Prediction of Efficacy of Vabicaserin, a 5-HT2C Agonist, for the Treatment of Schizophrenia Using a Quantitative Systems Pharmacology Model

Abstract

A quantitative systems pharmacology model that combines in vitro /preclinical neurophysiology data, human imaging data, and patient disease information was used to blindly predict steady‐state clinical efficacy of vabicaserin, a 5‐HT 2C full agonist, in monotherapy and, subsequently, to assess adjunctive therapy in schizophrenia. The model predicted a concentration‐dependent improvement of positive and negative syndrome scales (PANSS) in schizophrenia monotherapy with vabicaserin. At the exposures of 100 and 200 mg b.i.d., the predicted improvements on PANSS in virtual patient trials were 5.12 (2.20, 8.56) and 6.37 (2.27, 10.40) (mean (95% confidence interval)), respectively, which are comparable to the observed phase IIa results. At the current clinical exposure limit of vabicaserin, the model predicted an ~9‐point PANSS improvement in monotherapy, and <4‐point PANSS improvement adjunctive with various antipsychotics, suggesting limited clinical benefit of vabicaserin in schizophrenia treatment. In conclusion, the updated quantitative systems pharmacology model of PANSS informed the clinical development decision of vabicaserin in schizophrenia.
CPT Pharmacometrics Syst. Pharmacol . (2014) 3, e111; doi: 10.1038/psp.2014.7 ; published online 23 April 2014

Full text

Citation: CPT Pharmacometrics Syst. Pharmacol. (2014) 3, e111; doi:10.1038/psp.2014.7
© 2014 ASCPT All rights reserved 2163-8306/14
www.nature.com/psp
Vabicaserin (SCA-139) is a novel 5-HT2C agonist.1 The
5-HT2C agonism has been hypothesized to have therapeu-
tic potentials in a wide range of psychiatric disorders based
on evidence from preclinical animal models.2,3 Unlike most
agents currently developed for the treatment of schizophre-
nia, vabicaserin does not directly involve targeting dopamine
receptors. Vabicaserin has in vitro functional selectivity for
5-HT2C over 5-HT2A, 5-HT2B, and other receptors.1 Vabica-
serin decreases dopamine levels of nucleus accumbens
without affecting striatal dopamine in rodents. This profile is
consistent with potential efficacy in the treatment of psychotic
symptoms of schizophrenia.4,5 Chronic administration of vabi-
caserin significantly decreases the number of spontaneously
active mesocorticolimbic dopamine neurons without affecting
nigrostriatal dopamine neurons, consistent with the effects of
atypical antipsychotics.4 Unlike atypical antipsychotics, acute
administration of 5-HT2C agonists in rodents reduces meso-
corticolimbic dopaminergic activity,4 suggesting that vabica-
serin could have a rapid onset of action. Vabicaserin also
increases extracellular glutamate content in medial prefrontal
cortex of rats, an effect which may provide improved cogni-
tive function.5 Results from preclinical studies also suggest
that 5-HT2C agonists could be effective in improving mood
disorders and cognitive impairment associated with schizo-
phrenia, without producing extrapyramidal side effects or
weight gain.4 Therefore, vabicaserin offers the possibility of
a new antipsychotic medication with broader efficacy (e.g.,
cognitive symptoms) as well as improved safety and tolerabil-
ity over existing antipsychotic agents. However, in a recent
phase IIa clinical trial, vabicaserin demonstrated only moder-
ate efficacy in schizophrenia as monotherapy.5 These results
raised questions regarding whether additional clinical studies
in monotherapy and in adjunctive therapy of schizophrenia
should be conducted.
Given the limitations of animal models in predicting effi-
cacy in schizophrenia, such as fundamental differences in
neurotransmitter circuitry between rodents and humans and
the incomplete representation of the full human pathology,6
a quantitative systems pharmacology model of schizophre-
nia was used to predict the clinical efficacy of vabicaserin in
monotherapy and subsequently to assess the potential effi-
cacy in adjunctive therapy in schizophrenia. The model was
blinded to the phase IIa data to reduce prediction bias.
The quantitative systems pharmacology model of schizo-
phrenia is a computer-based mechanistic disease modeling
platform that combines in vitro/preclinical neurophysiology
information with human imaging, postmortem, and clinical
data.7,8 The model has successfully predicted the efficacy on
total positive and negative syndrome scales (PANSS) of a
phase II study of the novel investigative drug JNJ37822681,
a highly selective low-affinity dopamine D2 antagonist, in
schizophrenia.9 The model has also predicted the relative,
but not the absolute, clinical effect of another antipsychotic
drug, ocaperidone.9
The purpose of this study was to utilize a quantitative sys-
tems pharmacology model to predict steady-state clinical effi-
cacy of vabicaserin as a monotherapy and, subsequently, to
estimate the efficacy of vabicaserin as an adjunctive therapy in
schizophrenia and to inform whether additional clinical studies
with vabicaserin are warranted. To accomplish this, the previ-
ously developed quantitative systems pharmacology model
of PANSS7–9 was extended to incorporate 5-HT2C receptor
effects in three ways: (i) effect on dopamine firing frequency,
(ii) effect on cholinergic striatal interneurons, and (iii) effect
Received 30 September 2013; accepted 6 February 2014; advance online publication 23 April 2014. doi:10.1038/psp.2014.7
2163-8306
e111
CPT Pharmacometrics Syst. Pharmacol.
10.1038/psp.2014.7
Original Article
23April2014
3
30September2013
6February2014
2014
© 2014 ASCPT
QSP Modeling of Vabicaserin in Schizophrenia
Liu et al.
A quantitative systems pharmacology model that combines in vitro/preclinical neurophysiology data, human imaging data,
and patient disease information was used to blindly predict steady-state clinical efficacy of vabicaserin, a 5-HT2C full agonist,
in monotherapy and, subsequently, to assess adjunctive therapy in schizophrenia. The model predicted a concentration-
dependent improvement of positive and negative syndrome scales (PANSS) in schizophrenia monotherapy with vabicaserin.
At the exposures of 100 and 200 mg b.i.d., the predicted improvements on PANSS in virtual patient trials were 5.12 (2.20, 8.56)
and 6.37 (2.27, 10.40) (mean (95% confidence interval)), respectively, which are comparable to the observed phase IIa results.
At the current clinical exposure limit of vabicaserin, the model predicted an ~9-point PANSS improvement in monotherapy,
and <4-point PANSS improvement adjunctive with various antipsychotics, suggesting limited clinical benefit of vabicaserin in
schizophrenia treatment. In conclusion, the updated quantitative systems pharmacology model of PANSS informed the clinical
development decision of vabicaserin in schizophrenia.
CPT Pharmacometrics Syst. Pharmacol. (2014) 3, e111; doi:10.1038/psp.2014.7; published online 23 April 2014
1Clinical Pharmacology, Pfizer, Groton, Connecticut, USA; 2Neuroscience Research Unit, Pfizer, Cambridge, Massachusetts, USA; 3In Silico Biosciences, Lexington,
Massachusetts, USA; 4Oregon Health and Science University, Portland, Oregon, USA; 5Perelman School of Medicine, University of Pennsylvania, Philadelphia,
Pennsylvania, USA. Correspondence: J Liu (jing.liu@pfizer.com)
Prediction of Efficacy of Vabicaserin, a 5-HT2C Agonist,
for the Treatment of Schizophrenia Using a Quantitative
Systems Pharmacology Model
J Liu1, A Ogden1, TA Comery2, A Spiros3, P Roberts3,4 and H Geerts3,5
ORIGINAL ARTICLE

CPT: Pharmacometrics & Systems Pharmacology
QSP Modeling of Vabicaserin in Schizophrenia
Liu et al.
2
on γ-aminobutyric acid (GABA) interneurons. The model was
then recalibrated to determine the best coupling parameters
for these mechanisms. The computer-based quantitative sys-
tems pharmacology model also incorporated polypharmacol-
ogy of the existing antipsychotics that sometimes involves an
interaction with 5-HT2C receptors, which may interact in a non-
linear fashion with vabicaserin. The model was blinded to the
previously generated vabicaserin phase IIa efficacy data5 in
schizophrenia monotherapy to minimize prediction bias.
RESULTS
Model-predicted 5-HT2C agonist effects on striatal dopa-
mine neuron firing
The effect of full 5-HT2C receptor agonism on striatal dopa-
mine firing was calibrated based on literature data on the
interventions (knockout,10 antagonism,11 and agonism12) on
5-HT2C receptor and subsequent effects on striatal dopa-
mine levels. The activation levels of 5-HT2C receptors and
the reductions in dopamine firing were determined using the
receptor competition model (top-left oval in Figure 1, modified
from previously published models7–9,13). The model-predicted
relationship between dopamine firing frequency and either
free dopamine levels in the synaptic cleft or tracer binding at
the postsynaptic D2 receptors was determined to be −1.9%
dopamine firing change per 1% 5-HT2C receptor activation
increase. This average value of −1.9% for the slope is within
the biologically relevant range of −1.6 to −3.2% derived from
published preclinical data.10–12,14
Recalibration of the updated PANSS model
As the new 5-HT–mediated physiology was added to the
model platform, and because many antipsychotics that have
been clinically tested show pharmacological effects toward
some of the newly introduced receptors, the updated PANSS
model (version 2.3.1) was recalibrated as compared with the
previous version.7–9 As shown in Figure 2, the recalibration
resulted in a strong correlation between the observed and
the model-predicted PANSS (r2 = 0.6861).
Model-predicted effect of vabicaserin as monotherapy
Figure 3 shows the model-predicted change in PANSS relative
to a range of normalized 5-HT2C receptor activation from 0.5-
to 1.5-fold of the normal baseline receptor activation (63%).
At 5-HT2C receptor activation levels less than 0.9-fold of the
normal baseline, the model predicted worsening of PANSS. At
5-HT2C receptor activation levels greater than and equal to the
normal baseline levels (≥onefold), the model predicted mono-
tonic improvements of PANSS. The improvement of PANSS
was predicted to saturate when 5-HT2C receptor activation is
greater than 1.4-fold of its normal baseline activation, i.e., at
greater than 90% of total 5-HT2C receptor activation.
The predicted changes of PANSS over a range of vabica-
serin free drug concentrations are also shown in Figure 3.
At 18 nmol/l of vabicaserin, the highest average unbound
steady-state concentration achieved but not well tolerated in
all populations,15 the model predicted ~9-point improvement
in PANSS. A maximal anticipated clinical improvement of 12
points for PANSS was predicted to be achieved with 5-HT2C
receptor activation greater than 1.4-fold of the normal base-
line value, which would require a free vabicaserin concentra-
tion of ≥35 nmol/l.
Model-predicted effect of vabicaserin adjunctive with
antipsychotics
The potential additional improvements on PANSS by vabi-
caserin adjunctive to other antipsychotics in schizophre-
nia were predicted by the recalibrated PANSS model. In
Figure 1 Detailed model of the pharmacology in the ventral striatum that is related to the clinical PANSS outcome. 5-HT
2C
receptor activation
in the VTA affects dopaminergic firing and together with 5-HT
3
has direct effects on D
1
and D
2
receptor activations, whereas 5-HT
2C
, 5-HT
6
, and
5-HT
7
all affect muscarinic cholinergic M
1
receptor tone on MSN neurons through their effect on cholinergic interneurons and 5-HT
2C
affects
GABA interneurons which reduces excitability of the MSN. DA, dopamine; GABA, γ-aminobutyric acid; MSN, medium spiny neuron; PANSS,
positive and negative symptoms scale in schizophrenia; VTA, ventral tegmentum area.
VTA/SN
5HT2C
5HT2C
5HT3
M2M2
D2-RD2-R
D2-R
Ksi L K-A
5HT3
D2-R
D2-R
Gating
signal
Stimulating
pyramidal
signal
Glutamatergic spike trains
Glutamatergic spike trains GABA spiny neuron
Striatum
Drugs may affect any
receptors shown
Measure
Kir2
L-Ca L-C1
D1-R
α1A
M1
5HT6
5HT3PFC
5HT2C
5HT7
Background
pyramidal
signal
Spike
train
Info
process
function
D3-R
D1-R
DA competition
model
Amygdala
Hippocampus
DA

www.nature.com/psp
QSP Modeling of Vabicaserin in Schizophrenia
Liu et al.
3
general, the model predicted <4-point additional improve-
ments of PANSS (black-shaded areas in Figure 4, except
for placebo) when vabicaserin was added to the standard-
of-care antipsychotics as cotreatments with haloperidol,
olanzapine, asenapine, or zotepine. No improvement was
predicted with ziprasidone, aripiprazole, quetiapine, risperi-
done, or paliperidone at the unbound exposure of 18 nmol/l,
the current clinical exposure limit of vabicaserin. The gray-
shaded areas in Figure 4 represent the predicted PANSS
improvements by either placebo or the standard care of
antipsychotics alone ranging from 10 to 23 points, which
are generally consistent with the reported clinical efficacy of
these antipsychotics.16,17
Virtual patient trial simulation with vabicaserin
To gain confidence in the model, a comparison of the model
prediction of virtual trials with the observed phase IIa clinical
study results in vabicaserin monotherapy was performed.
The predicted mean improvements of PANSS and the 95%
confidence intervals of the 10 virtual trials were averaged
for the 100 and 200 mg b.i.d. of vabicaserin, respectively,
and are shown in Table 1. The predicted mean (95% con-
fidence interval) improvements of PANSS from the 10 vir-
tual trials were 5.12 (2.20, 8.56) and 6.37 (2.27, 10.40) for
100 and 200 mg b.i.d. of vabicaserin, respectively. These
quantitative systems pharmacology–predicted average
effects are in general agreement with the actual observed
clinical improvements of PANSS in the phase IIa study
(last observation carried forward mean (95% confidence
interval)), 8.57 (1.98, 15.15) and 5.91 (−0.84, 12.2), at 100
and 200 mg b.i.d. of vabicaserin, respectively.5 It is impor-
tant to note that the typical SDs on PANSS improvements
range from 20 to 30 points in the literature.17 Therefore,
the differences between predicted and observed PANSS
improvements are relatively minor, and the 95% confidence
intervals of the predicted PANSS improvements are within
the corresponding observed 95% confidence intervals at
100 and 200 mg b.i.d. of vabicaserin, respectively. Although
the simulations do not fully recapitulate the apparent higher
mean response at the lower dose, the apparent numerical
difference on the observed mean PANSS improvements at
100 and 200 mg b.i.d. of vabicaserin was likely an artifact
of data variability.
DISCUSSION
This report describes the implementation of 5-HT2C receptor
neurophysiology and the impact of 5-HT2C activation changes
by vabicaserin in the updated quantitative systems pharma-
cology model of schizophrenia, the output of which correlates
with the observed PANSS improvements in a phase IIa clini-
cal study. Vabicaserin is a relatively specific 5-HT2C modulator
with modest additional effect on 5-HT2B. The 5-HT2C recep-
tor is an interesting novel target in schizophrenia because
the therapeutic effects of its modulation are presynaptic on
the dopaminergic projections to the striatum and the human
pathology is thought to be presynaptic, rather than postsyn-
aptic.18 Almost all of the currently marketed antipsychotics
interact with the postsynaptic dopamine D2 receptor on stria-
tal medium spiny neurons, thus providing a unique mecha-
nism of action for vabicaserin and potential for differentiation
from existing therapeutic agents.
In this updated PANSS model, the neurophysiology of
5-HT2C was implemented at two different levels: (i) modula-
tion of dopaminergic ventral tegmentum area (VTA) firing and
(ii) modulation of striatal cholinergic tone. The coupling factor
of a 1.9% decrease in dopamine firing per 1% change in
5-HT2C receptor activation is at the lower range of a number
of previously published data on the in vivo coupling, which
ranges from 1.6 to 3.2% decreases in dopamine firing rate per
percent 5-HT2C receptor activation change.10–12,14 However,
this relationship is within the calculated range (1.3–2.3%)
from preclinical and in vitro experiments with vabicaserin,19
which gives a greater confidence to the determination of the
relationship between 5-HT2C receptor activation and changes
in VTA dopamine firing.
The effects of different serotonergic receptor levels,
such as 5-HT2C, 5-HT6, and 5-HT7 on striatal cholinergic
and GABA interneurons,19–21 were further implemented into
the model using preclinical data. The coupling parameters
for the new neurophysiological model are constrained with
calibration using human clinical data.7 Although intuitively
increased 5-HT2C receptor activation could worsen PANSS
through its effect on the Cl− leak,22 the model incorpo-
rates the increased 5-HT2C receptor activation through its
effect on D2 receptor activation and muscarinic M1 receptor
Figure 2 Calibration between model-predicted and clinically
reported PANSS responses for 42 drug–dose combinations with the
updated PANSS model, including the specific neurophysiology of
5-HT
2C
, 5-HT
6
, and 5-HT
7
receptors. PANSS, positive and negative
symptoms scale in schizophrenia.
−35 −30 −25 −20
−5
0
R2 = 0.6861
Predicted PANSS
Observed PANSS
0
−10
−15
−20
−25
−30
−35
−15 −10 −5
Figure 3 Effect of relative changes (normalized to the baseline
value) in 5-HT
2C
receptor activation on the model-predicted changes
in PANSS by vabicaserin as a monotherapy. PANSS, positive and
negative symptoms scale in schizophrenia.
−15
15
−10
10
−5
5
0
5-HT2CR activation change (× normal baseline)
0.4 0.5 0.6 0.7 0.8 0.9 1.1 1.2 1.3 1.4
3816
Vabicaserin (nmol/l)
35
1.5 1.6
Change in PANSS

CPT: Pharmacometrics & Systems Pharmacology
QSP Modeling of Vabicaserin in Schizophrenia
Liu et al.
4
activation, resulting in a monotonic dose–response predic-
tion on PANSS improvement.
In the prediction of the changes of PANSS vs. 5-HT2C
receptor activation by vabicaserin monotherapy in Figure 3,
fixed-effect simulations were performed without random vari-
ability as the objective was to understand the underlying
trend of the relationship. At the clinical exposure limit of 18
nmol/l free vabicaserin, the predicted PANSS improvement
was ~9 points, which is lower than the typical reported effi-
cacy of marketed antipsychotics,16 and the exposure was not
well tolerated in all subject populations.
In the subsequent virtual trial simulations to predict the
PANSS improvement by vabicaserin in the phase IIa study,
the variability of observed vabicaserin exposures and the
30% variability around the biological coupling parameters
were incorporated into the simulations. The predicted mean
and 95% confidence intervals of PANSS improvements from
10 virtual trials of 200 subjects/trial are in the range of cor-
responding observed values at 100 and 200 mg b.i.d. of vabi-
caserin, respectively. It is noted that the observed PANSS
improvements were based on last observation carried for-
ward imputation because there were significant dropouts
due to adverse events, and thus, the observed cases may
not adequately represent the underlying treatment effect of
vabicaserin. The wide ranges of 95% confidence intervals of
the observed PANSS improvements by vabicaserin might be
due to the intrinsic variability of the PANSS end point17 and
further amplified by the wide variability of vabicaserin expo-
sure. Therefore, the apparent higher mean PANSS improve-
ment of 8.57 points at 100 mg b.i.d. than the mean value of
5.91 points at 200 mg b.i.d. was likely an artifact of data vari-
ability with ~70 subjects in each group, and this difference is
considered minimal and not statistically significant given the
wide ranges and the overlap of the 95% confidence intervals
of (1.98, 15.15) and (−0.84, 12.2) at 100 and 200 mg b.i.d.
of vabicaserin, respectively. Overall, the model predicted the
observed ranges of PANSS improvements at the two vabi-
caserin dose levels. In addition, the model-predicted mean
placebo effect was 1.8, which is consistent with the observed
mean placebo response of 2.7 in the phase IIa study. The
virtual trial simulation results provided confidence on the
updated PANSS model and the model predictions.
Although the predicted mean PANSS improvements by
vabicaserin monotherapy were higher than the predicted
mean placebo effect, the values were much lower than the
predicted response by olanzapine (23 points), which is also
consistent with the observed phase IIa clinical results. The
predicted vabicaserin mean treatment effects at 100 and
200 mg b.i.d. were also lower than that typically observed for
currently marketed antipsychotics in monotherapy,16 suggest-
ing that vabicaserin monotherapy may not provide compa-
rable efficacy as currently marketed antipsychotics.
The overall effect of vabicaserin in adjunctive therapy is a
result of a number of nonlinear interactions and is therefore
difficult to predict in an intuitive and qualitative way. For exam-
ple, some antipsychotics are relatively strong 5-HT2C recep-
tor antagonists, whereas others affect muscarinic receptors
or presynaptic 5-HT1B receptors that drive free 5-HT levels
and other 5-HT receptor activation levels.23 The prediction
of adjunctive treatment effects was performed using fixed-
effect simulations as the main objective was to predict the
mean improvements on PANSS for vabicaserin with cotreat-
ment with antipsychotics. In general, the model predicted
additional PANSS improvements of <4 points with all antipsy-
chotics, although slightly better with haloperidol, olanzapine,
quetiapine, or zotepine than the improvements with other
antipsychotics. However, the predicted improvements of <4
points in vabicaserin adjunctive therapy is generally consid-
ered to be clinically nondetectable.
It is acknowledged that there are some limitations for this
mechanism-based computer model. The current PANSS
model does not take into account the specific downstream
interactions of the direct and the indirect pathway but math-
ematically combines the output of a D1 receptor–positive
medium spiny neuron (MSN) and a D2 receptor–positive MSN
neuron.7 In principle, a more elaborate basal ganglia model
could be developed following earlier work.23–25 The calibra-
tion of this model with the clinical data was quite successful,
probably because all antipsychotics modulate D2 receptors,
an effect driven by the MSN neurons.7 However, the effect
of vabicaserin as a modulator of VTA dopamine firing is
upstream of the dopaminergic hyperactivity and is therefore
closer to the pathology.18 This also suggests that effects of
vabicaserin are in the same global circuit as the antipsychot-
ics. Therefore, there is a high level of confidence that the pre-
dicted clinical effect of vabicaserin based upon this model is a
reasonable approximation of the biological system.
Figure 4 Model-predicted PANSS improvements by vabicaserin
adjunctive with various antipsychotic co-medications (black-shaded
areas, except for placebo). The antipsychotics are used at their
relevant clinical doses as baseline (gray-shaded areas, except
for placebo). PANSS, positive and negative symptoms scale in
schizophrenia.
0
Co-medications
18 nmol/l vabi
Baseline
Placebo
Aripiprazole
Asenapine
Haldol
Olanzapine
Paliperidone
Perphenazine
Quetiapine
Risperidone
Ziprasidone
Zotepine
5
10
15
Improvement in PANSS
20
25
30
Table 1 Summary of model-predicted PANSS improvements in vir tual trial
simulations using 10 trials of 200 subjects at vabicaserin doses of 100 and
200 mg b.i.d. and in comparison to the observed phase IIa clinical study
results
Groups
Vabicaserin
(100 mg b.i.d.)
Vabicaserin
(200 mg b.i.d.)
Model-predicted PANSS
improvement, mean (95% CI)
5.12 (2.20, 8.56) 6.37 (2.27, 10.40)
Observed PANSS improvement in
phase IIa, LOCF mean (95% CI)
8.57 (1.98, 15.15) 5.91 (−0.84, 12.2)
CI, confidence interval; LOCF, last observation carried forward; PANSS,
positive and negative symptoms scale in schizophrenia.

www.nature.com/psp
QSP Modeling of Vabicaserin in Schizophrenia
Liu et al.
5
It is also noted that this simulation is based on a model that
is calibrated with the average outcome for a group of patients
on the same drug–dose combinations and therefore does not
take into account the individual variability within each group.
Another limitation of the current PANSS model is the lack
of time–course changes over the 4–12-week duration of the
trial. However, the main purpose of this project was to predict
(quasi–)steady-state treatment effect on PANSS. Extensive lit-
erature exists showing that the onset of antipsychotic treatment
effects on PANSS occurs rapidly within the first two weeks of
treatment,26 quasi–steady-state reaches after 4–6 weeks of
treatment.27–29 The model was calibrated with data from 4- to
12-week trials, which is sufficient to answer the key question of
(quasi–)steady-state treatment effect on PANSS in this study.
In conclusion, the updated PANSS model was implemented
using published knowledge of the subcortical neuroanatomy
and neurophysiology and was supplemented with insights
from the schizophrenia pathology as determined from human
patient populations. This computer model was calibrated
using a large collection of published clinical data with 42 drug–
dose combinations, including compounds that were effective
in preclinical animal models, but failed in clinical studies.7–9
In addition, the correlation between the model-predicted
and model-reported clinical outcomes has been shown to be
two- to threefold better than the correlation between the D2
receptor occupancy and the same clinical data.7
The model predicts that the 5-HT2C receptor activation–
mediated dopamine reduction in the ventral striatum, at the
exposures achieved in the reported clinical study, is likely to be
of insufficient magnitude to provide clinical improvement as a
stand-alone medication to the same magnitude as postsynaptic
D2 receptor antagonists. The model also predicts that vabicase-
rin would provide minimal additional improvement in PANSS in
adjunctive therapy with antipsychotics. Virtual trial simulations
using the updated PANSS model predicted the ranges of the
observed clinical results of vabicaserin and the controls by pla-
cebo and olanzapine, which provides confidence on the model
and on the decision making based on the model predictions.
This case study provides an example that quantitative systems
pharmacology model of biophysically realistic and humanized
brain circuits may be a novel approach for quantitative clinical
efficacy predictions in neuroscience disease areas.
METHODS
The PANSS model has been extensively described pre-
viously.7–9 Basically, the level of functional antipsychotic
concentration was derived from a simulation of the raclo-
pride positron emission tomography displacement stud-
ies in humans, and this intrasynaptic concentration was
subsequently used to estimate the impact of the drug on
other receptor activation level using the appropriate affini-
ties and dynamics of the neurotransmitters. These modified
postsynaptic receptor activation levels on a number of neu-
rotransmitter systems impacted the appropriate ion-channel
conductances throughout the circuit, modulated the mem-
brane potential, and therefore changed the firing dynamics.
The model was then calibrated by adjusting six biological
coupling parameters using published changes in PANSS
from 42 retrospective drug–dose combinations to optimize
the correlation between the model predictions and the pub-
lished clinical trial data.
Pharmacology of antipsychotics
Table 2 illustrates the affinities for D2 receptor (D2R), 5-HT2C
receptor (5-HT2CR), 5-HT1B receptors (5-HT1BR), and clinically
relevant unbound concentrations of a large number of antipsy-
chotics. These values were used to predict the co-medication
effects of vabicaserin.
Pharmacology of 5-HT2C
The following preclinical data were utilized to calibrate the
effect of full 5-HT2C receptor agonism on striatal dopamine fir-
ing. Microdialysis in 5-HT2C receptor knockout mice suggest
a 25% increase in dopamine levels in the ventral striatum,10
whereas treatment with the 5-HT2C receptor antagonist SB
206553 reduces the binding of 11C-raclopride in rats by 20%
in ventral striatum.11 Furthermore, treatment with the 5-HT2C
receptor agonist Ro 60–0175 into the medial prefrontal cor-
tex in rats decreases accumbal dopamine outflow by 40%.12
On the basis of these experimental studies, the effects of the
interventions (knockout, antagonism, and agonism) on 5-HT2C
receptor activation levels were determined using the receptor
competition model7,13 and the reductions in dopamine firing
associated with observed changes in either raclopride binding
or free dopamine levels.
Preclinical data on vabicaserin
The following preclinical data on vabicaserin were used to
predict PANSS improvement by vabicaserin. In rodents,
Table 2 Pharmacology of various antipsychotics against D2, 5-HT2C, and
5-HT1B receptors relevant in the calculation of the effect of vabicaserin on
antipsychotic-mediated clinical outcomes ranked in order of increasing ratio
of 5-HT2CR/D2R affinities
Drug
D2R
(nmol/l)
5-HT2CR
(nmol/l)
Ratio 5-
HT2CR/D2R
5-HT1BR
(nmol/l)
Clinical
exposure
(nmol/l)
Clozapine 220 5.59 0.025 398 250
Ziprasidone 5.2 0.9 0.173 2.5 32
Zotepine 8 4.2 0.525 67 50
Olanzapine 27.3 17 0.623 509 90
Chlorpromazine 5.5 6.1 1.119 1,498 35
Sertindole 4.3 6 1.40 60 25
Asenapine 2 10 5.00 55 13
Paliperidone 9.4 48 5.11 17.5 60
Quetiapine 406 2,500 6.16 2,050 800
Melperone 180 2,100 11.7 N/A 440
Risperidone 3.1 49 15.8 6 20
Remoxipride 295 5,500 18.6 N/A 1,400
Aripiprazole 3.3 76 23.0 830 115
Iloperidone 3 146 48.7 89 22
Perphenazine 1.4 130 92.9 N/A 9
Flupenthixol 0.8 102 127 N/A 5
Pimozide 5.1 874 171 N/A 35
Fluphenazine 0.32 579 1,809 334 5
Haloperidol 1.21 4,474 3,698 210 8
All affinity values are in nmol/l and derived from the standardized Psycho-
active Drug Screening Database.31 Functional brain concentration (clinical
exposure) derived previously is for an average clinical dose (e.g., 6 mg/day
risperidone).7 Aripiprazole is a special case as it acts as a D2 partial agonist
with 20% maximum efficacy.13

CPT: Pharmacometrics & Systems Pharmacology
QSP Modeling of Vabicaserin in Schizophrenia
Liu et al.
6
17 mg/kg i.p. decreased striatal DA levels by 39%. Dopamine
VTA firing was decreased by 40, 50, and 65%, respectively,
at 3, 10, and 17 mg/kg i.p. Dopamine synthesis was reduced
by 27 and 44%, respectively, at 3 and 10 mg/kg s.c. DOPA
(3:4-dihydroxyphenylalanine) synthesis was lowered by 39%
at 3.2 mg/kg and ventral striatal raclopride was increased by
30 and 40%, respectively, at 10 and 17 mg/kg s.c. Dorsal stria-
tum raclopride was increased by 17% at the highest dose of
17 mg/kg s.c.19 The free vabicaserin levels in the brain, in the
plasma, and in cerebrospinal fluid in rats were within 1.5-fold
of each other (unpublished data). In vitro data also showed
free equilibrium in permeability assay, suggesting that vabica-
serin is not a human P-gp substrate. Based on these data, it
was reasonable to assume that free vabicaserin concentration
in the brain is the same as the free concentration in plasma.30
Implementation of the updated PANSS model with 5-HT2C
The details of incorporating direct 5-HT2C receptor mecha-
nisms from the VTA dopamine firing and the cholinergic inter-
neurons in the updated PANSS model are described below.
Using the receptor competition model together with the
preclinical data on vabicaserin19 (see above) in combina-
tion with published data on the effect of 5-HT2C modulation
on DA activity,2,10–12 a mathematical function was derived to
describe the effect of 5-HT2C receptor activation on dopamine
firing frequency and therefore the effects on striatal D1 and D2
receptor activation. Next, the effect of serotonin neurotrans-
mission (including 5-HT2C) on striatal cholinergic interneurons
was derived. From preclinical studies19 and assuming that the
antagonists RS102221 (5-HT2C), SB258585 (5-HT6), and SB
269970 (5-HT7) were used at maximal concentrations, the
data suggest a maximal reduction in M1 receptor activation of
33.1% for total 5-HT2C block, 50.5% for total 5-HT6 block, and
82.3% for total 5-HT7 block. Although we did not model explic-
itly cholinergic interneurons in the striatum, their effects on
MSN neurons were implemented using the following equation:
MHTC
xHT C
HT C
HT
HT CMadj
Act
Con
Act
1
52 1
15 2
152
52 15 156
=−
−




+−.556 25 157
57HT
HT
HT
Con
Act
Con




+−










.
where 5HT2CM1adj is a coupling parameter that is deter-
mined during recalibration, 5HTXAct is the activation of the
5-HTX receptor with drug, and 5HTXCon is the activation of
the 5-HTX receptor in control situations. The different coef-
ficients reflect the effects of total 5-HTX receptor block on
M1 activity, relative to 5-HT2C receptor. M1 receptor activation
changes the conductance of the leak-chloride current in the
MSN as follows:
∆gMMM
M
Cl
ConHTC Act
Con
=−
03 11
52 1
1
.
where
M
Con
1 is the activation of the M1 receptor in control
situations and MAct
1 is the activation of the M1 receptor with
drug. Notice that when all 5-HT receptors are at their control
activations, MHT C
1
52
1
= so that there is no effect on MAct
1.
Furthermore, 5-HT2C receptor affects GABA interneurons in
the striatum.21 Because there are no explicit GABA interneu-
rons in the simulation, we model their effects as an impediment
to the excitability of the MSN by modifying the chloride conduc-
tance. Increased GABA interneuron firing increases the back-
ground chloride conductance in the MSN as:
gHTC HT C
HT Cg
Cl Adj
GABA Act
Con
Cl
=− −














⋅⋅15 21
52
52 ˆ∆∆gCl
where in general, XAct and XCon are actual and baseline
control receptor activation levels, 52HT CAdj
GABA is a coupling
parameter that is determined during recalibration, and ˆ
gCl is
the default chloride leak conductance.
Clinical trial with vabicaserin
The phase IIa clinical study of vabicaserin in schizophre-
nia was published previously.5 Briefly, vabicaserin was
evaluated in a 6-week randomized, double-blind, placebo-
controlled phase IIa study with olanzapine as an active
comparator. Hospitalized subjects with acute exacerbations
of schizophrenia were enrolled and randomized into one of
the four treatment arms: vabicaserin 100 or 200 mg b.i.d.,
olanzapine 15 mg/day, or placebo for a 6-week treatment.
Blinded, independent central raters performed the PANSS
and the Clinical Global Impression–Severity scale (CGI-S)
assessment via videoconferencing at screening, baseline
and each of the 6 weekly postbaseline visits. Central rated
PANSS positive was the primary end point, PANSS and
PANSS negative were secondary end points. The observed
improvements of PANSS (last observation carried forward
mean (95% confidence interval)) were 8.57 (1.98, 15.15)
and 5.91 (−0.84, 12.2) at 100 and 200 mg b.i.d., respectively.
Vabicaserin concentration data was collected between 10
and 15 h after the dose was given. The average total plasma
concentration and SD were 3.93 ± 3.30 and 8.58 ± 8.43 ng/
ml, for the doses of 100 and 200 mg b.i.d., respectively.
Simulation of virtual patient trial
Vabicaserin plasma concentration distribution was described
by a γ distribution based on the observed concentration
data. This distribution was used to generate the vabicaserin
plasma levels of virtual patients. The plasma levels were then
converted to nmol/l brain concentrations by using a factor of
0.968 (nmol/l)/(ng/ml). This conversion is based on the free
fraction in plasma protein binding and the molecular weight
of vabicaserin and assuming free equilibrium in brain pen-
etration as described in the “Preclinical data on vabicaserin”
section. The corresponding brain concentration determined
for every virtual patient was applied to determine the pre-
dicted improvements on PANSS. It is worthwhile to note that
at steady-state twice-daily dosing, the peak-to-trough ratio of
vabicaserin concentrations is less than twofold,15 and there-
fore, the observed concentration data is a reasonable repre-
sentation of steady-state vabicaserin exposure.
A total of 4,000 virtual patients with individual drug lev-
els and subsequent brain target engagement levels gener-
ated from the observed mean plasma level and variability
were simulated for 10 trials of 200 patients each with vabica-
serin daily doses of 100 and 200 mg b.i.d. In addition, Gauss-
ian distributions with variability of 30% around the biological
coupling parameters determined from the calibration were
used for the prediction of individual PANSS improvements.

www.nature.com/psp
QSP Modeling of Vabicaserin in Schizophrenia
Liu et al.
7
Acknowledgments. The authors of In Silico Biosciences
acknowledge the contributions of the late Leif Finkel (Univer-
sity of Pennsylvania), whose vision and support helped them
to develop this modeling and simulation tool.
Author Contributions. J.L. and H.G. wrote the manuscript.
J.L., A.O., T.A.C., and H.G. designed the research. J.L., A.S.,
P.R., and H.G. performed the research. J.L., A.O., T.A.C.,
A.S., P.R., and H.G. analyzed the data.
Conflict of Interest. A.S., P.R., and H.G. are employees of In
Silico Biosciences, a company providing mechanism-based
computer simulation solutions for neuroscience research
and development. J.L., A.O., and T.A.C. are employees of
Pfizer.
1. Dunlop, J. et al. Characterization of vabicaserin (SCA-136), a selective
5-hydroxytryptamine 2C receptor agonist. J. Pharmacol. Exp. Ther. 337, 673–680
(2011).
2. Siuciak, J.A. et al. CP-809,101, a selective 5-HT2C agonist, shows activity in animal
models of antipsychotic activity. Neuropharmacology 52, 279–290 (2007).
3. Wang, R. et al. The antidepressant effects of curcumin in the forced swimming test involve
5-HT1 and 5-HT2 receptors. Eur. J. Pharmacol. 578, 43–50 (2008).
4. Rosenzweig-Lipson, S., Comery, T.A., Marquis, K.L., Gross, J. & Dunlop, J. 5-HT(2C)
agonists as therapeutics for the treatment of schizophrenia. Handb. Exp. Pharmacol. 213,
147–165 (2012).
5. Shen, J. et al. A 6-week randomized, double-blind, placebocontrolled, comparator-
referenced, multicenter trial of vabicaserin in subjects with acute exacerbation of
schizophrenia. Neuropsychopharmacology 36, S106 (2011).
6. Geerts, H. Of mice and men: bridging the translational disconnect in CNS drug discovery.
CNS Drugs 23, 915–926 (2009).
7. Spiros, A., Roberts, P. & Geerts, H. A Quantitative systems pharmacology computer model
for schizophrenia efficacy and extrapyramidal side effects. Drug Dev. Res. 73, 196–213
(2012).
8. Geerts, H., Spiros, A., Roberts, P. & Carr, R. Quantitative systems pharmacology as an
extension of PK/PD modeling in CNS research and development. J. Pharmacokinet.
Pharmacodyn. 40, 257–265 (2013).
9. Geerts, H., Spiros, A., Roberts, P., Twyman, R., Alphs, L. & Grace, A.A. Blinded
prospective evaluation of computer-based mechanistic schizophrenia disease model for
predicting drug response. PLoS ONE 7, e49732 (2012).
10. Abdallah, L. et al. Impact of serotonin 2C receptor null mutation on physiology and
behavior associated with nigrostriatal dopamine pathway function. J. Neurosci. 29,
8156–8165 (2009).
11. Egerton, A., Ahmad, R., Hirani, E. & Grasby, P.M. Modulation of striatal dopamine release
by 5-HT2A and 5-HT2C receptor antagonists: [11C]raclopride PET studies in the rat.
Psychopharmacology (Berl.) 200, 487–496 (2008).
12. Leggio, G.M. et al. In vivo evidence that constitutive activity of serotonin2C receptors in the
medial prefrontal cortex participates in the control of dopamine release in the rat nucleus
accumbens: differential effects of inverse agonist versus antagonist. J. Neurochem. 111,
614–623 (2009).
13. Spiros, A., Carr, R. & Geer ts, H. Not all par tial dopamine D(2) receptor agonists are the
same in treating schizophrenia. Exploring the effects of bifeprunox and aripiprazole using a
computer model of a primate striatal dopaminergic synapse. Neuropsychiatr. Dis. Treat. 6,
589–603 (2010).
14. Gobert, A. et al. Serotonin(2C) receptors tonically suppress the activity of mesocor tical
dopaminergic and adrenergic, but not serotonergic, pathways: a combined dialysis and
electrophysiological analysis in the rat. Synapse 36, 205–221 (2000).
15. Mako, B., Poola, N.R., Posener, J.A., & Paul, J. Safety, tolerability, and pharmacokinetics
of multiple ascending oral doses of vabicaserin, a serotonin 2C agonist, in subjects with
schizophrenia and schizoaffective discorder. Clin. Pharmacol. Ther. 83 (suppl. 1), S21
(2008).
16. Citrome, L. A systematic review of meta-analyses of the efficacy of oral atypical
antipsychotics for the treatment of adult patients with schizophrenia. Expert Opin.
Pharmacother. 13, 1545–1573 (2012).
17. Davis, J.M., Chen, N. & Glick, I.D. A meta-analysis of the efficacy of second-generation
antipsychotics. Arch. Gen. Psychiatry 60, 553–564 (2003).
18. Howes, O.D., Fusar-Poli, P., Bloomfield, M., Selvaraj, S. & McGuire, P. From the prodrome
to chronic schizophrenia: the neurobiology underlying psychotic symptoms and cognitive
impairments. Curr. Pharm. Des. 18, 459–465 (2012).
19. Marquis, K.L., et al. SCA-136: a novel 5-HT2C receptor agonist possessing atypical
antipsychotic-like effects in preclinical models. 36th Annual Meeting for the Society of
Neuroscience, Washington, D.C., 14–18 October 2006.
20. Bonsi, P. et al. Endogenous serotonin excites striatal cholinergic interneurons via
the activation of 5-HT 2C, 5-HT6, and 5-HT7 serotonin receptors: implications for
extrapyramidal side effects of serotonin reuptake inhibitors. Neuropsychopharmacology 32,
1840–1854 (2007).
21. Blomeley, C.P. & Bracci, E. Serotonin excites fast-spiking interneurons in the striatum. Eur.
J. Neurosci. 29, 1604–1614 (2009).
22. Perez-Rosello, T. et al. Cholinergic control of firing pattern and neurotransmission in rat
neostriatal projection neurons: role of CaV2.1 and CaV2.2 Ca2+ channels. J. Neurophysiol.
93, 2507–2519 (2005).
23. Sari, Y. Serotonin1B receptors: from protein to physiological function and behavior.
Neurosci. Biobehav. Rev. 28, 565–582 (2004).
24. Rubin, J.E. & Terman, D. High frequency stimulation of the subthalamic nucleus eliminates
pathological thalamic rhythmicity in a computational model. J. Comput. Neurosci. 16,
211–235 (2004).
25. Guo, Y. & Rubin, J.E. Multi-site stimulation of subthalamic nucleus diminishes
thalamocortical relay errors in a biophysical network model. Neural Netw. 24, 602–616
(2011).
Study Highlights
WHAT IS THE CURRENT KNOWLEDGE ON THE
TOPIC?
3 There are limited examples of systems phar-
macology approaches that incorporate relevant
brain targets and circuits, knowledge of schizo-
phrenia pathology, and polypharmacology of
treatments.
WHAT QUESTION DID THIS STUDY ADDRESS?
3 This study confirmed the predictive capability
of humanized quantitative systems pharma-
cology computer simulations with the actual
phase IIa clinical results of a novel 5-HT2C
agonist, vabicaserin. This study provides pre-
dictions regarding the potential lack of utility
of vabicaserin treatments in schizophrenia as
monotherapy and as antipsychotic augmenta-
tion, which guided the early clinical develop-
ment decision of vabicaserin.
WHAT THIS STUDY ADDS TO OUR KNOWLEDGE
3 This case study provides an example that the
quantitative systems pharmacology model of
biophysically realistic and humanized brain
circuits may be a novel approach for quantita-
tive clinical efficacy predictions in neuroscience
disease areas.
HOW THIS MIGHT CHANGE CLINICAL
PHARMACOLOGY AND THERAPEUTICS
3 The quantitative systems pharmacology model
of biophysically realistic and humanized brain
circuits may provide an alternative approach to
traditional animal models in selecting novel tar-
gets in neuroscience discovery and development.

CPT: Pharmacometrics & Systems Pharmacology
QSP Modeling of Vabicaserin in Schizophrenia
Liu et al.
8
26. Agid, O., Kapur, S., Arenovich, T. & Zipursky, R.B. Delayed-onset hypothesis of
antipsychotic action: a hypothesis tested and rejected. Arch. Gen. Psychiatry 60,
1228–1235 (2003).
27. Sherwood, M., Thornton, A.E. & Honer, W.G. A meta-analysis of profile and time-course
of symptom change in acute schizophrenia treated with atypical antipsychotics. Int. J.
Neuropsychopharmacol. 9, 357–366 (2006).
28. Ahadieh, S., Corrigan, B., Riley, S. & Lalonde, R. Model based meta-analysis of positive
and negative syndrome scale (PANSS) in schizphrenic patients. Clin. Pharmacol. Ther. 85
(suppl. 1), S64 (2009).
29. Pilla Reddy, V. et al. Pharmacokinetic-pharmacodynamic modeling of antipsychotic drugs
in patients with schizophrenia Part I: the use of PANSS total score and clinical utility.
Schizophr. Res. 146, 144–152 (2013).
30. Brodfuehrer, J., Rong, H. & Liu, J. Confidence in predicted human brain penetration. 2010
FIP PSWC/AAPS Annual Meeting and Exposition, New Orleans, LA, 14–18 November
2010.
31. Psychoactive Drug Screening Program database. <http://pdsp.med.unc.edu/indexR.html>
CPT: Pharmacometrics & Systems Pharmacology is an
open-access journal published by Nature Publishing
Group. This work is licensed under a Creative Commons Attribution-
NonCommercial-NoDerivative Works 3.0 License. To view a copy of
this license, visit http://creativecommons.org/licenses/by-nc-nd/3.0/