Abstract
Simple Summary
The recent success of immunotherapy treatments against cancer relies on
helping our own body’s defenses in the fight against tumours, namely
reinvigorating the cancer killing action of T cells. Unfortunately, in
a large proportion of patients these therapies are ineffective, in part
due to the presence of other immune cells, macrophages, which are
mis-educated by the cancer cells into promoting tumour growth. Here we
start from an existing model of macrophage polarization and extend it
to the specific conditions encountered inside a tumour by adding
signals, receptors, transcription factors and cytokines that are known
to be the key components in establishing the cancer cell-macrophage
interaction. Then we use a mathematical Boolean model applied to a gene
regulatory network of this biological process to simulate its temporal
behaviour and explore scenarios that have not been experimentally
tested so far. Additionally, the KO and overexpression simulations
successfully reproduce the known experimental results while predicting
the potential role of regulators (such as STAT1 and EGF) in preventing
the formation of pro-tumoural macrophages, which can be tested
experimentally.
Abstract
The tumour microenvironment is the surrounding of a tumour, including
blood vessels, fibroblasts, signaling molecules, the extracellular
matrix and immune cells, especially neutrophils and monocyte-derived
macrophages. In a tumour setting, macrophages encompass a spectrum
between a tumour-suppressive (M1) or tumour-promoting (M2) state. The
biology of macrophages found in tumours (Tumour Associated Macrophages)
remains unclear, but understanding their impact on tumour progression
is highly important. In this paper, we perform a comprehensive analysis
of a macrophage polarization network, following two lines of enquiry:
(i) we reconstruct the macrophage polarization network based on
literature, extending it to include important stimuli in a tumour
setting, and (ii) we build a dynamical model able to reproduce
macrophage polarization in the presence of different stimuli, including
the contact with cancer cells. Our simulations recapitulate the
documented macrophage phenotypes and their dependencies on specific
receptors and transcription factors, while also unravelling the
formation of a special type of tumour associated macrophages in an in
vitro model of chronic lymphocytic leukaemia. This model constitutes
the first step towards elucidating the cross-talk between immune and
cancer cells inside tumours, with the ultimate goal of identifying new
therapeutic targets that could control the formation of tumour
associated macrophages in patients.
Keywords: Boolean model, tumour associated macrophage, macrophage
polarization, nurse-like cells, chronic lymphocytic leukaemia
1. Introduction
As all living cells, macrophages perceive and respond to intra- and
extracellular signals in order to maintain their functions (endocytic,
phagocytic and secretory, for example) by displaying a wide spectrum of
specific phenotypes (polarizations) in different inducer environments.
Based on their activity and the expression of specific proteins,
markers and chemokines, two major subsets of macrophages have been
identified, namely classically activated macrophages (M1) exhibiting a
pro-inflammatory response, and alternatively activated macrophages (M2,
themselves subdivided into 4 subclasses: M2a, M2b, M2c, M2d
[[40]1,[41]2,[42]3]) exhibiting an anti-inflammatory response.
Additionally, multiple studies support the idea that M1 and M2
macrophages represent, in fact, the extremes of a continuous
polarization spectrum of cells deriving from the differentiation of
monocytes [[43]4]. Their plastic gene expression profile is determined
by the type, concentration and duration of exposure to the polarization
stimuli in an inflammatory environment [[44]3,[45]5,[46]6,[47]7,[48]8].
Macrophages are also found inside tumors, as part of the tumor
micro-environment (TME), a complex collection of cells that are found
surrounding cancer cells including also other immune cells, such as
lymphocytes and neutrophils, as well as other normal cells. In many
tumors, infiltrated macrophages display mostly an M2-like phenotype,
which provides an immunosuppressive microenvironment. In cancer, these
tumor associated macrophages (TAMs) secrete several cytokines,
chemokines and proteins which promote tumor angiogenesis, growth and
metastasis [[49]9,[50]10,[51]11,[52]12]. Interestingly, it was observed
that in established tumors, signals originating from cancer cells can
cause phenotypic shifts in macrophages, leading to alternative
functions that do not correspond to either M1 or M2 phenotypes
[[53]13]. Several studies have demonstrated that TAMs directly suppress
CD8
[MATH: + :MATH]
T cell activation in vitro [[54]14,[55]15,[56]16,[57]17]. Mechanisms
that orchestrate this process, either directly or indirectly, remain
unclear [[58]18] and warrant further exploration due to macrophages’
important impact on tumor progression.
The TME can be defined as an ecological system given the presence of
diverse types of cells that interact in specific ways with each other.
To name a few, cytotoxic T lymphocytes can attack cancer cells and kill
them, while macrophages can phagocyte apoptotic or dead cells. However,
complex feedback relations exist between the signals produced by some
cell types and the phenotypic transitions that are induced by these
signals in other cells. For example, due to the high density of
leukemic B cells, monocytes residing in lymph nodes from CLL patients
tend to differentiate into a pro-tumoral state, which is able in turn
to promote survival of cancer cells through the secretion of
anti-apoptotic signals, rather than eliminating them. These complex
interrelations between cells are typical of predator-prey systems in
ecology and can be modelled using similar approaches in which the
possible final states (attractors) of the system define which
populations will dominate [[59]19].
In any given environment, the cellular processes that determine a
cell’s phenotype consist in a cascade of interactions, which can be
represented as a regulatory network, in which nodes represent proteins,
enzymes, chemokines, etc., while the connections represent the type
(activation or inhibition) and direction of interactions of different
types (transcriptional and post-translational activations). Network
modelling has found numerous applications in studying the structure and
dynamic behaviour of different biological systems in response to
environmental stimuli and internal perturbations
[[60]20,[61]21,[62]22,[63]23]. Several computational models of
different pathways involved in the inflammatory immune response have
been previously published, such as: continuous, logical and multi-scale
models of T cell differentiation [[64]24,[65]25,[66]26], logical models
of macrophage differentiation in pro- and anti-inflammatory conditions
[[67]27], multi-scale models of innate immune response in tumoral
conditions [[68]28], etc.
Macrophages are extremely plastic cell types, whose phenotypes can
easily switch depending on conditions. Since the specific macrophage
state that protects cancer cells from undergoing apoptosis is
fundamental in the development of resistance to treatments in CLL and
solid tumors, we turned to study it as a polarization state. An
important computational model of macrophage polarization was able to
detect 3 different M2 subgroups of macrophages, as a result of various
combinations of pro- and anti-inflammatory extra-cellular signals
[[69]27], using exclusively literature-based knowledge of the
intra-cellular regulatory interactions and pathways involved in the
polarization process. In a more recent work, Ramirez et al. [[70]29]
used temporal expression profiles of in vitro macrophage cytokines to
infer logical models of macrophage polarization (M1 and 3 subcategories
of M2: M2a, M2b and M2c) in the presence of different stimuli.
Nevertheless, many important questions remain to be explored regarding
the polarization states, especially in a tumor setting. More
specifically, it is important to identify the pathways involved in TAM
formation and to understand to what extent the macrophage plasticity
facilitates this process inside a tumor. On the other hand, despite the
wealth of quantitative information from bulk and single-cell sequencing
datasets, the inference of regulatory networks based on experimental
data remains a difficult challenge, with most approaches proposing a
combination of both literature- and data-driven methods
[[71]29,[72]30,[73]31,[74]32].
In Chronic Lymphocytic Leukemia (CLL), a B-cell malignancy in which
patients accumulate large quantities of malignant CLL cells in their
lymph nodes, an interesting ecology of cancer cells and immune cells is
established. CLL cells are able to educate surrounding monocytes,
through direct contact and cytokine signals, turning them into TAMs,
which in this disease are referred to as Nurse Like Cells (NLCs)
[[75]33]. NLCs are derived from CD14
[MATH: + :MATH]
monocytes and are characterised by a distinct set of antigens (CD14lo,
CD68hi, CD11b, CD163hi) [[76]34,[77]35]. Moreover, NLCs express
stromal-derived-factor-1alpha, a chemokine which promotes chemotaxis
and activates mitogen activated protein kinases, ultimately leading to
more aggressive cancers and better survival of these cells in vitro.
Through direct contact, the NLCs are able to protect the cancer CLL
cells from apoptotic signals, and stimulate environment mediated drug
resistance. Interactions between NLCs and CLL cells appear to be
mediated by the B cell receptor, which, when stimulated, activates
production of CCL3/4, initiating the recruitment of other cells,
including CD4
[MATH: + :MATH]
T cells and more NLCs. Another pathway that has been associated with
NLCs and TAMs more in general is that of CSF-1 (MCSF). Patients with
high expression of this factor usually show faster CLL progression and
this gene was implicated in the production of NLCs. Also the more M1-
or M2-like profile of NLCs in specific patients correlates with active
and controlled disease, respectively. Analyses of the transcriptomic
profile of NLCs suggest their high similarity to the macrophage M2
profile described in solid tumors, which makes studying the formation
of NLCs all the more relevant in the quest of controlling TAMs in other
malignancies.
The main characteristics of these 3 types of macrophages are given in
[78]Table 1. Considering the close phenotypic similarity between M2 and
NLC/TAM macrophages, we consider that the NLC/TAM components should
include the M2 ones. Here we indicate in blue the components that have
been used in our model and in bold the ones that were taken as
signature components for each phenotype. A schematic diagram of M1 and
M2 (with 4 subcategories) macrophages can be found in [[79]2], whereas
a short description of the profiles for the main macrophage phenotypes
is given in [80]Appendix A. More detailed explanations of the
mechanisms, pathways and components involved in the polarization
process can be found in the cited papers and the references therein.