J Pharmacokinet Pharmacodyn (2014) 41:461–478

DOI 10.1007/s10928-014-9386-9

ORIGINAL PAPER

Modeling cancer-immune responses to therapy

L. G. dePillis • A. Eladdadi • A. E. Radunskaya

Received: 1 April 2014 / Accepted: 17 September 2014 / Published online: 4 October 2014
Ó Springer Science+Business Media New York 2014

Abstract Cancer therapies that harness the actions of the aim of this chapter is to highlight the importance of
immune response, such as targeted monoclonal antibody mathematical modeling and simulation in the design of
treatments and therapeutic vaccines, are relatively new and immunotherapy protocols for cancer treatment. The results
promising in the landscape of cancer treatment options. demonstrate the power of these approaches in explaining
Mathematical modeling and simulation of immune-modi- determinants that are fundamental to cancer-immune
fying therapies can help to offset the costs of drug dis- dynamics, therapeutic success, and the development of
covery and development, and encourage progress toward efficient therapies.
new immunotherapies. Despite advances in cancer immu-
nology research, questions such as how the immune system Keywords Mathematical model  Tumor 
interacts with a growing tumor, and which components of Chemotherapy  Immunotherapy  Mixed chemo-
the immune system play significant roles in responding to immunotherapy  Dosimetry  Pharmaceutical  Optimize
immunotherapy are still not well understood. Mathematical
modeling and simulation are powerful tools that provide an
analytical framework in which to address such questions. A Introduction
quantitative understanding of the kinetics of the immune
response to treatment is crucial in designing treatment The world-wide incidence of cancer and cancer-related
strategies, such as dosing, timing, and predicting the deaths increases each year, despite medical progress in the
response to a specific treatment. These models can be used diagnosis and treatment of many types of cancer [1].
both descriptively and predictively. In this chapter, various According to recent studies by the American Association
mathematical models that address different cancer treat- for Cancer Research (AACR), the most promising avenues
ments, including cytotoxic chemotherapy, immunotherapy, in cancer research are epigenetics, cancer prevention and
and combinations of both treatments, are presented. The immunotherapy [2].
Recent progress in cancer immunology and advances in
immunotherapy suggest that the immune system plays a
L. G. dePillis (&)
fundamental role in combating tumors, and hence can be
Department of Mathematics, Harvey Mudd College, Claremont,
CA, USA used as a vehicle to prevent or cure cancer [3]. Despite
e-mail: [email protected] these developments, a thorough understanding of the
complex dynamics of the tumor-immune system remains
A. Eladdadi
tenuous. The multidimensional nature of tumor–immune
Department of Mathematics, The College of Saint Rose, Albany,
NY, USA interactions requires cross-disciplinary approaches to cap-
e-mail: [email protected] ture more realistic dynamics of the essential biology [2].
One such approach combines cancer immunology with
A. E. Radunskaya
mathematics to model aspects of these multifaceted inter-
Department of Mathematics, Pomona College, Claremont, CA,
USA actions. Using mathematical models to explore the
e-mail: [email protected] dynamics of different components of the immune system

123
462 J Pharmacokinet Pharmacodyn (2014) 41:461–478

allows interdisciplinary teams of laboratory, computational Mathematical models of cytotoxic chemotherapy are
and clinical researchers to investigate the mechanisms usually more straightforward than are models of immuno-
underlying an effective immune response to cancer as well therapy. For example, kill rate parameters needed for
as to design improved therapy protocols. These therapeutic chemotherapy models can often be measured directly using
strategies typically combine chemotherapy with immuno- in vitro cell lysis assays. The immune system cannot be
therapy to deplete cancer cells and simultaneously boost ignored, however, when designing chemotherapeutic pro-
the immune response. tocols, since the ability of the host to mount a prolonged
Mathematical modeling is a powerful tool that can help defense is crucial to a positive long-term outcome. The
researchers and clinicians understand complex mechanisms immune system is a complex network of cells and chemical
of cancer development and progression. Ideally, modeling signals, and mathematical models of immunotherapy must
should lead to improved therapy strategies, and stimulate describe multiple immune populations. Multiple self-reg-
the development of new ones. For example, pharmacoki- ulating mechanisms can prevent the sustained enhancement
netics/pharmacodynamics (PK/PD) modeling has been of a particular part of the immune response, so the design
used effectively to predict the effect and efficacy of drug of immunotherapeutic strategies involves tracking many
dosing over time. PK/PD has also been an essential com- interacting factors.
ponent of drug discovery and development. PK/PD models In this chapter, models of cytotoxic therapy that take
are typically compartment models that use ordinary and/or into account the health of the immune system are pre-
delay differential equations. These systems of equations sented, as well as how these models are adapted to describe
describe temporal changes within the compartments, and treatments that affect the immune response itself.
trafficking between compartments. One goal of mathe- The breadth and depth of the work carried out by many
matical models is dosimetry, or the accurate prediction of different researchers in modeling tumor–immune interac-
dose-response by simulating different dosing schedules. tions over the past two to three decades is vast. This
Mathematical modeling has been useful in guiding che- chapter is by no means a comprehensive review of this
motherapy and radiotherapy treatments [4–7] and com- body of work. Instead, it provides an overview and a brief
bined radiopharmaceutical therapy [8]. Mathematical description of some of the existing mathematical models on
models can also illuminate the underlying mechanisms of tumor–immune dynamics and therapies. Our hope is that
disease and drug response as well as the stochastic nature the reader will be inspired by this chapter to investigate
of these processes [9–11]. further both the mathematical models mentioned herein,
In the pharmaceutical industry, it normally takes and other related works that space constraints prevent us
10–15 years to develop and manufacture a new drug, at a from summarizing. Interested readers are encouraged to
cost of 10M$–1.2B$. One in ten new drugs in clinical trials consult the cited literature for detailed model development,
is likely to be approved [12]. Of necessity, the pharma- analysis and simulation. A sample of other worthwhile
ceutical industry is becoming more innovative in reducing works can be found in references [14–33].
the time and the cost of drug discovery and development of The aim of this chapter is to highlight the importance of
new therapies. This development to production process is mathematical modeling and simulation in immune-modu-
streamlined by devoting greater resources to mathematical lating therapies for cancer. Sect. 2 addresses mathematical
and computational (in silico) modeling. Mathematical models of cytotoxic chemotherapy. Sect. 3 covers some
models can also be used to personalize treatments by topics in immunotherapy. Sect. 4 presents a sampling of
identifying key system parameters that most strongly affect models that combine chemotherapy and immunotherapy
treatment outcomes. Knowing these parameters is espe- treatments of cancer.
cially important when designing immunotherapies for
cancer, since effective protocols require ‘‘treatment per-
sonalization,’’ whereby doses and administration schedules Modeling cytotoxic chemotherapy
are tailored to the individual patient [13].
This chapter introduces the reader to mathematical Ordinary differential equations (ODE) models
models of chemotherapy and immunotherapy. Immuno-
therapy can be considered a special type of chemotherapy. In the early 1990s, Kuznetsov et al. [34] proposed a
However, traditional cytotoxic chemotherapy targets and mathematical model of the cytotoxic T lymphocyte
kills tumor cells directly. The goal of immunotherapy, on response to the growth of an immunogenic tumor. The
the other hand, is to increase the effectiveness of the model was a two-population system of Ordinary Differ-
immune response against a cancer. Thus, immunotherapy ential Equations (ODEs) that included exponential tumor
is a treatment that indirectly targets tumor cells. growth and Michaelis–Menten dynamics for tumor cell

123
J Pharmacokinet Pharmacodyn (2014) 41:461–478 463

Fig. 1 Simulations of the model in [34] illustrating the possibility first trajectory (IC1) is in the basin of attraction of the low tumor
that a tumor could remain dormant for a time, but then escape equilibrium, and the tumor population approaches a small value. The
immune surveillance. This phenomenon is known as sneaking second trajectory (IC2) is in the basin of attraction of the large tumor
through. Two simulations are shown: one with an initial population equilibrium. The IC2 trajectory rapidly approaches the large tumor
of 6 9 106 effector cells (labeled ‘‘IC1’’ in red) and one with an equilibrium after coming through the initial dormant period. Right
initial population of 6.9 9 106 effector cells (labeled ‘‘IC2’’ in blue). Panel The same two trajectories are shown in the state space. The
Both simulations start with the same number of tumor cells (T(0) = dashed line shows the boundary between the basins of attraction of
1.3 9 107 ). Left Panel The tumor populations in the two simulations the two equilibria. Note the logarithmic scale in the right panel.
are shown over time. In both cases, the tumor values initially get very Figure from [35] where more details can be found (Color figure
small and remain there for a while: this is the ‘‘dormant’’ period. The online)

recruitment of effector cells. Model parameters were the Kuznetsov system by replacing the tumor cell popula-
determined based on the kinetics of growth and regression tion T with T 2=3 . This substitution takes place in the
of B-lymphoma BCL1 tumor cells in the spleen of mice. Michaelis-Menten effector cell recruitment term, as well as
Model analysis and simulations showed that for a certain in the mass-action terms that represent the mutually
set of parameters, the course of tumor growth and its destructive interaction between effector cells and tumor
clinical manifestation had a recurring profile. This recur- cells. Model simulations are then presented to confirm the
ring profile took place over a 3- to 4-month cycle, a pattern hypothesis that overly aggressive chemotherapy dosing can
that is qualitatively similar to certain leukemias. A signif- damage the positive action of the immune response and
icant contribution of this model was that it captured the consequently reduce the degree of tumor control.
clinically observed and not well understood phenomena of Studies by Thomlinson et al. [40] show that some cancer
tumor dormancy and sneaking through. This relatively treatments reveal an unexpected asynchronous response to
simple two-population model was able to provide one chemotherapy treatments in some patients. Although one
possible explanation for these mysterious tumor behaviors. might expect a tumor to shrink initially after the adminis-
The activity of the immune system against the tumor pro- tration of a chemotherapy dose, some tumors were actually
vides a feasible explanation that addresses both tumor observed to grow directly after treatment administration,
dormancy and sneaking through. See Figure 1. but then shrink at a later time. To address the question of
A follow up study to Kuznetsov’s model was recently what could be driving such an asynchronous tumor
conducted by Roesch et al. [36]. The authors fit their model response to cytotoxic chemotherapy, de Pillis et al. [41, 42]
using data from [37–39]. This study shows that chemo- developed a mathematical model that describes the inter-
therapy treatments for lymphoma that are too intense result actions of the tumor-immune system and tracks three cell
in less effective tumor control. The authors suggest that populations using a nonlinear system of deterministic
cytotoxic chemotherapy that is too aggressive damages the ODEs. The three cell populations are tumor cells, immune
immune system and undermines tumor control. Roesch and cells, and normal cells. For simplicity, this model assumes
colleagues propose a modification of Kuznetszov’s two- homogeneity within each cell population, even though each
population ODE model that adds first-order loss terms, of these three cell types contains multiple subtypes. The
allowing chemotherapy to damage both tumor and immune hypothesis was that competition for resources and the
cells. They then modify Kuznetsov’s model by assuming presence of the immune response could be partly respon-
that effector–tumor interactions are no longer well-mixed, sible for the observed asynchronicity in the data. The
but that effector cells interact with only the surface of the purpose of the study was to determine how chemotherapy
growing tumor, which in turn is assumed to be spherical. In outcomes could be affected both by competition for
practice, this means that the authors modify three terms in resources among the three cell populations and by the

123
464 J Pharmacokinet Pharmacodyn (2014) 41:461–478

predator-prey behavior between the immune and tumor in Eqs. (1–3)] varied, with normal cells being damaged at
cells. the lowest rate, and tumor cells being damaged at the
A dynamical systems analysis of the model of [41, 42] highest rate. With treatment terms included, simulation
determined the different types of equilibria that existed in the with hypothetical dosing schedules was possible. The
system. One interesting finding was that the only scenario relative mathematical simplicity of this model allowed for
under which a completely tumor-free equilibrium would the application of optimal control theory. Optimal control
persist was when the immune system was present and suf- theory is a mathematical approach that can automatically
ficiently strong. In other words, the tumor-free equilibrium search for improved treatment protocols. In an optimal
was stable as long as the per-capita growth rate of the tumor control problem, an objective must be chosen and repre-
was dominated by the strength of the immune response. sented by a function. In the case of this tumor model, the
Competition for resources by normal cells also played an objective used in [41] was to minimize the total tumor
important role in maintaining the stability of the tumor-free size within a pre-determined time frame while ensuring
state. If, on the other hand, tumor growth was too aggressive, that the normal cell population was not too heavily
then according to this model, no amount of chemotherapy damaged. In this case, the objective was obtainable.
would be able to eliminate the tumor successfully. However, simulations yielded oscillations in cell popula-
Other types of biologically interesting equilibria also tions sizes, a side effect that would not be particularly
exist in this model. For example, there are co-existing desirable in the clinic.
equilibria, which are states in which all three cell types Fortunately, the choice of the objective function is not
exist in a stable state together in the system. Analysis unique, and a different choice can lead to very different
showed that there were zero, one, two or three of these outcomes. For example, it is possible to minimize the
equilibria. Depending on the parameter values, there could tumor size while also minimizing oscillations in the cell
also be stable co-existing states in which the tumor cell populations, as was done in [42]. In other cases, the
population is relatively small, and not growing larger. This objective has been to minimize both tumor size and total
would be considered a harmless state. The goal of che- chemotherapy dosing simultaneously, as in [43]. In that
motherapy, then, would be to drive the system either into a study, the authors carry out an in-depth analysis and
zero-tumor state, or alternately, into one of these harmless comparison of linear and quadratic objective functions.
equilibrium states (see Fig. 2 in [41]). Additionally, they explore how treatment outcomes differ
The model developed in [41] was extended to include if, in addition to minimizing tumor burden, the time
chemotherapy dosing. In addition to the three cell popula- endpoint of treatment is also optimized. Interestingly,
tions, a fourth quantity representing a drug concentration in although optimal treatment protocols differed depending
the system was added. Each cell population equation was on whether the objective functional was linear or qua-
then modified with a term to describe damage done to cells dratic and on how the final time was chosen, one quali-
by chemotherapy. The system of equations that represents tative commonality emerged in all cases. In every case,
the tumor–immune–chemo interaction was given by: the most effective treatments were comprised of an initial
large (i.e. maximal tolerated dose) treatment bolus fol-
dE rET lowed by lower dose, longer term infusions of chemo-
¼ s  dE  c1 ET  k1 ð1  eu Þ ð1Þ
dt rþT therapy (see Figs. 3 and 4 in [43]). The authors point out
that this optimal control generated protocol actually
dT reflects certain clinical findings.
¼ a1 Tð1  b1 TÞ  c2 ET  c3 NT  k2 ð1  eu Þ ð2Þ
In [44], the models of [41, 42] are summarized in an
dt
overview of best practices in mathematical modeling. In
dN chapter 4 of [44], the authors include figures from a
¼ a2 Nð1  b2 NÞ  c4 NT  k3 ð1  eu Þ ð3Þ numerical simulation of the system of Eqs. (1–4) that
dt
compares cases of tumor growth with no treatment, with
traditional ‘‘pulsed’’ chemotherapy given in regular doses
du
¼ mðtÞ  d2 u: ð4Þ at equal intervals, and with optimally controlled doses of
dt chemotherapy (see Fig. 2).
Here, E represents the population of effector immune In some applications, it is not sufficient to represent the
cells (such as CD8þ T cells); T is the population of tumor immune response with a single homogeneous population of
cells; N represents a normal cell population. Parameter u effector killer cells. It may be necessary to capture the
is the concentration of chemotherapy. The parameters and different dynamics of more than one immune cell popula-
their values are described in detail in [41, 42]. The rate of tion. For example, in order to predict a system’s response
damage to each cell population [represented by k1 ; k2 ; k3 to a particular treatment that can boost the effectiveness of

123
J Pharmacokinet Pharmacodyn (2014) 41:461–478 465

No Chemotherapy
No Chemotherapy: I 0=0.15 No Chemotherapy: I 0=0.1
1 1
Immune Immune
0.9 Tumor 0.9 I0=0.1, T0=0.25, N0=1 Tumor
I =0.15, T =0.25, N =1
0 0 0 Normal
s=0.33, R=0.01, A=0.3 Normal s=0.33, R=0.01, A=0.3
0.8 0.8

0.7 0.7

Number of Cells
Number of Cells

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
0 50 100 150 0 50 100 150
Time in Days Time in Days
Traditional Pulsed Chemotherapy

Traditional pulsed chemotherapy: I0=0.15 Traditional pulsed chemotherapy: I0=0.1

2 2
Immune
Immune 1.8 Tumor
1.8 I =0.15, T =0.25, N =1 I0=0.1, T0=0.25, N0=1
0 0 0 Tumor Normal
s=0.33, R=0.01, A=0.3 Normal s=0.33, R=0.01, A=0.3
1.6 1.6 Drug Dose
Drug Dose
1.4 1.4
Number of Cells
Number of Cells

1.2 1.2

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 50 100 150 0 50 100 150
Time in Days Time in Days

Solution of Optimization Problem: Irregularly-space, Extended Doses

Optimal control chemotherapy: I0=0.15 Optimal control chemotherapy: I0=0.1
2 2

1.8 I0=0.15, T0=0.25, N0=1 1.8 I =0.1, T =0.25, N =1

0 0 0
s=0.33, R=0.01, A=0.3
s=0.33, R=0.01, A=0.3
1.6 1.6

1.4 1.4
Number of Cells

Number of Cells

1.2 1.2

1 1

0.8 0.8
Immune
Immune
0.6 0.6 Tumor
Tumor
Normal
Normal
0.4 0.4 Drug Dose
Drug Dose
0.2 0.2

0 0
0 50 100 150 0 50 100 150
Time in Days Time in Days
I0 = E(0) = .15 I0 = E(0) = .10

123
 466 J Pharmacokinet Pharmacodyn (2014) 41:461–478

b Fig. 2 A comparison of chemotherapy treatment protocols. This MHC-I receptors. However, this down-regulation will
figure depicts three different solutions of the model system given by make them potential NK cell targets.
Eqs. (1) through (4). Top row The dose function, mðtÞ, which
describes the rate of drug delivery, is zero (no treatment). The left In [45], the authors develop a three-population model
graph shows a solution with an initial value describing a patient with that tracks tumor cells and two separate killer immune cell
stronger initial immune strength (E(0) = 1.5 9 105 cells), while the populations: the NK cells, and the CD8þ T (killer T, or
right graph shows a patient with weaker initial immune strength (E(0) cytotoxic T) cells. The model is validated using laboratory
= 1 9 105 cells). Without treatment, both solutions show the tumor
population converging to a high tumor equilibrium. Center row The mouse data published in [47], and modeling terms are
dose function, mðtÞ, describes regularly spaced treatments: a bolus introduced that reflect the distinct dynamics governing the
injection every 3 days for 90 days. The left panel shows that this NK and CD8þ T cell populations. In particular, the authors
treatment does control the tumor in the patient with the stronger found that the laboratory data required a new functional
immune system, but it is ineffective in the case of the patient with the
weaker immune system, shown in the right panel. Bottom row The form to represent how the T cells killed the tumor cells. If
dose function, mðtÞ, is the solution to an optimal control problem in we let T(t) represent the tumor cell population and N(t) the
which the objective is to minimize final tumor size while keeping the NK cell population at time t, then a mass-action dynamic is
immune cell population above a specified threshold. Function mðtÞ in described by the following differential equation:
this scenario is an irregularly varying function, and in both the case of
the stronger and the weaker immune response, shown in the left and dT
right panels, the tumor is reduced to zero. Note that the total amount ¼ aNT: ð5Þ
dt
of drug delivered (the area under the graph of the function mðtÞ) is the
same in the second and third rows. Figures reproduced from Fig.12 of Here, a is a parameter that can be adjusted to fit to a
[44] particular data set. Fits to two sets of data are shown in
Fig. 3. These data are from [47], and result from two lysis
assays. In both assays, initial immune cell to tumor cell
the cytotoxic immune cell response, it is necessary to ratios ranged from 6:1 up to 100:1. The percent of tumor
model the kinetics of effector-cell lysis of target cells. In cell lysis after four hours was measured. For Data Set 1,
[45], the authors show that different cytotoxic immune cell which has lysis data for control cells, the parameter a is
types operate using distinct cell lysis kinetics. 1.9 9 10-4 Data Set 2 has lysis data for tumor cells that
There are two main branches of the cellular immune have been transduced with immune-stimulating ligands. In
response: the innate and the adaptive response. Natural this case, the value of a is 1.4 9 10-3, nearly ten times
Killer (NK) cells are a type of cytotoxic lymphocyte higher than in the control case.
critical to the innate branch of the immune response. A The mass-action form of Eq. (5) was sufficient to
normal cell will express self-antigens through Major describe the data reflecting the NK killing of tumor cells in
Histocompatibility Complex I (MCH-I) receptors on its the mouse models published in [47]. However, this form
surface. When the NK cell contacts the target cell, if an could not adequately describe the action of the CD8þ T
MHC-I-self-antigen connection is made to the target cell, cells. In order to address this issue, the authors developed
the kill signal in the NK cell is deactivated. The NK cell what later came to be known as the dePillis-Radunskaya
and the target cell then separate and go their respective Law. This law states that the ratio of the immune cells to
ways. However, if the target cell is not expressing self- the tumor cells should be taken as the dependent variable in
antigens in the MHC-I receptor, or if the MHC-I receptor the per-cell kill rate.
has been down-regulated on the cell, as may be the case In particular, if we let L represent the CD8þ T lympho-
for mutated cancer cells, then the MCH-I connection cytes (‘‘L’’ for lymphocyte), the kill term in the differential
cannot be made with the NK cell. In this event, the NK equation for T takes the form:
cell’s cytolytic response is triggered, and the target cell dT ðL=TÞk
can be eliminated. ¼ d T: ð6Þ
dt s þ ðL=TÞk
Part of the adaptive (or specific) branch of the cellular
immune response is carried out by CD8þ T cells. Unlike the The exponent k represents how the lysis rate depends on
NK cells, these T cells must first be activated before going the lymphocyte to tumor ratio. The parameter s is the value
after target cells. In the case of cancer, the T cells have to at which the tumor size curve reaches half its maximum.
be trained to recognize tumor-specific antigen that is Therefore, increasing s shifts the curve to the right.
present in MHC-I receptors that are on the surface of the Parameter d gives the maximum cell lysis rate. The ratio-
tumor cell. When the T cell binds to the MHC-I-tumor- dependent kill rate was validated using data from several
antigen complex, a kill signal is triggered, either through independent sources, in particular those published in [47,
the secretion of granulocytes, or via the Fas/FasL binding 48]. For detailed parameter estimates see [45]. A compar-
mechanism (see, e.g., [46]). Some tumor cells are able to ison of the mass-action and ratio-dependent kill rates is
evade T-cell attack by down-regulating expression of shown in Fig. 4.

123
J Pharmacokinet Pharmacodyn (2014) 41:461–478 467

Fig. 3 Graphs showing a least-

squares fit to data from [47]
using the mass-action kill rate
given in Eq. (5)

Fig. 4 A comparison of the

Mass Action Law (Eq. (5)) and
the dePillis-Radunskaya Law
(Eq. (6)). The immune cell
population is given as a multiple
of the tumor cell population

Agent-based and cellular automata models. An allergic reaction from a patient can force treatment to
be halted. The authors use a Cellular Automata (CA) model
Mathematical modeling techniques other than ODEs have of the immune system to investigate how the allergic drug
also been used for modeling cytotoxic chemotherapy. A schedule, dose fractionation, dosing interval, and rate of
cellular automaton (CA) model assigns a state to each administration affect the resulting symptoms of hypersen-
element of a grid, and all states are updated according to sitivity. The authors explore a wide variety of other dose
rules that depend on the current state of the grid element as scheduling scenarios that can help to reduce the allergic
well as the state of neighboring grid elements. Each com- response in a patient. For example, simulations show that
putational grid element might represent one biological cell, for an equal total drug dose split into two doses, lower
or a small portion of physical space containing a collection levels of histamine are released when the first dose is lower
of cells. A CA model is quite flexible, since it can describe than the second dose (and similarly, higher histamine levels
the interactions of individual cells with their neighbors, are released when the first dose is higher). They point out,
rather than treating the populations as homogeneous. however, that a larger second dose is not advisable unless
In a study by Castiglione et al. [49], the authors examine the sensitization is made with a very low first dosage.
the processes by which chemotherapeutic agents such as Simulations also showed that controlling the dosing inter-
paclitaxel can induce hypersensitive responses in a patient. val can modify histamine release. The standard dosing

123
468 J Pharmacokinet Pharmacodyn (2014) 41:461–478

interval of two weeks actually maximizes histamine equations, but one that can be analyzed using techniques
release. Simulations provide dose-dependent predictions from dynamical systems theory.
for how far apart dosing should be in order to minimize These studies show that a difference equation approxi-
histamine release. Overall, their results suggest that there mation of the agent-based model can closely replicate the
may be an optimal dosing schedule that should be used behavior of the original agent based model under a wide
during anti-cancer therapies that will minimize the allergic range of conditions and can run in a few minutes as
response in patients. opposed to a few days. Furthermore, the reduced com-
An agent-based model (ABM) is a generalization of a plexity of the system allowed for a sensitivity analysis of
CA model. In an agent-based model, each cell is assigned the parameters of the original system, identifying those that
characteristics (cell type, stage in cell cycle), and the state were most important in determining treatment outcomes.
is updated according to rules that describe the interaction of The authors then extended the model to derive an
cells with one another, as well as with the environment or analogous partial differential equation model of CML with
treatments. There are several distinctions between ABM imatinib treatment. The authors used parameters from [50]
and CA models. First, ABM models are often ‘‘off-lattice,’’ based on human data of stem cell turnover and the response
which means that the ‘‘agents’’ don’t necessarily position of CML under treatment by Gleevec. In [54], the authors
themselves regularly through space. In CA models, on the used a delay differential equation of a T cell response to
other hand, the state variables are values, possibly multi- chronic myelogenous leukemia during Gleevec treatment
dimensional, that are assigned to nodes of a lattice or a to investigate whether strategic treatment interruptions can
graph. In particular, in an ABM, multiple agents can improve therapeutic outcome. This study concluded that
occupy the same position in space, and agents’ locations strategic treatment interruptions might drive a potent and
relative to one another can change over time. In a CA, the prolonged anti-cancer T cell response, leading to extended
time step is regular, and all cells are updated at the same cancer remission. Thus, strategic treatment interruptions
time. In an ABM, the agents’ states can be updated at might serve as a feasible treatment approach.
arbitrary times, potentially triggered by the environment or
by other cues. Thus, ABMs can describe a much richer
variety of processes. While ABMs allow for a great deal of Modeling immunotherapies
flexibility, they usually must also track of a huge number of
variables. Both CA and ABM models can make use of The goal of immunotherapy is to boost the system’s own
probabilistic rules for describing a cell’s propensity for immune defense against the tumor. The benefits to this
changing from one state to another. However, modeling approach over cytotoxic chemotherapies include relatively
individual cells is computationally costly. Just the lower toxicity, and, in some cases, the ability to target
description of the initial state requires specifying the tumor cells that cytotoxic drugs are unable to reach. Some
individual states of millions of (simulated) cells. of the earliest successes with immunotherapy have been in
Roeder et al. [50] present an agent-based model of the treatment of metastatic kidney cancer and metastatic
chronic myelogenous leukemia (CML), extending a pre- melanoma, particularly in cases when the disease did not
vious model due to [51]. Subsequently the authors in [52, respond to chemotherapy [55]. In the case of glioblastoma
53] simplified the model presented in [50] to study Gleevec multiform (GBM), a common and very aggressive brain
(imatinib), a treatment for CML. In this study, in order to cancer, it is difficult to get drugs through the blood brain
address some issues of computational cost, individual cells barrier, and the GBM cells are resistant to radio-therapy. A
were no longer tracked. Instead, an age-structured model variety of immunotherapies for GBM are now being tested
was developed in which the number of cells of each type in in clinical trials [56].
each stage are the system variables. Figure 5 illustrates In their seminal work [57], Kirschner and Panetta pro-
how an ABM could be reconfigured as an age-structured pose a three-population ODE model to capture IL-2 treat-
model. The structuring of the population allows the model ment of cancer. The model tracks tumor cells, a population
to describe the action of imatinib on only the proliferating of activated immune-effector cells (meant to capture the
cells in the G1 stage of the cell-cycle. The lumping of the activity of cytotoxic T cells, macrophages, and natural
agents, or cells, into compartments, dramatically reduces killer cells in one population), and a concentration of the
the number of computations required at each time step. A cytokine IL-2 (Interleukin-2). IL-2 is one of a number of
further reduction in complexity was achieved by replacing cytokines that have been evaluated in cancer immuno-
the stochastic rules in [50] by a deterministic rule that therapy. IL-2 is produced within a host (endogenously) as
describes the average behavior of the cell populations in part of natural immune activity. In addition, concentrations
each stage. This results in a large system of difference of cloned IL-2 have been used as part of external

123
J Pharmacokinet Pharmacodyn (2014) 41:461–478 469

Fig. 5 An agent-based model

consisting of N cells, each with
its own characteristics
(‘‘states’’) can be approximated
by an age-structured model,
where the N cells are classified
into a relatively small number of
types. This reduces
computational complexity, and
can make the model tractable to
analysis

(exogenous) cancer treatment regimens [58, 59]. Rosen- stimulator of cellular immunity against cancer, and the
berg and colleagues found that in the presence of high authors outline a strategy for selecting IL-21 immuno-
concentrations of cloned IL-2, large numbers of activated therapy as a function of tumor immunogenicity.
lymphoid cells were generated that could kill certain tumor Bellomo et al. have produced a number of works and
cells [58]. In the Kirschner and Panetta model, endogenous reviews related to modeling tumor–immune interactions,
IL-2 is stimulated through interactions between the effector c.f. [61–65]. For example, in [63], the authors provide a
and tumor cell populations, but is tumor-cell saturation review of certain generalized kinetic models of the cellular
limited through the use of a Michaelis–Menten term. immune response to tumor. The authors present details of
Exogenous IL-2 treatment can be introduced into the sys- progression models (in which endothelial cells lose dif-
tem through an independent source term. Additionally, ferentiation and begin to progress toward tumor behavior),
adoptive cellular immunotherapy (ACI) - representing and mean field models (in which cells can sense the pre-
either tumor infiltrating lymphocyte (TIL) or lymphokine sence of other cells within a certain distance, and in which
activated killer cell (LAK) injections - can be simulated cell proliferation or death processes are determined by
through a source term in the effector cell equation. The interactions with these neighboring cells). They also dis-
authors observe that according to their model, treatment cuss some approaches to modeling immunotherapeutic
with IL-2 can be somewhat unstable: too little IL-2 does strategies.
not sufficiently boost immune activity, while too much can Immunotherapies for cancer operate via several different
cause pathological effects. The model is able to evidence pathways. Some therapies aim to promote the response
both short-term oscillations in tumor size, as well as longer from cytotoxic T-lymphocytes, such as CD8þ T cells, by
term tumor relapse. The authors find that IL-2 treatment attaching immune-stimulating adjuvants such as a virus or
alone at any dose is not able to control tumor growth, but bacteria to a patient’s own irradiated tumor cells. Alter-
that the right amount of IL-2 in combination with some nately, the immune cascade can be triggered by injecting
form of ACI can lead to successful tumor clearance. either trained dendritic cells or peptides found on tumor
In another study by Cappuccio et al. [60], the authors cells into an individual. A third option is to transfer acti-
investigate the effect of IL-21 immunotherapy treatments vated immune cells directly to the patient [55]. Antibodies
on tumor control. IL-21 accelerates the transition from the can also be used to stimulate NK cells through antibody-
innate NK-cell-mediated immune response to the adaptive dependent cell-mediated cytotoxicity (ADCC), a mecha-
CD8þ T cell-mediated anti-tumor immune response, the nism that allows the NK cells to recognize tumor cells
latter of which has been shown to be more effective in bound to specific antibodies [3]. Other immunotherapies
tumor control. The model is a six-population system of attempt to down-regulate the activity of immuno-suppres-
ODEs that tracks IL-21 concentration, NK cell population sive cells, such as regulatory T cells (T-regs) [66].
in the spleen, CD8þ T cell activity in the lymph nodes, a In [67], the authors combine laboratory experiments
CD8þ T cell memory element, a cytotoxic protein that with mathematical models to further an understanding of
affects tumor lysis, and overall tumor mass. The model is the effect of immunotherapies on cancer progression. In
able to capture experimental growth dynamics in B16 particular, their work delves into the details of the kinetics
melanoma as well as in MethA and MCA205 fibrosarco- of the adaptive cellular response. In [67], two types of
mas. The authors show that there is a strong dependence of immune cells in the adaptive response are modeled: the
the NK-cell/CD8þ T-cell balance on tumor immunogenic- CD8þ ‘‘killer’’ T-cells and the CD4þ ‘‘helper’’ T-cells.
ity. Model simulations support the clinical use of IL-21 as a These two types of T-cells respond to a cancer vaccine in

123
470 J Pharmacokinet Pharmacodyn (2014) 41:461–478

Fig. 6 Flow diagram

illustrating the five phases of the
T-cell cascade: (1) naive, (2)
rapidly proliferating, (3)
circulating, (4) apoptotic, (5)
memory. The proliferation
cascade is triggered by the
APCs (upper compartment).
The three connection times
appear as delays in the
differential equations. In the full
model, these five phases are
tracked for two types of T cells:
CD8þ and CD4þ T-cells. From
[67], Fig. 1

the form of a peptide injected into the system. The peptide Predicting optimal treatment protocols
vaccine activates Antigen-Presenting Cells (APCs) that
travel to the lymph organs and stimulate the proliferation of Once a model is calibrated to data, it can be used to test
T-cells that are trained to recognize the specific peptide different treatment protocols. Since cancer vaccines are
that is found on the tumor cells. Some of the new cells will weak antigens, one vaccination is not sufficient to elicit an
move into the system and locate the tumor cells. The effective immune response. It is important to ask, then,
proliferation, or expansion, phase is followed by a con- when the best time would be to administer a boost, i.e., a
traction phase, where the new, active T-cells begin to die subsequent vaccine dose. For the model developed in [67],
quickly. Some of these T-cells will become memory cells, a heuristic optimization technique was implemented to
able to be quickly stimulated when the same antigen is determine an optimal post-vaccination boosting time. The
presented at a later time. results of the mathematical optimization showed that a
In order to capture the delays inherent in this cellular boost should be administered 3 days after the initial vac-
immune response, as well as the different phases of the cination. The ‘‘Day 3 hypothesis’’ was then tested in the
cascade, the mathematical model in [67] consists of a laboratory. Laboratory results confirmed that indeed, the
system of eleven differential equations, with five delays. cellular response was greater after a boost on Day 3 than it
Each equation represents one of the five phases in the was after a later boost. For example, a boost on Day 6,
T-cell cascade: Naive, Proliferating, Apoptotic (dying which was just before the peak of the initial response, was
quickly), Circulating and Memory cells. This gives ten significantly less effective. See Fig. 7.
equations to describe the dynamics of the two cell types:
Helper Cells (CD4þ ) and Killer Cells (CD8þ ). The flow Modeling immune cell trafficking & dendritic cell
between cell phases for one cell type, as well as the trig- vaccines
gering of the immune cascade by APCs, is shown in Fig. 6.
The model focuses on the kinetics of these cells in the With an understanding of the kinetics of the cellular
spleen, in response to the presence of a weak antigen, such response to a weak antigen, a model of the effect of a
as a cancer vaccine. The eleventh equation describes the Dendritic Cell (DC) vaccine, such as the DC vaccine cur-
dynamics of the APCs. An important assumption in the rently studied for the treatment of prostate cancer [68], can
introduction of the delays is that once the APCs are in be developed. In [69], the authors extend a calibrated
contact with the T-cells, they do not disassociate. Fur- model of dendritic cell trafficking developed by Ludewig
thermore, the rate of cell death in the naive, proliferating et al. [70] by adding a tumor compartment to blood and
and memory phases is assumed to be much slower than the spleen compartments. The model bears similarities to the
connection times represented by the delays. This model model described in [67], in that it also includes delay dif-
was informed by laboratory experiments on mice, where ferential equations to describe dendritic cell activation in
labeled immune cells could be counted at various time the spleen. However, in [69] the kinetics in the spleen are
points (data shown in [67]). These experiments led to the simplified, and only the CD8þ T cells (the ‘‘killer cells’’)
question of whether there was a way to maximize the are tracked. After calibrating the model to published
immune response to a vaccine treatment. We address this experimental data, the model was used to investigate dif-
question in Sect. 3.1. ferent treatment strategies. In addition to exploring the

123
J Pharmacokinet Pharmacodyn (2014) 41:461–478 471

Fig. 7 Model simulations showing the increased cellular response (in after the initial vaccination (right panel). The vertical lines show the
total number of CD8þ cells) if a vaccine boost is given 3 days after timing of the vaccinations, time is in days. From [67], Fig. 8
the initial vaccination (left panel), compared to a boost given 6 days

Fig. 8 Fractionated Dosing comparison. Intratumoral injections. Compare DC dosing schedule from [71] (left panel) to hypothetical fractionated
dosing schedule (right panel). Figure reproduced from [69]

Fig. 9 Fractionated Dosing comparison. Intravenous injections. Compare DC dosing schedule from [71] (left panel) to hypothetical fractionated
dosing schedule (right panel). Figure reproduced from [69]

effect of varying dose, the model can also allow for a blood stream were the most effective strategy for slowing
comparison of the effects of injecting the DCs into the the growth of the tumor. Fig. 8 illustrates that fractionated
blood, or directly into the tumor. The conclusions of this in intratumoral dosing does not improve outcomes, but Fig. 9
silico study were that fractionated doses injected into the shows that fractionated intravenous dosing is more

123
472 J Pharmacokinet Pharmacodyn (2014) 41:461–478

Fig. 10 Prophylactic
vaccination and the effect of
varying immune strength
parameter d. Vaccinate with DC
treatments, days 0 and 7,
1 9 105 DCs per dose. Tumor
challenge on day 21, with
2 9 105 tumor cells. Dosing
follows Preynat [72]
experiment. With d = 1, tumor
is controlled as a result of
vaccination. Figure reproduced
from [69]

effective in controlling tumor growth. Model simulations in a few compartments, such as the spleen, blood or tumor.
also revealed an important parameter, the maximal kill More recent studies in modeling immunotherapies include
rate, d, given in Eq. (6). If this parameter is above a critical [78], in which the authors develop a space and time
value, then the tumor can be eradicated with a sufficiently dependent model representing tumor–lymphocyte interac-
large dose of the DC vaccine; if d is less than this critical tions, and they also explore the effects of immunotherapy.
value, then the tumor will always escape immune surveil- They find that chemotactic lymphocyte guidance plays
lance (see Fig. 10). Thus, model simulations uncovered a only a minor role in tumor control, but a lymphocyte
potential target for therapy: adjuvants that increase the blockade of advancing tumor cells can effectively control
effectiveness of the CD8þ T cells should be administered in tumor growth. The authors suggest that including an anti-
conjunction with the DC vaccine. migratory agent as part of tumor therapy could be critical
A model that explores immunotherapy targeting to tumor control.
immuno-suppressive cells was described in [73]. This In [79], the authors investigate Interleukin-27 as a
model includes regulatory T-cells (T-regs), which suppress potential anti-tumor agent. Tumor cells transfected with
the response of effector cells, such as CD8þ T-cells. The IL-27 have been shown to cause CD8þ T cells to promote
model also includes a variable that describes T-reg down- the secretion of the anti-tumor cytokine IL-10, but at the
regulating treatment with the drug sunitinib. The model same time IL-27 tends to inhibit CD8þ T secretion of
successfully replicates the results of clinical studies from Interferon-c, which then inhibits anti-tumor activity. The
[74, 75]. Simulated scenarios highlight certain patient authors present a mathematical model consisting of a sys-
characteristics that are most important in predicting a tem of partial differential equations that assumes a spher-
positive response to sunitinib of treatment. ical tumor and tracks IL-27, IL-10, Interferon-c, tumor cell
Heterogeneity of patients’ responses is the focus of the density, and tumor-antigen specific CD8þ T cell density.
model described in [76]. B Cell Chronic Lymphocytic Tumor cell proliferation is assumed to follow a logistic
Leukemia (B-CLL) is a disease of the immune system. growth curve, immune inhibition of tumor growth is satu-
Predicting a patient’s response to existing therapies for ration-limited, and all chemical and cell concentrations are
B-CLL has been extremely difficult, if not impossible. A allowed to diffuse through space. The numerical simula-
careful parameter fitting study using data from [77], as well tions presented are in qualitative agreement with experi-
as a sensitivity analysis, revealed that the recruitment rate mental data. The authors propose continuous injection
of T-cells, along with the turnover rate of the tumor cells, protocols that tend to reduce tumor load more effectively
were the most important prognostic indicators. Thus, the than do intermittent injections. However, based on the
model suggests that treatments should target the T-cell underlying PDE model, tumor load reduction is temporary,
recruitment rate, and that assays that can estimate tumor and the tumor burden is only decreased while treatment is
cell proliferation and death rates could help in guiding ongoing.
patient-specific treatments. In [80], the authors build on the 1994 two-population
The models presented so far in this section assume ODE system of Kuznetsov [34], which tracks CD8þ T
populations of tumor and immune cells that are well-mixed effector cell and tumor cell populations. In [80], the authors

123
J Pharmacokinet Pharmacodyn (2014) 41:461–478 473

add a third population representing helper T cells. Helper T results reflected certain clinical findings [85] in which
cells are called HTCs in that work, but as we have seen in immunotherapy was shown to be most effective when
other works, the CD4þ T cells are referenced as synony- administered in conjunction with chemotherapy. See
mous with the helper T cell population, as was done in Fig. 11.
[67]. The authors of [80] use a simplified version of the In follow up work [86], the authors further refine the six-
Kuznetsov model as proposed by Galach [81], in which the population model, expand the action of IL-2 on other cell
Michaelis-Menten immune recruitment term is replaced populations, and include CD8þ T cell self-regulation. The
with a simpler bilinear term. This simplification makes the authors carefully calibrated all model parameters to data
dynamical systems analysis more tractable. In addition, from [48] to reflect metastatic melanoma and known
following the treatment modeling approach of Kirschner immune levels for two patient data sets. The authors
and Panetta [57], the authors add a constant source term in showed that patient-specific efficacy of immunotherapy
the effector cell population. This term represents the depended on parameters that could actually be determined
incorporation of TIL injections as a treatment to boost anti- by existing assays.
tumor immune activity. Dynamical systems analysis and An important question to ask when exploring combi-
simulation results show that the activity of HTCs could nation treatment approaches is whether there is a way to
affect the long term periodic oscillation behaviors of improve or even optimize treatment scheduling. In [87], the
tumor–immune system interactions. With this model, authors modify the model of [82] slightly to account for
tumor cells can be completely eradicated from the patient’s endogenous as well as exogenous IL-2, and carry out a
body using TIL treatment. detailed mathematical analysis of a theoretical optimal
control strategy for administering the three treatments
introduced in [82]. In a subsequent work [88], the authors
Modeling combination chemo-immunotherapy build upon the mathematical analysis of [87], and deter-
mine conditions under which an optimal solution combin-
Combining chemotherapy with immunotherapy can be ing the three treatments can be found. In this work,
important in at least two ways: protecting the patient from numerical experiments that explore the interactions among
opportunistic infection, and combating the cancer itself. the three treatments are presented. The authors find that for
Cytotoxic treatments make the patient prone to dangerous tumors that may be too small to be detectable, treating
infections since chemotherapy depletes the immune sys- mainly with TIL therapy is most effective at eliminating
tem. For this reason alone, it is desirable to strengthen the the tumor while reducing side effects. However, for larger
immune system after an immune-depleting course of che- tumors, the combined actions of chemotherapy and TIL
motherapy. However, as we have described previously, it is therapy were needed for most effective tumor control.
also desirable to harness the body’s own defenses in the Interestingly, TIL and chemotherapy treatments were
fight against the cancer. Maintaining a strong immune found to be far more effective than IL-2 treatments in the
system, especially during chemotherapy, may be essential scenarios tested, and the optimal strategy rarely included
to successfully controlling tumor growth. IL-2 in the proposed protocol.
In [82], the authors built a model based on the model in Another treatment approach includes targeted therapies
[45] to explore possibilities for therapy protocols that in the form of monoclonal antibodies (mAb), which have
combine cytotoxic chemotherapy with immune therapy. been used in a variety of immunotherapeutic approaches
This model tracks six populations: Tumor cells, NK cells, for treating different cancers. Monoclonal antibodies are
CD8þ T lymphocytes, circulating lymphocytes, cytotoxic able to target cancer cells in at least two ways: mAbs can
chemotherapy concentration, and exogenous IL-2 concen- target tumor cells directly (by blocking a growth signal to
tration. Recall that IL-2 is a naturally occurring cytokine the cancer cell by binding to the growth factor receptors),
that has been evaluated as a cancer immunotherapy [58]. or indirectly (by making tumor cells much more suscepti-
With this model, the authors were able to explore combi- ble to destruction by both NK cells and chemotherapy). For
nation therapy protocols. Immune therapy consisted of example, in about 20–30 percent of women with metastatic
boosting both the CD8þ T cells [83] and injecting IL-2 [84]. breast cancer, the cancer cells have mutated so that there is
Chemotherapy was represented by a systemic, cytotoxic, an increased expression of HER2, an epidermal growth
non-cell-cycle specific medication, such as some alkylating factor like receptor. Herceptin, an anti-HER2 monoclonal
agents. The authors were able to simulate scenarios in antibody, was created to target the HER2 receptor, and is
which chemotherapy alone or immunotherapy alone was now commercially available to treat HER2 positive breast
not sufficient to eradicate a tumor, but in combination, the cancers [58]. In [89], the authors extend the six-population
treatments successfully eliminated all tumor cells. Their model of [86] to allow an exploration of mAb treatments

123
474 J Pharmacokinet Pharmacodyn (2014) 41:461–478

Fig. 11 Model parameters for these simulations are taken from fitting Immunotherapy. Immune system response to high tumor with the
the model to mouse tumor growth data. Top Left No treatment. administration of immunotherapy from days 7 to 8. Bottom Right
Immune system without intervention where the tumor reaches Combination therapy. Chemotherapy and immunotherapy as previ-
carrying capacity and the mouse ‘‘dies’’. Top Right Chemotherapy. ously described given simultaneously effectively control of the tumor.
The immune system response to high tumor with chemotherapy Figures reproduced from Fig. 6 in [82]
administered for one day in a fourteen day cycle. Bottom Left

for colorectal cancer. Parameter values and simulated shows a comparison between reported response rates and
scenarios were constructed so they could compare the simulated response rates in clinical trials for irinotecan,
effectiveness of two separate mAb treatments, cetuximab cetuximab and panitumumab. These clinical trial outcomes
and panitumumab, either alone, or in combination with the can be found in [90–93].
chemotherapy treatment irinotecan. The model also Hypothetical treatment scenarios that yield a response
allowed for a range of patient-specific parameters to rate in the virtual patient population are presented in
account for individual variations in immune system Fig. 13. In this figure, results of two hypothetical treat-
strength and sensitivity to chemotherapy. Treatment out- ments combining irinotecan and cetuximab are compared
comes were obtained not only for individuals, but also for to a combination treatment scenario that was tested in a
groups of virtual patients represented by a biologically clinical trial [92]. Details of the different dosing schedules
reasonable range of patient-specific parameters. Numerical are outlined in Table 1.
experiments produced results that closely matched pub- In other studies such as in [94], the authors develop a
lished clinical trials data for tested protocols. Figure 12 delay differential equation model of an anti-cancer T cell

123
J Pharmacokinet Pharmacodyn (2014) 41:461–478 475

Fig. 12 Clinical trial simulations compared to reported clinical trial distributions of parameters that reflect individual response strengths to
results for irinotecan monotherapy, cetuximab monotherapy, and chemo and immunotherapies. For irinotecan monotherapies, the
panitumumab monotherapy. Simulation results closely match pub- reduced response seen in our simulations is intended, since the
lished results for both cetuximab and panitumumab monotherapies. patients receiving mAb therapy are often not as responsive as most
Sixty four different patients were simulated by sampling from patients to other treatments. Figure reproduced from [89]

Table 1 Response Rates: Combination Therapies

Treatment Medication & dosing

Standard [92] Irinotecan: 125 mg/m2 once weekly; Cmab:

400 mg/m2 load; 250 mg/m2 weekly
Hypoth. Treatment 3 Irinotecan: 125 mg/m2 once weekly; Cmab:
from [89] 400 mg/m2 load; starting day 4, 250 mg/m2
weekly
Hypoth. Treatment 4 Irinotecan: 350 mg/m2 once every three
from [89] weeks; Cmab: 500 mg/m2 once every two
weeks
A comparison of published and hypothetical dosing schedules. Pmab
panitumumab, Cmab cetuximab. See Fig. 13 for a graphical com-
Fig. 13 Response rates from clinical trial simulations, comparing parison of outcomes
standard treatment to an experimental treatment schedule combining
irinotecan and cetuximab. NR, No Response. PR, Partial Response.
CR, Complete Response. 320 individuals simulated. Dosing details in populations over time. This study simulates different pos-
Table 1. Figure reproduced from [89]
sible vaccination schedules during imatinib treatment and
concludes that combining imatinib treatment with cancer
vaccination could greatly improve treatment outcome. The
response against Chronic Myelogenous Leukemia (CML) same research group in a different study [95] considers a
during Gleevec (imatinib) treatment. This study suggests system of nonlinear delay-differential equations (DDEs)
that imatinib treatment induces a transient anti-cancer T that models the dynamics of the interaction between CML,
cell response as leukemia enters remission and that the imatinib, and the anti-leukemia immune response. The
transient immune response can be augmented and pro- group develops several computational methods that can
longed using properly-timed cancer vaccines. The authors help identify the model parameters that affect the dynamics
utilize experimental data to calibrate the model, which of the anti-leukemia immune response during imatinib
describes the rise and fall of anti-cancer immune treatment. The parameters representing T-cell proliferation

123
476 J Pharmacokinet Pharmacodyn (2014) 41:461–478

after stimulation by the presence of the leukemic cells, as 5. Besse IM, Madsen MT, Bushnell DL, Juweid ME (2009) Mod-
well as the background rate of production of the T-cells, eling combined radiopharmaceutical therapy: a linear optimiza-
tion framework. Technol Cancer Res Treat 8(1):51–60
are found to be the most important in predicting the disease 6. Marqa MF, Mordon S, Betrouni N (2012) Laser interstitial
outcome. These results differ somewhat from the parameter thermotherapy of small breast fibroadenomas: numerical simu-
sensitivity analysis in [96], which considered a model of a lations. Lasers Surg Med 44(10):832–839
solid tumor with a dendritic cell vaccine. In that analysis, 7. Berris T, Mazonakis M, Stratakis J, Tzedakis A, Fasoulaki A,
Damilakis J (2013) Calculation of organ doses from breast cancer
the progression of the disease is most sensitive to the radiotherapy: a Monte Carlo study. J Appl Clin Med Phys
parameter representing the kill rate of the tumor cells by 14(1):133–146
the immune cells. 8. Satti J (2009) The emerging low-dose therapy for advanced
cancers. Dose-Response 7(3):208–220
9. Traina TA, Theodoulou M, Feigin K, Patil S, Tan KL, Edwards C
et al (2008) Phase I study of a novel capecitabine schedule based
Summary and conclusion on the Norton-Simon mathematical model in patients with met-
astatic breast cancer. J Clin Oncol 26(11):1797–1802
In this chapter, we have described a sampling of some of 10. Comen E, Morris PG, Norton L (2012) Translating mathematical
modeling of tumor growth patterns into novel therapeutic
the variety of work being done in mathematical modeling approaches for breast cancer. J Mammary Gland Biol Neoplasia
and simulation of chemotherapy, immunotherapy and 17(3—-4, SI):241–249
mixed chemo-immunotherapies for cancer. 11. Newton PK, Mason J, Bethel K, Bazhenova L, Nieva J, Norton L
Cancer is the leading cause of death worldwide. et al (2013) Spreaders and sponges define metastasis in lung
cancer: a Markov Chain Monte Carlo mathematical model.
Enhancing the immune response by the effective admin- Cancer Res 73(9):2760–2769
istration of immunotherapies is a promising avenue for 12. The annual 2013 Pharmaceutical Industry Profile. http://www.
increasing survival times as well as improving the quality phrma.org/industryprofile2013; 2013. Accessed March 23, 2014
of life for cancer patients. However, the interaction of these 13. Agur Z, Vuk-Pavlovic S (2012) Mathematical modeling in
immunotherapy of cancer: personalizing clinical trials. Mol Ther
therapies with other treatments, such as surgery, radiation 20(1):1
and cytotoxic drugs, is still poorly understood. Further- 14. Eftimie R, Bramson JL, Earn DJ (2011) Interactions between the
more, a patient’s response to treatments varies widely, and immune system and cancer: a brief review of non-spatial math-
it is important to develop predictive tools to determine the ematical models. Bull Math Biol 73:2–32
15. Ledzewicz U, Faraji M, Schaettler H (2012) On Optimal Proto-
best treatment strategy for each patient. Mathematical cols for Combinations of Chemo- and Immunotherapy. Pro-
models can help to unravel the interactions between ceedings of 51st IEEE Conference on Decision and Control, Maui
treatments, understand dose-response profiles, and safely 2012;pp. 7492–7497
test a variety of treatments in silico, as a precursor to 16. d’Onofrio A, Ledzewicz U, Schaettler H (2012) New challenges
for cancer systems biology. Springer Verlag, New York
clinical trials. The ultimate goal of these modeling studies 17. d’Onofrio A (2008) Metamodeling tumor-immune system inter-
is to provide a framework for a new age in dosimetry. action, tumor evasion and immunotherapy. Math Comput Model
In collaboration with pharmaceutical scientists, we hope 47(5):614–637
to use mathematical modeling and simulation in conjunc- 18. d’Onofrio A (2005) A general framework for modeling tumor-
immune system competition and immunotherapy: mathematical
tion with laboratory tests and clinical trials to efficiently analysis and biomedical inferences. Physica D 208(3):220–235
answer two central questions in therapy design: How much 19. Adam JA, Bellomo N (1997) A survey of models for tumor
and how often? immune systems dynamics. Springer, New York
20. Agur Z, Arakelyan L, Merbl Y, Daugulis P, Ginosar Y, Vainstein
V, et al. (2003) Cancer Modeling and Simulation. CRC Press/
Chapman & Hall, ed: Luigi Preziosi; 2003. p. 185–219
21. Bar-Or R (2000) Feedback mechanisms between T helper cells
References and macrophages in the determination of the immune response.
Math Biosci 163(1):35–58
1. Bray F, Jemal A, Grey N, Ferlay J, Forman D (2012) Global 22. Castiglione FF, Piccoli B (2006) Optimal control in a model of
cancer transitions according to the Human Development Index dendritic cell transfection cancer immunotherapy. Bull Math Biol
(20082030): a population-based study. Lancet Oncol 68(2):255–274
13:790–801 23. De Boer RJ, Hogeweg P, Dullens HF, De Weger RA, Den Otter
2. Biemar F, Foti M (2013) Global progress against cancer-chal- W (1985) Macrophage T lymphocyte interactions in the anti-
lenges and opportunities. Cancer Biol Med 10:183–6 tumor immune response: a mathematical model. J Immunol
3. Topalian SL, Weiner GJ, Pardoll DM (2011) Cancer Immuno- 134(4):2748–2758
therapy Comes of Age. J Clin Oncol 29:4828–4836 24. Rihan F, Abdel Rahman D, Lakshmanan S, Alkhajeh A (2014) A
4. Gorelik B, Ziv I, Shohat R, Wick M, Hankins WD, Sidransky D time delay model of tumourimmune system interactions:Global
et al (2008) Efficacy of weekly docetaxel and bevacizumab in dynamics, parameter estimation, sensitivity analysis. Appl Math
mesenchymal chondrosarcoma: a new theranostic method com- Comput 232:606–623
bining xenografted biopsies with a mathematical model. Cancer 25. Kronik N, Kogan Y, Vainstein V, Agur Z (2008) Improving al-
Res 68(21):9033–9040 loreactive CTL immunotherapy for malignant gliomas using a

123
J Pharmacokinet Pharmacodyn (2014) 41:461–478 477

simulation model of their interactive dynamics. Cancer Immunol 45. dePillis LG, Radunskaya AE, Wiseman CL (2005) A validated
Immunother 57(3):425–439 mathematical model of cell-mediated immune response to tumor
26. Kuznetsov V (1997) Basic models of tumor-immune system growth. Cancer Res 65(1):7950–7958
interactions - identification, analysis and predictions. In: Adam J, 46. Lai R, Jackson T (2004) A mathematical model of receptor-
Bellomo N (eds) A survey of models for tumor-immune system mediated apoptosis: dying to know why FasL is a trimer. Math
dynamics. Springer, New York, pp 237–294 Biosci Eng 1(2):325–338
27. Lin A (2004) A model of tumor and lymphocyte interactions. 47. Diefenbach A, Jensen E, Jamieson A, Raulet D (2001) Rae1 and
Discrete Contin Dyn Syst Ser B 4(1):241–266 H60 ligands of the NKG2D receptor stimulate tumor immunity.
28. Mallet DG, de Pillis LG (2006) A cellular automata model of Nature 413:165–171
tumor-immune system interactions. J Theor Biol 239(3):334–350 48. Dudley ME, Wunderlich JR, Robbins PF, Yang JC, Hwu P,
29. Nazari S, Basirzadeh H (2014) Natural Killer or T-lymphocyte Schwartzentruber DJ et al (2002) Cancer regression and auto-
cells: which is the best immune therapeutic agent for cancer? an immunity in patients after Clonal repopulation with antitumor
optimal control approach. Int J Control Autom Syst 12:84–92 lymphocytes. Science 298(5594):850–854
30. Takayanagi T, Ohuchi A (2001) A mathematical analysis of the 49. Castiglione FF, Castiglione V, Agur Z (2003) Cancer Model-
interactions between immunogenic tumor cells and cytotoxic T ling and Simulation. Chapman & Hall/CRC Mathematical and
lymphocytes. Microbiol Immunol 45(10):709–715 Computational Biology, Luigi Preziosi, ed.; 2003. p. 333–366
31. de Vladar HP, Gonzalez JA (2004) Dynamic response of cancer 50. Roeder I, Horn M, Glauche I, Hochhaus A, Mueller M, Loeffler
under the influence of immunological activity and therapy. M (2006) Dynamic modeling of imatinib-treated chronic myeloid
J Theor Biol 227(3):335–348 leukemia: functional insights and clinical implications. Nat Med
32. Wein LM, Wu JT, Kirn DH (2003) Validation and analysis of a 12:11811184
mathematical model of a replication-competent oncolytic virus 51. Michor F, Hughes T, Iwasa Y, Branford S, Shah N, Sawyers C
for cancer treatment: implications for virus design and delivery. et al (2005) Dynamics of chronic myeloid leukemia. Nature
Cancer Res 63(6):1317–1324 435:1267–1270
33. Yueping Dong Rinko Miyazaki YT (2014) Mathematical mod- 52. Kim PS, Lee PP, Levy D (2008) Modeling imatinib-treated
eling on helper T cells in a tumor immune system. Discret Contin chronic myelogenous leukemia: reducing the complexity of
Dyn Syst Ser B 19:55–72 agent-based models. Bull Math Biol 70(3):728–744
34. Kuznetsov V, Makalkin I, Taylor M, Perelson A (1994) Non- 53. Kim PS, Lee PP, Levy D (2008) A PDE model for imatinib-
linear dynamics of immunogenic tumors: Parameter estimation treated chronic myelogenous leukemia. Bull Math Biol
and global bifurcation analysis. Bull Math Biol 56(2):295–321 70(7):1994–2016
35. de Pillis L, Radunskaya AE (2014) Modeling of tumor-immune 54. Paquin D, Kim PS, Lee PP, Levy D (2011) Strategic treatment
dynamics. In: Eladdadi A, Kim P (eds) Mathematical modeling of interruptions during imatinib treatment of chronic myelogenous
tumor-immune dynamics. Springer, New York, pp 67–115 leukemia. Bull Math Biol 73(5):1082–1100
36. Roesch K, Hasenclever D, Scholz M (2014) Modelling Lym- 55. Rosenberg SA (2004) Development of effective immunotherapy
phoma Therapy and Outcome. Bull Math Biol 76(2):401–430 for the treatment of patients with cancer. J Am Coll Surg
37. Pfreundschuh M, Trmper L, Kloess M, Schmits R, Feller AC, 198(5):685
Rudolph C et al (2004) Two-weekly or 3-weekly CHOP che- 56. Wainwright DA, Nigam P, Thaci B, Dey M, Lesniak MS (2012)
motherapy with or without etoposide for the treatment of young Recent developments on immunotherapy for brain cancer. Expert
patients with good-prognosis (normal LDH) aggressive lympho- Opin Emerg Drugs 17:181–201
mas: results of the NHL-B1 trial of the DSHNHL. Blood 57. Kirschner DD, Panetta JC (1998) Modeling immunotherapy of
104(3):626–633 the tumor - immune interaction. J Math Biol 37(3):235–252
38. Pfreundschuh M, Trmper L, Kloess M, Schmits R, Feller AC, 58. Goldsby RA, Kindt TJ, Osborne BA, Kuby J (2003) Immunology,
Rbe C et al (2004) Two-weekly or 3-weekly CHOP chemother- 5th edn. W. H. Freeman, New York
apy with or without etoposide for the treatment of elderly patients 59. Rosenberg S, Yang J, Schwartzentruber D, Hwu P, Marincola F,
with aggressive lymphomas: results of the NHL-B2 trial of the Topalian S et al (1999) Prospective randomized trial of the
DSHNHL. Blood 104(3):634–641 treatment of patients with metastatic melanoma using chemo-
39. Pfreundschuh M, Schubert J, Ziepert M, Schmits R, Mohren M, therapy with cisplatin, dacarbazine, and tamoxifen alone or in
Lengfelder E et al (2008) Six versus eight cycles of bi-weekly combination with interleukin-2 and interferon alfa-2b. J Clin
CHOP-14 with or without rituximab in elderly patients with Oncol 17(3):968–975
aggressive CD20 þ B-cell lymphomas: a randomised controlled 60. Cappuccio A, Elishmereni M, Agur Z (2006) Cancer immu-
trial (RICOVER-60). Lancet Oncol 9:105–116 notherapy by interleukin-21: potential treatment strategies
40. Thomlinson R (1982) Measurement and management of carci- evaluated in a mathematical model. Cancer Res 66(14):
noma of the breast. Clin Radiol 33(5):481–493 7293–7300
41. de Pillis L, Radunskaya A (2001) A mathematical tumor model 61. Bellomo N, Bellouquid A, Delitala M (2004) Mathematical topics
with immune resistance and drug therapy: an optimal control on the modelling complex multicellular systems and tumor
approach. J Theor Med 3:79–100 immune cells competition. Math Models Methods Appl Sci
42. de Pillis L, Radunskaya A (2003) The dynamics of an optimally 14(11):1683–1733
controlled tumor model: a case study. Math Comput Model 62. Bellomo N, Preziosi L (2000) Modelling and mathematical
(Special Issue) 37:1221–1244 problems related to tumor evolution and its interaction with the
43. de Pillis L, Gu W, Fister K, Head T, Maples K, Murugan A et al immune system. Math Comput Model 32(3):413–452
(2007) Chemotherapy for tumors: an analysis of the dynamics 63. Bellomo N, Bellouquid A, DeAngelis E (2003) The modelling of
and a study of quadratic and linear optimal controls. Math Biosci the immune competition by generalized Kinetic (Boltzmann)
209:292–315 MModel: review and research Perspectives. Math Comput Model
44. de Pillis L, Radunskaya A (2012) Best practices in mathematical 37:65–86
modeling. In: Mayeno A, Reisfeld B (eds) Computational toxi- 64. Bellomo N, Bertotti ML, Motta S (2003) Cancer Modelling and
cology, methods in molecular biology, part 2. Springer, New Simulation. Chapman & Hall/CRC Mathematical and Computa-
York, pp 51–74 tional Biology, Luigi Preziosi, ed.; 2003. p. 299–332

123
478 J Pharmacokinet Pharmacodyn (2014) 41:461–478

65. Bellomo N, Forni G (2006) Looking for new paradigms towards a 81. Galach M (2003) Dynamics of the tumor-immune system com-
biological-mathematical theory of complex multicellular systems. petition - the effect of time delay. Int J Appl Math Comput Sci
Math Model Methods Appl Sci 16:1001–1029 13:395–406
66. Antony P, Restifo N (2005) CD4?CD25? T regulatory cells, 82. de Pillis L, Gu W, Radunskaya A (2006) Mixed immunotherapy
immunotherapy of cancer, and interleukin-2. J Immunother and chemotherapy of tumors: modeling applications and biolog-
28(2):120–128 ical interpretations. J Theor Biol 238(4):841–862
67. Radunskaya A, Hook S (2012) Modeling the Kinetics of the 83. Yee C, Thompson JA, Byrd D, Riddell SR, Roche P, Celis E et al
immune response. In: d’Onofrio A, Cerrai P, Gandolfi A (eds) (2002) Adoptive T cell therapy using antigen-specific CD8? T
New challenges for cancer systems biomedicine. Springer-Ver- cell clones for the treatment of patients with metastatic mela-
lag, New York, pp 267–282 noma: In vivo persistence, migration, and antitumor effect of
68. Cheever MA (2011) PROVENGE (Sipuleucel-T) in Prostate transferred T cells. Proc Natl Acad Sci USA 99(25):16168–16173
Cancer: The first FDA-Approved Therapeutic Cancer Vaccine. 84. Dillman RO, DePriest C, McClure SE (2014) High-Dose IL2 in
Clin Cancer Res 17(11):3520–3526 metastatic melanoma: better survival in patients immunized with
69. de Pillis L, Gallegos A, Radunskaya A (2013) A model of den- antigens from autologous tumor cell lines. Cancer Biother Ra-
dritic cell therapy for melanoma. Front Oncol 3(56):1–14 diopharm 29(2):53–57
70. Ludewig BB, Krebs P, Junt T, Metters H, Ford NJ, Anderson RM 85. Machiels J, Reilly R, Emens L, Ercolini A, Lei R, Weintraub D
et al (2004) Determining control parameters for dendritic cell- et al (2001) Cyclophosphamide, doxorubicin, and paclitaxel
cytotoxic T lymphocyte interaction. Eur J Immunol 34(9): enhance the antitumor immune response of granulocyte/macro-
2407–2418. doi:10.1002/eji.200425085 phage-colony stimulating factor-secreting whole-cell vaccines in
71. Lee TH, Cho YH, Lee MG (2007) Larger numbers of immature HER-2/neu tolerized mice. Cancer Res 61(9):3689–3697
dendritic cells augment an anti-tumor effect against established 86. de Pillis L, Fister KR, Gu W, Collins C, Daub M, Gross D et al
murine melanoma cells. Biotechnol Lett 29(3):351–357 (2009) Mathematical model creation for cancer chemo-immu-
72. Preynat-Seauve O, Contassot E, Schuler P, French LE, Huard B notherapy. Comput Mat Methods Med 10(3):165–184
(2007) Melanoma-infiltrating dendritic cells induce protective 87. de Pillis LG, Fister KR, Gu W, Collins C, Daub M, Gross D et al
antitumor responses mediated by T cells. Melanoma Res (2007) Seeking bang-bang solutions of mixed immuno-chemo-
17:169–176 therapy of tumors. Electron J Diff Eqns 171:1–24
73. de Pillis L, Caldwell T, Sarapata E, Williams H (2013) Mathematical 88. de Pillis LG, Fister KR, Gu W, Head T, Maples K, Neal T et al
modeling of the regulatory T cell effects on renal cell carcinoma (2008) Optimal control of mixed immunotherapy and chemo-
treatment. Discret Contin Dyn Syst Ser B 18(4):915–943 therapy of tumors. J Biol Syst 16(1):51–80
74. Meropol N, Barresi G, Fehniger T, Hitt J, Franklin M, Caligiuri 89. de Pillis LG, Radunskaya AE, Savage H (2014) Mathematical
M (1998) Evaluation of natural killer cell expansion and activa- model of colorectal cancer with monoclonal antibody treatments.
tion in vivo with daily subcutaneous low-dose interleukin-2 plus Br J Med Med Res 4(16):3101–3131
periodic intermediate-dose pulsing. Cancer Immunol Immunother 90. Lenz HJ (2007) Cetuximab in the management of colorectal
46:318326 cancer. Biol: Targ Ther 2:77–91
75. Jea Ko (2009) Sunitinib mediates reversal of myeloid-derived 91. Grothey AM (2006) Defining the role of panitumumab in colo-
suppressor cell accumulation in renal cell carcinoma patients. rectal cancer. Commun Oncol 3:10–16
Clin Cancer Res 6:2148–2157 92. Gravalos C, Cassinello J, Garcia-Alfonso P, Jimeno A (2010)
76. Nanda S, dePillis LG, Radunskaya AE (2013) B cell chronic Integration of panitumumab into the treatment of colorectal
lymphocytic leukemia — a model with immune response. Discret cancer. Crit Rev Oncol/Hematol 74(1):16–26
Contin Dyn Syst Ser B 18(4):1053–1076 93. De Vita VJ, Hellman S, Rosenberg S (2000) Cancer: principles
77. Messmer BT, Messmer D, Allen SL, Kolitz JE, Kudalkar P, Cesar and practice of oncology, 7th edn. Lippincott Wiliams & Wilkins,
D et al (2005) In vivo measurements document the dynamic Philadelphia
cellular kinetics of chronic lymphocytic leukemia B cells. J Clin 94. Kim PS, Lee PP, Levy D (2008) Dynamics and potential impact
Invest 115:755–764 of the immune response to chronic myelogenous leukemia. PLoS
78. Ramos RA, Zapata J, Condat CA, Deisboeck TS (2013) Modeling Comput Biol 4(6):e1000095
cancer immunotherapy: assessing the effects of lymphocytes on 95. Peet MM, Kim PS, Niculescu SI, Levy D (2009) New compu-
cancer cell growth and motility. Physica A 392:2415–2425 tational tools for modeling chronic myelogenous leukemia. Math
79. Kang-Ling Liao Xue-Feng Bai AF (2014) Mathematical model- Model Nat Phenom 4:119–139
ing of Interleukin-27 induction of anti-tumor T cells response. 96. Radunskaya A, de Pillis L, Gallegos A (2013) A model of den-
PLoS ONE 9(3):e91844 dritic cell therapy for melanoma. Front Oncol 3:56
80. Dong Y, Miyazaki R, Takeuchi Y (2014) Mathematical modeling
on helper T cells in a tumor immune system. Discret Contin Dyn
Syst Ser B 19:55–72

123