pharmacometric models for antibody drug conjugates and...
TRANSCRIPT
ACTAUNIVERSITATIS
UPSALIENSISUPPSALA
2016
Digital Comprehensive Summaries of Uppsala Dissertationsfrom the Faculty of Pharmacy 217
Pharmacometric Models forAntibody Drug Conjugates andTaxanes in HER2+ and HER2-Breast Cancer
BRENDAN BENDER
ISSN 1651-6192ISBN 978-91-554-9603-6urn:nbn:se:uu:diva-292617
Dissertation presented at Uppsala University to be publicly examined in B/B42, BMC,Husargatan 3, Uppsala, Friday, 2 September 2016 at 13:15 for the degree of Doctor ofPhilosophy (Faculty of Pharmacy). The examination will be conducted in English. Facultyexaminer: Professor Oscar della Pasqua (University College London School of Pharmacy;GlaxoSmithKline plc).
AbstractBender, B. 2016. Pharmacometric Models for Antibody Drug Conjugates and Taxanesin HER2+ and HER2- Breast Cancer. Digital Comprehensive Summaries of UppsalaDissertations from the Faculty of Pharmacy 217. 87 pp. Uppsala: Acta UniversitatisUpsaliensis. ISBN 978-91-554-9603-6.
In oncology, there is a need to optimize drug treatment for efficient eradication of tumors,minimization of adverse effects (AEs), and prolonging patient survival. Pharmacometric modelscan be developed to streamline information between drug development phases, describe andquantify response to treatment, and determine dose regimens that balance toxicity and efficacy.In this thesis, data from trastuzumab emtansine (T-DM1) and taxane drug treatment were usedto develop pharmacometric models of pharmacokinetics (PK), AEs, anti-tumor response, andsurvival, supporting drug development.
T-DM1 is an antibody-drug conjugate (ADC) for treatment of human epidermal growth factorreceptor 2 (HER2)–positive breast cancer. ADCs are a relatively new class of oncologic agents,and contain multiple drug-to-antibody ratio (DAR) moieties in their dose product. The complexdistribution of T-DM1 was elucidated through PK models developed using in vitro and invivo rat and cynomolgus monkey DAR data. Mechanism–based PK/pharmacodynamic (PKPD)models were also developed for T-DM1 that described the AEs thrombocytopenia (TCP) andhepatotoxicity in patients receiving T-DM1. Variable patterns of platelet and transaminase (ALTand AST) response were quantified, including an effect of Asian ethnicity that was relatedto higher incidences of TCP. Model simulations, comparing dose intensities (DI) and Grade3/4 incidences between the approved T-DM1 dose (3.6 mg/kg every three weeks) and weeklyregimens, determined that 2.4 mg/kg weekly provided the highest DI.
Docetaxel and paclitaxel are taxane treatment options for HER2–negative breast cancer.Tumor response data from these treatments were used to develop a mechanism–based modelof tumor quiescence and drug–resistance. Subsequently, a parametric survival analysis foundthat tumor baseline and the model–predicted time to tumor growth (TTG) were predictors ofoverall survival (OS). This tumor and OS modeling approach can be applied to other anticancertreatments with similar patterns of drug–resistance.
Overall, the pharmacometric models developed within this thesis present new modelingapproaches and provide understanding on ADC PK and PKPD (TCP and hepatotoxicity), aswell as drug–resistance tumor response. These models can inform simulation strategies andclinical study design, and be applied towards dose finding for anticancer drugs in development,especially ADCs.
Keywords: pharmacometric, PKPD, model, breast cancer, T-DM1, thrombocytopenia,hepatotoxiticy, HER2
Brendan Bender, Department of Pharmaceutical Biosciences, Box 591, Uppsala University,SE-75124 Uppsala, Sweden.
© Brendan Bender 2016
ISSN 1651-6192ISBN 978-91-554-9603-6urn:nbn:se:uu:diva-292617 (http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-292617)
List of Papers
This thesis is based on the following papers, which are referred to in the text by their Roman numerals.
I. Bender BC, Schaedeli–Stark F, Koch R, Joshi A, Chu YW, Rugo H, Krop IE, Girish S, Friberg LE, Gupta M. A population pharma-cokinetic/pharmacodynamic model of thrombocytopenia character-izing the effect of trastuzumab emtansine (T-DM1) on platelet counts in patients with HER2–positive metastatic breast cancer. Cancer Chemother Pharmacol. 2012 Oct;70 (4):591–601.
II. Bender BC, Quartino A, Li C, Chen S–C, Smitt M, Strasak A, Chernyukhin N, Wang B, Jin J, Girish S, Friberg LE. An integrated pharmacokinetic–pharmacodynamic (PKPD) modeling analysis of trastuzumab emtansine (T-DM1)–induced thrombocytopenia (TCP) and hepatotoxicity in patients with HER2–positive metastatic breast cancer (In manuscript)
III. Bender BC, Leipold DD, Xu K, Shen BQ, Tibbitts J, Friberg LE. A mechanistic pharmacokinetic model elucidating the disposition of trastuzumab emtansine (T-DM1), an antibody–drug conjugate (ADC) for treatment of metastatic breast cancer. AAPS J. 2014 Sep;16(5):994–1008.
IV. Bender BC, Jin J, Friberg LE. A mechanism–based model of tumor quiescence and drug–resistance in HER2–negative metastatic breast cancer patients receiving docetaxel or paclitaxel. (In manuscript)
Reprints were made with permission from the respective publishers.
Contents
Introduction ..................................................................................................... 9 Model–based framework for drug development in oncology .................... 9 Pharmacometric Background in Oncologic Drug Development .............. 11
Population PK and PKPD modeling .................................................... 11 Preclinical PK and PKPD modeling .................................................... 13 Pharmacometric Models of Continuous Type Data in Oncology ........ 14 Time-to-Event Modeling in Oncology ................................................ 19 Pharmacometric analyses incorporating tumor SLD and TTE modeling .............................................................................................. 19 Utility of a modeling and simulation framework in oncology ............. 21
Metastatic Breast Cancer Treatment ............................................................. 23 Paclitaxel and Docetaxel .......................................................................... 23 Trastuzumab emtansine (T-DM1) ............................................................ 24
Trastuzumab ........................................................................................ 25 DM1 ..................................................................................................... 25 T-DM1 Dose Product .......................................................................... 25 T-DM1 Mechanism of Action ............................................................. 26
Aims .............................................................................................................. 27
Data and Methods ......................................................................................... 28 Clinical Data ............................................................................................. 30
T-DM1 ................................................................................................. 30 Docetaxel and Paclitaxel...................................................................... 30
Preclinical Data ........................................................................................ 30 Preparation of T-DM1DAR 3.1 and T-DM1DAR 1.5 Dose Products ........... 31 Rats ...................................................................................................... 31 Cynomolgus Monkeys ......................................................................... 32
Measurements........................................................................................... 32 Platelet, ALT, and AST ....................................................................... 32 Platelet, ALT, and AST Toxicity Grades ............................................ 32 T-DM1 and Total Trastuzumab Concentrations .................................. 33 T-DM1 Drug-to-antibody Ratio (DAR) Concentrations ..................... 33 Tumor Sum of Longest Diameter (SLD) ............................................. 34
Model Building and Data Analysis .......................................................... 34 Post-hoc Bayesian estimates ................................................................ 34
Paper I: T-DM1 PKPD Model of TCP ................................................ 35 Paper II: T-DM1 PKPD Model of TCP and Hepatotoxicity ............... 37 Paper III: T-DM1 PK Models .............................................................. 40 Paper IV: Taxane Tumor Model .......................................................... 41 Model building methods ...................................................................... 43 Covariate Analyses .............................................................................. 43 Model Evaluation and Simulation ....................................................... 44
Results/Discussion ........................................................................................ 47 Papers I and II .......................................................................................... 47 Paper II Simulations ................................................................................. 54 Paper III .................................................................................................... 58
T-DM1 Mechanistic PK Model ........................................................... 58 Reduced T-DM1 PK Model................................................................. 63
Paper IV.................................................................................................... 67
Conclusions ................................................................................................... 74
Future perspectives ....................................................................................... 75
Acknowledgements ....................................................................................... 79
References ..................................................................................................... 82
Abbreviations
AE adverse effects ALT alanine transaminase AST aspartate transaminase AUC area under the curve AUCss steady state area under the curve BM biomarker BSL baseline C concentration Ce effect compartment concentration Circ circulation CL clearance CLd distributional clearance Cmax,ss steady state maximum concentrations Cp central compartment concentration C/P carboplatin/paclitaxel CRC colorectal cancer DAR drug-to-antibody ratio DEC deconjugation DI dose intensity ECD extracellular domain ECOG Eastern Cooperative Oncology Group Emax maximum effect ELISA enzyme linked immunosorbent assay Exp exponential FDA Food and Drug Administration FO first order FOCE first order conditional estimate FnR fraction resistant GIST gastro-intestinal stromal tumor HER2 human epidermal growth factor receptor 2 IACUC Institutional Animal Care and Use Committee IIV interindividual variability L liter LC–MS liquid chromatography mass spectrophotome- try
LDH lactate dehydrogenase LLN lower level of normal log(G) log of the tumor growth rate MOA mechanism of action MTT mean transit time NONMEM nonlinear mixed effects modeling NSCLC non-small cell lung carcinoma OFV objective function value ORR overall response rate OS overall survival PD pharmacodynamic PK pharmacokinetic PFS progression–free survival PKPD pharmacokinetic–pharmacodynamic PLT platelet compartment PP platelet proliferation PPC posterior predictive check Prol proliferation q1w, q3w, q4w weekly, every 3 weeks, every 4 weeks R resistant RCC renal cell carcinoma RDI relative dose intensity RECIST Response Evaluation Criteria in Solid Tumors RSE relative standard error sKIT soluble stem cell factor SLD sum of longest diameter SMCC N–succinimidyl–4–(N–maleimidomethyl)– cyclohexane–1–carboxylate sVEGFR-3 soluble vascular endothelial growth factor receptor 3 T transit t time–course TCP thrombocytopenia TGI tumor growth inhibition TT total trastuzumab TTE time-to-event TSR tumor size ratio TTG time to tumor growth μL microliter ULN upper level of normal V1 central volume of distribution V2 peripheral volume of distribution 2 V3 peripheral volume of distribution 3 VPC Visual Predictive Check
Introduction
Cancer remains an unmet medical need [1]. Not only is there a need to de-velop new drugs in oncology, but also to improve the speed and efficiency of preclinical to clinical oncology drug development [2]. In oncology, optimiz-ing drug treatment to effectively kill tumors, minimize adverse effects (AE), and prolong patient survival is paramount. Pharmacometric models and strategies can be developed based on anti-cancer drug treatment data, and then applied to new anti-cancer drugs and their effects (e.g. for efficacy, biomarker, and toxicity endpoints), in order to improve the speed and effi-ciency of analysis. Pharmacometric models are also powerful tools towards the design of clinical trials (e.g. the most informative doses, relevant varia-bles, data collection time points, number of patients, etc.), to ultimately de-termine the dose regimen that will provide the best balance between desired effects and side–effects.
Model–based framework for drug development in oncology Figure 1 illustrates a proposed framework for pharmacometric drug devel-opment in oncology. As shown, pharmacometric models may be developed early and used to inform Phase I–III study design and target endpoints. From development of a population pharmacokinetic (PK) model, PK metrics may be tested and implemented into PK–pharmacodynamic (PKPD) models de-scribing various responses, such as adverse effects, tumor and biomarker response, and also be assessed as predictors for survival. Once developed, these data–driven PKPD models can be used to explore and support dose and regimen changes, as well as to provide model–based metrics that can be assessed as “drivers” for other PD responses and as predictors for survival.
The similarity among measurements and endpoints in oncology, regardless of cancer type, makes this an applicable framework for oncologic drug de-velopment programs. For example: 1) anti–tumor treatment efficacy is typi-cally established using tumor bearing mice; 2) in the clinic, tumor sum of longest diameter (SLD) measurements are used as an indication of treatment
9
F
igur
e 1.
Mod
el–b
ased
fram
ewor
k fo
r dr
ug d
evel
opm
ent i
n on
colo
gy, a
dapt
ed fr
om p
ublic
atio
ns [
3, 4
].
from
bas
elin
e;
AU
C: a
rea
unde
r the
dru
g co
ncen
tratio
n–tim
e cu
rve;
AU
CB
M: a
rea
unde
r the
bio
mar
ker–
time
curv
e; k
GR
OW
: tu
mor
gro
wth
rate
con
stan
t; O
S:
over
all s
urvi
val;
PFS:
pro
gres
sion
–fre
e su
rviv
al;
PKPD
: pha
rmac
okin
etic
–pha
rmac
odyn
amic
; (t):
tim
e–co
urse
; TS
R: t
umor
size
ratio
; TTG
: tim
e to
tum
or g
row
th.
Phas
eI–
III C
linic
al T
rials
Surv
ival
Mod
el
PFS,
OS
Clin
ical
Dos
eRe
gim
en
chan
ges
AE M
etric
s Ba
selin
e,Bl
ood
Cell
Circ
ulat
ion(
t)
Biom
arke
r Re
spon
se
PKPD
Mod
el
Tum
orM
etric
s TS
R, T
umor
(t)
TTG
, kG
ROW
Biom
arke
r M
etric
sAU
C BM
,Bi
omar
ker(
t)
PK M
etric
sAU
C, D
rug
Conc
entr
atio
n(t)
Tum
orRe
spon
sePK
PD M
odel
Adve
rse
Effe
cts
(AE)
PK
PD M
odel
Popu
latio
n PK
Mod
el
Tum
orst
atic
Co
ncen
trat
ion
(TSC
), an
d/or
Dos
ere
spon
se
Mou
se T
umor
Resp
onse
PK
PD M
odel
Prec
linic
alPK
Mod
els
Biom
arke
r an
d To
xici
tyPK
PD M
odel
s
Info
rmCl
inic
al D
rug
Expo
sure
Targ
et
Info
rmSt
udy
Des
ign,
Bi
omar
ker T
arge
t
effect for solid tumors; 3) circulating biomarkers, associated with drug mechanism of action, may be assessed as early indicators of treatment effect; 4) adverse effects, such as chemotherapy–induced myelosuppression, are noted across many cancer treatments; and 5) progression–free survival (PFS) and overall survival (OS) are primary clinical endpoints for evaluating treatment success.
This pharmacometric framework serves as a template to introduce analyses within this thesis work of pharmacometric models for antibody drug conju-gates (ADCs) and taxanes in human epidermal growth factor receptor 2 (HER2)–positive and HER2–negative breast cancer. The following models have been developed: 1) a PKPD model describing platelet response to T-DM1 treatment in HER2–positive breast cancer patients; 2) a PKPD model simultaneously describing platelet, ALT, and AST response to T-DM1 treatment in HER2–positive breast cancer patients; 3) PK models elucidating and conceptualizing T-DM1 disposition based on preclinical experimental data; and 4) a model for tumor response to taxane treatment in HER2–negative breast cancer patients, and a subsequent parametric time-to-event analyses of survival data.
Pharmacometric Background in Oncologic Drug Development Pharmacometrics is the science of quantitative drug development, and in-cludes concepts utilized in this thesis work regarding population PK [5], PKPD [6-9], and model–based drug development [3, 10]. In the following sections, general pharmacometric concepts and analyses are introduced that have guided this thesis work. Specifically: 1) population PK and PKPD modeling; 2) preclinical PK and PKPD models in oncology; 3) pharmaco-metric models of continuous type data in oncology; and 4) time-to-event (TTE) modeling in oncology.
Population PK and PKPD modeling In drug development, pharmacometricians develop and fit PK models to drug plasma concentration–time data so that typical PK parameters and their variabilities can be estimated. Patient characteristics (covariates), such as weight, creatinine clearance, disease status, etc., can be tested and imple-mented into the model to explain the between patient PK variability [11]. Development of a population PK model that well describes the drug concen-tration–time data implies an understanding of how the body is processing the drug.
11
Following this analysis, pharmacometricians may attempt to incorporate PK model–based metrics to describe the drug concentration–effect, or “PKPD”, relationship. In oncology, PD measurements are typically associated with how the patient responds to treatment, such as the tumor burden, a bi-omarker, or the development of adverse effects. Shown in Figure 2 is a gen-eral PKPD model structure, in which predicted drug concentrations over time (Cp(t)) from a PK model are directly linked to PD responses.
Figure 2. Compartmental representation of a general PKPD system. CL: drug clearance; CLd: distributional clearance; kin: zero–order rate constant for production of response; kout: first–order rate constant for loss of the response; Slope: drug effect scaling parameter; V1: central volume of distribution; V2: peripheral volume of distribution
As shown, the PK portion of the model system is a two compartmental mod-el, developed from drug concentration–time data, and the typical population PK parameters (i.e. clearances (CL, CLd) and volumes of distribution (V1 and V2)) and their variabilities are estimated. The PK parameters for each individual subject can be obtained in a ‘posthoc’ step.
In the PD portion of the model, a direct link using the model–predicted drug concentration–time curve (Cp(t)), from the central compartment (V1), acts to stimulate the production of PD Response1 and inhibit the production of PD Response2. Slope1 and Slope2 are drug–related scaling parameters to be es-timated by the PKPD model when fit to the PD data. Kin and kout are system–related parameters which are also estimated by the model [12]. Following development of a PKDP model, simulations can be done as illustrated in Figure 3.
kout, PD1
kout, PD2
Dose
CL
CLd
V1V2
PD Response1
PDResponse2
Pharmacokinetics (PK) Pharmacodynamics (PD)
12
Figure 3. Model–predicted curves for a general PKPD system. Model–predicted time–courses of PD response for stimulation of kin (stippled line) and inhibition of kin (solid line); the model–predicted drug concentration time–course (bottom curve) is shown for once every 3 week (q3w) dose regimen
Preclinical PK and PKPD modeling As shown in Figure 1, pharmacometric models in oncology may be devel-oped and applied early in preclinical drug development. For PK modeling, the approach is the same as described in the previous section, though covari-ates are typically not explored; in vivo animal experiments typically do not have the heterogeneity in body composition or physiological function as seen in the healthy human or diseased patient population.
Most relevant for clinical support, preclinical PKPD models for efficacy and toxicity may be developed to support and inform Phase I–III dose selection. In oncology, myelosuppression is one of the most common adverse effects (AE) following chemotherapy. Notably, Friberg et al. [13] has proposed a strategy for scaling myelosuppression data from rats to humans via PKPD modeling. For efficacy, PKPD modeling of tumor response from tumor bear-ing mice has been used to derive a tumorstatic concentration–the drug con-centration at which there is no net tumor growth, thereby providing a target exposure that may be used to support clinical dose selection [14, 15].
13
0
50
100
150
200
250
300
350
400
450
0
25
50
75
100
125
150
0 3 6 9 12 15
Drug Concentraion
Phar
mac
odyn
amic
Res
pons
e
Time (Weeks)
PD2: Inhibition of kin
PD1: Stimulation of kin
Pharmacometric Models of Continuous Type Data in Oncology For this thesis work, platelet counts, liver enzyme levels (e.g. alanine trans-aminase (ALT) and aspartate transaminase (AST)), and tumor SLD were the measurements available for development of PKPD models. Prior to model development, literature searches were done to establish starting points. Elevations in ALT and AST levels provide indication of liver damage, and PKPD models have not been readily applied, and only two models were found for consideration. Fetterly et al. [16] presented a precursor–dependent mechanism-based model for ALT response to the anti-tumor drug trabedec-tin. Pollak et al. [17] presented a linear effect compartment approach to pre-dict ALT response to amiodarone, a treatment for atrial fibrillation. In con-trast to ALT and AST, PKPD models for platelet and tumor SLD have been routinely applied in oncologic drug development and are discussed further here: 1) the myelosuppression model and 2) the clinical tumor growth inhibi-tion (TGI) model.
Pharmacometric Model for Myelosuppression For chemotherapy, myelosuppression is one of the most common adverse effects (AE). Myelosuppression is a condition in which the normal produc-tion of blood cells in the bone marrow is reduced, resulting in fewer circulat-ing red blood cells, leukocytes (60–70% neutrophils), and/or platelets. Given the importance for neutrophils and platelets in fighting infections and in maintenance of blood clotting, respectively, myelosuppression is dose limit-ing for many anticancer agents.
In 2002, a myelosuppression PKPD model was presented by Friberg et al. [18] based on leukocyte and neutrophil data from six chemotherapeutic agents. This model has been used to characterize neutrophil [19-25], leuko-cyte [24], and platelet [20, 26] responses. As shown in Figure 4, the myelo-suppression model is a physiological–based model, with components of myelocyte proliferation, maturation, blood circulation, and feedback.
14
Figure 4. Compartmental representation of the myelosuppression model. Prol: pro-liferation cell pool compartment; T1, T2, and T3: transit compartments of cell matu-ration; Circ: blood circul o-sure: e.g., the drug–concentration time–course ; kel: elimination rate constant; kPROL: proliferation rate constant; ktr: intercompartmental transit rate constant; Slope: drug inhibition constant; Circ0: : feedback term; MTT: mean transit time, derived as ktr/(n+1), where n is the number of transit compartments.
The model consists of a proliferation cell pool (Prol) compartment, three transit compartments (T1, T2, and T3) mimicking the maturation of the non–proliferative cells, and a blood circulation compartment (Circ) where the measurements (e.g. neutrophil counts) have been observed over time. The
proliferation rate as circulating cell levels are depleted and slows down pro-liferation when the circulating neutrophil counts are above the baseline val-ue. The drug effect is the product of the Slope parameter and the individual exposure metric, where exposure can be AUC, Dose, or concentration (t), etc.
The mechanism–based nature of the myelosuppression model readily allows modifications to the structure. For example, Quartino et al. [27] incorporated granulocyte stimulating colony factor (G–CSF) measurements. A turnover model of G–CSF, where G–CSF elimination was dependent on the circulat-ing neutrophil counts, replaced the feedback function and provided a more mechanistic interpretation which may allow for improved predictive capaci-ty.
With regard to platelets, van Kesteren et al. [20] presented a PKPD model based on this myelosuppression structure to describe platelet response to the anti-cancer agent indisulam. Chalret du Rieu et al. [26] also added modifica-
Slope Exposure
Prol
ktr
kPROL= ktr
T1 T2 ktr
T3
Circ
ktr = 4 / MTT ktr
kel = ktr
Transit Compartments
CircCircFeedback 0
Drug Effect
15
tions to the myelosuppression model, using an extra feedback term and al-ternate platelet baseline parameterization, to describe platelet response to abexintostat.
Figure 5 illustrates components of the myelosuppression model, showing simulations of neutrophil and drug effect time–course concentration (t) in this example), for a once every 3 week intravenous drug treatment.
Figure 5. Myelosuppression model–predicted neutrophil (top curve) and drug effect (bottom curve) time–courses for a once every 3 week (q3w) drug treatment. e-line may be calculated from the model–predicted (baseline–nadir)/baseline.
As shown, neutrophil counts oscillate in response to drug treatment, reach-
the next dose.
Pharmacometric Model for Tumor Response In oncologic drug development for solid tumors, the tumor sum of longest diameter (SLD) measurement is used as an indication of treatment effect. These measurements are sparse, typically once every 6–8 weeks. The Re-sponse Evaluation Criteria in Solid Tumors (RECIST; currently version 1.1) is then used to categorize tumor response to treatment [28]. However, RE-CIST 1.1 is not an optimal assessment of drug efficacy, as information is lost when the continuous tumor SLD data is categorized.
In 2009, the tumor growth inhibition (TGI) model was presented by Claret et al. [29] on data from colorectal cancer patients receiving capecitabine. The
0
10
20
30
40
50
60
70
80
90
100
0
1
2
3
4
5
0 3 6 9 12 15
Drug
Effect9 /
L)
Time (Weeks)
16
TGI model offered an improved understanding for response versus the standard RECIST criteria, by modeling longitudinal tumor SLD data on a continuous scale. This structural model (Figure 6) has been applied by inves-tigators to several cancer types and drugs.
Figure 6. Compartmental representation of the TGI model. kGROW: tumor growth rate constant; Exposure: drug exposure metric; kKILL: tumor kill rate constandrug resistance parameter; k: drug expo the k parame-ter was not applied in the original publication (k=0) [29].
The TGI model predicts that, in the absence of treatment, tumor SLD in-creases exponentially according to a disease–specific first–order growth rate constant kGROW. The tumor SLD doubling rate is derived as ln(2)/kGROW. Drug–related tumor shrinkage is accounted for by a metric for drug exposure (e.g. Dose, AUC) and a drug–specific cell kill rate constant (kKILL). Loss of drug effect is described by a time–dependent mono–exponential function defined by the parameter ( ). Figure 7 shows a representative plot of drug effect and tumor SLD time–courses simulated using the TGI model.
TumorSLD
kGROW
Drug Effect
kKILL Exposure e - time
k
17
Figure 7. Model–predicted PKPD curves for the TGI model. TGI model–predicted tumor SLD (top curve) and drug effect (bottom curve) time–courses for a once every 3 week (q3w) drug treatment. TSR: tumor size ratio from baseline, typically as-sessed after 1 or 2 treatment cycles (6–8 weeks); TTG: time to tumor growth.
In this simulation, the drug exposure decreases mono–exponentially within treatment cycles via the k parameter, and tumor begins regrowth around week 9 as the drug effect diminishes.
Several tumor metrics can be derived from the TGI model for assessment as predictors for OS. These metrics include the tumor time–course (red line), the tumor size ratio (TSR), and the time to tumor growth (TTG). TSR re-flects the relative change in tumor SLD from baseline at a certain time point, e.g. after one or two treatment cycles (6–8 weeks). There are drawbacks with TSR in that it is determined at a fixed time point, irrespective of the ongoing tumor response. TSR can thereby be the same value for two patients, where both patients initially respond to treatment, but one patient develops tumor regrowth; it is not possible to ascertain whether tumor SLD is in a declining or increasing phase at week 6–8.
TTG corresponds to the time of a patient’s tumor SLD nadir, at which there is no net tumor growth. The TTG metric shares some limitations with TSR; it is a constant value metric that is also used to predict the hazard of survival time–points before the metric has occurred, and TTG can have the same value for patients with vastly different tumor responses. Time-varying pre-
(resistance)
TSR
kGROWTTG
18
dictors, such as the tumor time–course may have advantages, especially if the model is to be used prospectively [3, 30].
Time-to-Event Modeling in Oncology In the current paradigm for oncology, overall survival (OS) – the time from randomization until death from any cause, is the preferred endpoint to evalu-ate treatment benefit [31]. However, it may take many years for OS data to reach the number of events so that statistical conclusions can be drawn, and thus preliminary drug approval may be granted based on an improvement in progression-free survival (PFS) [31]. “Progression” refers to tumor growth, and is defined as a >20% increase in tumor SLD over baseline (or from best response), with a minimum 5 mm absolute increase, or appearance of any new lesions [28].
As illustrated in Figure 1, pharmacometricians may evaluate and implement “metrics” from PK and PKPD models into parametric TTE models as pre-dictors for OS, in order to improve the assessment of treatment efficacy. Concepts regarding parametric modeling of survival are presented below. A recent tutorial on TTE analysis for pharmacometricians [32] and a textbook on modeling survival data [33] provide further background.
Shown in the following equation is the hazard function, h(t), which describes the instantaneous rate at which an event (in oncology = death) occurs:
h(t) = h0(t) e 1 1 2 2 n n
The baseline hazard h0 (t) is defined by one or more estimated parameters, and x1, x2, … xn represent a set of predictors that affect the hazard. As shown in Figure 1, these predictors may be a derived constant value for each patient (e.g. TSR, TTG), a model parameter estimate (e.g. kGROW), or a predicted time varying metric (e.g. Tumor(t), Biomarker(t)). Additionally, patient baseline characteristics (e.g. ECOG status, observed tumor SLD at baseline, number of lesions, etc.) are frequently assessed as predictors. The impact of the predictors is determined by the size of the respective coeffi-
1 2 n, which are estimated from the data [34].
Pharmacometric analyses incorporating tumor SLD and TTE modeling Table 1 shows representative analyses [29, 35-43], for a diversity of solid tumor types and treatment drugs, in which the time–course of longitudinal tumor SLD was described by population models, and metrics therefrom were assessed as predictors for survival.
19
Tabl
e 1.
Pha
rmac
omet
ric a
naly
ses
inco
rpor
atin
g tu
mor
SLD
and
TTE
Tum
or ty
peTr
eatm
ent
PDm
easu
rem
ent (
s)D
river
of P
D
Res
pons
eM
odel
Typ
ePr
edic
tors
for S
urvi
val
Mod
el-B
ased
Oth
erR
efer
ence
NSC
LCVa
rious
chem
othe
rapi
estu
mor
SLD
Non
e; d
ose-
inde
pend
ent
Line
ar g
row
th;
exp
shrin
kage
TSR
(wee
k 8)
ECO
G, t
umor
SLD
0W
ang
2009
[35]
NSC
LCC
/P o
r C/P
+
Beva
cizu
mab
or
C/P
+ M
otes
anib
tum
or S
LDN
one;
dos
e-in
depe
nden
tLi
near
gro
wth
;ex
p sh
rinka
geTS
R (w
eek
8)EC
OG
, tum
or S
LD0
Cla
ret 2
012
[36]
CR
CC
apec
itabi
ne;
Fluo
rour
acil
tum
or S
LDD
aily
dos
eTG
ITS
R (w
eek
7)tu
mor
SLD
0C
lare
t 200
9 [2
9]
Thyr
oid
canc
erM
otes
anib
tum
or S
LDAU
Css
TGI
EC
OG
, tum
or S
LD0
Lu 2
010
[37]
Cla
ret 2
010
[38]
Met
asta
tic
brea
st c
ance
rC
apec
itabi
ne +
do
ceta
xel
tum
or S
LDD
ose
TGI
TSR
(wee
k 6)
tum
or S
LD0,
ECO
G,
num
ber o
f m
etas
tase
s, s
tudy
ef
fect
Brun
o 20
12 [3
9]
GIS
TSu
nitin
ibtu
mor
SLD
Dai
ly A
UC
, sV
EGFR
-3(t)
, sK
IT(t)
TGI
tum
or S
LD0
Han
sson
201
3 [4
0]
CR
CBe
vaci
zum
ab +
ch
emot
hera
pytu
mor
SLD
Con
stan
t ex
posu
reTG
I with
co
nsta
nt
expo
sure
TTG
, TS
R (w
eek
6),
log(
G)(
*),
EC
OG
, tum
or S
LD0
(*),
treat
men
t arm
(*),
num
ber o
f org
ans
Cla
ret 2
013
[41]
NSC
LCM
otes
anib
tum
or S
LDD
ose
TGI w
ith
cons
tant
ex
posu
re
Log(
TTG
)As
ian
ethn
icity
, Sm
okin
g hi
stor
y,
tum
or S
LD0
Cla
ret 2
014
[42]
RC
CTe
msi
rolim
us,
suni
tinib
, axi
tinib
, in
terfe
ron,
orso
rafe
nib
tum
or S
LDD
ose
TGI w
ith
cons
tant
ex
posu
re
TTG
, TS
R (w
eek
8)EC
OG
, num
ber o
f m
etas
tase
s, L
DH
0,he
mog
lobi
n,
corr
ecte
d ca
lciu
m
Cla
ret 2
015
[43]
AU
Css
: ste
ady
stat
e ar
ea u
nder
the
drug
con
cent
ratio
n cu
rve;
C/P
: car
bopl
atin
/pac
litax
el; C
RC
: col
orec
tal c
ance
r; E
CO
G: E
aste
rn C
oope
rativ
e O
ncol
ogy
Gro
up;
exp:
expo
nent
ial;
GIS
T: g
astro
-inte
stin
al s
trom
al tu
mor
; LD
H0:
base
line
lact
ate
dehy
drog
enas
e;lo
g(G
): lo
g of
the
tum
or g
row
th ra
te; N
SC
LC: n
on-s
mal
l-cel
l lun
g ca
rcin
oma;
PD
: pha
rmac
odyn
amic
; RC
C: r
enal
cel
l car
cino
ma;
sK
IT: s
olub
le s
tem
cel
l fac
tor;
SLD
: sum
of l
onge
st d
iam
eter
s; S
LD0:
base
line
sum
of l
onge
st d
iam
eter
s;
sVE
GFR
-3: s
olub
le v
ascu
lar e
ndot
helia
l gro
wth
fact
or re
cept
or 3
; TG
I: tu
mor
gro
wth
inhi
bitio
n m
odel
; TS
R: t
umor
siz
e ra
tio; T
TG:t
ime
to tu
mor
regr
owth
. (*)
Sur
viva
l an
alys
is w
as p
erfo
rmed
usi
ng n
on-p
aram
etric
met
hods
.
As shown in Table 1, most published tumor modeling incorporating TTE analyses since 2009 have utilized the TGI. Other than the TGI model, Wang et al. [35] presented a simple empirical model with exponential drug–dependent shrinkage and linear growth with time to describe tumor SLD data from four clinical trials of patients with NSCLC treated with various chemotherapies or placebo. This model structure was also used to describe data from NSCLC patients treated with carboplatin/paclitaxel alone or in combination with bevacizumab or motesanib [36].
In contrast to the model first presented by Wang et al., the TGI model typi-cally includes dose or drug exposure as PK drivers, and therefore has the potential to simulate tumor response at other dose levels. In addition, metrics derived from biomarker PKPD models have been implemented as “drivers” for tumor response and OS. For example, Hansson et al. [40] found that bi-omarkers time–courses of sVEGFR–3 and sKIT were drivers of tumor re-sponse, and sVEGFR-3 response was the best predictor of OS. Biomarker analyses such as these can confirm or drive new hypotheses as to how the drug is reducing tumor burden and may provide an early indication of anti–tumor efficacy.
From Table 1, the tumor baseline SLD (SLD0) and ECOG status are con-sistent patient characteristics predictive for OS in most tumor types. The model–based metric of TSR (at week 6–8) was identified as a predictor for OS in NSCLC [35, 36], colorectal cancer [29, 41], and metastatic breast cancer [39]. In addition to TSR, the relatively new model–based metric of TTG was shown as a predictor for survival in a parametric TTE analysis of colorectal cancer data [41] and renal cell cancer [43]. The interpretation of these survival predictors is as follows: 1) an increased tumor baseline (SLD0), reflective of higher initial tumor burden, is associated with poorer survival prognosis; 2) a smaller TSR (week 6–8) and a shorter TTG, reflec-tive of treatment efficacy, is associated with poorer survival prognosis; and 3) an increased ECOG status, reflective of poorer health, is associated with poorer survival prognosis.
Utility of a modeling and simulation framework in oncology Development of population PKPD and TTE models within the modeling framework aid in the understanding of the link between drug exposure, PD response, and survival, and provide a tool for optimizing drug treatment. The application of this modeling framework has been demonstrated, with notable examples: 1) the long–term clinical outcome in Phase III studies was predicted based on short–term Phase II data [29, 36]; 2) PFS and overall response rate (ORR) were predicted for thyroid cancer patients treated with motasenib [37, 38]; and 3) the capecitabine dose that would show non–
21
inferiority to the dose currently registered, when used in combination with docetaxel, was predicted [39].
22
Metastatic Breast Cancer Treatment
Shown in Figure 8 is a timeline of the US Food and Drug Administration (FDA) approval of drugs for metastatic breast cancer, adapted from Cortazar et al. [44]. In this thesis work, pharmacometric model have been developed from clinical trials of paclitaxel [45], docetaxel [46], and trastuzumab em-tansine (T-DM1) [47-51] from metastatic breast cancer patients.
Figure 8. FDA approval timeline of drugs for metastatic breast cancer
Paclitaxel and Docetaxel Paclitaxel and docetaxel are members of the taxane drug class, anti–mitotic chemotherapies that work by interfering with spindle microtubules leading to cell cycle arrest and apoptosis. Paclitaxel (Taxol®, Abraxane®) is a drug approved for treatment of ovarian, breast, lung, pancreatic and other cancers. Docetaxel (Taxotere®, Docecad®) is approved for treatment of locally ad-vanced or metastatic breast cancer, head and neck cancer, gastric cancer, hormone–refractory prostate cancer and non-small cell lung cancer. Shown in Figure 9 are the chemical structures for docetaxel and paclitaxel.
Paclitaxel1994 Gemcitabine
2004
2006Albumin-bound
Paclitaxel2010
Eribulin
Lapatinib,Ixabepilone
2007
Trastuzumab emtansine (T-DM1)2013
Trastuzumab,Capecitabine
1998
1996Docetaxel
2012Pertuzumab
23
Figure 9. Doceta el and Paclita el
Trastuzumab emtansine (T-DM1) T-DM1 (trastuzumab emtansine, Kadcyla®) is an ADC approved by the FDA in 2013 for the treatment of HER2–positive cancers [52, 53]. ADCs are a relatively new promising class of compounds in clinical development for the treatment of cancer [54-57]. ADCs harness the targeting capability and long half–life of antibodies to deliver potent small molecule cytotoxins that may be too toxic to directly administer and/or have poor PK properties. Shown in Figure 10 is a schematic of T-DM1.
Figure 10. T-DM1 structure
T-DM1 is composed of 1) trastuzumab, a humanized monoclonal antibody directed against the extracellular region of the HER2 antigen, 2) DM1, an anti–microtubule agent derived from the plant extract maytansine which inhibits cell division through tubulin binding, and 3) a lysine–SMCC link-
ed to conjugate DM1 to trastuzumab.
Docetaxel Paclitaxel
24
Trastuzumab The antibody backbone of T-DM1 is trastuzumab (Herceptin®), a recombi-nant, humanized anti–HER2 monoclonal IgG antibody approved in 1998 for the treatment of HER2–positive metastatic breast cancer. HER2 is a protein overexpressed in approximately 30% of breast cancer patients, and these patients are denoted as HER2–positive. HER2 overexpression is associated with uncontrolled cell growth, poor prognosis, and shorter survival time [58]. Trastuzumab targets the HER2 receptor, and binding inhibits down-stream signaling associated with tumor growth. HER2–positive breast can-cers can have up to 25–50 copies of the HER2 gene and up to 40–100–fold increase in HER2 protein resulting in 2 million receptors expressed at the tumor cell surface [59].
DM1 DM1 is a potent, small molecule cytotoxic derivative of the microtubule inhibitor maytansine, which was abandoned as a chemotherapeutic due to a narrow therapeutic index [56]. DM1 is chemically linked to lysine residues on trastuzumab via a stable non–reducible (“non– cleavable”) thioether bond using the SMCC (N–succinimidyl–4–(N–maleimidomethyl)–cyclohexane–1–carboxylate) linker.
T-DM1 Dose Product ADCs made by conventional drug conjugation strategies (using antibody lysine side–chain amines or sulfhydryl groups for conjugation) are complex, heterogeneous mixtures of various drug–to–antibody ratio (DAR) moieties. During the manufacturing process for T-DM1, DM1 may conjugate to one or more lysine residues on trastuzumab, yielding a T-DM1 dose product containing a mixture of different numbers of DM1 (from 0 to 8) per trastuzumab antibody. Understanding the PK underlying these ADCs has been elusive, primarily due to the lack of analytical techniques capable of measuring the individual DAR moieties comprising the ADCs. Shown in Figure 11 is an illustration of the T-DAR moieties.
25
Figure 11. T-DM1 dose product containing multiple DARs. DAR: drug–to–antibody ratio; The total trastuzumab concentration is the sum of each individual DAR moie-ty and free trastuzumab (i.e., unconjugated DAR0) concentrations; The T-DM1 concentration is sum of DAR moieties with at least one DM1 conjugated to trastuzumab.
T-DM1 Mechanism of Action Following uptake via the HER2 receptor, T-DM1 is degraded in the lyso-some. DM1, and DM1–containing catabolites, are released intracellularly and bind tubulin, thereby disrupting microtubule assembly/disassembly and selectively killing HER2–overexpressing tumor cells [60]. Notably, HER2+ patients may benefit from a dual pharmacologic effect as T-DM1 retains the anti–tumor mechanism of trastuzumab [61].
Figure 12. T-DM1 Mechanism of Action. HER2: human epidermal growth factor receptor 2; Mab: monoclonal antibody
HER2 receptor
26
Aims
The aim of this thesis work was to develop pharmacometric models and strategies for analyzing 1) antibody drug conjugate (ADC) disposition, 2) ADC–driven hepatotoxicity and thrombocytopenia, and 3) clinical tumor response/survival from taxane treatment. The developed models and strate-gies, derived from HER2–positive and HER2–negative metastatic breast cancer breast cancer treatment and data, provide important pharmacometric tools to apply within the modeling and simulation framework proposed for oncologic drug development.
The specific aims were:
1. To develop a pharmacokinetic–pharmacodynamic (PKPD) model describing platelet response to T-DM1 treatment in HER2–positive breast cancer patients.
2. To develop a PKPD model describing simultaneous platelet, ALT, and AST response to T-DM1 treatment in HER2–positive breast cancer patients.
3. To develop a PKPD model simulation strategy of alternative T-DM1 dose regimens, incorporating dose modification rules for platelet, ALT, and AST responses, towards supporting T-DM1 dose and scheduling.
4. To develop PK models elucidating and conceptualizing T-DM1 dis-position from preclinical experiments, that may guide study design, PKPD modeling, and drug development for other ADCs.
5. To develop models for tumor response to taxane treatment in HER2–negative breast cancer patients, characterizing patterns of tumor quiescence and drug–resistance, and apply results into a par-ametric time-to-event analyses of survival data.
27
Data and Methods
In this thesis, clinical trial data from T-DM1 (trastuzumab emtansine, Kadcyla®), docetaxel (Taxotere®, Docecad®), and paclitaxel (Taxol®, Abraxane®) were used for the development of pharmacometric models. For the five T-DM1 clinical studies used in this thesis, protocols were ap-proved by the institutional review boards of all participating institutions and were carried out in accordance with the Declaration of Helsinki, current US FDA Good Clinical Practices, and applicable local laws. Patients provided written informed consent The docetaxel and paclitaxel studies were conducted in accordance with the Declaration of Helsinki, the Good Clinical Practice guidelines of the Interna-tional Conference on Harmonization, and the laws and regulations of the countries involved. The protocol was approved by local ethics committees; written informed consent was obtained from all patients before screening. The T-DM1 PK studies in rats were approved by Institutional Animal Care and Use Committee (IACUC) and conducted at Genentech, Inc. The T-DM1 PK studies in cynomolgus monkeys were approved by IACUC and conduct-ed by Covance Laboratories, Inc. (Covance; Madison, WI) Table 2 outlines the study designs and data within this the-sis.
28
Tabl
e 2.
Ove
rvie
wof
Stu
dyD
esig
ns a
nd D
ata
Pape
rD
rug
Trea
tmen
tSp
ecie
s /
Mat
rixn
Dru
gD
ose
Leve
lR
oute
of
adm
inis
trat
ion
Mea
sure
men
ts
IT-
DM
1H
uman
164
q3w
regi
men
s: 0
.3 (n
=3),
0.6
(n=1
), 1.
2 (n
=1),
2.4
(n=1
), 3.
6 (n
=127
), an
d 4.
8 m
g/kg
(n=3
)
q1w
regi
men
s: 1
.2 (n
=3),
1.6
(n=2
), 2.
0 (n
=1),
2.4
(n=1
3), a
nd 2
.9 m
g/kg
(n=2
)
Intra
veno
usin
fusi
on;
0.5–
1.5
hr
Plat
elet
s
IIT-
DM
1H
uman
658
q3w
regi
men
s: 0
.3 (n
=3),
0.6
(n=1
), 1.
2 (n
=1),
3.6
(n=6
29),
and
4.8
mg/
kg (n
=3)
q1w
regi
men
s: 1
.2 (n
=3),
1.6
(n=2
), 2.
0 (n
=1),
2.4
(n=1
3), a
nd 2
.9 m
g/kg
(n=2
)
Intra
veno
usin
fusi
on0.
5–1.
5 hr
Plat
elet
s,AL
T,AS
T
IIIT-
DM
1Pl
asm
a–
100 μg
/mL
in v
itro
incu
batio
nT-
DM
1,
Tota
l tra
stuz
umab
,
Dru
g:An
tibod
yR
atio
s
Rat
340.
3 (n
=7),
3.0
(n=8
), 1
0.0
(n=1
0), a
nd
20.0
mg/
kg (n
=9)
Intra
veno
us
Cyn
omol
gus
Mon
key
430
.0 m
g/kg
(n=4
) In
trave
nous
Cyn
omol
gus
Mon
key
1410
.0 m
g/kg
q3w
(n=
14)
Intra
veno
us
IVD
ocet
axel
Hum
an18
510
0 m
g/m
2q3
w (m
axim
um9
cycl
es)
Intra
veno
usIn
fusi
on; 1
hr
Tum
orSL
D
Pacl
itaxe
lH
uman
242
90 m
g/m
2on
day
s 1,
8, a
nd 1
5 q4
w
q1w
: wee
kly;
q3w
: eve
ry3
wee
ks; q
4w: e
very
4 w
eeks
; ALT
: ala
nine
trans
amin
ase;
AST
: asp
arta
te tr
ansa
min
ase;
SLD
: sum
oflo
nges
tdia
met
er
Clinical Data
T-DM1 As shown in Table 2, data was used from 164 HER2–positive breast cancer patients for the development of a PKPD model of platelet response to T-DM1 treatment in Paper I. These patients were from the Phase I TDM3569g [48] and the Phase II TDM4258g [51] clinical trials and received doses rang-ing from 0.3–4.8 mg/kg.
In Paper II, a PKPD model of simultaneous ALT, AST, and platelet response to T-DM1 was developed. This PKPD modeling analysis was extended from Paper I to include ALT and AST data, as well as additional patient data from the Phase II TDM4374g [49], the Phase III TDM4370g [50], and the Phase II TDM4450g [47] studies. The final dataset therefore included 658 HER2–positive breast cancer patients from 5 clinical T-DM1 trials. This dataset consisted of 99% females, with a median age of 53 years. Sixty percent of patients had ECOG status of 0, 39% ECOG 1, and 1% ECOG 2. Eleven percent of patients were Asian, 44% had liver metastases, and 88% had measurable disease.
Docetaxel and Paclitaxel In Paper IV, tumor SLD data and survival times were available from 185 of 241 HER2–negative breast cancer patients in the docetaxel treatment group of the Phase III AVADO trial [46]. In this modeling dataset, 59% of patients were ECOG=0 and 41% were ECOG=1. Patients were women with a medi-an age of 55 years (range 29–83 years).
For paclitaxel, tumor SLD data and survival times were available from 242 of 326 HER2–negative breast cancer patients in the paclitaxel treatment group of the Phase III E2100 trial [45]. In this dataset, patients were women with a median age of 55 years (range 27–85 years); ECOG status was not available.
Preclinical Data In Paper III, for the development of PK models of T-DM1 disposition, two different T-DM1 dose products were dosed to rats and cynomolgus monkeys (T-DM1DAR3.1 and T-DM1DAR1.5).
30
Preparation of T-DM1DAR 3.1 and T-DM1DAR 1.5 Dose Products The two preclinical T-DM1 dose products were prepared by regulating the amount of SMCC linker added during the manufacturing process. Specifi-cally, less SMCC linker was added to generate the T-DM1DAR 1.5 dose prod-uct compared to more SMCC linker added to generate the T-DM1DAR 3.1 dose product. As shown in Figures 13 and 14 below, the T-DM1DAR 3.1 dose product (aver-age DAR=3.1) and T-DM1DAR 1.5 (average DAR=1.5) dose products con-tained various DAR moieties with the following percentages as measured by affinity capture liquid chromatography mass spectrophotometry (LC–MS).
Figure 13. T-DM1DAR 3.1 dose product
Figure 14. T-DM1DAR1.5 dose product
RatsTwo PK studies were conducted in vivo in rats. In the first study, 10 naïve Sprague–Dawley rats (n=5 rats/group) were administered a single intrave-nous dose of 10 mg/kg T-DM1DAR 3.1 or 10 mg/kg T-DM1DAR 1.5. Blood sam-ples were collected up to 21 days post–dose. T-DM1 DAR concentrations and total trastuzumab concentrations were determined.
For the second PK study, T-DM1DAR 3.1 was administered as a single IV dose at 0.3, 3.0, and 20.0 mg/kg (n=7–9 Sprague Dawley rats/group). Blood sam-ples were collected up to 42 days post–dose. Blood samples were processed for total trastuzumab and T-DM1 concentrations. Rats weighed between 193–283 g.
31
Cynomolgus Monkeys T-DM1DAR 3.1 was administered to cynomolgus monkeys both as a single IV dose (30 mg/kg; (n=4)), as well as every 3 weeks (10 mg/kg q3w (n=14)). Animals were approximately 3–5 years old and weighed 2.6–4.5 kg. Blood samples were collected via the femoral vein. Blood samples were processed for total trastuzumab, DAR, and T-DM1 concentrations.
Measurements Platelet, ALT, and AST In Papers I and II, laboratory data were collected at different time points following T-DM1 dosing and typically within a day before each new cycle treatment. For patients receiving q3w regimens in study TDM3569g, meas-urements were assessed on day 1, 2, 4, 7–8, 11, and 18 post-dose in cycle 1. Typical collection times in subsequent cycles included days 1, 4, and 7–8 post-dose. For patients receiving q1w regimens, measurements were as-sessed on day 1, 2, 4, 7–8, 11, 14–15, and 18 post–dose in cycle 1, with pre–dose weekly sampling beginning day 21. For TDM4258g, measurements were assessed every 7 days. For TDM4374g, measurements were assessed weekly during cycle 1 and on day 1 and day 8 post-dose of all subsequent cycles. For TDM4370g and TDM4450g, measurements were assessed week-ly. For Grade 3/4 assessment of ALT and AST, measurements were normal-ized to the upper level of normal (ULN) from the respective analytical lab; the median ALT ULN value was 45 units/L (range 22–85 units/L) and the median AST ULN value was 39 units/L (range 20–66 units/L).
Platelet, ALT, and AST Toxicity Grades Toxicity grades were based on the Common Terminology Criteria for Ad-verse Events (CTCAE) Version 4.0.
Table 3. Toxicity Grade Classification for ALT, AST, and PlateletsAdverse
EventGrade
1 2 3 4
ALT increased >ULN – >3.0 – >5.0 –
AST increased >ULN – >3.0 – >5.0 –
Platelet countdecreased
<LLN – 75 ( 1000/μL) <75 – 50 ( 1000/μL) <50 – 25 ( 1000/μL) <25 ( 1000/μL)
LLN: lower level of normal; ULN: upper level of normal
32
T-DM1 and Total Trastuzumab Concentrations Enzyme–linked immunosorbent assays (ELISA) were used to measure T-DM1 concentrations (trastuzumab bearing at least one T-DM1) and total trastuzumab concentrations (trastuzumab with or without DM1). T-DM1 concentrations were measured for Papers I–III, and total trastuzumab con-centrations were measured for Paper III. Figure 15 illustrates the ELISA process for T-DM1 and total trastuzumab.
Figure 15. ELISA format for T-DM1 and total trastuzumab, adapted from [62]. (Left panel): The total trastuzumab assay uses capture of ADC antibody (T-DM1 with or without DM1) via an antigen or target extracellular domain (ECD), with detection using labeled antibody to ADC antibody. (Right panel): The T-DM1 anti-body assay uses capture of ADC via an anti–DM1 antibody, with detection using labeled antigen or ECD.
T-DM1 Drug-to-antibody Ratio (DAR) Concentrations In Paper III, measurement of individual DAR moieties in plasma was done using an affinity capture liquid chromatography mass spectrophotometry (LC–MS) assay recently developed [63, 64]. The percentage of each DAR (DAR0 7) moiety was calculated by its relative signal intensity. The DAR plasma concentrations were then obtained by multiplying these indi-vidual percentages by the total trastuzumab concentration (from ELISA) at the respective time. Shown in Figure 16 is a representative LC–MS chroma-togram illustrating the signal intensities from individual T-DM1 DARs.
Total trastuzumab Assay T-DM1 Assay
ECDReceptor
Anti-human antibodydetection
Labeled ECDReceptordetection
33
Figure 16. LC–MS chromatogram illustrating the signal intensities from T-DM1 DARs. LC-MS was used for the in vitro plasma experiments and in vivo cynomolgus monkey and rat PK data.
Tumor Sum of Longest Diameter (SLD) In Paper IV, planned tumor SLD measurements were performed by investi-gators every 9 weeks until week 36 and every 12 weeks thereafter until dis-ease progression occurred. The docetaxel dataset contained 879 tumor SLD measurements (2–13 measurements/patient collected up to a maximum 138 weeks). The paclitaxel dataset contained 784 tumor SLD measurements (2–9 measurements/patient collected up to a maximum 105 weeks). For docet-axel, the median patient tumor baseline SLD was 69mm (range 10 – 308mm). For paclitaxel, the median patient tumor baseline SLD was 83mm (range 12 – 391mm).
Model Building and Data Analysis Post-hoc Bayesian estimates Two population PK models have previously been developed for T-DM1 by other investigators, and the post–hoc Bayesian estimates of the individual patient PK parameters were used for PKPD model development in Papers I and II. In the first PK analysis, T-DM1 plasma concentration–time data from the Phase I TDM3569g [48], the Phase II TDM4258g [51], and the Phase II study TDM4374g [49] were used to develop a linear two–compartment PK
DAR
DAR
DAR2
DAR3
DAR4
DAR5
DAR6
DAR7
34
model with first–order elimination from the central compartment [65]. In the second population PK analysis [66], the modeling was extended to include data from the Phase III TDM4370g [50], and the Phase II TDM4450g [47] studies. Table 4 outlines results from these population PK analyses.
Paper I: T-DM1 PKPD Model of TCP In Paper I, a PKPD model describing platelet response to T-DM1 treatment was developed. Model development started by considering the PKPD model of myelosuppression presented by Friberg et al. [18] (see Figure 4). During model development, modifications to the myelosuppression model (i.e., the addition of structural components and parameters) were made tak-ing into account two notable observations from the clinical studies.
First, in some patients, platelet–time profiles drifted slowly down over multiple cycles of T-DM1.
Second, platelet count nadirs were generally lowest after the first T-DM1 dose.
From 1–5 transit compartments were considered, as well as T-DM1 concen-trations acting directly on the platelet compartment. Shown in Figure 17 is a schematic of the final PKPD model for platelets.
Table 4. Summary of Post-hoc Bayesian estimates for PKPD model development
Paper Data Source
Typical Population Parameter Estimates
CL (L/day)(IIV%)
V1 (L)(IIV%)
CLd (L/day)(IIV%)
V2 (L)(IIV%)
I Phase I TDM3569g Phase II TDM4258g Phase II TDM4374g
0.70 (21)
3.33 (13.2)
0.78
IIV not estimated
0.89 (50)
II Phase I TDM3569g Phase II TDM4258g Phase II TDM4374g Phase III TDM4370gPhase II TDM4450g
0.676 (19.1)
3.13(11.7)
1.53(181)
0.66(75)
IIV: interindividual variability; CL: clearance; CLd: distributional clearance; L: liter; V1: central compartment volume; V2: peripheral compartment volume
35
Figure 17. Schematic of PKPD model for platelet response to T-DM1 treatment. BASE: baseline platelet count at time = 0, modeled as BASE1 + BASE2; BASE1: amount of baseline proliferating PP that is nondepletable; BASE2: amount of baseline proliferating PP that is depletable by the Cavg deplete rate; Cavg: average T-DM1 con-centration over dosing intervals; CL: clearance; CLd: intercompartmental clearance;
T-DM1 drug effect; GAM: feedback parameter that increases the prolif-eration rate when PLT count drops below BASE; kdeplete: rate of depletion of BASE2; kel: rate of physiologic elimination of circulating platelets; kPROL: rate of PP prolifera-tion; ktr: intercompartmental transit rate constant; MTT: mean transit time, equal to the number of intercompartmental transits divided by the transit rate constant (4/ktr); PLT: circulating platelet compartment; PP: platelet proliferation compartment; Slope: drug potency parameter; T1, T2, and T3: transit compartments; V1: T-DM1 central volume of distribution; V2 : T-DM1 peripheral volume of distribution.
As shown in Figure 17, individual patient PK parameters from the popula-tion PK model [65] provided T-DM1 central compartment concentrations (C), which acted as a linear inhibitory drug effect r-ation rate (kPROL) of the PP compartment.
The downward drift in some patient platelet–time profiles was modeled by the introduction of an additional slow T-DM1–related drug effect on a de-pletable fraction of the PP pool. Specifically, the PP compartment was mod-eled as a nondepletable (BASE1) proliferating pool and as a depletable (BASE2; derived by BASE- BASE2) proliferating pool of platelets. The drug effect (Cavg deplete) was incorporated to slowly deplete the BASE2 pool over time. The following equation was applied: BSL(t) = (BASE – BASE1 –Cavg deplete ) + BASE1.
CSlope 21
CirculatingPlatelets
ktr
kPROL= ktr
T1 T2 ktr
T3
ktr = 4 / MTTktr
kel = ktr
GAM
PLTBASEFeedback
Transit Compartments
V2
CLd
T-DM1Dose
V2
CLd
CL
Platelet Proliferation
Pool (PP)
BASE2
BASE1
CirculatingPlatelets
(PLT)
depleteavg kC
36
Two separate Slope parameters were used to capture the lower platelet nadir in the first cycle– Slope1 for the first dose, and Slope2 for all subsequent doses. System–related parameters that were estimated included MTT, BASE, BASE1, and GAM. Drug–related parameters that were estimated included Slope1, Slope2, and kdeplete.
Paper II: T-DM1 PKPD Model of TCP and Hepatotoxicity In Paper II, a PKPD model describing platelet, ALT, and AST response to T-DM1 treatment was developed. For ALT and AST, only two published models were available for consideration: 1) a precursor–dependent PKPD model for ALT response to trabedectin presented by Fetterly et al. [16], and 2) a linear effect compartment model predicting ALT response to amioda-rone presented by Pollak et al. [17]. The following patterns were also noted in the data.
Similar to the platelet–time profiles, some ALT–time and AST–time profiles gradually increased over multiple cycles of T-DM1.
Similar to the platelet count nadirs, ALT and AST levels were gen-erally highest after the first T-DM1 dose.
Ultimately, for ALT and AST modeling, the model structure and parameter-ization developed for platelets in Paper I was used as a starting point. This model utilizes a delay in maximum drug effect through transit compart-ments, and also a drug-induced drift in baseline. Employing structural simi-larities between platelet, ALT, and AST readily allows incorporation of pa-rameter correlations between PD responses.
s, available for Paper II provided robust data to refine the original modeling approach for platelets, and apply to ALT and AST. Shown in Figure 18 is a schematic of the final PKPD model for ALT, AST, and platelet response to T-DM1.
37
Figure 18. Schematic of PKPD model for platelet, ALT, and AST response to T-DM1. ALT: alanine transaminase; ALTcirc (pool): ALT circulating or pool compartment; AST: aspartate transaminase; ASTcirc (pool): AST circulating or pool compartment; BSL0: baseline measurement at time = 0; BSL2: secondary baseline value; BSL(t): baseline time–course; CL: clearance; CLd: distributional clearance; Cp(t): T-DM1 central compartment concentrations time–course; Ce(t): T-DM1 EC concentration time–course; Ce,50: T-DM1 EC concentration at 50% Emax; EC: effect compartment; Emax: maximum drug effect; GAM: feedback parameter; kEC: EC rate constant; kOUT: output rate; kPROL: rate of PLTprog proliferation; ktr: intercompartmental transit rate; n: Hill factor; PLTcirc: circulating platelet compartment; PLTprog: proliferative progenitor platelet pool compartment; Slope0: initial drug effect; SlopeSS: steady state drug effect; Slope(t): Slope time–course; T1, T2, and T3: transit compartments; TDEC: half–life decay parameter; V1: T-DM1 central volume of distribution; V2: T-DM1 peripheral volume of distribution.
As shown in Figure 18, the final model consisted of ALT, AST, and platelets submodels. For each PD response, the individually predicted T-DM1 central compartment concentrations (Cp(t)), based on parameter PK results from the extended population PK analysis [66], were used to drive the platelet, ALT, and AST responses. For platelets, Cp(t) acted as a linear inhibitory drug effect (Cp(t) (t)) on the proliferation rate (kPROL) of the PLTprog compartment. For ALT and AST, Cp(t) acted as a linear stimulation of the ALT or AST production rate (kIN = kOUT (t)), where BSL(t) is the baseline measure-ment at time=t.
ktr,L ktr,L
ktr,L ktr,L
ktr,L
ktr,S
ktr,S ktr,S
ktr,S ktr,S
kEC,X p(t)
kEC,X
ALTpool
T2
T3
ALTcirc
T1
ASTpool
T1
T2
T3
ASTcirc
EC
CLd
CL
V2
T-DM1Dose
V1
ktr,P ktr,P ktr,P ktr,P ktr,P
PLTprog T1 T2 T3 PLTcirc
38
In this analysis, the baseline time–course , (BSL(t)), for each PD response was described using an effect compartment (EC) approach in which theoret-ical T-DM1 EC concentrations (Ce(t)) typically either reduced (platelets), or increased (ALT and AST) baseline values slowly over time. Specifically, pretreatment baseline values at time = 0 (BSL0), slowly transitioned to a minimum (platelets) or maximum (ALT and AST) baseline (BSLmin/max) upon repeated T-DM1 dosing.
This cumulative dose effect was saturable and modeled using a sigmoidal Emax model. The following equation was applied for platelet, ALT, and AST BSL(t): BSL(t) = (BSL0 – BSL2 – e(t)
n )/( Ce,50n + Ce(t)
n )) + BSL2. BSL2 is a secondary baseline parameter for model fitting purposes. Ce(t) was predicted from the central compartment using an EC input rate of kEC Cp(t), and a first order output rate of kEC. This readily allowed assess-ment and implementation of correlations between platelet, ALT, and AST cumulative effects.
In Paper I PKPD platelet model, a Slope1 and Slope2 parameterization cap-tured the phenomenon of increased drug effect after the first dose. In Paper II, the modeling approach was refined and applied to ALT and AST re-sponses. The drug effect time–course , (Slope(t)), was modeled as an initial drug effect (Slope0) decaying to a steady state drug effect (SlopeSS), with a half–life decay parameter (TDEC). Correlations between platelet, ALT, and AST acute dose response were assessed and implemented.
After model building, dose modification rules for T-DM1 were incorporated into the PKPD model code to provide relevant simulations and VPCs. Dur-ing the model simulation process, a predose assessment, immediately before the scheduled dose, checked whether a 3 ALT, AST, or platelet was predicted, and model code incorporated the following dose modifications:
ALT or AST Platelet
Grade 3 Grade 4 Grade 3 Grade 4
Delay the next dose until concentrations recover to Grade 2
Dose reduce by 0.6 mg/kg for q3w regimens, and by 0.4 mg/kg for q1w regimens
DiscontinueT-DM1 Dosing
Delay the next dose until concentrations recover to Grade 1
Do not dose reduce
Delay the next dose until concentrations recover to Grade 1
Dose reduce by 0.6 mg/kg for q3w regimens, and by 0.4 mg/kg for q1w regimens
Discontinue T-DM1 treatment if more than 2 dose reductions are necessary
39
Paper III: T-DM1 PK Models In Paper III, PK models elucidating and conceptualizing T-DM1 disposition from preclinical experiments in rat and cynomolgus monkeys were devel-oped. Initially, a three–compartmental PK model was developed to best de-scribe total trastuzumab concentrations. Subsequently, the mechanistic T-DM1 PK model was fit simultaneously to total trastuzumab and DAR0DAR7 in vitro plasma stability and in vivo experiments. DM1 deconjugation occurring in V1,V2, and V3 compart-ments was considered. A model parameterization was also tested which as-sumed that DM1 deconjugation rates were independent of the number of DM1s attached or the conjugations site as described by Gibiansky et al. [67]; in this approach, k was the single deconjugation parameter, and the other intercompartmental deconjugation rate constants were factors of k , i.e., k , k , etc.
From the mechanistic PK model, a reduced T-DM1 PK model was devel-oped by fitting the model simultaneously to ELISA measurements of total trastuzumab and T-DM1 concentrati are schematics of the final T-DM1 PK models.
Figure 19. Schematic of the T-DM1 PK Models. (Left panel): Mechanistic model decribing concentration time–courses of individual DAR0 7 moieties and total trastuzumab (TT). CLTT: TT clearance from the central compartment; CLin vivo: in vivo antibody clearance from the central compartment; CLd2, CLd3: distributional clear-ances; DARn: n DM1 molecules bound to trastuzumab (drug-to-antibody ratio); Kplas-
ma: antibody degradation rate in plasma; K –1: DM1 deconjugation rate from higher to next DAR moiety; V1, V2, V3: volumes of distribution of central and peripheral compartments; (Right panel): Reduced T-DM1 PK model describing time–courses of T-DM1 and TT. T-DM1 is any trastuzumab molecule with at least one DM1 mole-cule. CLDEC: deconjugation rate from T-DM1 to unconjugated trastuzumab (DAR0).
CLAB
V2
V3
V1
CLD2
CLD3
K10
K21
K32
K43
K54
K76
K65
KAB
DAR7
DAR6
DAR5
DAR4
DAR3
DAR
DAR
DAR7
T-
DAR
CLTT
V2
V3
V1
CLD2
CLD3
CLDEC
CLTT
Mechanistic PK Model Reduced PK Model
40
As shown in the mechanistic PK model approach, individual DAR moieties are linked together through a catenary chain of sub–compartments within the central compartment (V1 model. DAR moieties under-go distributional clearance (CLd2 and CLd3) into two peripheral compart-ments (V2 and V3) and are cleared from the central compartment through antibody clearance (CLTT) processes. CLTT was parameterized as composed of a first order antibody degradation rate (Kplasma), supported by the in vitro plasma stability data, and other in vivo antibody clearance processes (CLin
vivo). Individual rates of DM1 deconjugation from DAR moieties were mod-eled as first order rate constants (K –1), where n is the higher DAR moiety, and n–1 is the subsequent DAR moiety in the catenary chain.
In the reduced PK model, T-DM1 is represented by a single compartment within the central compartment (V1
deconjugation parameter (CLDEC) described the conversion of T-DM1 to unconjugated trastuzumab (DAR0). T-DM1 and DAR0 distribute into pe-ripheral compartments and are cleared by antibody clearance (CLTT) pro-cesses.
Paper IV: Taxane Tumor Model Tumor Model In Paper IV, a general model for tumor response was developed base on taxane treatment, and results were applied in parametric TTE analyses of survival data. Tumor model development began by first considering the TGI model of tumor SLD proposed by Claret et al. [29] (see Figure 6). However, the TGI model did not adequately fit certain patterns of tumor response, primarily those in patients exhibiting long periods of tumor quiescence and regrowth. Therefore, modifications to the TGI model (i.e., the addition of structural components and parameters) were made. The schematic for the final tumor model is shown in Figure 20.
41
Figure 20. Schematic of the tumor model. SLD0: tumor SLD at time=0; FnR: frac-tion of tumor SLD0 quiescent at time=0; 1–FnR: fraction of tumor SLD0 sensitive to drug treatment; kGrow,Sens: growth rate constant of drug–sensitive tumor fraction; kGrow,Resist: growth rate constant of drug–resistant tumor fraction; kKill: tumor kill rate constant; kDelay: transit compartment delay rate constant from quiescent tumor to drug–resistant tumor; kDrug: drug effect elimination rate constant; R1, R2, and R3: transit delay compartments of nonproliferating drug–resistant tumor cells. The over-all tumor SLD is the sum of all six tumor compartments. The drug–resistant tumor SLD is the sum of the Tumor Quiescent, R1, R2, R3 and Proliferating Tumor Drug–Resistant compartments.
The final tumor model consisted of a chain of five compartments mimicking tumor drug–resistance, and a tumor drug–sensitive compartment susceptible to drug treatment. The resistant tumor compartments consisted of an initial tumor quiescent compartment, 3 transit compartments (R1, R2, and R3), and a proliferative drug–resistant tumor compartment. During model building, from 1–5 transit compartments were considered. The total tumor SLD is represented as the sum of all 6 compartments.
0, where SLD0 equals the observed tumor baseline and FnR is a model parame-ter describing the fraction of SLD0 which transitions to proliferative drug–resistant tumor; the drug–sensitive tumor SLD was derived by (1– SLD0. The drug–sensitive tumor and drug–resistant tumor compartments proliferate according to first order growth rate constants, kGrow,Sens and kGrow,Resist, respectively. The antitumor taxane drug effect was described by
kKill
kDrug
Drug
R1
R2
R3
kDelay
kDelay
kDelay
kDelay
kGrow,Sens
kGrow,Resist
Tumor Sensitive
(1-FnR) SLD0
Tumor ResistantProliferating
Tumor Resistant
Quiescent
FnR SLD0
Drug Effect
42
first order tumor kill rate (kKill) multiplied by the predicted taxane time–course Drug(t)). Drug(t) was based on the administration schedule and a first order drug loss rate (kDrug). Lastly, the first order kDelay rate parameter de-scribed the transition time from tumor quiescence to proliferative drug–resistant tumor.
Overall Survival (OS) Model In Paper IV, TTE survival models were developed according to the follow-ing equation for the hazard function: h(t) = h0(t) e 1 1 2 2 n n
The baseline hazard function, h0(t), was first determined considering expo-nential, Weibull, and log–normal distributions. The best model, determined by the lowest objective function value and VPCs, was carried forward in the covariate
[3, 34].
Model building methods For PK and PKPD model building, the First Order Conditional Estimation (FOCE) method with INTERACTION using NONMEM 7.3.0 was used [68]. Platelet, ALT, AST, and tumor SLD data were log–transformed. Log–normal parameter distributions were typically used for IIV, where the pa-rameter value for the ith patient was represented by Parameteri = Typical Value exp( i), where the CV of i represents the IIV. The residual error was modeled as a proportional error, which is additive in the log domain. For the kdeplete (Paper I), platelet keo (Paper II), and kdelay (Paper IV) parame-ters, the distributions were bimodal and the mixture model implementation in NONMEM was used. A difference in OFV of >6.63, corresponding to a significance level of P < 0.01, was used for discrimination between two nested models that differed in one parameter.
OS model building was performed using the FO method in NONMEM 7.3.0. A difference in OFV of >6.63, corresponding to a significance level of P < 0.01, was used for discrimination between two nested models that differed in one parameter.
Covariate Analyses For Papers I, II, and IV, analyses were used to identify covariates (e.g. pa-tient baseline characteristics, model–based metrics) that might explain sources of IIV on model parameters. Shown in the Table 5 are the covariates assessed for each paper.
43
In Papers I and II, the covariate analysis was first performed using the line-arized FOCE method with stepwise covariate model building [69], to screen the considered covariates for significance on any parameter. If significant covariate parameter correlations were found (p<0.01), these were assessed using the nonlinear FOCE method for inclusion into the model. In Paper IV, covariates were first screened separately for significance, and the most sig-nificant covariate by OFV was retained in the model. This procedure was repeated adding covariates stepwise until no significance remained.
Model Evaluation and Simulation For Papers I, II, and III, model simulations were done for model evaluation, dose comparison, and illustration of PK concepts. Table 6 introduces specif-ics of model simulations.
Table 5. List of covariates assessed in Papers I, II, and IVCovariate
Type Paper I Paper II Paper IV
Categorical Race, Prior myelosuppressive chemotherapies (paclitaxel, docetaxel, or carboplatin) (Yes/No)
Race=Asian (Yes/No), Liver metastases (Yes/No), Measurable disease (Yes/No)
Continuousa Age, Weight, Creatinine clearance, Serum creatinine, Albumin, Total protein, Platelet count, ALT, AST count, Total bilirubin, Tumor SLD, HER2 ECD
Age, Albumin, HER2 ECD, Bilirubin, Tumor SLD,Number of disease sites
Age,Tumor SLD
Model–Basedb — — kDrug , kKill , kDelay , kGrow,Sens, kGrow, Resist, FnR, TSR6, TTG
ALT: Alanine transaminase; AST: Aspartate transaminase; SLD: sum of longest diameter; ECD: extracellular domain; ECOG: Eastern Cooperative Oncology Group; HER2: human epidermal growth factor receptor 2; TSR6: tumor size ratio at week 6; TTG: time to tumor growtha Baseline valuesb
44
Ta
ble
6. S
imul
atio
n P
roto
colf
or P
aper
s I,
II, a
nd II
I
Pape
rM
odel
T-D
M1
Dos
e(s)
Sim
ulat
edPu
rpos
eSi
mul
atio
n N
otes
IPl
atel
etT-
DM
1 PK
PD3.
6 m
g/kg
q3w
Exte
rnal
eval
uatio
nda
tase
tTD
M43
74g
Stra
tific
atio
nof
Gra
de3
TCP
by p
late
let
base
line,
T-D
M1
AUC
, and
T-D
M1
Cm
ax
100
sim
ulat
edre
plic
ates
110
patie
nts/
repl
icat
e10
dos
ecy
cles
sim
ulat
edSi
mul
atio
ns s
tratif
ied
by k
depl
ete
IIAL
T, A
ST,
and
Plat
elet
T-D
M1
PKPD
3.6
mg/
kgq3
w;
2.4
mg/
kgq1
w;
1.2
mg/
kgq1
w
Inte
rnal
eva
ulat
ion
data
stet
TDM
4370
gPP
Cs
Com
paris
on o
f Dos
e R
egim
ens
for %
G
rade
3 A
Es, D
I, an
d R
DI
200
sim
ulat
edre
plic
ates
338
patie
nts/
repl
icat
e10
dos
ecy
cles
sim
ulat
edM
odel
incl
udes
dose
mod
ifica
tion
rule
spe
r T-D
M1
pres
crib
ing
info
rmat
ion
IIIM
echa
nist
icT-
DM
1 PK
DAR
3.1:
3.6
mg/
kgan
d3.
6 m
g/kg
q3w
DAR
1.5:
3.6,
5.16
, 5.8
, 7.1
,an
d 8.
0 m
g/kg
q3w
Illus
tratio
nof
PK s
yste
mSu
ppor
t hyp
othe
sis
test
ing
for e
ffica
cyan
d to
xici
tyM
OA
DAR
AVG, T
-DM
1, a
nd C
onju
gate
dD
M1
time
cour
ses
wer
eel
ucid
ated
Red
uced
T-D
M1
PKD
AR3.
1: 3.
6 m
g/kg
Illus
tratio
nof
CL D
ECSu
ppor
t hyp
othe
sis
test
ing
for e
ffica
cyan
d to
xici
ty
Sim
ulat
ions
ofC
L DEC
incr
ease
dC
L DEC
AE:a
dver
seef
fect
; ALT
: Ala
nine
trans
amin
ase;
AST
: Asp
arta
te tr
ansa
min
ase;
AUC
: are
a un
der t
he c
urve
; Cm
ax: m
axim
al c
once
ntra
tion;
D
ARAV
G: a
vera
ge d
rug:
antib
ody
ratio
; DI:
dose
inte
nsity
; MO
A: m
echa
nism
of a
ctio
n; P
PC: p
oste
riorp
redi
ctiv
e ch
eck;
TC
P: th
rom
bocy
tope
nia;
q1
w: w
eekl
y; q
3w: e
very
3 w
eeks
; RD
I: re
lativ
e do
sein
tens
ity
Model Evaluation Posterior Predictive Checks (PPCs) for Grade 3 and Grade 4 To icities As shown in Table 6, an external evaluation was performed in Paper I using platelet data from the TDM4374 trial. 100 simulation replicates were gener-ated from the PKPD model using the TDM4374g patient number (N=110; 3.6 mg/kg q3w). The ability of the model to predict the incidence of Grade
most patients were still on study, and Grade usually reached within three treatment cycles. The observed incidence of Grade with the simulated predictions. These probabilities were further separated by quartiles of observed baseline platelet count, T-DM1 Cmax (calculated by Dose/V1), and T-DM1 area under the curve (AUC) (calculated by Dose/CL).
In Paper II, an internal evaluation was done using the TDM4370g portion of the model building dataset (~ 50% of total patients); 200 simulation repli-cates were generated using the TDM4374g dataset as the template (N=338; 3.6 mg/kg q3w). PPC histograms of percentages of model–predicted Grade 3 and Grade 4 toxicities in the 200 datasets were compared with observed toxicities.
Visual Predictive Checks (VPCs) For VPCs, 90% prediction intervals with 95% confidence intervals were obtained by simulating 100 datasets from the model using the original da-tasets as templates. For Papers I, II, and IV, the model–predictions were examined to determine whether observed 5th, 50th, and 95th percentiles of data fell within the 95% confidence intervals of the corresponding percen-tiles derived from model simulations. For Paper III, due to the relatively small number of animals used, the model–predictions were examined to determine whether the observed 50th prediction of data fell within the 95% confidence intervals of the corresponding prediction derived from model simulations.
For OS model evaluation, VPCs were done by comparing 100 simulated Kaplan–Meier curves, based on the final OS model estimates, to the ob-served docetaxel and paclitaxel survival data [70].
46
Results/Discussion
Papers I and II This impetus for the original analysis in Paper I was the fact that thrombocy-topenia (TCP), a reduction in platelet counts, is the dose–limiting toxicity for T-DM1. Therefrom, a PKPD model was developed describing platelet re-sponse to T-DM1 treatment in HER2–positive breast cancer patients, and patient covariates were investigated in attempts of explaining factors that may contribute to TCP. In Paper II, the model was updated with data from 494 additional patients to simultaneously describe platelet, ALT, and AST response. Hepatotoxicity, in the form of elevated ALT and AST enzymes, is also a common Grade 3/4 AE for T-DM1 [71].
For PKPD modeling, there was a clear concentration-effect relationship and thus the time-course of T-DM1 was used to drive platelet, ALT, and AST responses. For platelet response, this has mechanistic support as trastuzumab, as a single agent, has not been associated with myelosuppres-sive effects [72]. Secondly, maytansine, the parent drug of DM1 and of simi-lar structure, showed no substantial myelosuppressive effects in Phase I [73] and Phase II [74] studies. Thirdly, TCP has not been reported as a significant toxicity with other DM1–containing antibody drug conjugates (ADCs) [75, 76]. A T-DM1 effect acting on the platelet progenitor compartment was chosen as the best model. Experimental support for this recently been shown by Uppal et al. [77], who concluded that T-DM1 is internalized into megakaryocytes (i.e. platelet precursors) in a HER2–independent, FcgRIIa–dependent manner, resulting in the intracellular cytotoxic release of DM1. Experimental support for T-DM1 concentrations as the PK driver for ALT and AST responses has been shown in vivo by Poon et al. [78], who deter-mined that elevations were driven by T-DM1 in a HER2 antigen–independent manner. Similarly, this drug effect was applied to the “presys-temic” pool compartment.
47
In Figures 21–23, VPCs and model simulations show that the PKPD models developed in Papers I and II well–described the data and incidences of Grade 3 toxicities. The model in Paper I well–predicted the platelet time–course data, as well as the ~8% incidence of Grade 3 TCP determined in the model building and evaluation datasets (Figure 21). Patients with lower platelet baseline were more likely to experience Grade 3 TCP, as a higher incidence (~25%) occurs when baseline platelet counts are <201 1000/ L.
Figure 21. VPC and model simulations from Paper I (T-DM1 3.6 mg/kg q3w). (Left panel): VPC: The solid red line represents the median of the observed platelet data (open circles). The red shaded region represents the 95% confidence interval of the model simulated median. The outer blue shaded regions represent the 95% confi-dence interval around the model simulated 5th and 95th percentiles. The stippled red lines represent the 5th and 95th percentiles of the observed platelet observations. (Right panel): Grade are stratified by “All” baseline platelet counts, and by quartiles of observed baseline platelet counts (lower 25%, 25–75%, and upper 25%).
In Figure 22, Paper I model simulations show that the model also well–described the two patterns of platelet response in the evaluation dataset: 1) those patients with “stable” platelet nadirs (kdeplete,POP1 patients), and 2) those patients with a downward drift in platelet nadirs and post nadir counts over time (kdeplete,POP2 patients).
48
Figure 22. Model simulations (T-DM1 3.6 mg/kg q3w) stratified by the two patterns of platelet response to T-DM1. (Left panel): kdeplete,POP1 patients, i.e. those with a stable platelet count nadir over time; (Right panel): kdeplete,POP2 patients, i.e. those with a downward drift in nadir and post nadir counts over time. In each plot, the solid black line represents the median of the model–building platelet data. The solid red line represents the median of the evaluation dataset platelets (open circles). The light blue shaded region represents the 95% confidence interval of the model simu-lated median. Blue lines represent the 90th prediction interval. The stippled grey lines represent the 95% confidence interval around the model simulated 5th and 95th percentiles. Horizontal grey line indicates the cutoff for Grade 3 1000/ L).
Shown in Figure 23 are the Paper II VPCs and posterior predictive checks (PPCs) for platelet, ALT, and AST responses. As shown for the VPCs, the model well–predicted the longitudinal ALT, AST, and platelet data simulta-neously. PPCs show that the model also predicted the incidences of Grade
3 events for platelet (~17%), ALT (~4%), and AST (~6%) in the evaluation dataset simultaneously.
49
Figure 23. Paper II VPCs and PPCs. VPCs (Left panels): The solid black line rep-resents the median of the platelet, ALT, and AST observations (black circles). The red shaded region represents the 95% confidence interval of the model simulated median. The outer blue shaded regions represent the 95% confidence intervals of the model simulated 5th and 95th percentiles. The stippled black lines represent the 5th and 95th percentiles of the observed platelet, ALT, or AST observations. Vertical lines indicate binning intervals for VPCs. (Right panels): PPC histograms showing the distribution of the model predicted percentages of patients i-ties from 200 simulated trials. The solid blue line is the 50th percentile of the model predicted percentages. The stippled blue lines represent the 5th and 95th percentiles of the model predicted percentages. The solid red line is the observed incidence of toxicity in the TDM4370g study as an internal evaluation dataset.
The VPCs also show that the higher drug effect after the first T-DM1 dose is well–described by the model. Most evident is the lowest platelet nadir occur-ring ~Day 8 (Figure 23). Similarly, ALT and AST concentrations are slight-ly higher at ~Day 8. Final parameter estimates from the two PKPD models developed in Paper I and II have been integrated into Table 7.
50
Tabl
e 7.
Fin
alP
KP
D M
odel
Est
imat
es fo
r Pap
er I
and
Pap
er II
Pape
rIPa
perI
I
Para
met
erD
escr
iptio
nU
nit
Plat
elet
Plat
elet
ALT
AST
Valu
eIIV
%Va
lue
IIV%
Valu
eIIV
%Va
lue
IIV%
Slop
e 1-3
)Sl
ope
forf
irstd
ose
L/m
g2.
9736
.1—
——
——
—
Slop
e 2-3
)Sl
ope
fors
ubse
quen
tdos
esL/
mg
1.82
56.3
——
——
——
Slop
e 0-3
)In
itial
Slop
eva
lue
L/m
g—
—52
.023
.140
.740
.5 b
42.6
40.5
b
Slop
e SS
-3) a
Slop
eat
ste
ady
stat
eL/
mg
——
2.08
47.9
4.28
83.0
b10
.283
.0 b
T DEC
-2)
Hal
f-life
for S
lope
0SS
wee
k—
—1.
3953
.72.
9785
.3 b
3.17
85.3
b
BASE
(or B
SL0)
Base
line
at ti
me
= 0
coun
tc25
532
.326
428
.224
.046
.427
.237
.7
BASE
1N
onde
plet
able
plat
elet
base
line
coun
tc11
837
.4—
——
——
—
BASE
2D
eple
tabl
epl
atel
etba
selin
eco
untc
137
——
——
——
—
BSL 2
Seco
ndar
yba
selin
eco
untc
——
1.41
52.9
43.7
67.9
56.6
42.4
P(1)
Prob
abilit
yfo
r Pop
ulat
ion
1—
0.55
4—
0.84
0—
1.0
—1.
0—
k dep
lete
, PO
P1-4
)BA
SE2
depl
etio
nra
te; P
OP1
L/m
g-1
6.25
88.1
e—
——
——
—
k dep
lete
, PO
P2-4
)BA
SE2
depl
etio
nra
te; P
OP2
L/m
g-1
84.2
88.1
e—
——
——
—
k EC
, PO
P1-5
)EC
rate
con
stan
t; PO
P1hr
-1—
—0.
889
86.9
0.91
784
.60.
164
74.0
k EC
, PO
P2-5
)EC
rate
con
stan
t; PO
P2hr
-1—
—7.
6451
.0—
——
—
Cef
f,50
dT-
DM
1[E
C] a
t 50%
Em
axm
g/L
——
0.90
00
0.30
30
0.15
60
Emax
Max
imum
dru
gef
fect
from
EC
L/m
g—
—0.
635
54.2
1.20
111
3.70
67.2
nH
ill fa
ctor
——
—2.
67—
2.67
—1.
66—
GAM
Plat
elet
feed
back
para
met
er—
0.13
5—
0.15
0—
——
——
k tr
-2) e
Inte
rcom
partm
enta
ltra
nsit
rate
hr-1
——
——
2.35
31.6
3.29
28.5
MTT
Mea
ntra
nsit
time
hr37
.424
.559
.523
.8—
——
—
Res
. Err
.R
esid
uale
rror
—18
.4%
—19
.2%
—15
.1%
9.85
%
EC =
effe
ctco
mpa
rtmen
t; [E
C] =
effe
ctco
mpa
rtmen
tcon
cent
ratio
n; P
OP
= po
pula
tion
aAs
ian
ethn
icity
was
a s
igni
fican
t cov
aria
te fo
r Pla
tele
t Slo
pess
bSh
ared
ETA
impl
emen
ted
due
to a
ppro
xim
atel
y 10
0% c
orre
latio
n; s
cale
d va
lues
= 1
.93,
1.5
8, a
nd 1
.24
for A
LT S
lope
0, AL
T Sl
ope s
s, an
d AL
T T D
EC,
resp
ectiv
ely
(see
Pap
erII
for d
etai
ls)
c dIIV
% fi
xed
to 0
ek P
RO
Lse
t equ
ival
ent t
o k t
r(M
TT/4
) for
pla
tele
t sub
mod
el;k
OU
Tse
t equ
ival
ent t
o k t
rfo
r ALT
and
AST
sub
mod
els
fO
MEG
ASA
ME
optio
n us
ed fo
r IIV
In both Paper I and II, a higher drug effect for platelets was noted after the first T-DM1 dose. In Paper I, a simple Slope1 (for the first dose) and Slope2 (for all subsequent doses) parameterization was used. Paper II refined this approach, using a Slope0 transitioning to SlopeSS according to the TDEC pa-rameter. For both approaches, the initial Slope value (Slope0 and Slope1) was highest, effectively capturing the low platelet nadir after the first dose. Most patients in the model–building dataset demonstrated this pattern of lowest platelet nadir in the first cycle, with variable degrees of increased nadir counts in the second and subsequent cycles.
Paper II also described the phenomena that ALT and AST show the highest responses after the first dose (i.e. SlopeSS > Slope0). Parameters of Slope0, SlopeSS, and TDEC were highly correlated between ALT and AST, but not with platelets. These correlations are supported by physiology, given that both AST and ALT are likely released into circulation upon cellular damage from T-DM1 treatment. Slope0 was not correlated with SlopeSS for platelets, ALT, or AST, suggesting that the steady state treatment response cannot be predicted from the initial dose response.
In Paper II, the downward drift in platelet nadir was also refined from Paper I, utilizing BSL0 and BSL2 instead of BASE1 and BASE2. In Paper I, the following equation was applied for BSL(t):
BSL(t) = (BASE – BASE1) e p(–Cavg kdeplete time) BASE1
In Paper II, the equation for BSL(t) was updated to: BSL(t) = (BSL0 – BSL2) e p(( – Ema Ce(t)
n )/( Ce,50n Ce(t)
n )) BSL2
In this approach, the model more appropriately predicted data following dose delays that had been built into the model, as “time” was replaced by an effect compartment approach. Ce(t) was predicted from the EC compart-ment, using an input rate of kEC Cp(t), and first order elimination of kEC.
In each approach for BSL(t), a population approach on a key parameter (kdeplete in Paper I, and kEC in Paper II) determined the probability of a patient belonging to 1) a population (POP1) with a stable platelet nadir, and 2) a population (POP2) with a downward drift in platelet nadir and post-nadir counts over time. In each modeling analyses, the platelet–time profiles for the typical patient in POP2 was shown to stabilize after 24 weeks (eight treatment cycles), and above Grade 3 TCP.
In addition to Slope0, SlopeSS, and TDEC, kEC was also highly correlated be-tween ALT and AST, but not with platelets. Thus, patients showing gradual elevations in ALT concentrations over time coincided with AST elevations,
52
but not with platelet decline. The mixture model was not supported for ALT and AST, and patient profiles were typically stable over time.
In the covariate analysis, the Paper II analysis revealed that Asian patients are more sensitivity than non-Asians with regard to platelet response. Asian ethnicity was a significant covariate for platelet SlopeSS, the steady state drug effect parameter which describes the extent of platelet nadir in the later T-DM1 dosing cycles. This ethnic difference has previously been reported by Dieras et al. [71]. In this dataset, 38% of Asian patients had Grade 3 TCP and 5 % Grade 4 TCP versus 12.5% Grade 3 TCP and 2% Grade 4 TCP in non-Asians. The covariate analysis did not find any significant predictor for ALT or AST response, and therefore an individual ALT or AST response to T-DM1 cannot be predicted a priori from any baseline characteristics tested.
Overall, the PKPD modeling analysis quantified the T-DM1 concentration-effect relationship simultaneously for platelet, ALT, and AST responses. The correlations that were found and implemented for model simulations, in addition to the dose modification rules, was instrumental for accurate model evaluation and simulation.
Figure 24 illustrates additional seminal conclusions, with regard to platelets, from the Paper I and Paper II modeling results: 1) the typical Asian patient (top panel) has lower platelet nadirs versus Non–Asian patients (bottom panel); 2) there are two typical patterns of platelet response in patients re-ceiving T-DM1; 3) typical patient platelet responses stabilize above Grade 3 TCP; and 4) the first-dose drug effect corresponds to a typical platelet count nadir of 115 1000/ L in the first cycle and then a higher platelet nadir of 145 1000/ L in the second cycle (for Non–Asians).
53
Figure 24. Typical platelet–time profiles following T-DM1 administration (3.6 mg/kg q3w). Black lines show the model–predicted platelet response for the typical Asian versus Non–Asian patient (kEC,POP1 patients). Blue lines are the corresponding model–predicted platelet responses for the patient subpopulation (kEC,POP2 patients) with faster platelet nadir decline. Grey points indicate the individual patient platelet data from the model building dataset. The solid red line indicates the Grade 3 threshold for platelets.
Paper II Simulations In Paper II, a PKPD model simulation strategy was developed towards sup-porting T-DM1 dose and scheduling. The simultaneous modeling approach, parameter correlations, and incorporation of T-DM1 dose modification rules, allowed for prediction of dose intensity (DI) and relative dose intensity (RDI), which may be used as endpoints for overall treatment safety assess-ment and as a target drug exposure for clinical decision–making. The match-ing of T-DM1 steady state exposure (AUCSS; 1.2 mg/kg q1w) and steady
54
state maximum concentrations (Cmax,SS; 2.4 mg/kg q1w) to the approved dose (3.6 mg/kg q3w) may be relevant to consider for drug development, in which a target exposure or concentration may direct dose optimization. Shown in Figure 25 are the model predictions for PK and PD for these dose regimens based on the typical PKPD parameter estimates (see Tables 4 and 7).
Figure 25. Model–predicted T-DM1 concentration time–courses (top) and ALT, AST, and platelet time–courses (bottom) for the typical patient receiving 1.2 mg/kg q1w, 2.4 mg/kg q1w, and 3.6 mg/kg q3w T-DM1. q1w: weekly; q3w: every 3 weeks
55
As shown in Figure 25, the shortened dose interval with T-DM1 q1w dosing prevents full return towards baseline prior to the next dose for ALT, AST, and platelets, in contrast to T-DM1 q3w dosing. As dose interruptions are based on predose measurements, more dose interruptions may be expected for q1w dosing. With q3w dosing, more Grade 3 events may be expected during the treatment cycle compared to q1w, but the recovery to baseline prevents the need for dose modifications. Shown in Table 8 are the results of T-DM1 PKPD model simulations, comparing the percentages of simulated patients with Grade 3/4 events, based on weekly assessments, and dose mod-ifications for these dose regimens.
56
Tabl
e 8.
PK
PD
Mod
elS
imul
atio
ns o
fGra
de3/
4 A
dver
seE
vent
s an
d D
ose
Mod
ifica
tions
a
Even
tD
escr
iptio
n
T-D
M1
Dos
eR
egim
en
3.6
mg/
kg q
3w1.
2 m
g/kg
q1w
2.4
mg/
kg q
1w
Med
ian
%R
ange
%(5
th95
th)
Med
ian
%R
ange
%(5
th95
th)
Med
ian
%R
ange
%(5
th95
th)
Even
ts
Plat
elet
, ALT
, or A
ST b
33.0
14.5
44.5
Plat
elet
23.6
7.67
27.1
ALT
4.42
2.06
6.49
AST
10.9
5.60
17.7
ALT
or A
ST12
.47.
0820
.4
Gra
de 4
Ev
ents
Pl
atel
et, A
LT, o
r AST
5.31
(0
(2.
07(
Plat
elet
5.01
7.08
)0
1.77
ALT
00
0
AST
00
0
ALT
or A
ST0
00
Dos
e M
odifi
catio
ns1
Dos
e R
educ
tion
2.07
(7.
08(
21.8
(
4.87
14.5
44.5
Dos
e di
scon
tinua
tions
0.
295
1.18
6.78
Dos
e In
tens
ity(m
g/kg
/wk)
c1.
181.
152.
07
Rel
ativ
e(%
) d98
.395
.786
.2
aTh
e in
cide
nces
ofG
rade
3/4
eve
nts
are
base
d on
wee
kly
asse
ssm
ents
.Dos
e m
odifi
catio
n an
d do
se in
tens
ity c
alcu
latio
ns d
eriv
efro
m
pred
ose
asse
ssm
ents
per
clin
ical
guid
elin
es.
bPl
atel
et, A
LT, o
r AST
Gra
de 3
/4 e
vent
s m
ay b
e le
ss th
an th
e su
m o
f the
indi
vidu
al in
cide
nces
(i.e
. ALT
+AST
+Pla
tele
t), a
s G
rade
3 e
vent
s m
ay o
ccur
sim
ulta
neou
sly
cD
ose
inte
nsity
is c
alcu
late
d by
the
tota
l dos
e gi
ven
divi
ded
by th
e tre
atm
entd
urat
ion
dR
elat
ive
dose
inte
nsity
is c
alcu
late
d by
the
ratio
of t
he to
tal d
ose
rece
ived
and
the
tota
l dos
e in
tend
eddu
ring
the
treat
men
t dur
atio
n
Table 8 shows that, as expected, the 2.4 mg/kg q1w regimen resulted in the highest dose intensity (2.07 mg/kg/wk) compared to the 1.2 mg/kg q1w reg-imen (1.15 mg/kg/wk) and the approved dose (1.18 mg/kg/wk). The q1w dose regimens are associated with more dose modifications due to more frequent Grade 3 events predose. This leads to the lower RDIs of 95.7% and 86.2% for the weekly regimens, versus the approved dose (98.3%). These simulations may provide guidance when considering alternative dose regi-mens for T-DM1.
Paper III For Paper III, an LC-MS assay provided the crucial analytical measurements of the individual T-DM1 drug-to-antibody ratio (DAR) moieties. Two PK models, elucidating and conceptualizing T-DM1 disposition, were developed from preclinical in vitro and in vivo rat and cynomolgus monkey data: 1) a mechanistic PK model simultaneously fit to total trastuzumab and individual DAR concentrations, and 2) a reduced PK model simultaneously fit to total trastuzumab and T-DM1 concentrations.
T-DM1 Mechanistic PK Model As shown in the following VPCs, the mechanistic PK model well–described the total trastuzumab and DAR0 7 concentration–time data from the in vitro and in vivo studies in rats (Figure 26) and monkeys (Figure 27). Table 9 presents the final parameter estimates for the mechanistic PK model.
58
F
igur
e 26
. VP
Cs
of to
tal t
rast
uzum
ab (T
T) a
nd D
AR
0D
AR
7 con
cent
ratio
ns fo
r th
e m
echa
nist
ic P
K m
odel
fit t
o da
ta fr
om: (
Top
Pane
l) in
vitr
o ra
t pla
sma
stab
ility
stu
dy w
ith 1
00 μ
g/m
L T-
DM
1 DA
R 3
.1; (
Mid
dle
Pane
l) R
at P
K s
tudy
with
10
mg/
kg T
-DM
1 DA
R 3
.1; (
Bot
tom
Pan
el) R
at P
K
stud
y w
ith10
mg/
kg T
-DM
1 DA
R 1
.5.
The
shad
ed a
rea
is th
e m
odel
–pre
dict
ed m
edia
n w
ith 9
5% c
onfid
ence
inte
rval
, and
the
solid
line
is th
e m
edia
n of
the
obs
erve
d da
ta (
circ
les)
. Bot
h ax
es a
re s
how
n on
log
sca
le f
or v
isua
lizat
ion
purp
oses
. DA
Rn
= n
DM
1 m
olec
ules
bou
nd t
o tra
stuz
umab
F
igur
e 27
. VP
Cs
of t
otal
tra
stuz
umab
(TT
) an
d D
AR
0D
AR
7 co
ncen
trat
ions
for
the
mec
hani
stic
dec
onju
gatio
n m
odel
fit
to d
ata
from
: (T
op
Pane
l) in
vitr
o cy
nom
olgu
s mon
key
plas
ma
stab
ility
stud
y w
ith 1
00 μ
g/m
L T-
DM
1 DA
R 3
.1; (
Bot
tom
Pan
el) C
ynom
olgu
s m
onke
y PK
stud
y w
ith
30 m
g/kg
T-D
M1 D
AR
3.1. T
he s
hade
d ar
ea is
the
mod
el–p
redi
cted
med
ian
with
95%
con
fiden
ce in
terv
al, a
nd th
e so
lid li
ne is
the
med
ian
of th
e ob
serv
ed d
ata
(circ
les)
. Bot
h ax
es a
re sh
own
on lo
g sc
ale
for v
isua
lizat
ion
purp
oses
; DA
Rn
= n
DM
1 m
olec
ules
bou
nd to
tras
tuzu
mab
Tabl
e 9.
Fin
al P
aram
eter
Est
imat
es fo
r the
T-D
M1
Mec
hani
stic
PK
Mod
elR
atC
ynom
olgu
s M
onke
y
Para
met
erD
escr
iptio
nU
nit
Valu
e
R
SE (%
)IIV
%R
SE (%
)Va
lue
RSE
(%)
IIV %
RSE
(%)
CL T
Ta
Tota
l tra
stuz
umab
clea
ranc
em
L/da
y2.
42(5
.7)
24.0
(18)
17.4
(13)
24.8
(22)
V 1C
entra
l vol
ume
mL
11.0
(4.3
)18
.5(1
1)14
8(4
.8)
11.7
(12)
CLd
2D
istri
butio
nal
Cle
aran
ce 2
mL/
day
49.0
(35)
——
25.5
(35)
——
V 2Pe
riphe
ral v
olum
e 2
mL
3.44
(58)
49.4
(41)
57.2
(38)
46.8
(33)
CLd
3D
istri
butio
nal
Cle
aran
ce 3
mL/
day
12.0
(16)
81.2
(19)
——
V 3Pe
riphe
ral v
olum
e 3
mL
16.7
(6.2
)16
.8(2
1)12
7(1
5)—
—
K pla
sma
Plas
ma
degr
adat
ion
rate
con
stan
tda
y-1
0.15
6(1
.2)
——
0.09
39(3
.0)
——
CL i
nvi
vob
in v
ivo
antib
ody
clea
ranc
em
L/da
y0.
704
——
—3.
50—
——
K, K
K, K K
DAR
7–D
AR3
deco
njug
atio
n ra
te
cons
tant
s c
day-
10.
543
(15)
21.8
(40)
0.34
1(7
.7)
——
KD
AR2–
DAR
1de
conj
ugat
ion
rate
co
nsta
nts
day-
10.
388
(7.1
)—
—0.
255
(6.7
)—
—
Kda
y-1
0.11
4(3
0)15
.1(4
8)0.
0939
(5.8
)—
—
Res
. Err
.R
esid
ual e
rror
%11
.1(6
.1)
——
15.3
(3.3
)—
—
IIV:i
nter
indi
vidu
al v
aria
bilit
y; R
SE: r
elat
ive
stan
dard
err
or; a
Com
pose
d of
CL i
nvi
vo a
nd K
plas
ma
1; b
Der
ived
by:
CL i
nvi
vo =
CL t
rast
uzum
ab-K
plas
ma
1c R
ates
wer
e de
term
ined
to b
e eq
ual f
rom
mod
el b
uild
ing
As shown, values for typical compartmental PK parameters (CLTT, V1, CLd2, V2, CLd3, and V3) were estimated. The conjugation of DM1 (Mol. wt.=737 Da) does not appear to alter the disposition of the trastuzumab antibody (Mol. wt.=146 kDa), as the individual DAR1 7 moieties are well de-scribed using the same distributional volumes and clearance pathways as DAR0 (unconjugated trastuzumab).
Notably, DM1 per trastuzumab in both rats and cynomolgus monkeys. It can be hy-pothesized that 1) higher conjugated DAR moieties have more probability to lose DM1 than less conjugated DAR moieties, or 2) that they have DM1 molecules that are conjugated to more solvent–accessible lysine residues on trastuzumab, making them less resistant to deconjugation [79, 80]. DM1 deconjugation from DAR1 was the slowest step in the deconjugation chain,
2 in rats and cyno-molgus monkeys, respectively.
Following development of the mechanistic model, simulations were done to illustrate all components of the PK system. In addition to total trastuzumab and the individual DAR moieties, the T-DM1 concentration, conjugated DM1 concentration, and the DARAVG time–courses were derived (see Paper III for details). Shown in Figure 28 are simulations of the PK system follow-ing a T-DM1 single dose of 3.6 mg/kg T-DM1DAR 3.1 and the cynomolgus monkey parameter estimates.
62
Figure 28. Mechanistic PK model predictions of total trastuzumab (TT), T-DM1, conjugated DM1, and DAR0 DAR7 concentration time curves, and DARAVG time curve, based on the final parameter estimates from cynomolgus monkeys analysis. The T-DM1 curve is the composite of the 7 curves representing DAR1 7 moie-ties. The TT curve is the composite of the 8 curves representing DAR0 7 moie-ties. Predictions are for a single dose of 3.6 mg/kg T-DM1DAR 3.1.
The development of the mechanistic PK model provides guidance when considering experimental designs regarding PD responses from T-DM1 treatment. For example, with regard to T-DM1 anti–tumor efficacy, potential PK drivers include: 1) T-DM1 concentrations, where the binding of the anti–HER2 region of T-DM1 drives response, such that the anti–tumor ef-fects between DAR moieties are equal; 2) the individual DAR moiety con-centrations, where more heavily loaded DARs have more anti–tumor effect; 3) conjugated DM1 concentrations, where the overall DM1 drug load still attached to antibody may more appropriately drive response.
Reduced T-DM1 PK Model The reduced model, based on T-DM1 and total trastuzumab, is more practi-cal analytically in the clinical setting, as only ELISA measurements are nec-essary. As shown in the Figure 29 VPCs, the reduced PK model well–described both the total trastuzumab and T-DM1 concentration–time data from the in vivo rat and cynomolgus monkey studies. Final parameter esti-mates are shown in Table 10.
63
F
igur
e 29
. VP
Cs
of t
otal
tra
stuz
umab
(TT
) an
d T-
DM
1 co
ncen
trat
ions
for
the
red
uced
PK
mod
el f
it to
dat
a, f
rom
(to
p pa
nel)
rat P
K s
tudy
w
ith 0
.3, 3
.0, a
nd 2
0 m
g/kg
T-D
M D
AR
3.1 a
nd (b
otto
m p
anel
s) c
ynom
olgu
s m
onke
y PK
stu
dy w
ith 1
0 m
g/kg
q3w
and
30
mg/
kg T
-DM
1 DA
R 3
. Fo
r eac
h V
PC, t
he s
hade
d ar
ea is
the
mod
el–p
redi
cted
med
ian
with
95%
con
fiden
ce in
terv
al, a
nd th
e so
lid li
ne is
the
med
ian
of th
e ob
serv
ed
data
(circ
les)
.
Tabl
e 10
. Fin
al P
aram
eter
Est
imat
es fo
r the
T-D
M1
Red
uced
PK
Mod
elR
atC
ynom
olgu
s M
onke
y
Para
met
erD
escr
iptio
nU
nit
Valu
e
R
SE (%
)IIV
%R
SE (%
)Va
lue
RSE
(%)
IIV %
RSE
(%)
CL T
TTo
tal t
rast
uzum
abcl
eara
nce
mL/
day
2.37
(4
.9)
24.6
(19)
19
.9(9
.8)
19.8
(23)
V 1C
entra
l vol
ume
mL
10.7
(4.0
)19
.8(1
2)
154
(4.8
)7.
65(1
8)
CLd
2D
istri
buiti
onal
Cle
aran
ce 2
mL/
day
59.7
(3
2)56
.8(4
6)
V 2Pe
riphe
ral v
olum
e 2
mL
2.52
(2
8)68
.0(3
9)
50.0
(60)
CLd
3D
istri
buiti
onal
Cle
aran
ce 3
mL/
day
13.9
(1
1)60
.4(3
5)
V 3Pe
riphe
ral v
olum
e 3
mL
15.5
(4
.9)
16.5
(18)
84
.7(5
0)27
.3(4
3)
CL D
ECD
econ
juga
tion
clea
ranc
em
L/da
y2.
24
(3.4
)15
.3(1
8)22
.0(6
.9)
11.9
(28)
CL T
-DM
1aT-
DM
1cle
aran
cem
L/da
y4.
61
41.9
TT t 1
/2b
Tota
l tra
stuz
umab
term
inal
t 1/2
day
8.84
10.5
T-D
M1
t 1/2
bT-
DM
1 te
rmin
al t 1
/2da
y4.
775.
21
Res
Err
Res
idua
l err
or%
10.9
(5
.6)
9.64
(16)
IIV, i
nter
indi
vidu
al v
aria
bilit
y; R
SE, r
elat
ive
stan
dard
err
or
a D
eriv
ed b
y: C
L T-D
M1 =
CL t
rast
uzum
ab+
CL D
ECb
Der
ived
by
3-co
mpa
rtmen
t PK
equa
tion
for t
erm
inal
hal
f-life
As shown in Tables 9 and 10, the two modeling approaches yielded similar estimates for CLTT and V1 in both rats and cynomolgus monkeys but slightly different estimates for CLd2, V2, CLd3, and V3; the difference between the distributional parameters was attributed to the different data used for each analysis.
The reduced PK model employed T-DM1 clearance via 2 mechanisms: anti-body clearance mechanisms (CLTT) and deconjugation clearance mecha-nisms (CLDEC). Notably, CLDEC is approximately 50% of CLT-DM1 in both rats and cynomolgus monkeys, indicating that the overall T-DM1 deconjugation is similar between species. For both species, CLDEC was approximately equal to that of CLTT, indicating that deconjugation is a major pathway of T-DM1 clearance. Overall, deconjugation is supported as the mechanism for the increased clearance of T-DM1 over its parent trastuzumab antibody. For antibodies, allometric scaling of clearance from cynomolgus monkeys to humans has been successful given that cynomolgus monkeys are a similar antibody binding species [15, 81]. The reduced PK model can also be uti-lized towards allometric scaling for T-DM1, investigating approaches for scaling of clearance (e.g. CLTT and CLDEC) to human data.
For ADCs, efficacy and toxicity is influenced in part by the stability of the linker, in which systemic cytotoxin release from less stable linkers results in loss of efficacy and potential for toxicity. CLDEC can be viewed as a quanti-tative measure of linker stability, where higher CLDEC values describe a less stable linker, and can inform ADC design. Shown in Figure 30 is an illustra-tion using the reduced PK model, with CLDEC values of 22 mL/day (for T-DM11) and a higher 44 mL/day (for T-DM12) predictions. In this prediction, higher CLDEC values (T-DM12) in patients may correlate with higher inci-dence of free DM1–related toxicities, and lower anti-tumor efficacy.
66
Figure 30. Reduced PK model predictions of total trastuzumab antibody with T-DM11 (CLDEC = 22 mL/day) and T-DM12 (CLDEC = 44 mL/day)
The reduced T-DM1 PK model has clinical implications and utility. An op-timal sampling strategy for the ELISA analyses (i.e., for antibody, ADC, or both) can be used streamline clinical development [82]. T-DM1 may also be considered for front–line treatment of metastatic breast cancer, as HER2+ patients may benefit from a dual pharmacologic effect given that T-DM1 retains the anti–tumor mechanism of trastuzumab [61]. Here, the measure-ment of both total trastuzumab and T-DM1 concentrations is warranted, and a single population model of these two analytes can be developed using the reduced PK modeling approach. Moreover, total trastuzumab and T-DM1 concentrations can be evaluated as dual PK drivers of anti–tumor response.
Paper IV In Paper IV, a model describing tumor response following taxane treatment in HER2–negative breast cancer patients was developed. This model charac-terized various patterns of tumor quiescence and resistance, and results were applied into a parametric time-to-event analyses of survival data. This tumor model provides an alternative approach for modeling tumor data, and shows advantages over the TGI model (Figure 6), especially when fitting patterns of tumor quiescence followed by tumor regrowth. Tumor model VPCs (Fig-ure 31) for docetaxel and paclitaxel treatments show that the tumor model developed well–predicted the longitudinal tumor SLD measurements. Final parameter estimates are presented in Table 11.
Total trastuzumabT-DM11
T-DM12
CLDEC
67
Fig
ure
31. V
PC
s of
the
final
tum
or m
odel
sim
ulat
ions
for
doce
tael
(lef
t pan
el) a
nd p
aclit
ael
(rig
ht p
anel
). Th
e so
lid b
lack
line
repr
esen
ts th
e m
edia
n of
the
obse
rved
tum
or S
LD m
easu
rem
ents
(ci
rcle
s). T
he in
ner
shad
ed r
egio
n re
pres
ents
the
95%
con
fiden
ce i
nter
val o
f th
e m
odel
si
mul
ated
med
ian.
For
doc
etax
el, t
he o
uter
sha
ded
regi
ons
repr
esen
t the
95%
con
fiden
ce in
terv
als
of th
e m
odel
sim
ulat
ed 5
th a
nd 9
5th p
erce
n-til
es, a
nd th
e st
ippl
ed b
lack
line
s re
pres
ent t
he 5
th a
nd 9
5th p
erce
ntile
s of
the
obse
rved
tum
or S
LD m
easu
rem
ents
. For
pac
litax
el, t
he o
uter
sh
aded
regi
ons r
epre
sent
the
95%
con
fiden
ce in
terv
als o
f the
mod
el si
mul
ated
10th
and
90th
per
cent
iles,
and
the
stip
pled
line
s rep
rese
nt th
e 10
th
and
90th
per
cent
iles o
f the
obs
erve
d tu
mor
SLD
mea
sure
men
ts. V
ertic
al li
nes i
ndic
ate
binn
ing
inte
rval
s for
VPC
s.
Ta
ble
11. F
inal
Par
amet
er E
stim
ates
for t
he T
umor
Mod
el
Doc
etax
elPa
clita
xel
Para
met
erD
escr
iptio
nU
nit
Valu
e
R
SE
(%)
IIV %
RSE
(%
)Va
lue
RSE
(%
)IIV
%R
SE (%)
k Gro
w,S
ens
-1)
grow
th ra
tese
nsiti
ve
tum
or fr
actio
nw
eek-
10.
0506
29.2
0—
0.11
323
.413
024
.8
k Gro
w,R
esis
t -1
)gr
owth
rate
resi
stan
t tu
mor
frac
tion
wee
k-1
1.19
21.1
118
28.0
0.12
125
.517
633
.7
k Kill
-3)
tum
or k
ill ra
tem
g-1
-10.
714
15.9
164
21.6
0.80
620
.114
126
.5
k Dru
gdr
ug e
limin
atio
n ra
tew
eek-
10.
248
22.5
391
24.9
0.49
826
.225
329
.4
FnR
logi
tLo
git t
rans
form
atio
nFn
R—
0.01
6335
.910
917
.60.
600
21.3
103
20.0
FnR
bfra
ctio
n tu
mor
resi
stan
t—
0.50
4—
——
0.64
6—
——
1-Fn
Rfra
ctio
n tu
mor
sen
sitiv
e—
0.49
6—
——
0.35
4—
——
k Del
ay,p
op1
-2)
trans
it de
lay
rate
con
stan
t; Po
pula
tion
1w
eek-
10.
0042
631
.223
930
.30.
699
32.2
0—
k Del
ay,p
op2
trans
it de
lay
rate
con
stan
t; Po
pula
tion
2w
eek-
10.
0209
24.4
239
25.9
0.05
2523
.421
134
.3
P(1)
k del
ay,p
op1
prob
abilit
y—
0.62
914
.5—
19.5
0.53
219
.5—
—
Res
. Err
.R
esid
uale
rror
%15
.24.
40—
5.37
6.34
5.37
——
IIV =
inte
rindi
vidu
al v
aria
bilit
y; R
SE =
rela
tive
stan
dard
err
ora
addi
tive
erro
r b
Der
ived
par
amet
er; F
nR =
exp
(FnR
logi
t) / (
exp(
FnR
logi
t) +
1)
Model predictions for the two typical tumor patient responses (kDelay,pop1 and kDelay,pop2) as reported in Table 1are illustrated for patients receiving docetaxel treat-ment in Figure 32. As shown, the typical docetaxel patient had a tumor base-
––
resistant tumor SLD. The drug–sensitive tumor fraction was estimated to grow at the kGrow,Sens rate of 0.00506 week-1 (tumor SLD doubling rate = 137 weeks). Docetaxel had a typical elimination rate of 0.248 week-1 (2.8 week half-life) that was scaled by a tumor kill constant (kKill = 0.000714 mg-1 week-1), and this drug effect acted to eliminate the drug–sensitive tumor fraction as shown by 24 weeks.
The transition rate from quiescent drug–resistant tumor cells to a proliferat-ing state was quantified using the kDelay parameter. The kDelay parameter was bimodal, with one probability (0.629; P(1)) of having a much slower rate (kDelay,pop1 = 0.0000426 week-1) compared to a faster kDelay,pop2 rate of 0.0209 week-1. As shown in Figure 32, patients were stratified into a stable response group (Population 1), and into a group initially responding and then develop-ing resistance (Population 2). For the kDelay,pop1 group, the slower rate pre-dicted no tumor regrowth over the study length. The tumor SLD profile for the typical patient in the kDelay,pop2 group was predicted to begin to regrow at the kGrow,Resist rate (0.119 week-1; tumor SLD doubling = 5.8 weeks) after the predicted tumor nadir at ~30 weeks. A similar figure and discussion for paclitaxel is presented in Paper IV.
70
Figure 32. Model predictions for the two typical tumor patient responses (kDelay,pop1 and kDelay,pop2) receiving doceta el treatment, derived from population parameter estimates shown in Table 11. The typical tumor SLD for each group is shown in the solid red line. The solid green line is the time–course of drug–sensitive tumor. The solid blue line is the time–course of drug–resistant tumor. The tumor SLD (red line) is the sum of the drug–resistant and drug–sensitive tumor time–courses. Observed data (black circles) are connected for each patient by grey lines.
Shown in the Figure 33 are Kaplan–Meier VPCs for docetaxel and paclitaxel datasets. The observed docetaxel and paclitaxel survival data fall within the simulated 95th confidence interval.
71
Fig
ure
33. K
apla
n–M
eier
VP
C o
f the
ve
rall
Surv
ival
(S)
mod
el fo
r D
ocet
ael
and
Pac
lita
el. T
he s
hade
d re
gion
s re
pres
ent t
he 9
5% c
onfi-
denc
e in
terv
als o
f the
mod
el si
mul
ated
dat
a. T
he so
lid li
ne is
the
obse
rved
surv
ival
dat
a, a
nd th
e ve
rtica
l lin
es in
dica
te c
enso
red
patie
nts.
Table 12 presents the parameter estimates for the survival models. In the survival analysis, an increased tumor baseline, reflective of higher tumor burden, and a shorter TTG, reflective of treatment efficacy, was associated with poorer survival prognosis. This aligns with results from other cancer treatments and indications as shown in Table 1. Notably, TSR and ECOG status were not statistically significant predictors for OS.
.
Table 12. Parameter Estimates for Survival ModelsTreatment
GroupSurvivalModel Parameter Value (RSE)
CovariateSearch
OFVDrop
Docetaxel Log-Normal
SD 0.894 (8.70)
Mu 2.64 (7.80)
TTG 0.134 (18.1) TTG — -51.4
BSL -0.00447 (28.9) TTG BSL -12.9
Paclitaxel Weibull Lambda 0.0350 (10.9)
Shape 1.49 (6.06)
TTG -0.0342 (20.4) TTG — -18.2
BSL 0.00398 (24.8) TTG BSL -12.0
OFV: objective function value; RSE: relative standard error; TTG: time to tumor growth; BSL: tumor baseline
73
Conclusions
In this thesis work, pharmacometric models and methodology, from HER2–positive and HER2–negative metastatic breast cancer treatment, were devel-oped. These pharmacometric tools can be applied within the modeling and simulation framework proposed for oncologic drug development, with pri-mary applications to antibody-drug conjugates (ADCs) and anticancer treatments exhibiting tumor drug–resistance.
Specifically:
A PKPD model describing longitudinal platelet response to T-DM1 treatment in HER2–positive breast cancer patients was developed. The model successfully predicts rates of Grade 3 thrombocytopenia (TCP).
A PKPD model describing simultaneously the time–course of platelet, ALT, and AST response to T-DM1 treatment in HER2–positive breast cancer patients was developed. Correlations between parameters were implemented, and the model determined that Asian ethnicity was related to higher incidences of TCP due an increased drug effect on platelet production.
Dose modification rules were incorporated into the PKPD model, and successfully predicted incidences of Grade 3 TCP, ALT, and AST. Model simulations of alternative T-DM1 dose regimens quantified the
s and dose intensities, as compared to the approved dose, supporting T-DM1 drug development.
Two PK models were developed from preclinical experiments and elu-cidated the underlying pharmacokinetics (PK) of T-DM1. The modeling approaches describe and quantify the T-DM1 PK system, and provide modeling flexibility given available analytical techniques for ADCs.
A model for tumor response to taxane treatment in HER2–negative breast cancer patients was developed. This model shows advantages over the standard TGI modeling approach, primarily in characterizing patterns of tumor quiescence and drug–resistance. Time-to-event analy-sis indicates that tumor size at baseline and TTG are predictors for OS.
74
Future perspectives
ADCs are emerging as novel, exciting therapeutics in oncology [83, 84]. In a recent review on the preclinical and clinical toxicities of 35 ADCs, TCP was reported for 5 ADCs, elevated aminotransferases in 6 ADCs, and neutro-penia in 13 ADCs [85]. In this thesis work, the model approaches developed for T-DM1 PKPD of AE (Papers I and II), and for T-DM1 PK to elucidate its complex disposition (Paper III), may be applicable to other ADCs given the similarity in the synthesis, antibody properties, and mechanism of action of these drugs. Specifically:
The modeling work described and quantified particular response patterns
of AE for T-DM1– a “first dose” effect and the cumulative dose effect. These phenomena may not be specific to T-DM1.
Regarding hepatotoxicity, models for ALT and AST are not routinely applied in pharmacometrics as yet, and the PKPD models developed herein may provide guidance.
The mechanistic and reduced PK models can be applied to any ADC given the analytical measurements of DAR, ADC, and total antibody.
Tumor drug–resistance is a primary obstacle for cancer treatment, and this was described in Paper IV using a mechanism–based approach. The tumor model, developed n-cer, can be applied to any drug treatment and indication showing patterns of tumor quiescence and drug–resistance. Results from this analysis add to the growing body of literature for predictors of OS in various cancers and cancer treatments (Table 1).
Ultimately, PKPD model development in oncology should encompass both AE and efficacy, in order to provide simulations balancing both risk and benefit. Within this thesis work, the pharmacometric analyses presented can be expanded. For example,
A tumor model (preferably incorporating individual dose exposure) de-scribing tumor response to T-DM1 treatment can be integrated into the current T-DM1 PKPD model of AEs.
75
PKPD models of docetaxel and paclitaxel AE, e.g. neutropenia from taxane treatment [24, 25]), can be integrated with the developed models of taxane tumor response.
Tumor model metrics and AE metrics can simultaneously be assessed as predictors for OS in a parametric TTE analysis, as illustrated in Figure 1.
With more complete models describing both AE and tumor response, model–based simulation strategies can developed to weigh different tox-icity and efficacy outcomes. For example, 1) comparing outcomes be-tween AUCss–matched and Cmax,SS–matched dose regimens to an ap-proved dose, or 2) comparing standard population drug treatment to a model–based dose adaptation approach, in which doses are titrated with-in an individual to provide maximum acceptable exposure, as has been described [86].
A recently conducted T-DM1 clinical trial illustrates applications for the models developed within this thesis work. Shown in Table 13 are treatment arm details for a clinical T-DM1 trial in patients with gastric cancer.
Using the pharmacometric models developed herein, and their proposed extensions, the following predictions may have been used to support clinical decision making: 1) predictions of Grade 3/4 platelet, ALT, and AST for ARM A vs ARM B; 2) the relative differences in toxicity rates between ARM A and B, given that Gastric cancer patients may respond differently than MBC patients; 3) expected PFS and OS between ARMS A, B, and C. Results from the clinical trial in HER2–positive gastric cancer may serve as a model evaluation dataset for the model predictions based on HER2–positive and HER2–negative breast cancer (e.g. toxicity incidences, DI and RDI, tumor growth rates, OS predictors).
Overall, the pharmacometric models developed within this thesis work have supplemented the available methodologies for application within the model–based framework for drug development in oncology (Figure 1), especially for ADCs. The population PK, PKPD, and TTE modeling analyses support
Table 13. Phase II/III Study To Evaluate The Efficacy And Safety Of T-DM1 Versus Taxane (Docetaxel Or Paclitaxel) In Patients With Previously Treated Locally Advanced Or Metastatic Her2-Positive Gastric Cancer
ARM A ARM B ARM C
T-DM1 3.6 mg/kg q3w
T-DM1 2.4 mg/kg q1w
docetaxel or paclitaxel
(standard taxane therapyper investigatorchoice)
q1w: weekly; q3w: every 3 weeks
76
efforts to link drug exposure with PD response, and PK and PD response with survival. This, in order to explain and forecast patient response to drug treatment in oncology, with the ultimate goal of predicting and improving survival by bringing the “right drug to the right patient at the right dose”.
77
Acknowledgements
The work presented in this thesis was carried out at the Department of Pharmaceutical Biosciences, Faculty of Pharmacy, Uppsala Universitet, Uppsala, Sweden. This thesis was funded by Genentech, Inc, South San Francisco, California. I would like to first and foremost thank Dr. Amita Joshi, Vice President and Global Head of Clinical Pharmacology at Genentech, Inc, for the approval and support of my pursuit of a doctoral thesis while employed at Genentech. AND Prof. Lena Friberg, my mentor at Uppsala, for her inspiration, advice, and timely reviews of all the work we managed here, and for her patience with my forgetting what time we meet, what day it is, downloading Pirana, dis-cussing the definition of a “biomarker” , etc. AND The many breast cancers patients who allowed their data to be shared…. Thank you to Prof. Mats Karlsson, for creating and leading such a world class group, and for being a fierce competitor to push me past my limits… for the ski weekends, go-karting, egg races, sponsoring my attendance at TACA, and for volunteering me for numerous events. Thank you to all my managers past and present at Genentech, for inspiring me to become a pharmacometrician and supporting me in this endeavor at Uppsala. Drs. Shasha Jumbe, Manish Gupta, and Jin Jin. Thank you to the Anna Maria Lundins stipendifond at Smålands Nation for awarding me a travel grant to attend the Sils Maria PKPD conference. Thanks to Dr. Gregory Sivolapenko, for hosting us at the University of Pa-tras. I look forward to future collaborations and more ouza.
79
Thanks to the organizers of PEGS, AAPS, and SUP Meetings for the invited speakerships. Now some random thoughts, thanks, and shout outs for some pepes I ran into in the last 5.4 years. Thank you to Karin, Ulrica, and Marina R. for all the administrative help through the years! …but more importantly for all the party entertainment (with Jessica), we had back in the day, and hopefully one more time. Thanks all my officemates for the good times, with some managing longer than others: Agneta, Bettan, Emma, Angelica, Anne-Gaelle, Salim, Rob-in, Benjamin, and Gustaf. Hmmm…this seems like a lot. Thank you to my housemates Anne-Gaelle (late night conversations, corn in the fridge), Ana K (Mad Men, Castle, and you writing the history of PK), Mirjam (the kayak trip, Bulsufan, cooking together without fighting, and the trips to Austria), Paolo (learning to cut veggies and late night pasta every night), Chayan and Lani (breaking 1 wine glass per day ), Ana K. and A.G. for being considerate for my only 2 rules: 1) no boyfriends and 2) no hair in the shower….well at least you tried. Thanks to the first PHD group I met here, Kristin, Martin, Angelica, Em-ma, Elodie, Sebastian, Bettan, Klas, and Jocke. It was one big nonstop dissertation party for awhile…thank god you all finally left. Kajsa and Ri-kard for all the PSN updates..I stopped counting Andy, Marija K and Siv for keeping us in line when the boss is away. Siv and her dance moves. Marina S (you are a good friend the last couple years, late night arguments and random music and reading my mscrpt right now), Ida (a good friend as well, thanks for helping me to move and also arguing with me for some rea-son), Nebojasa– I really tried to be your wingman, “Finish HER”, and pass the ball so you don’t fall on your head, Eirini, thanks for the Greece trip, the food, and all the Greek hospitality, My master’s students, Ari, Aurelia, Bjorn, and Johanna…you guys got lucky. Chunli (I ate your salad sorry), Yang (I cannot believe you can dunk!), Anders (color coordinated all of a sudden), Moustafa (you have 3 more years to learn to tie a tie), David you were always there to start my car and spot me on the weights, Master’s Stu-dents: Viktor and Susanna..what a crew we were and some good times…oh and Emma, Ilse, Josefina, Mina (just remember that puppet show), and Giulia (thanks for hosting us in Italy). Anders K making the beer, Gunnar (the best name ever), Srinath (thanks for the VPC help, you have potential), Emilie H (hosting me in Lyon), Akash (where did you go that night in Athens?), Vijay (Istanbul and Oktoberfest), Paul (buddyyy…just
80
one more pint), Ron (you owe me still), Waqas (chasing Ron around the island with me and MT), Tina (who stabbed your pear?) Julia (sorry I touched the hamburger), Elke (you made me look PC), Yasunori (no more dinner and discussion ), Robin (you kept us in the soccer game, but we could never score), Elodie and Sebastian (power couple on the rise), Anni-ka (in the end, you beat me to graduation, even with a baby and this guy Mattias around), Åsa (the kiss I will never forget), Steve and Eric (there can be only one cameraman / group, but you 2 pulled it off), Benjamin (bicycle wheelies and maybe my long lost son), Emilie (noone should mess with you, especially not at Oktoberfest), Salim (helping me keep USA safe during Thanksgiving and laughing during Precious), Ari (my firstborn and prodigy, don’t let me down), Elin (the best kettlebeller in the group that’s not me, bring Thomas !), the Oskars and Jergen, damn we almost won that race–not that B.S. one in Stockholm, the castle one. Siti (where’s your daughter), Jesmin (see you one day in California), Gustav (try acting if pharmacomet-rics doesn’t work out for you), Jaou (please don’t touch me), Gopichand (when’s the party?....seriously when’s the party?). Bruna (wait for me in Brazil and sell me some ice cream), Sofia (I won’t forget that kiss either), and Erik M and Irena (what do you guys do over there?) To all the babies I will never meet again but I got you a gift anyway. To my mother from another mother Britt, thanks for diagnosing my crystals and showing me that life can be fun forever. And now, since I know I neglected some of you, or maybe you are fortunate enough to get a copy of my thesis, here’s a little extra: Dear ________________, I will never forget the time when ____________ ____________________________________________________________. I hope that ____________________________________________________ _____________________________________________________________. P.S., Apologies for ____________________________________________
81
References
1. Siegel, R.L., K.D. Miller, and A. Jemal, Cancer statistics, 2015. CA: A Cancer Journal for Clinicians, 2015. 65(1): p. 5-29.
2. Kummar, S., et al., Drug development in oncology: classical cytoto ics and molecularly targeted agents. Br J Clin Pharmacol, 2006. 62(1): p. 15-26.
3. Bender, B.C., E. Schindler, and L.E. Friberg, Population pharmacokinetic-pharmacodynamic modelling in oncology: a tool for predicting clinical re-sponse. Br J Clin Pharmacol, 2015. 79(1): p. 56-71.
4. Bruno, R. and L. Claret, On the use of change in tumor size to predict survival in clinical oncology studies: toward a new paradigm to design and evaluate phase II studies. Clin Pharmacol Ther, 2009. 86(2): p. 136-8.
5. Mould, D.R. and R.N. Upton, Basic concepts in population modeling, simula-tion, and model-based drug development. CPT Pharmacometrics Syst Pharma-col, 2012. 1: p. e6.
6. Pasqua, O.E.D., PKPD and Disease Modeling: Concepts and Applications to Oncology, in Clinical Tiral Simulations. AAPS Advances in the Pharmaceuti-cal Sciences Series I, K. HHC and P. CC, Editors. 2011.
7. Karlsson, M.O., et al., Pharmacokinetic/pharmacodynamic modelling in onco-logical drug development. Basic Clin Pharmacol Toxicol, 2005. 96(3): p. 206-11.
8. Duffull, S.B., D.F. Wright, and H.R. Winter, Interpreting population pharma-cokinetic-pharmacodynamic analyses - a clinical viewpoint. Br J Clin Pharma-col, 2011. 71(6): p. 807-14.
9. Wright, D.F., H.R. Winter, and S.B. Duffull, Understanding the time course of pharmacological effect: a PKPD approach. Br J Clin Pharmacol, 2011. 71(6): p. 815-23.
10. Keizer, R.J., et al., Pharmacodynamic biomarkers in model-based drug devel-opment in oncology. Curr Clin Pharmacol, 2011. 6(1): p. 30-40.
11. Joerger, M., Covariate pharmacokinetic model building in oncology and its potential clinical relevance. Aaps j, 2012. 14(1): p. 119-32.
12. Dayneka, N.L., V. Garg, and W.J. Jusko, Comparison of four basic models of indirect pharmacodynamic responses. J Pharmacokinet Biopharm, 1993. 21(4): p. 457-78.
13. Friberg, L.E., M. Sandstrom, and M.O. Karlsson, Scaling the time-course of myelosuppression from rats to patients with a semi-physiological model. Invest New Drugs, 2010. 28(6): p. 744-53.
14. Jumbe, N.L., et al., Modeling the efficacy of trastuzumab-DM1, an antibody drug conjugate, in mice. J Pharmacokinet Pharmacodyn, 2010. 37(3): p. 221-42.
15. Wang, J., et al., Projecting human pharmacokinetics of monoclonal antibodies from nonclinical data: comparative evaluation of prediction approaches in early drug development. Biopharm Drug Dispos, 2016. 37(2): p. 51-65.
82
16. Fetterly, G.J., et al., Semimechanistic pharmacokinetic/pharmacodynamic model for hepatoprotective effect of de amethasone on transient transaminitis after trabectedin (ET-743) treatment. Cancer Chemother Pharmacol, 2008. 62(1): p. 135-47.
17. Pollak, P.T. and S.L. Shafer, Use of population modeling to define rational monitoring of amiodarone hepatic effects. Clin Pharmacol Ther, 2004. 75(4): p. 342-51.
18. Friberg, L.E., et al., Model of chemotherapy-induced myelosuppression with parameter consistency across drugs. J Clin Oncol, 2002. 20(24): p. 4713-21.
19. Hansson, E.K. and L.E. Friberg, The shape of the myelosuppression time pro-file is related to the probability of developing neutropenic fever in patients with doceta el-induced grade IV neutropenia. Cancer Chemother Pharmacol, 2012. 69(4): p. 881-90.
20. van Kesteren, C., et al., Semi-physiological model describing the hematologi-cal to icity of the anti-cancer agent indisulam. Invest New Drugs, 2005. 23(3): p. 225-34.
21. Latz, J.E., et al., Clinical application of a semimechanistic-physiologic popula-tion PK/PD model for neutropenia following pemetre ed therapy. Cancer Chemother Pharmacol, 2006. 57(4): p. 427-35.
22. Gupta, P., et al., A semi-mechanistic model of CP-690,550-induced reduction in neutrophil counts in patients with rheumatoid arthritis. J Clin Pharmacol, 2010. 50(6): p. 679-87.
23. Joerger, M., et al., Population pharmacokinetics and pharmacodynamics of paclita el and carboplatin in ovarian cancer patients: a study by the European organization for research and treatment of cancer-pharmacology and molecu-lar mechanisms group and new drug development group. Clin Cancer Res, 2007. 13(21): p. 6410-8.
24. Quartino, A.L., L.E. Friberg, and M.O. Karlsson, A simultaneous analysis of the time-course of leukocytes and neutrophils following doceta el administra-tion using a semi-mechanistic myelosuppression model. Invest New Drugs, 2012. 30(2): p. 833-45.
25. Kloft, C., et al., Population pharmacokinetic-pharmacodynamic model for neutropenia with patient subgroup identification: comparison across anti-cancer drugs. Clin Cancer Res, 2006. 12(18): p. 5481-90.
26. Chalret du Rieu, Q., et al., Pharmacokinetic/Pharmacodynamic modeling of abe inostat-induced thrombocytopenia across different patient populations: application for the determination of the ma imum tolerated doses in both lym-phoma and solid tumour patients. Invest New Drugs, 2014. 32(5): p. 985-94.
27. Quartino, A.L., et al., Characterization of endogenous G-CSF and the inverse correlation to chemotherapy-induced neutropenia in patients with breast can-cer using population modeling. Pharm Res, 2014. 31(12): p. 3390-403.
28. Eisenhauer, E.A., et al., New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer, 2009. 45(2): p. 228-47.
29. Claret, L., et al., Model-based prediction of phase III overall survival in colo-rectal cancer on the basis of phase II tumor dynamics. J Clin Oncol, 2009. 27(25): p. 4103-8.
30. Ribba, B., N. Holford, and F. Mentre, The use of model-based tumor-size met-rics to predict survival. Clin Pharmacol Ther, 2014. 96(2): p. 133-5.
31. FDA. Guidance for industry. Clinical trial endpoints for the approval of can-cer drugs and biologics. 2007; Available from: http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/ucm071590.pdf.
83
32. Holford, N.H., A Time to Event Tutorial for Pharmacometricians. CPT: Phar-macometrics & Systems Pharmacology, 2013: p. 8.
33. Collett, D., Modelling Survival Data in Medical Research. second ed, ed. C. Hall/CRC. 2003.
34. Holford, N., A time to event tutorial for pharmacometricians. CPT Pharmaco-metrics Syst Pharmacol, 2013. 2: p. e43.
35. Wang, Y., et al., Elucidation of relationship between tumor size and survival in non-small-cell lung cancer patients can aid early decision making in clinical drug development. Clin Pharmacol Ther, 2009. 86(2): p. 167-74.
36. Claret, L., et al., Simulations using a drug-disease modeling framework and phase II data predict phase III survival outcome in first-line non-small-cell lung cancer. Clin Pharmacol Ther, 2012. 92(5): p. 631-4.
37. Lu, J.F., et al., Population pharmacokinetic/pharmacodynamic modeling for the time course of tumor shrinkage by motesanib in thyroid cancer patients. Cancer Chemother Pharmacol, 2010. 66(6): p. 1151-8.
38. Claret, L., et al., Development of a modeling framework to simulate efficacy endpoints for motesanib in patients with thyroid cancer. Cancer Chemother Pharmacol, 2010. 66(6): p. 1141-9.
39. Bruno, R., et al., Simulations to Assess Phase II Noninferiority Trials of Differ-ent Doses of Capecitabine in Combination With Doceta el for Metastatic Breast Cancer. CPT Pharmacometrics Syst Pharmacol, 2012. 1: p. e19.
40. Hansson, E.K., et al., PKPD Modeling of VEGF, sVEGFR-2, sVEGFR-3, and sKIT as Predictors of Tumor Dynamics and Overall Survival Following Sunitinib Treatment in GIST. CPT Pharmacometrics Syst Pharmacol, 2013. 2: p. e84.
41. Claret, L., et al., Evaluation of tumor-size response metrics to predict overall survival in Western and Chinese patients with first-line metastatic colorectal cancer. J Clin Oncol, 2013. 31(17): p. 2110-4.
42. Claret, L., et al., E ploratory modeling and simulation to support development of motesanib in Asian patients with non-small cell lung cancer based on MO-NET1 study results. Clin Pharmacol Ther, 2014. 95(4): p. 446-51.
43. Claret, L., et al., Modeling and simulations relating overall survival to tumor growth inhibition in renal cell carcinoma patients. Cancer Chemother Pharma-col, 2015. 76(3): p. 567-73.
44. Cortazar, P., et al., US Food and Drug Administration approval overview in metastatic breast cancer. J Clin Oncol, 2012. 30(14): p. 1705-11.
45. Miller, K., et al., Paclita el plus bevacizumab versus paclita el alone for meta-static breast cancer. N Engl J Med, 2007. 357(26): p. 2666-76.
46. Miles, D.W., et al., Phase III study of bevacizumab plus doceta el compared with placebo plus doceta el for the first-line treatment of human epidermal growth factor receptor 2-negative metastatic breast cancer. J Clin Oncol, 2010. 28(20): p. 3239-47.
47. Hurvitz, S.A., et al., Phase II randomized study of trastuzumab emtansine versus trastuzumab plus doceta el in patients with human epidermal growth factor receptor 2-positive metastatic breast cancer. J Clin Oncol, 2013. 31(9): p. 1157-63.
48. Krop, I.E., et al., Phase I study of trastuzumab-DM1, an HER2 antibody-drug conjugate, given every 3 weeks to patients with HER2-positive metastatic breast cancer. J Clin Oncol, 2010. 28(16): p. 2698-704.
84
49. Krop, I.E., et al., A phase II study of trastuzumab emtansine in patients with human epidermal growth factor receptor 2-positive metastatic breast cancer who were previously treated with trastuzumab, lapatinib, an anthracycline, a ta ane, and capecitabine. J Clin Oncol, 2012. 30(26): p. 3234-41.
50. Verma, S., et al., Trastuzumab emtansine for HER2-positive advanced breast cancer. N Engl J Med, 2012. 367(19): p. 1783-91.
51. Burris, H.A., 3rd, et al., Phase II study of the antibody drug conjugate trastuzumab-DM1 for the treatment of human epidermal growth factor recep-tor 2 (HER2)-positive breast cancer after prior HER2-directed therapy. J Clin Oncol, 2011. 29(4): p. 398-405.
52. Lewis Phillips, G.D., et al., Targeting HER2-positive breast cancer with trastuzumab-DM1, an antibody-cytoto ic drug conjugate. Cancer Res, 2008. 68(22): p. 9280-90.
53. Burris, H.A., 3rd, et al., Trastuzumab emtansine (T-DM1): a novel agent for targeting HER2 breast cancer. Clin Breast Cancer, 2011. 11(5): p. 275-82.
54. Chen, J., et al., Antibody-cytoto ic agent conjugates for cancer therapy. Expert Opin Drug Deliv, 2005. 2(5): p. 873-90.
55. Wu, A.M. and P.D. Senter, Arming antibodies: prospects and challenges for immunoconjugates. Nat Biotechnol, 2005. 23(9): p. 1137-46.
56. Lopus, M., Antibody-DM1 conjugates as cancer therapeutics. Cancer Lett, 2011. 307(2): p. 113-8.
57. Chari, R.V., Targeted cancer therapy: conferring specificity to cytoto ic drugs. Acc Chem Res, 2008. 41(1): p. 98-107.
58. Slamon, D.J., et al., Human breast cancer: correlation of relapse and survival with amplification of the HER-2/neu oncogene. Science, 1987. 235(4785): p. 177-82.
59. Kallioniemi, O.P., et al., ERBB2 amplification in breast cancer analyzed by fluorescence in situ hybridization. Proc Natl Acad Sci U S A, 1992. 89(12): p. 5321-5.
60. Shen, B.Q., et al., Catabolic fate and pharmacokinetic characterization of trastuzumab emtansine (T-DM1): an emphasis on preclinical and clinical ca-tabolism. Curr Drug Metab, 2012. 13(7): p. 901-10.
61. Junttila, T.T., et al., Trastuzumab-DM1 (T-DM1) retains all the mechanisms of action of trastuzumab and efficiently inhibits growth of lapatinib insensitive breast cancer. Breast Cancer Res Treat, 2011. 128(2): p. 347-56.
62. Lin, K. and J. Tibbitts, Pharmacokinetic considerations for antibody drug conjugates. Pharm Res, 2012. 29(9): p. 2354-66.
63. Xu, K., et al., Characterization of intact antibody-drug conjugates from plas-ma/serum in vivo by affinity capture capillary liquid chromatography-mass spectrometry. Anal Biochem, 2011. 412(1): p. 56-66.
64. Xu, K., et al., Characterization of the drug-to-antibody ratio distribution for antibody-drug conjugates in plasma/serum. Bioanalysis, 2013. 5(9): p. 1057-71.
65. Gupta, M., et al., Clinical implications of pathophysiological and demographic covariates on the population pharmacokinetics of trastuzumab emtansine, a HER2-targeted antibody-drug conjugate, in patients with HER2-positive meta-static breast cancer. J Clin Pharmacol, 2012. 52(5): p. 691-703.
66. Lu, D., et al., Population pharmacokinetics of trastuzumab emtansine (T-DM1), a HER2-targeted antibody-drug conjugate, in patients with HER2-positive metastatic breast cancer: clinical implications of the effect of covari-ates. Cancer Chemother Pharmacol, 2014. 74(2): p. 399-410.
85
67. Gibiansky, L. and E. Gibiansky, Target-mediated drug disposition model and its appro imations for antibody-drug conjugates. J Pharmacokinet Pharmaco-dyn, 2014. 41(1): p. 35-47.
68. Beal SL, S.L., Boeckmann AJ, and Bauer RJ (eds), NONMEM 7.3.0 Users Guides. (1989–2013). Hanover, MD.
69. Khandelwal, A., et al., A fast method for testing covariates in population PK/PD Models. Aaps j, 2011. 13(3): p. 464-72.
70. Hooker, A. and K. MO., The Kaplan-Meier Mean Covariate (KMMC) plot: a new diagnostic for covariates in time-to-event models. Population Approach Group Europe Meeting. 2012, Venice, Italy.
71. Dieras, V., et al., Trastuzumab emtansine in human epidermal growth factor receptor 2-positive metastatic breast cancer: an integrated safety analysis. J Clin Oncol, 2014. 32(25): p. 2750-7.
72. Perry, M.C., The chemotherapy source book. 2008: Lippincott Williams & Wilkins.
73. Blum, R.H. and T. Kahlert, Maytansine: a phase I study of an ansa macrolide with antitumor activity. Cancer Treat Rep, 1978. 62(3): p. 435-8.
74. Blum, R.H., et al., A therapeutic trial of maytansine. Cancer Clin Trials, 1978. 1(2): p. 113-7.
75. Rodon, J., et al., Cantuzumab mertansine in a three-times a week schedule: a phase I and pharmacokinetic study. Cancer Chemother Pharmacol, 2008. 62(5): p. 911-9.
76. Tolcher, A.W., et al., Cantuzumab mertansine, a maytansinoid immunoconju-gate directed to the CanAg antigen: a phase I, pharmacokinetic, and biologic correlative study. J Clin Oncol, 2003. 21(2): p. 211-22.
77. Uppal, H., et al., Potential mechanisms for thrombocytopenia development with trastuzumab emtansine (T-DM1). Clin Cancer Res, 2015. 21(1): p. 123-33.
78. Poon, K.A., et al., Preclinical safety profile of trastuzumab emtansine (T-DM1): mechanism of action of its cytoto ic component retained with improved tolerability. Toxicol Appl Pharmacol, 2013. 273(2): p. 298-313.
79. Shen, B.Q., et al., Conjugation site modulates the in vivo stability and thera-peutic activity of antibody-drug conjugates. Nat Biotechnol, 2012. 30(2): p. 184-9.
80. Erickson, H.K. and J.M. Lambert, ADME of antibody-maytansinoid conju-gates. Aaps j, 2012. 14(4): p. 799-805.
81. Deng, R., et al., Projecting human pharmacokinetics of therapeutic antibodies from nonclinical data: what have we learned? MAbs, 2011. 3(1): p. 61-6.
82. Lu, D., et al., An integrated multiple-analyte pharmacokinetic model to char-acterize trastuzumab emtansine (T-DM1) clearance pathways and to evaluate reduced pharmacokinetic sampling in patients with HER2-positive metastatic breast cancer. Clin Pharmacokinet, 2013. 52(8): p. 657-72.
83. Wang J, S.W., Zaro JL (eds), Antibody-Drug Conjugates, The 21st Century Magic Bulletsw for Cancer. 2015: AAPS Advances in the Pharmaceutical Scences Series.
84. Sievers, E.L. and P.D. Senter, Antibody-drug conjugates in cancer therapy. Annu Rev Med, 2013. 64: p. 15-29.
85. Donaghy, H., Effects of antibody, drug and linker on the preclinical and clini-cal to icities of antibody-drug conjugates. MAbs, 2016. 8(4): p. 659-71.
86
86. Wallin, J.E., L.E. Friberg, and M.O. Karlsson, Model-based neutrophil-guided dose adaptation in chemotherapy: evaluation of predicted outcome with differ-ent types and amounts of information. Basic Clin Pharmacol Toxicol, 2010. 106(3): p. 234-42.
87
Acta Universitatis UpsaliensisDigital Comprehensive Summaries of Uppsala Dissertationsfrom the Faculty of Pharmacy 217
Editor: The Dean of the Faculty of Pharmacy
A doctoral dissertation from the Faculty of Pharmacy, UppsalaUniversity, is usually a summary of a number of papers. A fewcopies of the complete dissertation are kept at major Swedishresearch libraries, while the summary alone is distributedinternationally through the series Digital ComprehensiveSummaries of Uppsala Dissertations from the Faculty ofPharmacy. (Prior to January, 2005, the series was publishedunder the title “Comprehensive Summaries of UppsalaDissertations from the Faculty of Pharmacy”.)
Distribution: publications.uu.seurn:nbn:se:uu:diva-292617
ACTAUNIVERSITATIS
UPSALIENSISUPPSALA
2016