Keywords
cerebral aneurysms - thrombosis - in vitro - thrombin–fibrinogen
Introduction
Thrombosis and thrombi formation are closely associated with cerebral aneurysms, which
are balloon-like structures on vessels of the brain resulting from a weakening of
the vessel wall layers.[1]
[2]
[3] Aneurysm risk can be assessed through image-based screening on a population basis,
of high-risk populations, clinical populations, or registries of patients. Thrombi
have been observed both before and after intervention, leading to a wide variability
of outcomes in patients with cerebral aneurysms. The main conundrum is that thrombi
can be beneficial or destructive in unruptured cerebral aneurysms, depending on the
type of clot formed.[4]
[5]
[6]
[7]
[8]
[9]
[10] The current hypothesis is that clots which partially occlude the aneurysm sac tend
to lead to further vascular wall degradation, with increased risk of rupture, while
those that fill the aneurysm sac contribute to stability.[3]
In an attempt to harness the benefits of constructive clotting, endovascular treatment
of cerebral aneurysms was designed to aid complete occlusion of the aneurysm sac.
This was achieved by placement of a high surface area device in the aneurysm sac (through
endovascular coiling) or by redirecting flow to the parent vessel and out of the aneurysm
sac (using flow diversion devices),[11]
[12] always with the aim to encourage development of a stable thrombus that completely
occludes the aneurysm. In the case of the latter, occlusion is not always immediate
and may be observed over a period of approximately 12 months. The type of clot that
forms in the aneurysm sac depends on several variables, including the form of the
aneurysm, the hematological/clotting profile of the individual, and the administration
of drugs during or after intervention. As a result, the process is highly patient-
and protocol-specific, and a solution that works for one individual may not necessarily
be suitable across the board.
The broad variability of patient outcomes has led to the development of various methodological
approaches, in an attempt to better understand thrombosis in cerebral aneurysms. Clinical
cases and reports have given insight into in vivo development of cerebral aneurysm
thrombosis in humans, and have focused on both device-induced and spontaneous clots.[2]
[3]
[9] In vivo animal models have given insight into some of the key biological events
that occur during clot formation.[13]
[14]
[15]
[16]
[17] In vitro studies aim to understand some of the general features of aneurysm thrombosis,
which would otherwise be difficult to grasp in a pathology which has such wide variability
among different patients.[18]
[19]
The attempt to manage the wide variability of outcomes has led to the development
of various computational models of cerebral aneurysm thrombosis.[20]
[21]
[22]
[23]
[24]
[25]
[26] The intended goal of these models is interventional planning, where prediction of
outcomes such as clotting can provide additional information to the clinician about
a specific patient's prospects. The current generation of models is patient specific
insofar as the geometry is concerned. The use of patient-derived geometries, obtained
from computed tomography and magnetic resonance imaging scans, enables calculation
of flow fields that are unique to that patient.[27] The boundary conditions are not always unique, hence some amount of error is present
in simulations. The clotting calculations are typically based on a single “physiological”
individual's parameters, which are derived from literature obtained largely from the
biochemistry community. The main challenge with several computational models is the
verification and validation of predicted clotting outcomes.
To address this shortcoming, various steps have been taken toward validating in silico
models. The data for validation are usually obtained from in vitro studies designed
to understand the general features of cerebral aneurysm thrombosis. Ou et al presented
a computational model focusing on fibrin accumulation, which is validated by data
from fibrin concentration measurements in the right common carotid artery of a rat
model.[24] Sarrami-Foroushani et al made use of Gester et al's model to validate their computational
models, where platelets play a key role in thrombosis outcome.[18]
[25] Tsuji et al compared their computational model of coil embolization with in vivo
clinical data.[26]
Even though significant progress has been made toward validation of computational
models, the main limitation with most of the in vitro experiments, on which results
are based, has been the use of nonhuman tissue. The challenge of obtaining sufficient
quantities of human blood to run macroscale flow experiments, over a sufficiently
long time period, has meant that the most viable alternative is the use of porcine
tissue and models.[13]
[14]
[18] While many studies have confirmed that porcine models demonstrate the key features
of human aneurysmal disease generally, it has been shown that there are incompatibilities
between human and porcine coagulation.[15]
[17]
[28]
[29]
[30]
[31] Some of the main differences include initiation, propagation and lysis of clots,
and relative contributions of constituent parts to clot strength.[28] Furthermore, most of our current models of cerebral aneurysm thrombosis, which occurs
under pathological conditions, are based on physiological biochemical frameworks.[32]
In this study, we develop a simplified, macroscale, thrombin–fibrinogen flow system,
based on commercially available, purified human-derived plasma proteins, which enables
thrombus growth in an idealized cerebral aneurysm geometry. Soluble fibrinogen (factor
I) is a glycoprotein complex that is the main clotting protein in blood. It is enzymatically
converted to insoluble fibrin fibers (which form the main structural building block
of a clot) by thrombin. We acknowledge that whole blood from a patient would contain
platelets, erythrocytes, white blood cells, and perhaps, a plethora of circulating
inflammatory biomarkers (e.g., cytokines); however, obtaining sufficient quantities
of human whole blood for a macroscale experiment would be unfeasible. Our thrombin–fibrinogen
model, albeit reduced, will allow us to regulate and mimic clot formation in a defined
manner, without the influence of the formed blood elements, allowing us to create
a well-controlled model that will generate repeatable results. In this study, we use
the model to examine how mechanical and biochemical variables contribute to clot formation
in an idealized cerebral aneurysm geometry. The model presented here would also be
useful for validating computational models of cerebral aneurysm thrombosis or for
testing endovascular device thrombogenicity, in future studies.
Materials and Methods
This controlled experiment aimed to develop a flow phantom for a simulated cerebral
aneurysm and parent vessel configuration and a host of postprocessing techniques for
the data generated, by using human-derived clotting proteins (purified fibrinogen
and thrombin) to recreate clot formation.
Flow Phantom
The clotting experiment was conducted in a reusable three-dimensional (3D) printed
geometry of an idealized cerebral aneurysm and parent vessel configuration, as illustrated
in [Fig. 1A]. The idealized geometry was adapted from the work of Mulder et al.[33] The phantom was printed with a Formlabs Form 2 printer, using the standard Formlabs
clear photopolymer resin. Once printed, the phantom was cleaned with isopropyl alcohol,
placed in the sun, sanded, and then sprayed with a clear lacquer-based spray paint.
More specific details relating to phantom preparation can be found in Ho et al.[34] The phantom included a screw mechanism at the thrombin inlet to provide a seal that
prevents leaking during experimentation.
Fig. 1 (A) Flow setup illustrating the positioning of the different components relative
to the flow phantom. (B) Design of the flow phantom which includes screw top to prevent
leaks.
Preparation of Clotting Factors
Human plasma-derived fibrinogen (35–65% protein) was obtained from Sigma-Aldrich (Merck
Group, Missouri, United States). A fibrinogen solution of 1 mg/mL was prepared by
dissolving the protein in sterile-filtered Dulbecco's phosphate buffered saline (PBS)
(without calcium and magnesium). Given that the bulk of the solution comprised saline,
the density was approximated at 1,000 kg/m3 and the viscosity at 0.001 kg/ms. Human plasma-derived thrombin (SAE0006-150UN) was
also obtained from Sigma-Aldrich and was prepared to a solution of 1 mg/mL using PBS.
The prepared solutions, which were both clear, were stored in a −18°C freezer when
not in use.
Flow Setup
The flow setup is illustrated in [Fig. 1B]. Fibrinogen solution was pumped into the inlet of the flow phantom, using an IsmatecReglo
CC pump (Ismatec, Glattbrugg, Switzerland), at different flow rates (0, 40, and 80 mL/min).
Cerebral aneurysms are commonly found on blood vessels of the circle of Willis, such
as the internal carotid artery, which experiences Reynolds numbers (Re) ranging from
approximately Re = 200 to Re = 531.[35] A flow rate of 40 mL/min corresponds to Re = 212, while a flow rate of 80 mL/min
corresponds to Re = 424, thus falling within this physiological range. Once the solution
was flowing throughout the entire phantom (and all air bubbles had been removed),
10 or 100 μL of thrombin solution was injected at the top of the aneurysm sphere,
using a syringe pump. The solution was not recirculated, as the fibrinogen which had
been in contact with thrombin turned to fibrin, resulting in different properties.
A Nikon D3300 DSLR camera (Nikon, Tokyo, Japan) was used to record the progression
of the growing clot into the flow field. The experiment was repeated five times, using
the same flow phantom, which was flushed out with water between experimental runs.
Estimation of Clot Area
The clot area within the aneurysm sac was approximated using Simpleware 2020.03 (Synopsys,
Exeter, United Kingdom). The photographic stills were extracted from the video recording
and images were imported into Simpleware. Each image was cropped (x minimum 300 mm, x maximum 633 mm, y minimum 230 mm, y maximum 410 mm). The image was then rescaled (scale factor 0.15) to ensure that the
x direction matched the size of the physical phantom (49.5 mm). The image was cropped
a second time (y maximum 67 mm). A mask was then created and grayscale threshold values of 76 to 96
were used to select part of the clot region. Any islands greater than 200 pixels were
removed and then the remainder of the clot region, not captured by the threshold,
was manually selected using a paint function. Once the clot region had been selected,
the surface area of the mask was calculated. This process was followed for almost
all the photographs; however, some required more manual intervention than others.
Statistical Analysis of Clot Area Results
To determine the impact of different variables on final clot growth area, two-way
analysis of variance (ANOVA) for unbalanced designed was performed using MATLAB's
statistics and machine learning toolbox (MathWorks, Natick, United States).
Results
[Table 1] clearly illustrates the effects of varying the fibrinogen flow rate and thrombin
volume. The former alters the speed at which fibrinogen is delivered (mechanical variable),
while the latter focuses on changing the amount of thrombin delivered (biochemical
variable). An increase in the amount of thrombin, for a fixed flow rate of 0 mL/min,
results in a drastic increase in occlusion outcome. Changes in flow rate present a
more complex picture. An increase from 0 to 40 mL/min, for 10 μL of thrombin, results
in an increase in occlusion percentage. An increase to 80 mL/min results in a decrease
in occlusion percentage when compared with both 0 and 40 mL/min. The results for two-way
ANOVA for unbalanced design, depicted in [Table 2], demonstrate that both fibrinogen flow rate and thrombin volume are significant.
Table 1
Occlusion percentage for experimental runs using different fibrinogen flow rates and
thrombin concentrations
|
Case ID
|
Fibrinogen flow rate (mL/min)
|
Thrombin volume (μL)
|
Area (mm2)
|
Occlusion percentage (%)
|
Mean (%)
|
SD (%)
|
|
0_10_1
|
0
|
10
|
37.4
|
45.8
|
43.5
|
2.0
|
|
0_10_2
|
0
|
10
|
34.2
|
41.9
|
|
0_10_3
|
0
|
10
|
35.0
|
42.8
|
|
0_100_1
|
0
|
100
|
81.7
|
100.0
|
100.0
|
0.0
|
|
0_100_2
|
0
|
100
|
81.7
|
100.0
|
|
0_100_3
|
0
|
100
|
81.7
|
100.0
|
|
40_10_1
|
40
|
10
|
45.8
|
56.1
|
57.6
|
2.3
|
|
40_10_2
|
40
|
10
|
48
|
58.8
|
|
40_10_3
|
40
|
10
|
49.2
|
60.2
|
|
40_10_4
|
40
|
10
|
45.3
|
55.4
|
|
80_10_1
|
80
|
10
|
7.1
|
8.7
|
7.7
|
0.9
|
|
80_10_2
|
80
|
10
|
5.6
|
6.9
|
|
80_10_3
|
80
|
10
|
6.2
|
7.6
|
Abbreviation: SD, standard deviation.
Table 2
Results of two-way ANOVA for unbalanced design
|
Source
|
Sum of squares
|
Degrees of freedom
|
Mean squares
|
F-statistic
|
p-Value
|
|
Fibrinogen flow rate
|
4,377.76
|
2
|
2,188.88
|
779.45
|
8.32 × 10−11
|
|
Thrombin volume
|
4,788.38
|
1
|
4,788.38
|
1,705.12
|
1.43 × 10−11
|
|
Fibrinogen flow rate × thrombin volume
|
0
|
0
|
0
|
0
|
|
|
Error
|
25.27
|
9
|
2.81
|
|
|
|
Total
|
13,154.03
|
12
|
|
|
|
Abbreviations: ANOVA, analysis of variance; NaN, Not a Number.
An unexpected, yet interesting finding is illustrated in [Fig. 2]. The fibrinogen and thrombin solutions which were prepared were clear in color,
resulting in the formation of a clear clot for almost every run. For the fourth 40
ml/min experimental run (40_10_4), the clot unexpectedly developed a slightly murky
hue, making it possible to visualize the progression of the clot clearly, as illustrated
in [Fig. 2]. At 0 second, the thrombin solution is injected into the flowing fibrinogen solution.
By 20 seconds, a small clot is already seen in the top left-hand corner of the sphere.
As the clot grows, it propagates downward, toward the center of the sphere, and also
toward the top right-hand side of the sphere. At 60 seconds, there is a clear asymmetric
bias, toward the left-hand side of the sphere, where clot growth was initiated. From
80 seconds, it becomes clear that the clot front propagates toward the right-hand
side, as the leftmost peak sets up a flow barrier which interrupts the circulation
of flow within the sac. At the same time, the leftmost peak propagates downward (at
a much slower rate than the march toward the right) and this effect is evident at
∼160 and 180 seconds. Once the clot fills a significant proportion of the upper half
of the aneurysmal sac, the peak, which had migrated toward the center at 160 seconds,
is seen to propagate toward the left (and downward) for the remainder of the time
steps.
Fig. 2 Cloth growth over time. The zero time point marks the time at which the thrombin
solution is injected into the aneurysm sac and comes into contact with the flowing
fibrinogen solution. The red arrow at 0 second indicates the direction of fibrinogen
flow.
[Fig. 3] illustrates the approximate area of the growing clot over time for 40_10_4. The
approximation does not take the 3D nature of the clot into account; however, it enables
the calculation of an approximate occlusion percentage. The area of the aneurysm sac
is 81.65 mm2, hence the clot area at 240 seconds (45.3 mm2) gives an occlusion percentage of ∼55%. Even though [Figs. 2] and [3] are not statistically significant, they demonstrate the type of information that
could potentially be gained from this methodology. To be able to achieve such visualization
for every experimental run, fluorescent fibrinogen and an appropriate laser source
would be required.
Fig. 3 The estimated clot area from the two-dimensional images. The clot is seen to increase
steadily in area over time.
Discussion
The results from the experiment clearly demonstrate that thrombus growth in an idealized
cerebral aneurysm geometry is a complex interplay between biochemical and mechanical
factors. Thrombin is injected at the start of the experiment only and the clot begins
as a small mass where thrombin first comes into contact with fibrinogen, resulting
in the formation of an insoluble fibrin network. As more fibers are generated, the
clot propagates out into regions where the flow has been slowed by the growing insoluble
fiber network. In the absence of flow, diffusion is the dominant mechanism of reactant
transport and the amount of thrombin in the system has a significant impact on clotting
outcome. The introduction of flow, as observed in the 40 mL/min case, assists with
transport of thrombin within the aneurysmal sac, enabling greater occlusion than in
the 0 mL/min case. Interestingly, a further increase in flow rate (to 80 mL/min) does
not amplify this effect, indicating that higher flow speeds are not necessarily supportive
of clot development. The area where the clot is initiated has to have sufficiently
slow flow for the reaction between thrombin and the soluble fibrinogen to take place
and for the insoluble clot to develop into a fibrin fiber network that will not be
destabilized by the flow. This suggests that the initiation of clotting is largely
dominated, or influenced, by mechanical factors. As the clotted mass propagates into
the aneurysmal sac, the flow is increasingly slowed and the rate of propagation is
then limited by the reaction rate and availability of reactants. Convective flow can
support clot development during the propagation process, but speeds that are too high
limit the contact time between thrombin and fibrinogen, thus disrupting the fibrin
formation process. While the use of a thrombin–fibrinogen in vitro model is a significant
simplification of an otherwise very complex system, it is beneficial for obtaining
an overall view of the main factors at play, particularly where both flow and biochemical
reactions are present.
The balance between complexity and reductionism has long been debated in the study
of biological systems.[36] Many of the discoveries that led to the present-day understanding of the hemostatic
system were based on reductionist in vitro models that studied the impact of individual
factors. From this understanding, a more complex network of the hemostatic system
could be developed. Modern computing has enabled the modeling of complex biochemical
signaling and flow, and has been used to explore some of the fundamental questions
that remain unanswered.[37] The obvious downside of a simplified system is that it fails to account for all
the in vivo variables in a patient with an aneurysm (including platelets, erythrocytes,
and white blood cells, as well as circulating inflammatory biomarkers that will bind
to and make the soluble fibrinogen hypercoagulable even before it interacts with thrombin).
We believe that this complex comprehensive interplay of so many biological entities
cannot be accounted for outside the human body. Even though the increase in complexity
has proven to be largely beneficial, particularly for the hemostatic system, where
many of the gaps in understanding have already been filled, there is still a place
for simpler systems, such as our proposed model. In the case of the cerebral aneurysm
thrombosis, there remain many questions, including the extent to which the signaling
pathways of this particular disease mirror those of physiological hemostasis. Different
studies have shown that the biochemical pathways of clotting are often altered by
pathological states, particularly those marked by inflammation, such as cerebral aneurysms.[38]
[39]
[40] As stands, the best in vivo data that we have is percentage of occlusion at the
end of endovascular device placement, as reported in clinical studies.[3] We have no data relating to the composition of the clot that forms at the end of
such an intervention. As such, creating a more sophisticated in vitro model, which
incorporates more blood cells, may be misguided as we would include blood cells based
on physiological models of clotting. The easiest solution would be to use blood from
cerebral aneurysm patients; however, we would not be able to obtain sufficient quantities
for a macroscale model. The model we present here therefore gives a picture of occlusion
outcome but would need to be used in conjunction with other modalities for an accurate
picture of clot formation. As with other reductionist models, the greatest relevance
might be in testing under very specific conditions or answering particular questions
that contribute to a bigger picture.[36]
[41]
The longer term goal of our work is to develop a virtual interventional planning tool
that can give an indication of clotting outcome, based on proposed interventions,
on a patient-specific, image-informed basis. The main question we wish to answer,
therefore, is whether or not the placement of a device (e.g., flow diverter) results
in the formation of a clot.[32] Of equal importance is whether or not the clot fills the aneurysm sac completely.
The in vitro method developed here provides a technique that will enable validation
of computational models developed to this end and will also enable simplified in vitro
study of human clotting in flow, especially within resource constrained environments.
In vivo observations of clotting in humans can be performed in the period immediately
after device placement; however, it can prove that it is difficult to identify individual
components and interactions within the system. Also, follow-up imaging (3, 6, and
9 months), that can be used to track development of occlusion (and inform anticoagulant
pharmaceutical regimes), may be prohibitively expensive. The use of animal models
has been beneficial for elucidating many of the key features of aneurysm evolution;
however, pigs (and porcine models), which develop similar aneurysm characteristics
to humans, are very expensive (in the region of $10,000 for recovery trials per animal).[29] The method developed here is based on a human-derived thrombin–fibrinogen model,
which comprises parts that are easy to procure, and is relatively easy to implement
and control. An improvement on the presented experiment would be the use of fluorescent
fibrinogen and a laser source, which would ensure that the progression can be visualized
for every experimental run.
Aside from the simplicity of the model, one of the main shortcomings of the method
presented in this study is the lack of accurate 3D image acquisition and subsequent
quantification, at this stage. The final outcome of the experiment resembles that
seen in clinical practice (occlusion percentage).[12] While this is ideal for matching outcomes between virtual predictions and clinical
workflows, quantification of the flow field or biochemical concentrations would be
much more beneficial for model validation. It would also give greater insight into
the exact contributions of different systems during clot formation. The other limitation
is the time period observed during experimentation. The period defined as “immediately
after endovascular device placement” is not well defined in the literature. It is
therefore difficult to predict how long it would take for a clot to form immediately
after intervention. On the other hand, the simplicity of the phantom and overall setup
would allow the implantation of flow diverters in this setup, and thus compare directly
the effect of such devices both with the preinterventional thrombus evolution and
with computational models that can incorporate virtual device implantation. The rate
of clot development would be influenced by, among other things, flow rate and protein
concentration. Given that the work presented here is a proof of concept, optimization
of these variables was not a goal at this stage, and a considerable amount of work
has already gone into guidelines relating to in vitro clot experiments.[42] Rather, our aim was to determine whether or not we could grow a clot successfully,
using a human thrombin–fibrinogen model in a simplified system. We also wanted to
see how well we could predict occlusion outcome based on data from the experiment.
This system is therefore beneficial for estimating initial angiographic occlusion
and could also assist device manufacturers in optimizing designs. The model would
not be suitable for studying longer term clot evolution and maturation following flow
diverter placement, as more cells (e.g., platelets, erythrocytes, white blood cells,
and even circulating inflammatory biomarkers such as cytokines) would be required
to adequately represent the complexity of that process, particularly if we want to
investigate the pathophysiological changes that occur in cerebral aneurysms.
The in vitro model presented in this article creates links between mechanics, biochemistry,
and clinical outcome, using a relatively low-cost system based on human-derived proteins.
The model clearly demonstrates that clot formation in cerebral aneurysms is a complex
interplay between mechanical and biochemical factors. The model would be of use to
device manufacturers carrying out tests of device thrombogenicity and for computational
modelers seeking to validate aspects of their computational models, within the cerebral
aneurysm community.
What Is Known on the Topic?
-
The extent to which a cerebral aneurysm is occluded by a thrombus has an impact on
aneurysm outcome.
-
Full occlusion is preferable to partial occlusion.
-
Placement of endovascular devices for treatment of cerebral aneurysms results in occlusion
of the sac.
-
Computational models of cerebral aneurysm thrombosis have been developed to predict
occlusion outcome.
-
Validation of these computational models has been performed with animal models.
What Does This Article Add?
-
A simplified, low-cost, in vitro method, based on human-derived clotting proteins,
for predicting occlusion of cerebral aneurysms.
-
Prediction of occlusion outcome, similar to clinical outcomes currently used.
-
Data which can be used to validate computational models.
