by the AUC ratios for CSF (Fig. 5c) exposures are displayed in Fig. 5. For anti-
three of four compartments being equal to 1. The minimal body exposure in plasma, the most sensitive parameter was
model slightly overpredicts exposure in the ventricular the tissue endosomal volume (VTE) followed by the
CSF compartment (AUC ratio = 1.2). reflection coefficient for the tissue vasculature (rTv). The
The model reduction method enabled a significant next sensitive parameters relate to FcRn binding (konFcRn/
improvement in computational speed. A single simulation koffFcRn), the fraction of the antibody FcRn complex that
on average took 37.5 s for the full model, whereas the recycles (FR), and the endosomal degradation rate (kdeg).
minimal model took 3.4 s (Fig. 4). Hence, simulations of Antibody exposure in brain is governed primarily by two
the reduced model were approximately 11 times faster. parameters, the BBB and BCSFB reflection coefficients.
Although this seems like a modest improvement for a The BBB reflection coefficient (rBBB) is most sensitive
single simulation, the computational burden for parameter towards brain ISF exposures, whereas the BCSFB reflec-
estimation, Monte Carlo simulations, and other sampling tion coefficient (rBCSFB) is most sensitive towards brain
methodologies could significantly improve from the order CSF exposures. Other notable sensitive parameters include
of days to hours. processes related to FcRn mediated binding and recycling,
endosomal volume and degradation rate, and the rate of
pinocytosis at brain barriers. Several parameters were

123
868 Journal of Pharmacokinetics and Pharmacodynamics (2021) 48:861–871

Fig. 3 Minimal versus full PBPK model predictions. Antibody (blue), 30 (purple), and 100 (pink) mg/kg for a duration of 1000 h.
concentrations in human a serum, b brain interstitial fluid (ISF), Dotted and solid lines represent minimal and full PBPK model
c brain CSF in lateral ventricle (LV), and d brain CSF in subarachnoid simulations, respectively (Color figure online)
space (SAS). Five IV doses were simulated 1 (red), 3 (green), 10

insensitive and result in less than a 1% change in AUC (left Discussion

of the line in Fig. 5). The sensitivity of model parameters
in the full PBPK model for altering plasma, brain ISF, and We have constructed a mPBPK model with detailed brain
brain CSF exposures are displayed in Supplementary physiology. We have shown that the model reduction
Fig. S4. The rank order and magnitude for the sensitivity of methodology did not lead to any appreciable change in
parameters between the minimal and full PBPK models model predictability, since the predicted plasma and brain
were similar. However, parameters related to the tissue concentration–time profiles from the minimal model are
compartment (VTe, rT, VTi, VTv) were more sensitive in comparable to the full model. The minimal model also
the minimal model compared to the full model. This could displayed significant improvements in computational
be due to the fact that the tissue compartment in the min- speed, which could save time when conducting parameter
imal model represents all combined tissues and the tissue estimation, sensitivity analyses, Monte Carlo simulations,
compartment used in the sensitivity analysis of the full population analyses, and other methodologies. The sim-
model only represents a single tissue. plicity of the mPBPK model could improve upon model
transparency, ease of understanding, and future utilization.
A minimalistic pharmacokinetic model could enable an

123
Journal of Pharmacokinetics and Pharmacodynamics (2021) 48:861–871 869

Table 2 AUC ratios of minimal vs full PBPK model simulations decreasing the frequency of drug administration [20]. The
Mouse Rat NHP Human
role of differences in FcRn binding due these mutations on
antibody half-life could be evaluated using the model.
Plasma 0.96 1.00 1.06 1.05 However, FcRn binding improvements do not always
ISF 0.94 1.00 1.06 1.04 translate into longer half-life as other physicochemical
CSF (LV) 1.17 1.19 1.22 1.21 characteristics that govern endosomal trafficking dynamics
CSF (SAS) 0.96 1.00 1.06 1.04 also play a role in drug elimination [21, 22]. This is a
current limitation of the model as improving FcRn affinity
will always result in PK predictions with improved half-
life. Understanding the structural and physicochemical
characteristics of antibodies that govern uptake into the
endosome, FcRn binding and recycling, and endosomal
degradation would provide insights into inter-antibody
pharmacokinetic differences. Another consideration is the
competition for binding to FcRn by endogenous IgG. There
is a lack of experimental measurements for the concen-
tration of FcRn, which was originally estimated, and the
fraction of antibody recycled via FcRn [18]. Additionally,
there is some uncertainty around the parameter for endo-
somal volume, which could range from 0.034 to 0.5% of
total tissue volume [23, 24]. These uncertainties should be
considered when including endogenous IgG competition.
Currently, the model is only applicable to describe the
pharmacokinetics of antibodies. The model could be
repurposed for other biologics, however there are a few
Fig. 4 Computational speed improvement. Simulation time differ- points to consider. First, antibodies are able to bind to FcRn
ences between the original/full (red) and reduced/minimal (blue) to prevent degradation via the lysosome, which may not be
PBPK models. Simulations were performed for 5 doses over 1000 h applicable to other biologics. Therefore, one would need to
for each species, using a solver step size of 0.01 h. The differential determine potential interactions with FcRn or other recy-
equation solver used was ode15s using a relative and absolute
tolerance of 2.3E-14 and 1E-22, respectively. Each bar represents cling mechanisms and potential differences in the endo-
the average simulation time across four simulations, one simulation somal degradation rate. Second, biologics that are small in
per species. Processor: Intel i7-8700K CPU @ 3.70 GHz (Color molecular weight could be subject to clearance via
figure online) glomerular filtration, which has been described in a recent
paper investigating clearance differences between antibody
easier integration with platform quantitative systems fragments and IgG in mice [25]. Biologics smaller than
pharmacology (QSP) models of neurological disease, albumin (66.5 kDa) begin to exhibit enhanced clearance
which have been increasing in popularity over the last few via glomerular filtration [26]. Third, the vascular perme-
years [19]. The model is available as MATLAB code and a ability of biologics could be different than antibodies.
SimBiology project file (MATLAB version 2020b). For Therefore, parameters that govern the uptake of antibody
SimBiology, the parameter sets for the individual species into tissues, such as the tissue and brain reflection coeffi-
are captured as separate variants while dosing is handled cients, would have to be re-estimated. In future, as data
through the Dose tab. Drug specific parameters can be accumulates, these parameters could be defined as a
added as a new variant and combined with the species function of molecular size and other relevant physico-
variant of interest for simulations. chemical characteristics.
Model applications include a priori predictions of anti- One of the most significant applications is expanding the
body pharmacokinetics in plasma and brain in mouse, rat, mPBPK brain model to include targets of interest, which
monkey, and human, which could be used to evaluate drug would expand model utility and enable predictions to
exposure differences for various dosing regimens. More- understand target engagement, determine efficacious dos-
over, the model could be expanded to describe in detail ing regimens for clinical studies, support affinity opti-
distribution into any other tissue/organ of choice which mization by understanding desirable affinity ranges, and
was included in the full model. Engineering Fc regions to evaluate levels of target engagement amongst several drug
extend the half-life of circulating antibodies has become a candidates. Expanding the model to include target-medi-
popular method for increasing drug exposure and ated processes introduces additional parameters that need

123
870 Journal of Pharmacokinetics and Pharmacodynamics (2021) 48:861–871

Fig. 5 Sensitivity analysis of minimal brain PBPK model for antibody represents 1000 simulations. Each parameter (one at a time) was
exposure in a plasma, b brain ISF, and c brain CSF. The change in multiplied by a value sampled from a log-normal distribution (l = 0,
AUC for each parameter is displayed as a box and whisker that r = 0.25)

to be defined, such as target expression and turnover as Multiple clinical trials investigating therapeutic anti-
well as antibody-target association and dissociation rates bodies for the treatment of neurodegenerative disease have
[27]. Various experimental technologies, such as Biacore used concentrations of antibody and target engagement in
and KinExA, can be used to measure antibody affinity [28]. the CSF as a surrogate for expected concentrations and
Careful consideration should be taken in the experimental engagement in the brain [10]. However, this may not be an
designs and use of these affinity measurements as there can entirely appropriate assumption as there could be differ-
be a significant amount of variability between/within ences in drug pharmacokinetics and target dynamics (ex-
assays and potential disconnects between in vitro mea- pression and turnover) in brain ISF, the site of drug action,
surements and in vivo observations. Common drug targets compared to CSF. Since it is experimentally impractical to
for neurodegenerative diseases are proteins that self-ag- sample drug concentrations and engagement in the human
gregate. Applications could include evaluating relative brain, a PBPK model of the brain expanded to include the
target engagement to monomeric, oligomeric, and insol- target of interest enables drug exposure and target
uble species. The target of interest could exist in plasma, engagement predictions in an otherwise unobservable
non-brain tissues, brain ISF and CSF. If the target exists in compartment. The minimal PBPK model presented here
plasma, the model should also be expanded to include the could be expanded to include drug targets to support pre-
target in plasma as well as brain and tissue vascular com- clinical and clinical drug development programs investi-
partments. The target and antibody target complex could gating antibody therapeutics for the treatment of
follow the same kinetic processes as antibodies or a sim- neurological diseases.
plifying assumption could be implemented where the target
Supplementary Information The online version contains
and antibody target complex don’t distribute between
supplementary material available at https://doi.org/10.1007/s10928-
compartments. The antibody target complex could follow 021-09776-7.
the same elimination as a typical antibody, which is often
the case for soluble targets that have a relatively lower Acknowledgements We would like to acknowledge the University at
molecular weight than an antibody. For membrane targets, Buffalo Center for Protein Therapeutics consortium and the labora-
tory of Dr. Dhaval Shah for providing access to the platform brain full
a key parameter to determine is the antibody-receptor PBPK model code.
complex internalization rate as the target could impact
antibody elimination.

123
Journal of Pharmacokinetics and Pharmacodynamics (2021) 48:861–871 871

Open Access This article is licensed under a Creative Commons 15. Jones H, Rowland-Yeo K (2013) Basic concepts in physiologi-
Attribution 4.0 International License, which permits use, sharing, cally based pharmacokinetic modeling in drug discovery and
adaptation, distribution and reproduction in any medium or format, as development. CPT Pharmacomet Syst Pharmacol 2:e63. https://
long as you give appropriate credit to the original author(s) and the doi.org/10.1038/psp.2013.41
source, provide a link to the Creative Commons licence, and indicate 16. Cao Y, Jusko WJ (2012) Applications of minimal physiologi-
if changes were made. The images or other third party material in this cally-based pharmacokinetic models. J Pharmacokinet Pharma-
article are included in the article’s Creative Commons licence, unless codyn 39(6):711–723. https://doi.org/10.1007/s10928-012-9280-
indicated otherwise in a credit line to the material. If material is not 2
included in the article’s Creative Commons licence and your intended 17. Chang HY, Wu S, Meno-Tetang G, Shah DK (2019) A transla-
use is not permitted by statutory regulation or exceeds the permitted tional platform PBPK model for antibody disposition in the brain.
use, you will need to obtain permission directly from the copyright J Pharmacokinet Pharmacodyn 46(4):319–338. https://doi.org/10.
holder. To view a copy of this licence, visit http://creativecommons. 1007/s10928-019-09641-8
org/licenses/by/4.0/. 18. Shah DK, Betts AM (2012) Towards a platform PBPK model to
characterize the plasma and tissue disposition of monoclonal
antibodies in preclinical species and human. J Pharmacokinet
Pharmacodyn 39(1):67–86. https://doi.org/10.1007/s10928-011-
References 9232-2
19. Bloomingdale P, Karelina T, Cirit M, Muldoon SF, Baker J,
1. Nelson AL, Dhimolea E, Reichert JM (2010) Development trends McCarty WJ, Geerts H, Macha S (2021) Quantitative systems
for human monoclonal antibody therapeutics. Nat Rev Drug pharmacology in neuroscience: Novel methodologies and tech-
Discov 9(10):767–774. https://doi.org/10.1038/nrd3229 nologies. CPT Pharmacomet Syst Pharmacol. https://doi.org/10.
2. Lu RM, Hwang YC, Liu IJ, Lee CC, Tsai HZ, Li HJ, Wu HC 1002/psp4.12607
(2020) Development of therapeutic antibodies for the treatment 20. Mackness BC, Jaworski JA, Boudanova E, Park A, Valente D,
of diseases. J Biomed Sci 27(1):1. https://doi.org/10.1186/ Mauriac C, Pasquier O, Schmidt T, Kabiri M, Kandira A,
s12929-019-0592-z Radosevic K, Qiu H (2019) Antibody Fc engineering for
3. Voge NV, Alvarez E (2019) Monoclonal antibodies in multiple enhanced neonatal Fc receptor binding and prolonged circulation
sclerosis: present and future. Biomedicines. https://doi.org/10. half-life. MAbs 11(7):1276–1288. https://doi.org/10.1080/
3390/biomedicines7010020 19420862.2019.1633883
4. Tso AR, Goadsby PJ (2017) Anti-CGRP monoclonal antibodies: 21. Maas BM, Cao Y (2018) A minimal physiologically based
the next era of migraine prevention? Curr Treat Options Neurol pharmacokinetic model to investigate FcRn-mediated mono-
19(8):27. https://doi.org/10.1007/s11940-017-0463-4 clonal antibody salvage: Effects of Kon, Koff, endosome traf-
5. Frampton JE (2020) Eculizumab: a review in neuromyelitis ficking, and animal species. MAbs 10(8):1322–1331. https://doi.
optica spectrum disorder. Drugs 80(7):719–727. https://doi.org/ org/10.1080/19420862.2018.1506648
10.1007/s40265-020-01297-w 22. Gurbaxani B, Dostalek M, Gardner I (2013) Are endosomal
6. Proceedings from the Pharmacia Symposium No. 1. Free radicals trafficking parameters better targets for improving mAb phar-
in medicine and biology. Uppsala, Sweden, October 8–9, 1979 macokinetics than FcRn binding affinity? Mol Immunol
(1980). Acta Physiol Scand Suppl 492:1–168 56(4):660–674. https://doi.org/10.1016/j.molimm.2013.05.008
7. Group GBDNDC (2017) Global, regional, and national burden of 23. Garg A, Balthasar JP (2007) Physiologically-based pharmacoki-
neurological disorders during 1990–2015: a systematic analysis netic (PBPK) model to predict IgG tissue kinetics in wild-type
for the Global Burden of Disease Study 2015. Lancet Neurol and FcRn-knockout mice. J Pharmacokinet Pharmacodyn
16(11):877–897. https://doi.org/10.1016/S1474-4422(17)30299-5 34(5):687–709. https://doi.org/10.1007/s10928-007-9065-1
8. Plotkin SS, Cashman NR (2020) Passive immunotherapies tar- 24. Chen Y, Balthasar JP (2012) Evaluation of a catenary PBPK
geting Abeta and tau in Alzheimer’s disease. Neurobiol Dis model for predicting the in vivo disposition of mAbs engineered
144:105010. https://doi.org/10.1016/j.nbd.2020.105010 for high-affinity binding to FcRn. AAPS J 14(4):850–859. https://
9. Panza F, Lozupone M, Logroscino G, Imbimbo BP (2019) A doi.org/10.1208/s12248-012-9395-9
critical appraisal of amyloid-beta-targeting therapies for Alzhei- 25. Li Z, Krippendorff BF, Shah DK (2017) Influence of molecular
mer disease. Nat Rev Neurol 15(2):73–88. https://doi.org/10. size on the clearance of antibody fragments. Pharm Res
1038/s41582-018-0116-6 34(10):2131–2141. https://doi.org/10.1007/s11095-017-2219-y
10. Pardridge WM (2019) Blood-brain barrier and delivery of protein 26. Norden AG, Lapsley M, Lee PJ, Pusey CD, Scheinman SJ, Tam
and gene therapeutics to brain. Front Aging Neurosci 11:373. FW, Thakker RV, Unwin RJ, Wrong O (2001) Glomerular pro-
https://doi.org/10.3389/fnagi.2019.00373 tein sieving and implications for renal failure in Fanconi syn-
11. Abbott NJ, Patabendige AA, Dolman DE, Yusof SR, Begley DJ drome. Kidney Int 60(5):1885–1892. https://doi.org/10.1046/j.
(2010) Structure and function of the blood-brain barrier. Neuro- 1523-1755.2001.00016.x
biol Dis 37(1):13–25. https://doi.org/10.1016/j.nbd.2009.07.030 27. Mager DE, Jusko WJ (2001) General pharmacokinetic model for
12. Freskgard PO, Urich E (2017) Antibody therapies in CNS dis- drugs exhibiting target-mediated drug disposition. J Pharma-
eases. Neuropharmacology 120:38–55. https://doi.org/10.1016/j. cokinet Pharmacodyn 28(6):507–532. https://doi.org/10.1023/a:
neuropharm.2016.03.014 1014414520282
13. Howard R, Liu KY (2020) Questions EMERGE as Biogen claims 28. Drake AW, Abdiche YN, Papalia GA (2018) Biophysical con-
aducanumab turnaround. Nat Rev Neurol 16(2):63–64. https:// siderations for development of antibody-based therapeutics. In:
doi.org/10.1038/s41582-019-0295-9 Tabrizi MA, Bornstein GG, Klakamp SL (eds) Development of
14. Knopman DS, Jones DT, Greicius MD (2020) Failure to antibody-based therapeutics. Springer, Singapore, pp 71–132
demonstrate efficacy of aducanumab: an analysis of the
EMERGE and ENGAGE trials as reported by Biogen, December
2019. Alzheimers Dement. https://doi.org/10.1002/alz.12213 Publisher’s Note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.

123