Dynamic dual-tracer MRI-guided fluorescence
tomography to quantify receptor density in vivo
Scott C. Davisa,1, Kimberley S. Samkoeb, Kenneth M. Tichauera,c, Kristian J. Sextona, Jason R. Gunna,
Sophie J. Deharvengtd, Tayyaba Hasane, and Brian W. Poguea,b,e
aThayer School of Engineering, Dartmouth College, Hanover, NH 03755; Departments ofbSurgery anddPathology, Geisel School of Medicine at Dartmouth,
Hanover, NH 03755;cDepartment of Biomedical Engineering, Illinois Institute of Technology, Chicago, IL 60616; andeWellman Center for Photomedicine,
Harvard Medical School, Massachusetts General Hospital, Boston, MA 02114
Edited by Rakesh K. Jain, Harvard Medical School and Massachusetts General Hospital, Boston, MA, and approved April 15, 2013 (received for review
August 27, 2012)
The up-regulation of cell surface receptors has become a central
focus in personalized cancer treatment; however, because of the
complex nature of contrast agent pharmacokinetics in tumor tissue,
methods to quantify receptor binding in vivo remain elusive. Here,
we present a dual-tracer optical technique for noninvasive estimation
of specific receptor binding in cancer. A multispectral MRI-coupled
fluorescence molecular tomography system was used to image the
uptake kinetics of two fluorescent tracers injected simultaneously,
one tracer targeted to the receptor of interest and the other tracer
a nontargeted reference. These dynamic tracer data were then fit
to a dual-tracer compartmental model to estimate the density of
receptors available for binding in the tissue. Applying this approach
to micewith deep-seated gliomas that overexpress the EGF receptor
produced an estimate of available receptor density of 2.3 ± 0.5 nM
(n = 5), consistent with values estimated in comparative invasive
imaging and ex vivo studies.
molecular imaging|oncology|optical imaging|engineered proteins|
regularity gives rise to a natural tumor-to-normal tissue contrast
compounds to cancer cells. Additionally, because many of these
receptors are involved in cell signaling pathways that lead to en-
hanced cell proliferation, reduced apoptosis, and drug resistance,
they are increasingly being targeted with receptor-specific thera-
pies that interfere with these pathways to block signaling and in-
hibit tumor growth. As efforts to personalize cancer treatment by
targeting patients’ specific receptor expression profiles expand (1–
4), the ability to noninvasively quantify the availability of those
receptors in living tissue would be an important development with
broad implications for drug development programs and for moni-
toring a patient’s response to therapies.
Examining receptor expression currently requires analyzing
select tissue specimens extracted by biopsy or tumor tissue resection.
Some of these techniques, such as immunohistochemical staining,
can be used to estimate the total density of receptors in a given
cancer cell population; however, an important distinction must be
made between the total density of receptors in the cell population
and the density of receptors available for binding in living tissue.
The latter depends on the pharmacokinetics of the injected agent,
the vascular and ECM structures in the tumor, the arrangement
of the cancer cells, and the biochemical environment. Thus, ex vivo
measurements should be thought of as a maximum potential re-
ceptor density, and likely overestimate the receptors available for
targeting by therapeutic agents (5). Analyzing receptor status using
tissue specimens is also subject to intratumoral heterogeneity (6)
and precludes tracking of receptor density longitudinally, because
repeated harvesting of tissue samples is usually infeasible. Non-
invasive techniques, however, would enable monitoring of
changes in cellular receptor availability in response to therapy over
ancer pathogenesis often results in abnormal expression of cell
surface receptors on the diseased cells. This pathological ir-
time, a capability applicable in both preclinical animal research
and clinical practice.
Examples of receptor-targeted imaging of cancer in vivo are
numerous (7–11), although few address the capacity to quantify
binding kinetics and receptor availability. The bulk of these efforts
involves imaging a single contrast agent, termed a tracer here,
targeted to a specific receptor. Although single-tracer modalities
can often produce images with enhanced tumor-to-normal tissue
contrast, the measurement of tracer–receptor binding is con-
founded by the behavior of the tracer in the body—namely, the
rates at which the tracer transports between the vascular and ex-
travascular spaces and the rates at which it binds and becomes
dissociated from the receptor—among other factors (12–15). To
isolate receptor–tracer binding quantities, the works by Lam-
mertsma and Hume (16) and Logan et al. (17) introduced com-
partmental modeling approaches for dynamic PET imaging of
receptor kinetics in the brain. These strategies require referencing
the tracer kinetics in the tissue of interest to the kinetics in a ref-
erence tissue region that has similar hemodynamics but is devoid
tissue region largely precludes the application of this approach in
cancerimaging, becausesolid tumorshave unique vasculature (12,
robust method applicable in any tissue region, including solid
tumors, by adapting the compartmental model principle to a dual-
tracer model developed around dynamic imaging of two tracers in
the same tissue simultaneously (21). Although optical imaging
methods are well-suited for this multitracer approach because of
their ability to discriminate multiple compounds based on their
spectral signatures,imagingdeep-seatedtumorsis severelylimited
by photon scattering in tissue. Thus, previous experiments vali-
dating the dual-tracer modeling approach were completed using
invasive, surface-weighted imaging techniques.
In this study, we report on noninvasive quantification of binding
kinetics and available receptor density in deep tissue using dual-
tracer MRI-coupled fluorescence molecular tomography (MRI-
FMT). Previous studies have used multimodal FMT to report
single-agent fluorophore kineticsin thebloodpool (22) and tumor
xenografts (23). The present study applies this technique to ex-
amine the quantification of specific receptor binding in vivo. Es-
injected fluorescent tracers simultaneously, one tracer targeted to
a specific receptor and the other a nontargeted counterpart, and
Author contributions:S.C.D., K.S.S.,and B.W.P.designedresearch; S.C.D.,K.S.S., K.J.S., J.R.G.,
and S.J.D. performed research; S.C.D., K.M.T., T.H., and B.W.P. analyzed data; and S.C.D.
wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
Freely available online through the PNAS open access option.
1To whom correspondence should be addressed. E-mail: email@example.com.
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
| May 28, 2013
| vol. 110
| no. 22
fitting the time-dependent tracer concentrations to a dual-tracer
compartmental model describing the pharmacokinetics of the
tracers (Materials and Methods).Ofparticular interest isthe binding
potential (BP), a parameter defined as the product of the concen-
tration of receptors available for binding, Bavail, and the affinity of
the targeted tracer to the receptor, ka:
Thus, BP is a unitless quantity proportional to the receptor–
target affinity (in Molar−1) and density of receptors available
to bind with the targeted tracer (in Molar). Because kacan often
be determined a priori using in vitro techniques, the capacity to
dynamically image targeted and nontargeted tracers simulta-
neously facilitates the recovery of Bavailin living tissue.
A key requirement for deploying the dual-tracer strategy is the
ability to image the dynamics of two tracers in tissue. The MRI-
FMT system presented here is capable of distinguishing multiple
fluorophores measured simultaneously by using spectrally resolved
optical detection. This dual-tracer imaging approach is realized by
acquiring multiple projections of data through the tissue volume,
each of which is a highly resolved spectrum composed of fluores-
cence signals from the different fluorescent compounds in the tis-
sue, illustrated in Fig. 1E. Provided that the fluorophores have
spectrum extracts the contribution of each fluorescent compound
to the measured spectrum, effectively separating signals emitted
from the tissue autofluorescence and any number of fluorescent
tracers in the tissue (Fig. 1 F–H). The fluorescent dyes used in this
study were IRDye 800CW (LI-COR Biosciences) and Alexa Fluor
750 (Life Technologies), with emission peak maxima at ∼800 and
775 nm, respectively. After the signals are separated, the internal
distributionofeachtracercanbe estimatedby fitting theprojection
data to a numerical model based on the diffusion approximation of
light propagation in tissue (24–28). Applying these principles, im-
aging fluorescence activity in relatively large tissue volumes (over
5 cm) is feasible. Substantial evidence exists suggesting that volu-
metric images of fluorescence activity are more accurate if addi-
tional information, such as the anatomical features of the tissue, is
To that end, the hybrid MRI-FMT system used in this study inte-
grates the optical detection channels into a specialized MRI insert,
data and MR image stacks. An illustration of this process is shown
in Fig. 1 and described in Materials and Methods.
To show the ability to separate and image two colocalized
tracers in a realistic volume, a cylindrical tissue-simulating phan-
tom was imaged with the MRI-FMT system. The 27-mm-diameter
phantom was constructed of epoxy resin containing titanium di-
oxide and India ink to mimic the optical scattering and absorption
properties typically found in living tissue. To simulate a tumor
region, a 6-mm-diameterhole drilled axially between theedge and
the center of the cylinder was filled with 1% (vol/vol) intralipid,
IRDye 800CW was held constant at 100 nM, whereas the con-
and the phantom was imaged at each concentration. The image
series for both fluorescent dyes are presented in Fig. 2A, and show
that there was very little variation in fluorescence activity in the
channel had a clear dependence on changes in concentration. A
plot of the recovered values in the region of interest as a function
of Alexa Fluor 750 concentration is shown Fig. 2B, and confirms
a significant linear response in the Alexa Fluor 750 channel, R2=
0.99, whereas the IRDye800 channel remained relatively flat
resentation of the MRI-FMT animal interface inside the MRI bore. The head of
the rodent (blue) is surrounded by eight optical fibers in the coronal plane. (B)
Coronal and (C) sagittal slices of the MR images showing the segmented brain
and tumor tissues in green. A yellow arrow points to the tumor tissue and the
red dots correspond to the optical fiber positions. (D) Surface renderings of the
mouse head (blue), brain (green), and tumor (red) used to create a finite el-
ement mesh that serves as a template for the optical image reconstructions. (E)
A measured spectrum from a single source–detector projection is shown as
a black solid line. A spectral fitting algorithm determines the relative contri-
bution from the (F) tissue autofluorescence and fluorescence from the (G)
targeted and (H) untargeted tracers that compose the measured spectrum.
The fitted spectrum is shown as a gray dotted line in E. After spectral fitting of
the spectra from each projection, the MRI-based finite element mesh is com-
bined with the separated tracer data to produce volumetric images of fluo-
rescence yield. A single slice from this volume is shown in I and J for the
targeted and untargeted tracers, respectively.
Dual-tracer MRI-FMT for a single frame in a dynamic series. (A) Rep-
Recovered images of fluorescence yield (normalized to the mean of the
IRDye800 channel) from colocalized Alexa Fluor 750 and IRDye800CW. (B)
Normalized values of fluorescence yield in the region of interest plotted as
a function of Alexa Fluor 750 concentration. The IRDye800CW concentration
was held constant at 100 nM for each concentration of Alexa Fluor 750.
Dual-tracer MRI-FMT imaging of a tissue-mimicking phantom. (A)
| www.pnas.org/cgi/doi/10.1073/pnas.1213490110 Davis et al.
(SD = 5%). Because of the differences in the excitation absorbance
and the quantum yield between the two dyes, the dominant signal
in the measured spectra transitioned between IRDye 800CW
fluorescence and Alexa Fluor 750 fluorescence as the Alexa Fluor
concentrations were increased, showing the ability of the spectral
fitting technique to accurately extract weak signals of either dye
from dominant background signals. Finally, we showed that in-
terchannel cross-talk is between 0% and 4% by examining the
spectral fitting results from phantoms containing a single tracer,
which is described in SI Text.
To investigate the capacity of dynamic dual-tracer MRI-FMT to
quantify the density of available receptors in a deep-seated tumor
model, athymic nude mice with U251 brain tumors were imaged
preparation details). U251 tumor cells have been reported to have
cell proliferation, is commonly overexpressed in clinical cancers
(36–39), and thus is a natural drug target (38, 40). Anesthetized
mice positioned in the MRI-FMT system were scanned pre- and
postgadolinium (Gd) contrast with a T1-weighted MRI sequence
and then administered a solution containing 0.2 nmol of each op-
tical tracer (targeted and untargeted) in the tail vein. The targeted
tracer used in this experiment was Affibody anti-EGFR imaging
agent (Affibody) conjugated to Licor IRDye 800CW fluorescent
dye. This anti-EGFR imaging agent is a three-helix protein engi-
neered to bind to EGFR with low nanomolar affinity (41). The
untargeted counterpart was Affibody negative control, which
agent was conjugated to the AlexaFluor 750 fluorescent dye (SI
Text has details on tracer selection and preparation). Dynamic
FMT imaging began immediately after injection and continued
over the course of 1 h at a rate of approximately one frame per 2
each dye were separated into two optical data channels using the
spectral fitting technique. Applying the anatomical structure from
the MR image stack in the optical image reconstruction algorithm
provided volumetric images of fluorescence activity for each tracer
This process is illustrated by the diagram in Fig. 3 and detailed
further in Materials and Methods.
A sampling of the time series images recovered for one animal
is presented in Fig. 4. For display purposes, each frame in the
image series is presented as a 2D coronal slice interpolated from
the 3D volume. The position of this slice with respect to the tissue
volume is illustrated in Fig. 4A, and the corresponding MR image
is presented in Fig. 4B. The tumor region, as defined by contrast-
enhanced MRI, is outlined in green in Fig. 4B. A sample of the
FMT image time series for both targeted and untargeted fluo-
rescent tracers is shown in Fig. 4 C and D, respectively. Because
the images were acquired shortly after injection, the tracer con-
centrations were relatively high throughout the tissue. Although
thenormal tissuesoutside the brain showthe highest fluorescence
uptake levels, the tumor region in the brain shows obvious en-
hancement compared with the normal brain tissue. There was
observations were consistent between all five animals.
To determine the binding potential of the anti-EGFR Affibody
tracer, the kinetics of both tracers in the tumor region were
extracted from the image sequences. The coronal MRI slices
shown in Fig. 4E illustrate the anatomical structure near the im-
aging plane in each mouse. The overlaid outlines in Fig. 4E
identify the tumor region defined by Gd contrast enhancement.
animals are presented in Fig. 4F. These data are consistent be-
tween all animals and show very different pharmacokinetic be-
havior between the anti-EGFR and negative control tracers. The
Affibody negative control tracer experienced rapid uptake im-
mediately after injection that was not captured by the image se-
quence at the acquisition rate used. This uptake was followed by
a steady clearance that reduced the drug concentration by 52%
(mean for all animals, ±8% SD) by the end of the 1-h imaging
session. In contrast, the targeted anti-EGFR Affibody tracer
showed less rapiduptake and much slower clearance that reduced
the fluorescence activity in the tumor by less than 13% (mean for
all animals, ±4% SD) from maximum over the course of 1 h. The
time series data were used to fit the parameters of the dual-tracer
compartment model to determine values of binding potential in
the tumor. The result was a mean binding potential of 0.80 ± 0.17
in the population of mice (n = 5). Applying Eq. 1 reveals an
available receptor density of 2.3 ± 0.5 nM for ka= 0.36 × 109M−1
for the Affibody probe.
To validate the binding potential values calculated using the
MRI-FMT image series, exposed s.c. U251 tumors were imaged
using a planar fluorescence scanner in a separate population of
mice. Selected images from the dynamic imaging session for one
animal are shown in Fig. 5, with the targeted and untargeted
tracer image series in Fig. 5 A and B, respectively. The average
fluorescence values in the tumor region, delineated manually,
were calculated to produce the time curves shown in Fig. 5C.
Binding potential values were calculated in the same manner as
the BP values from the MRI-FMT imaging data, resulting in
a mean value of 0.90 ± 0.23 and thus, an available receptor
density of 2.5 ± 0.6 nM for all animals (n = 5).
Because receptor expression can be influenced by the site of
tumor growth, immunoblotting was used to compare EGFR ex-
pression between s.c. and intracranial U251 tumors (SI Text). Al-
though these data show very similar expressions between growth
4) tumors, respectively], total receptor expression is only loosely
point method to approximate binding potential was applied to ex
vivo slices of intracranial U251 tumors. This approximation, de-
tailed in Materials and Methods, results in a simple relationship
between binding potential and the ratio of the targeted to untar-
geted dye fluorescence. Available receptor density values re-
covered using this method were 2.3 ± 1.5 (n = 6).
The binding potential values recovered using MRI-FMT are in
close agreement with those values determined using planar imaging,
Procedure for recovering binding potential using dynamic dual-
Davis et al.PNAS
| May 28, 2013
| vol. 110
| no. 22
suggesting that accurate noninvasive imaging of binding poten-
tial in vivo is feasible. These results were further validated by
using the recovered binding potential values to compute the density
of available receptors and comparing them with previously repor-
ted values. Tichauer et al. (21) previously reported in vivo binding
potential values between EGFR and EGF for s.c. U251 tumors.
receptor than the anti-EGFR Affibody molecule used herein. Ap-
plying Eq. 1 to compensate for different affinities results in similar
values for available receptor density: Bavail= 2.0 ± 0.4, 2.3 ± 0.5,
2.3 ± 1.5, and 2.5 ± 0.6 nM for measurements acquired by
Tichauer at al. (21), MRI-FMT, ex vivo analysis of intracranial
tumors, and planar imaging reported herein, respectively.
The close proximity of Bavailvalues measured with different
imaging paradigms, with different targeted tracers, and in tumors
implanted in different tissue types exemplifies the robustness of
the dual-tracer approach. Subcutaneous tumors often have very
different vasculature than tumor models grown directly in their
native organ (in this case, the brain). Growth site-specific vascu-
lature may explain the differences in the observed pharmacoki-
netic behavior of the tracers in organ-implanted and s.c. tumors
(Figs. 4F and 5C, respectively). Despite these observed differ-
ences, both tracers inhabit the same vascular and extravascular
space, and the density of available receptors was found to be
consistent, regardless of tumor implantation site. Single-tracer
approaches under these conditions are unable to adequately
compensate for differences in vascularity, making reliable re-
covery of available receptor density nearly intractable.
Our findings suggest that dual-tracer quantification of binding
potential is a robust technique that mitigates many of the complex
calibration procedures required for quantitative single-tracer im-
aging. Because the model operates on the normalized ratios of two
tracers injected simultaneously, quantifying BP is independent of
many of the parameters that challenge imaging a single tracer,
a process that often requires absolute instrument calibration. Thus,
the approach is unaffected by commonly confounding factors, such
of the tissue, concentration of tracer administered, and fluorophore
similar pharmacokinetic behavior in vivo, excluding binding. The
former is certainly the case given the proximity of the fluorescence
emission wavelengths of the tracers. The latter requires careful
choice of the tracer compounds, with consideration given to factors
such as molecular weight, conjugate binding chemistry, and poten-
tial for nonspecific binding.
The observed robustness of the dual-tracer modeling approach
is valid provided that a number of assumptions are met. As dis-
cussed elsewhere (21), the model system assumes first-order
binding kinetics and therefore loses accuracy when a large per-
centage of receptors are bound with targeted tracer. Although
receptor saturation may obtain if large quantities of tracer are
administered, an analysis of plasma and binding kinetics, and
receptor density in the U251 tumor line shows that receptor sat-
uration at the injected dose used herein (0.2 nmol) is unlikely (SI
Text). In addition to first-order kinetics, the model assumes that
the concentrations of bound and unbound targeted tracer in the
extravascular space are in instantaneous equilibrium. Although
fundamental to the derivation of this model system, values of
binding potential have been shown to be robust even when this
assumption does not apply (42, 43). Additional physiological
processes,suchastracer metabolism andreceptorinternalization,
are not captured by the model and therefore represent potential
sources of error in the calculations of BP. However, because the
model operates on data acquired within a short time frame after
supported by previous work (5). Finally, although most system-
specific biases are mitigated, the significant depth-dependent
sensitivity of FMT remains a limitation for imaging smaller
structures and pathologies deep within tissue, even when guided
by complementary imaging modalities, and could introduce cross-
talk between tissue regions. As a result, visualizing submillimeter
spatial variations in binding potential within a deep-seated tumor
will likely be infeasible. Despite this limitation, quantifying the
average available receptor density in tumor masses is achievable
in research animals, and the recent history of FMT studies in the
human breast (24, 44) suggests that translating the technology to
clinical imaging is feasible for tissue organs accessible by FMT
The noninvasive quantification of binding potential, and thus
density of available cell surface receptors, shown here represents
an important advancement towards patient-specific cancer treat-
ment. Costly preclinical drug development studies would benefit
from longitudinal monitoring of receptor binding in response to
rendering of a study mouse with an intersecting plane illustrating the location
of the coronal image slices in B–D. (B) A coronal MRI image of the mouse head
with the tumor region outlined in green. This image slice corresponds to the
time series images showing fluorescence activity of the (C) targeted and (D)
untargeted tracers. (E) Coronal MRI images of all five study animals with
overlay outlines of the tumor region. (F) Time series of fluorescence activity of
targeted (● solid lines) and untargeted (○ dotted lines) tracers in the tumors of
all animals. The targeted and nontargeted curves for each animal are paired by
color, which also corresponds to the colors of the outlined tumor regions in E.
Dynamic dual-tracer MRI-FMT imaging of gliomas in mice. (A) 3D
cence images of exposed U251 tumor and surrounding normal tissue in a liv-
ing mouse for (A) targeted and (B) untargeted tracers. (C) Time series data for
the average fluorescence activity in the tumor region for targeted (dotted line
and open circles) and untargeted (solid line and solid circles) tracers.
Dynamic dual-tracer planar imaging of an s.c. tumor. Planar fluores-
| www.pnas.org/cgi/doi/10.1073/pnas.1213490110 Davis et al.
clinical trials based on their receptor expression profiles would
help economize those programs. Additionally, monitoring patient
response to approved target-specific drugs would enable the tai-
loring of treatment regimens and provide actionable information
on drug resistance. Although our work has focused on cancer
pathologies, the platform is broadly applicable to other disease
states that involve overexpression of cellular receptors or specific
binding in general, such as Alzheimer’s disease or infection, and
thus may play a significant role in the growing efforts towards
Materials and Methods
Dual-Tracer Binding Potential Model. Derivation of the dual-tracer model
system has been described elsewhere (21). Briefly, the behavior of the tar-
geted tracer in this model is defined by a series of rate constants that
describe the movement of the tracer between the blood plasma and ex-
travascular space and once the tracer is in the extravascular space, the rates at
which it will bind to and be released from the available receptors. Because
the nontargeted tracer is assumed to have no binding affinity, its behavior
is defined by rate constants that describe the tracer movement between the
blood plasma and extravascular spaces only. If the two tracers have similar
rates of extravasation, the differential equations that describe the behavior of
both tracers can be combined and reduced to a single expression that
depends only on the concentrations of both tracers (measureable), the ratio
of the extravasation rates of both tracers, the intravasation rate of the
targeted tracer, and the binding potential (BP) of the tracer–receptor pair:
where the fluorescence activity of the targeted and untargeted tracers is
given by FT and FUT, respectively, R1 is the ratio of rate constants that
describes the transport of the targeted and untargeted tracers from the
vasculature to the extravascular space, and k2is the rate constant for tar-
geted tracer to transport from the extravascular space back into the vascu-
lature. FTand FUTcan be measured by dynamic imaging of both tracers, and
the remaining parameters fitted using a nonlinear least-squares algorithm
applied to the expression above.
Dynamic MRI-FMT Experimental Procedure. The hybrid MRI-coupled FMT sys-
tem has been described extensively in previous publications (29, 30, 45, 46).
Each animal was imaged with Gd-MRI at least 2 d before the MRI-FMT
scans. Only animals with visible Gd-enhanced features in the brain were
included in the study. The MRI-FMT imaging procedure began by positioning
an anesthetized (1.5% isoflurane, 1 L/min oxygen) mouse in the MRI-FMT
rodent coil and placing the optical fibers in contact with the head. A series of
MRI scout scans were used to confirm that the plane of the optical fibers
intersected the tumor volume; in some cases, animal repositioning was re-
quired. After a satisfactory position was established, a full set of pre- and
post-Gd contrast T1-weighted images was acquired over the volume of the
head. Under static FMT imaging conditions, the FMT imaging would be
completed during the MR acquisition; however, to simplify the process, dy-
namic FMT images were started after the MR scans were completed without
changing the animal position.
Dynamic FMT imaging began immediately after tail vein administration of
a solution containing both fluorescent tracers (0.2 nmol of each). These
measurements were acquired continuously over the course of 1 h after ad-
ministration at a rate of approximately 1 frame per 2 min for a total of 27
frames. Each frame of data consisted of 56 source–detector pairs (8 sources
and 7 detectors). Under normal static FMT imaging conditions, a set of ex-
citation measurements corresponding to the fluorescence emission meas-
urements is acquired to calibrate the data to the model and normalize
internal heterogeneities in optical properties. To reduce the per-frame ac-
quisition time, only two excitation measurements were collected for each
animal: one before tracer administration and the other between frames 15
and 16 during the dynamic fluorescence emission measurements.
MRI-FMT Data Processing and Image Recovery. The dual-tracer MRI-FMT image
recovery procedure involves processing and merging MRI and optical data. A
diagram of this procedure for one imaging frame in a dynamic sequence is
presented in Fig. 1. For reference, Fig. 1A shows a rendering of an animal
subject [rendered from the Mouse Whole Body (MOBY) phantom (47) and
shown in blue] in the MRI–optical interface. Eight optical fibers surrounding
the head transmit light to and from the tissue. Image processing began by
segmenting the T1-weighted Gd-enhanced MRI images into three regions using
MIMICS software (Materialise Medical Software): the tumor region as defined
by Gd enhancement, a normal brain region, and the rest of the tissue volume.
Fig. 1 B and C show coronal and sagittal MRI slices of one of the study animals
with volumetric renderings of the brain and tumor regions. Also shown are the
fiber positions (Fig. 1 B, red dots and C, red dots). From these volumetric seg-
mented regions, shown together in Fig. 1D, a 3D finite element mesh (∼15,000
nodes) compatible with the optical image recovery algorithms was created.
Data for an optical imaging frame in the dynamic MRI-FMT sequence
consist of 56 fluorescence emission spectra similar to the one shown in Fig. 1E,
solid black curve. Each measured spectrum is composed of the overlapping
emission spectra from the two fluorophores and the tissue autofluorescence.
To image the internal distribution of each tracer independently, the emis-
sion signals from each must be extracted from the measured spectrum. This
spectral discrimination was accomplished using a linear least-squares spec-
tral fitting algorithm, which uses premeasured basis spectra to extract in-
dividual fluorescent sources, a procedure described in previous publications
(29, 45, 48–50) and illustrated in Fig. 1 E–H. The basis spectra used here in-
clude the tissue autofluorescence spectra measured from mice before dye
administration (Fig. 1F) and the fluorescence emission peaks of the two
tracers measured independently in tissue phantoms (Fig. 1 G and H). The
fitted spectrum shown in Fig. 1E, gray dotted curve, represents the sum of
the spectral components in Fig. 1 F–H extracted using the spectral fitting
routine. After the contribution of each tracer’s emission spectra was
extracted from the measured spectrum, the peaks were integrated to pro-
vide a single intensity value for each tracer. This analysis was repeated for
each source–detector pair, producing 56 optical data points for each tracer
and a single frame.
The optical and MRI data flows merge in the image recovery algorithm as
described extensively elsewhere (30, 45, 51). Briefly, this process involves fit-
ting the optical data measured on the tissue boundary to a numerical model
based on the diffusion approximation of photon propagation in tissue. The
finiteelementmeshcreatedfromtheMRIimage stack servesasthenumerical
template on which the optical images are recovered. Before image re-
construction, the fluorescence data were calibrated to the model using the
excitation measurements acquired during the imaging sequence, a process
described elsewhere (45). Images were recovered using the hard-priors ap-
proach, which operates under the assumption that the fluorescence activity is
homogeneously distributed in a given internal tissue region. This approach
dramatically reduces the number of unknowns in the image recovery process.
activity: one for each tracer per frame. Two dimensional slices from the 3D
volumes are shown in Fig. 1 I and J for the targeted and untargeted tracers,
respectively. Image reconstructions for both tracers were repeated for each
frame in the dynamic series, although because the animal did not move
during the length of the MRI and optical scans, a single MRI-derived Finite
Element Method (FEM) mesh was used for all frames from an animal subject.
In total, the fully processed imaging results for each animal consisted of 27
frames of 3D fluorescence yield images for both fluorescent tracers—a total
of 54 images per animal. The values reported in these images were used in
the reference tracer model to extract values of binding potential.
Planar Imaging Experiments. Imaging was performed by scanning exposed
tumors with a dual-channel Licor Odyssey fluorescence scanner (Licor). The
animal subjects were anesthetized by i.p. administration of 100:10 mg/kg
ketamine:xylazine, and an incision was made in the skin to expose the tumor
mass and surrounding tissue. The animals were then positioned on a glass
slide on the imaging surface of the fluorescence scanner, with the exposed
tissue contacting the glass. Because of the hardware configuration of the
Odyssey instrument, which acquires dual-channel fluorescence images at 700
and 800 nm, the Affibody negative control imaging agent was conjugated to
Licor IRDye 680RD rather than Alexa Fluor 750, which was used in the MRI-
was acquired, a 100 μL solution containing 0.1 nmol of each tracer was
injected into the tail vein, and imaging scans started immediately. Each dual-
channel scan of the animal was completed in 40 s, and these scans were
acquired every 3–5 min over the course of 50 min. A total of 13 dual-channel
frames was acquired for each animal.
After acquisition, the images were processed to produce average binding
potential values in the tumors of each animal. For each animal, a tumor
region was identified by manual inspection in the fluorescence image
sequences of both tracers. Average values of fluorescence intensity in the
tumor region were then extracted from each frame, producing a time series
of fluorescence intensity in the tumor for both fluorescence tracers. The time
Davis et al.PNAS
| May 28, 2013
| vol. 110
| no. 22
course data from both channels were then fit to the parameters of the
reference tracer model to determine binding potential.
Ex Vivo Quantification of Available Receptor Density in Orthotopic Tumors.
An ex vivo method to approximate binding potential (21) was applied
to intracranial tumors. This approximation is enabled by two additional
assumptions, namely that the contribution from the plasma is negligible
(which is generally the case several minutes after injection given the low
volume of plasma in the tissue) and that the two tracers behave identically
in the tissue, with the exception of binding. Provided that these assumptions
apply, it can be shown that the dual-tracer model reduces to
tracers from a single dual-channel image. Relaxing the requirement to acquire
dynamic measurements of tracer kinetics in the tumor allows the analysis to be
completed with ex vivo specimens, and thus, it provides an independent
method to validate BP values in tumors inaccessible to in vivo planar imaging.
This approach was applied to ex vivo slices of intracranial U251 tumors and
compared with the MRI-FMT imaging results. Details on the experimental
procedure and data processing used for this study can be found in SI Text, Ex
Vivo Quantification of Available Receptor Density in Orthotopic Tumors.
ACKNOWLEDGMENTS. This work was funded by National Institutes of
Health Grants R01CA109558 (to S.C.D. and B.W.P.), R01CA156177 (K.S.S.,
J.R.G., T.H., and B.W.P.), and U54CA151662 (to K.M.T., K.J.S., and B.W.P.) as
well as Department of Defense Award W81XWH-09-1-0661 (to S.C.D.).
1. Deschoolmeester V, Baay M, Specenier P, Lardon F, Vermorken JB (2010) A review of
the most promising biomarkers in colorectal cancer: One step closer to targeted
therapy. Oncologist 15(7):699–731.
2. Rosen MJ, Heniford BT (2005) Endoluminal gastric surgery: The modern era of mini-
mally invasive surgery. Surg Clin North Am 85(5):989–1007.
3. Sharma SV, Settleman J (2010) Exploiting the balance between life and death: Tar-
geted cancer therapy and “oncogenic shock.” Biochem Pharmacol 80(5):666–673.
4. van’t Veer LJ, Bernards R (2008) Enabling personalized cancer medicine through
analysis of gene-expression patterns. Nature 452(7187):564–570.
5. Samkoe Ks, et al. (2012) High vascular delivery of EGF, but low receptor binding rate is
observed in AsPC-1 tumors as compared to normal pancreas. Mol Imaging Biol 14:
6. Gerlinger M, et al. (2012) Intratumor heterogeneity and branched evolution revealed
by multiregion sequencing. N Engl J Med 366(10):883–892.
7. Gambhir SS (2002) Molecular imaging of cancer with positron emission tomography.
Nat Rev Cancer 2(9):683–693.
8. James ML, Gambhir SS (2012) A molecular imaging primer: Modalities, imaging
agents, and applications. Physiol Rev 92(2):897–965.
9. Massoud TF, Gambhir SS (2003) Molecular imaging in living subjects: Seeing funda-
mental biological processes in a new light. Genes Dev 17(5):545–580.
10. Weissleder R, Mahmood U (2001) Molecular imaging. Radiology 219(2):316–333.
11. Weissleder R, Ntziachristos V (2003) Shedding light onto live molecular targets. Nat
12. Jain RK (2001) Delivery of molecular and cellular medicine to solid tumors. Adv Drug
Deliv Rev 46(1-3):149–168.
13. Orcutt KD, Rhoden JJ, Ruiz-Yi B, Frangioni JV, Wittrup KD (2012) Effect of small-
molecule-binding affinity on tumor uptake in vivo: A systematic study using a pre-
targeted bispecific antibody. Mol Cancer Ther 11(6):1365–1372.
14. Schmidt MM, Wittrup KD (2009) A modeling analysis of the effects of molecular size
and binding affinity on tumor targeting. Mol Cancer Ther 8(10):2861–2871.
15. Thurber GM, Schmidt MM, Wittrup KD (2008) Factors determining antibody distri-
bution in tumors. Trends Pharmacol Sci 29(2):57–61.
16. Lammertsma AA, Hume SP (1996) Simplified reference tissue model for PET receptor
studies. Neuroimage 4(3 Pt 1):153–158.
17. Logan J, et al. (1996) Distribution volume ratios without blood sampling from
graphical analysis of PET data. J Cereb Blood Flow Metab 16(5):834–840.
18. Jain RK (2005) Normalization of tumor vasculature: An emerging concept in anti-
angiogenic therapy. Science 307(5706):58–62.
19. Maeda H, Wu J, Sawa T, Matsumura Y, Hori K (2000) Tumor vascular permeability and
the EPR effect in macromolecular therapeutics: A review. J Control Release 65(1–2):
20. Minchinton AI, Tannock IF (2006) Drug penetration in solid tumours. Nat Rev Cancer
21. Tichauer KM, et al. (2012) In vivo quantification of tumor receptor binding potential
with dual-reporter molecular imaging. Mol Imaging Biol 14(5):584–592.
22. Leuschner F, et al. (2011) Therapeutic siRNA silencing in inflammatory monocytes in
mice. Nat Biotechnol 29(11):1005–1010.
23. Lin Y, Thayer D, Nalcioglu O, Gulsen G (2011) Tumor characterization in small animals
using magnetic resonance-guided dynamic contrast enhanced diffuse optical to-
mography. J Biomed Opt 16(10):106015.
24. Corlu A, et al. (2007) Three-dimensional in vivo fluorescence diffuse optical tomog-
raphy of breast cancer in humans. Opt Express 15(11):6696–6716.
25. Milstein AB, et al. (2003) Fluorescence optical diffusion tomography. Appl Opt 42(16):
26. Ntziachristos V, et al. (2004) Visualization of antitumor treatment by means of
fluorescence molecular tomography with an annexin V-Cy5.5 conjugate. Proc Natl
Acad Sci USA 101(33):12294–12299.
27. Ntziachristos V, Tung CH, Bremer C, Weissleder R (2002) Fluorescence molecular to-
mography resolves protease activity in vivo. Nat Med 8(7):757–760.
28. Ntziachristos V, Weissleder R (2001) Experimental three-dimensional fluorescence
reconstruction of diffuse media by use of a normalized Born approximation. Opt Lett
29. Davis SC, et al. (2007) Image-guided diffuse optical fluorescence tomography im-
plemented with Laplacian-type regularization. Opt Express 15(7):4066–4082.
30. Davis SC, et al. (2010) Comparing implementations of magnetic-resonance-guided
fluorescence molecular tomography for diagnostic classification of brain tumors.
J Biomed Opt 15(5):051602–051610.
31. Hyde D, et al. (2009) Hybrid FMT-CT imaging of amyloid-beta plaques in a murine
Alzheimer’s disease model. Neuroimage 44(4):1304–1311.
32. Kepshire D, et al. (2009) A microcomputed tomography guided fluorescence to-
mography system for small animal molecular imaging. Rev Sci Instrum 80(4):043701.
33. Lin Y, et al. (2011) A photo-multiplier tube-based hybrid MRI and frequency domain
fluorescence tomography system for small animal imaging. Phys Med Biol 56(15):
34. McCann CM, et al. (2009) Combined magnetic resonance and fluorescence imaging of
the living mouse brain reveals glioma response to chemotherapy. Neuroimage 45(2):
35. Gibbs-Strauss SL, et al. (2010) Detecting epidermal growth factor receptor tumor
activity in vivo during cetuximab therapy of murine gliomas. Acad Radiol 17(1):7–17.
36. Carpenter G (1987) Receptors for epidermal growth factor and other polypeptide
mitogens. Annu Rev Biochem 56(1):881–914.
37. Ciardiello F, Tortora G (2001) A novel approach in the treatment of cancer: Targeting
the epidermal growth factor receptor. Clin Cancer Res 7(10):2958–2970.
38. Marshall J (2006) Clinical implications of the mechanism of epidermal growth factor
receptor inhibitors. Cancer 107(6):1207–1218.
39. Mendelsohn J, Baselga J (2003) Status of epidermal growth factor receptor antago-
nists in the biology and treatment of cancer. J Clin Oncol 21(14):2787–2799.
40. Ciardiello F, et al. (1996) Antitumor activity of combined blockade of epidermal
growth factor receptor and protein kinase A. J Natl Cancer Inst 88(23):1770–1776.
41. Friedman M, et al. (2008) Directed evolution to low nanomolar affinity of a tumor-
targeting epidermal growth factor receptor-binding affibody molecule. J Mol Biol
42. Lammertsma AA, et al. (1996) Comparison of methods for analysis of clinical [11C]
raclopride studies. J Cereb Blood Flow Metab 16(1):42–52.
43. Zhou M, et al. (1993) Real-time measurements of kinetics of EGF binding to soluble
EGF receptor monomers and dimers support the dimerization model for receptor
activation. Biochemistry 32(32):8193–8198.
44. van de Ven S, et al. (2010) A novel fluorescent imaging agent for diffuse optical to-
mography of the breast: First clinical experience in patients. Mol Imaging Biol 12(3):
45. Davis SC, et al. (2008) Magnetic resonance-coupled fluorescence tomography scanner
for molecular imaging of tissue. Rev Sci Instrum 79(6):064302.
46. Davis SC, et al. (2010) MRI-coupled fluorescence tomography quantifies EGFR activity
in brain tumors. Acad Radiol 17(3):271–276.
47. Segars WP, Tsui BMW, Frey EC, Johnson GA, Berr SS (2004) Development of a 4-D
digital mouse phantom for molecular imaging research. Mol Imaging Biol 6(3):
48. Mansfield JR, Gossage KW, Hoyt CC, Levenson RM (2005) Autofluorescence removal,
multiplexing, and automated analysis methods for in-vivo fluorescence imaging.
J Biomed Opt 10(4):41207.
49. Xu H, Kuo C, Rice B (2008) Improved sensitivity by applying spectral unmixing prior
to fluorescent tomography. Biomedical Optics (BIOMED) (Optical Society of America,
St. Petersburg, FL).
50. Zacharakis G, Favicchio R, Simantiraki M, Ripoll J (2011) Spectroscopic detection im-
proves multi-color quantification in fluorescence tomography. Biomed Opt Express
51. Dehghani H, et al. (2008) Near infrared optical tomography using NIRFAST: Algorithm
for numerical model and image reconstruction. Commun Numer Methods Eng 25(6):
| www.pnas.org/cgi/doi/10.1073/pnas.1213490110Davis et al.