2974 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 58, NO.
10, OCTOBER 2011
Multiscale Modeling of Intracranial Aneurysms:
Cell Signaling, Hemodynamics, and Remodeling
Harvey Ho*, Vinod Suresh, Wendy Kang, Michael T. Cooling, Paul N. Watton, and Peter J. Hunter
Abstract—The genesis, growth, and rupture of intracranial
aneurysms (IAs) involve physics at the molecular, cellular, blood
vessel, and organ levels that occur over time scales ranging from
seconds to years. Comprehensive mathematical modeling of IAs,
therefore, requires the description and integration of events across
length and time scales that span many orders of magnitude. In
this letter, we outline a strategy for mulstiscale modeling of IAs
that involves the construction of individual models at each rele-
vant scale and their subsequent combination into an integrative
model that captures the overall complexity of IA development. An
example of the approach is provided using three models operat-
ing at different length and time scales: 1) shear stress induced
nitric oxide production; 2) smooth muscle cell apoptosis; and
3) fluid-structure-growth modeling. A computational framework
for combining them is presented. We conclude with a discussion of
the advantages and challenges of the approach.
Index Terms—Apoptosis, biomechanics, cerebral aneurysms,
modeling, multiscale, signaling pathways.
I. INTRODUCTION
NTRACRANIAL aneurysms (IAs) are abnormal focal di-
I lations of cerebral arteries that are prevalent in 1%–5% of
population and rupture in about 1/10 000 [1]. Most IAs are
asymptomatic and cause no problems in a person’s lifetime.
Some are revealed in emergency scanning after rupture, when it
is often too late for their treatment: the morbidity and mortality
rate for aneurysm rupture is about 50%.
With the increased use of MRI and CT scanning more unrup-
tured IAs are found incidentally while scanning for unrelated
conditions. When aneurysms are detected treatment decisions
require an assessment of the likelihood of continued growth and
Manuscript received April 1, 2011; revised June 13, 2011; accepted June 17,
2011. Date of publication June 27, 2011; date of current version September
21, 2011. This work was partly funded by the @neurIST, the EC Framework 6
project for multidisciplinary aneurysm research. The work of P. N. Watton was
supported by The Centre of Excellence in Personalized Healthcare (funded by
the Wellcome Trust and EPSRC, under Grant WT 088877/Z/09/Z). The work
of W. Kang was supported from the Maurice Wilkins Centre for Molecular
Biodiscovery. Asterisk indicates corresponding author.
*H. Ho is with the Auckland Bioengineering Institute, The University of
Auckland, Auckland 1142, New Zealand (e-mail: harvey.ho@auckland.ac.nz).
V. Suresh is with the Auckland Bioengineering Institute and the Department
of Engineering Science, The University of Auckland, Auckland 1142, New
Zealand (e-mail: v.suresh@auckland.ac.nz).
W. Kang, M. T. Cooling, and P. J. Hunter are with the Auckland Bio-
engineering Institute, The University of Auckland, Auckland 1142, New
Zealand (e-mail: wkan014@aucklanduni.ac.nz; m.cooling@auckland.ac.nz;
p.hunter@auckland.ac.nz).
P. N. Watton is with the Institute of Biomedical Engineering, Department of
Engineering Science, University of Oxford, Oxford OX1 2JD, U.K. (e-mail:
Paul.Watton@eng.ox.ac.uk).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TBME.2011.2160638
0018-9294/$26.00 © 2011 IEEE
possible rupture of the aneurysm. In practice, such assessments
are largely subjective and based on clinicians’ experience with
the location, size, and shape of the aneurysm. However, the fact
that even small aneurysms can rupture and large aneurysms can
remain dormant demands more comprehensive and accurate cri-
teria for rupture risk assessment. The development of predictive
mathematical models of IA genesis will be useful in developing
a database of “virtual aneurysms” against which patient records
are compared in order to guide treatment decisions. A prereq-
uisite for the creation of such models is the identification of
various factors involved in IA development.
The genesis, growth, and rupture of IAs are associated with
multiple factors including the wall shear stress (WSS) acting on
the vessel wall [2], weakening of the wall due to loss of smooth
muscle cells (SMC) in the medial layer [3], and vascular tone
regulation via the signaling of endothelial cells (ECs) [4]. Math-
ematical models of each of these processes have been developed,
e.g., for hemodynamics [5], apoptosis [6], and EC signaling [7].
In many cases these models have been developed in isolation and
in clinical contexts unrelated to IAs. While they can be used as
building blocks in the construction of an integrated mathemat-
ical model of IA development and growth, certain challenges
must be overcome. First, the various physiological factors oper-
ate across multiple spatial and temporal scales. For example, the
pulsatile WSS varies on the scale of the heartbeat, i.e., seconds;
death of individual SMCs can occur a few hours after stimulus;
and aneurysm growth as a whole occurs over a much longer
time scale ranging from months to years. Similarly, spatial vari-
ation of WSS occurs over the length scale of the aneurysm,
i.e., 10−3 m, while intracellular signaling pathways operate on
scales of 10−6 m. In addition, these mechanisms are strongly
associated with geometric features of IAs such as the aspect
ratio that varies with different patients. Second, multiple levels
of feedback may exist between the different factors and such
feedback mechanisms must be explicitly considered when com-
bining simpler models. Finally, each model component and the
overall IA model need to be validated using experimental and
clinical data specific to IAs. In this letter, we discuss concepts
and strategies to deal with the issues of multiple scales and
feedback in the modeling of IAs. We exemplify our approach
with three models that operate over different spatial and tempo-
ral scales, and propose a computational cycle that binds these
models.
II. MODELING HIERARCHY AND TOOLS
A “systems biology” approach to medical problems involves
the study of the system (e.g., an organ) as a whole by considering
the different levels of biological organization (e.g., tissue, cell,
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 58, NO. 10, OCTOBER 2011
Fig. 1. (a) Patient specific cerebral arterial tree. (b) Circle of Willis. (c) Flow
simulation for an aneurysm arises from the anterior communicating artery
(ACoA). Note that ACoA aneurysms account for 30% of all IAs [1].
and organelle) and the complex interactions between them [8].
Here, we propose a systems biology approach for IAs based on
a hierarchy of models that encompass a range of spatial and
temporal scales.
1) Population: Population level differences in the incidence
of IAs have been observed and linked to lifestyle (diet and smok-
ing), genetic, and racial factors. For example, a 5-year clinical
study found that the incidence per 100 000 of the population for
all aneurysms was 14.3 for European New Zealanders and 25.7
for Maori [9]. Maori were also found to have an abnormally high
incidence of middle cerebral artery IAs. Population level mod-
els are empirical and may be used to identify key environmental
factors that influence the growth and rupture of IAs. The bio-
physical mechanisms by which environmental factors act can
be understood using the mechanistic models are described as
follows.
2) Patient: The analysis of an IA cannot be separated from
its hosting environment i.e., the brain and the cardiovascular
system of the patient. In particular, most IAs are found at or
in the vicinity of the Circle of Willis (CoW). At this level in
the modeling hierarchy patient-specific models of the arterial
tree are created from medical imaging data (see Fig. 1). From a
hemodynamic modeling perspective, local flow in an aneurysm
can be simulated more accurately by placing it in a larger vas-
cular context such as the CoW, the cerebral arterial tree, or even
the whole body arterial tree [5].
3) Aneurysm: Models at this level are the most widely used
and involve retrieving shape and morphological features from
medical images to construct an anatomically accurate compu-
tational domain of a single aneurysm. Hemodynamics and wall
mechanics simulations are used to evaluate the force and pres-
sure acting on the aneurysmal wall [5], [10]. Since the shape
and location of IAs have been the primary criteria for assessing
the risk of rupture and making treatment decisions, these mod-
els are of immediate interest from a clinical perspective. These
models can be coupled with models that describe remodeling at
the tissue and cellular levels in order to follow the evolution of
aneurysm morphology over time.
4) Tissue: Morphological changes occur at the tissue level
due to aneurysm remodeling, and vessel walls with thin, degen-
erated tissue with hyaline deposits are more likely to rupture
than thick intima-like walls [11]. The changes in tissue compo-
sition and organization are modeled via mixture theory [12] to
alter the constitutive law describing the mechanical properties
of the vessel wall. The changed mechanical state of the wall
2975
Fig. 2. Model of evolving IA and WSS distribution on the right anterior
cerebral artery.
affects models both upstream and downstream in the modeling
hierarchy: in aneurysm models it affects wall deformation and
hemodynamics, which in turn influence stress distributions that
serve as stimuli for cell signaling pathways.
5) Cell/Subcellular Structures: The development and growth
of aneurysms involve degeneration of the cell layers and ma-
trix proteins that comprise the arterial wall. Factors that induce
apoptosis and matrix degradation include hemodynamic stim-
uli such as altered WSS and pressure [2] and inflammatory
mediators [13]. Progression from stimulus to apoptosis/matrix
degradation occurs via a complex network molecular signaling
pathways. The extracellular stimuli are communicated to the
intracellular space via ligand-receptor binding and/or the acti-
vation of stress sensitive ion channels. This triggers a cascade of
reactions involving second messenger molecules, transcription
factors, and regulatory molecules that leads to the activation of
apoptosis effector and matrix degrading proteins [14]. Different
cell organelles are involved in this process. The mitochondria
play a key role in apoptotic pathways while the endoplasmic
reticulum is involved in the homeostasis of calcium ions that are
important in the regulation of both apoptotic and matrix degra-
dation pathways. Translocation to the nucleus of transcription
factors and certain proteins is also a key step in inducing apop-
tosis. Thus, a detailed description of the microscopic processes
underlying IA formation must account for the dynamics and
spatial localization within the cell of these signaling pathways.
Models that operate at different levels of biological organiza-
tion may be developed by different research groups often using
different model creation tools. Encoding models using open
standards facilitates model exchange, reuse, and extension [8].
Two such open standards that have been developed are CellML
which is designed for 0-D biophysical models [15] and FieldML
for n-dimensional spatial fields [8].
III. MODELING EXAMPLES
A. Fluid-Structure-Growth Model
The composition and structure of the arterial tissue is continu-
ally maintained by vascular cells. The functionality of the cells,
and thus, the growth and remodeling of the tissue, is intimately
linked to the local mechanical environment. Consequently, to
simulate IA evolution, it is important to quantify the spatial
distributions of mechanical stimuli that act on the cells. Re-
cent models of IA evolution incorporate such mechanobiology
in a phenomenological manner [10]. Fig. 2 illustrates a simu-
lation of an evolving aneurysm on the the right anterior cere-
bral artery (the colormap highlights the WSS distribution). The
model accounts for the distinct natural reference configurations
2976
Fig. 3. Overview schematic of the signaling pathways activated on shear stress
via increased calcium influx. Calcium binds to calmodulin (CaM) which in turn
influences the L-Arginine binding reactions of eNOS.
of the load bearing constituents of the arterial wall (collagen and
elastin) and can simulates how they adapt over time [16]. Hemo-
dynamic and structural analyses quantify the evolving mechani-
cal stimuli and growth and remodeling algorithms (phenomeno-
logically) simulate cells responding to mechanical stimuli and
adapting the tissue. In the example illustrated, a localized degra-
dation of elastin is prescribed to create a small outpouching of
the computational domain to perturb the hemodynamic envi-
ronment [17]. Subsequent elastin degradation is then linked to
low WSS in this localized region [10], while the collagen fabric
remodels to restore its strain to the homeostatic value and col-
lagen growth is linked to the magnitude of the cyclic stretch of
the fibroblast cells.
In a fully integrative model of IA evolution, this model would
be coupled to explicit representations of cellular level models:
the computed WSS would serve as an input to signaling path-
ways (see later), and production and removal of constituents
would be derived from the cellular models.
B. Shear Stress-Induced Signaling Pathways
WSS is one of the major mechanical stimuli for nitric oxide
(NO) production in ECs for vascular tone regulation. Various
models have been proposed, e.g., on the effect of WSS on ion
channel flux in ECs [18], and on the biochemical reactions
of endothelial NO synthase (eNOS) activation and subsequent
NO production [7]. A number of different signaling pathways
modulate eNOS activation and act over timescales ranging from
seconds to hours. For example, calcium signaling on WSS in the
short time scale is known to be an important modulator of eNOS
activation. A schematic of the signaling pathways activated on
shear stress via increased calcium influx is shown in Fig. 3.
Reactions are modeled as ordinary differential equations rep-
resenting the change in molecular species concentrations follow-
ing mass-action kinetics. Additional components representing
calmodulin/calcium binding according to a simple Hill curve are
added to model calcium sensitivity of catalysis reactions. These
reactions and models are modularized and coded in CellML to
be incorporated into a library of EC signaling components [19].
Ongoing work includes mapping the quantitative WSS data
calculated from Section III-A into the CellML models, and cou-
pling NO generation models to SMC apoptosis models, as de-
scribed next.
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 58, NO. 10, OCTOBER 2011
Fig. 4. Intrinsic apoptosis mediated by the release of cytochrome-c (cyt-c)
from the mitochondria (mito) leading to the activation of caspase-3 (Casp3) and
its modulation by NO [6]. Inputs to the model are the initial concentration of
NO and caspase-8 (Casp8). The module describing the NO pathway is enclosed
within dashed lines and is dynamically combined with the module describing
apoptosis to create combined model.
C. SMC Apoptosis
Loss of SMCs in the arterial wall due to apoptosis is a
key step in IA development [3]. An existing model of intrin-
sic (mitochondria-mediated) apoptosis and its modulation by
NO [6] was implemented in CellML (see Fig. 4). Apoptosis and
NO pathways were coded as independent modules using the
CellML 1.1 framework to facilitate model verification, reuse,
and extension. Equations bridging the NO and apoptosis path-
ways were specified and the overall model depicted in Fig. 4
was dynamically composed during runtime using the “imports”
feature of CellML 1.1 that enables values of variables to be
passed between models. In the same way, a model linking WSS
with apoptosis can be composed by combining the models of
Figs. 3 and 4 that share a single common component, NO. Loss
of SMCs changes the composition, thickness, and mechanical
properties of the aneurysm wall and, thus, changes the parame-
ter values in the fluid-structure-growth model of Section III-A.
IV. MULTISCALE MODELING CYCLE
In the previous section, we explicitly specified the link be-
tween three models of different scales or cell types, i.e., WSS
derived from hemodynamic analysis at the aneurysmal level
could be used in NO generation models in ECs, which could
then be coupled with apoptosis models of SMCs. Loss of SMCs
leads to a weakened wall and changes lumen diameters and,
thus, affects WSS and wall remodeling. This modeling cycle
encompasses multiple spatial and temporal scales at different
levels of biological organization, and its overall methodology is
illustrated in Fig. 5.
The cycle begins with a structural analysis (bottom box) to
quantify the loaded geometry, stress, stretch, and cyclic defor-
mation of the tissue. The structural analysis is then coupled to
a hemodynamic analysis via a fluid-structure interaction algo-
rithm [20]. The flow solutions give the WSS, WSS gradients
and an oscillatory shear index that provides the mechanical
environment of vascular cells. The resulting changes in tissue
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 58, NO. 10, OCTOBER 2011
Fig. 5. Multiscale modeling framework for IA evolution combining structural,
hemodynamic, and signaling pathway analysis.
composition and organization are modeled by subcellular sig-
naling pathways. The changed mechanical state of the wall is
then computed by solving the finite deformation elasticity equa-
tions and the new geometry of the aneurysm is used to repeat
the cycle, as the composition and structure of the tissue adapts
an aneurysm evolves.
V. DISCUSSION
In this letter, we proposed a modeling hierarchy for IAs based
on integrating individual biophysical models at different scales.
The rationale of this “divide-and-conquer” approach is that, al-
though the mechanisms underlying IA evolution are very com-
plex, it is feasible to model a limited set of key mechanisms at
each spatial or temporal scale and integrate these models to gain
a global picture. An important aspect not discussed here is the
experimental validation of the models. Carefully designed labo-
ratory experiments are necessary to determine model parameters
and verify model predictions. If relatively simple models lim-
ited to specific mechanisms are used at each scale, rigourous
validation is possible. For the integrated model describing IA
evolution, animal models of experimentally induced IAs will be
required for validation.
Finally, the mathematical framework of Fig. 5 may be viewed
as a subset of the modeling hierarchy described in Section II.
Extension of this framework is possible both vertically (i.e.,
larger/smaller scales) and horizontally (i.e., new biophysical
factors at the same scales). An example of vertical extension
would be the generation of a population-specific database of
“virtual aneurysms” that can be compared to and complement
the “physical aneurysm” database derived from patients in the
@neurIST project [21]. An example of horizontal extension
would be the modeling of the effect of inflammatory mediators
on apoptosis [13].
REFERENCES
[1] J. L. Brisman, J. K. Song, and D. W. Newell, “Cerebral aneurysms,” New
England J. Med., vol. 355, no. 9, pp. 928–939, 2006.
[2] D. M. Sforza, C. M. Putman, and J. R. Cebral, “Hemodynamics of cerebral
aneurysms,” Annu. Rev. Fluid Mech., vol. 41, no. 1, pp. 91–107, 2009.
2977
[3] S. Kondo, N. Hashimoto, H. Kikuchi, F. Hazama, I. Nagata, H. Kataoka,
and W. I. Rosenblum, “Apoptosis of medial smooth muscle cells in the
development of saccular cerebral aneurysms in rats,” Stroke, vol. 29,
no. 1, pp. 181–189, 1998.
[4] S. Fukuda, N. Hashimoto, H. Naritomi, I. Nagata, K. Nozaki, S. Kondo,
M. Kurino, and H. Kikuchi, “Prevention of rat cerebral aneurysm for-
mation by inhibition of nitric oxide synthase,” Circulation, vol. 101,
pp. 2532–2538, 2000.
[5] H. Ho, G. Sands, H. Schmid, K. Mithraratne, G. Mallinson, and P. Hunter,
“A hybrid 1D and 3D approach to hemodynamics modelling for a patient-
specific cerebral vasculature and aneurysm,” in Proc. Med. Image Comput.
Comput.-Assist. Intervent., 2009, pp. 323–330.
[6] E. Z. Bagci, Y. Vodovotz, T. R. Billiar, B. Ermentrout, and I. Bahar, “Com-
putational insights on the competing effects of nitric oxide in regulating
apoptosis,” PLoS ONE, vol. 3, no. 5, p. e2249, 2008.
[7] K. Chen and A. S. Popel, “Theoretical analysis of biochemical pathways
of nitric oxide release from vascular endothelial cells,” Free Rad. Biol.
Med., vol. 41, no. 4, pp. 668–680, 2006.
[8] A. S. Popel and P. J. Hunter, “Systems biology and physiome projects,”
Wiley Interdisciplinary Rev.: Syst. Biol. Med., vol. 1, no. 2, pp. 153–158,
2009.
[9] P. V. Marks, J. K. A. Hope, A. D. Cluroe, and C. E. Furneaux, “Racial
differences between Maori and European New Zealanders in aneurysmal
subarachnoid haemorrhage,” Br. J. Neurosurg., vol. 7, no. 2, pp. 175–181,
2009.
[10] P. N. Watton, A. Selimovic, N. B. Raberger, P. Huang, G. A. Holzapfel, and
Y. Ventikos, “Modelling evolution and the evolving mechanical environ-
ment of saccular cerebral aneurysms,” Biomech. Model. Mechanobiol.,
vol. 10, no. 1, pp. 109–132, 2010.
[11] J. Frosen, A. Piippo, A. Paetau, M. Kangasniemi, M. Niemela, J.
Hernesniemi, and J. Jaaskelainen, “Remodeling of saccular cerebral artery
aneurysm wall is associated with rupture: Histological analysis of 24 un-
ruptured and 42 ruptured cases,” Stroke, vol. 35, no. 10, pp. 2287–2293,
2004.
[12] J. D. Humphrey and K. R. Rajagopal, “A constrained mixture model for
growth and remodeling of soft tissues,” Math. Models Methods Appl. Sci.,
vol. 12, no. 3, pp. 407–430, 2002.
[13] T. Jayaraman, A. Paget, Y. S. Shin, X. Li, J. Mayer, H. W. Chaudhry,
Y. Niimi, M. Silane, and A. Berenstein, “TNF-a-mediated inflammation
in cerebral aneurysms: A potential link to growth and rupture,” Vascular
Health Risk Manage., vol. 4, no. 4, pp. 805–817, 2008.
[14] S. Orrenius, B. Zhivotovsky, and P. Nicotera, “Regulation of cell death:
The calcium-apoptosis link,” Nat. Rev. Mol. Cell Biol., vol. 4, no. 7,
pp. 552–565, 2003.
[15] A. A. Cuellar, C. M. Lloyd, P. F. Nielsen, D. P. Bullivant, D. P.
Nickerson, and P. J. Hunter, “An overview of CellML 1.1, a biological
model description language,” Simulation, vol. 79, no. 12, pp. 740–747,
2003.
[16] P. N. Watton, N. A. Hill, and M. Heil, “A mathematical model for
the growth of the abdominal aortic aneurysm,” Biomech. Model.
Mechanobiol., vol. 3, no. 2, pp. 98–113, 2004.
[17] P. N. Watton, N. B. Raberger, G. A. Holzapfel, and Y. Ventikos, “Coupling
the hemodynamic environment to the evolution of cerebral aneurysms:
Computational framework and numerical examples,” J. Biomech. Eng.,
vol. 131, no. 10, pp. 101 003-1–101 003-14, 2009.
[18] A. Comerford, M. J. Plank, and T. David, “Endothelial nitric oxide syn-
thase and calcium production in arterial geometries: An integrated fluid
Mechanics/Cell model,” J. Biomech. Eng., vol. 130, no. 1, pp. 011 010-
1–011 010-13, 2008.
[19] H. Ho, M. Cooling, and P. Hunter, “Towards a multiscale integrative model
of WSS-induced signaling pathways in cerebral aneurysms,” IFMBE
Proc., vol. 31, pp. 1159–1162, 2010.
[20] D. A. Nordsletten, P. J. Hunter, and N. P. Smith, “Conservative and non-
conservative arbitrary LagrangianEulerian forms for ventricular flows,”
Int. J. Numer. Meth. Fluids, vol. 56, pp. 1457–1463, 2007.
[21] S. Benkner, A. Arbona, G. Berti, A. Chiarini, R. Dunlop, G.
Engelbrecht, A. Frangi, C. Friedrich, S. Hanser, P. Hasselmeyer, R. Hose,
J. Iavindrasana, M. Kohler, L. Iacono, G. Lonsdale, R. Meyer, B. Moore,
H. Rajasekaran, P. Summers, A. Wohrer, and S. Wood, “@neurIST: In-
frastructure for advanced disease management through integration of het-
erogeneous data, computing, and complex processing services,” IEEE
Trans. Inf. Technol. Biomed., vol. 14, no. 6, pp. 1365–1377, Nov. 2010.