Methods that allow robust imaging of specific molecular targets and biological processes in vivo should have widespread applications in biology and clinical medicine. Here we use a quantitative, three-dimensional fluorescence-mediated tomographic technique (FMT) that enables rapid measurements of fluorochrome-based affinity tags in live xenograft models. We validate the method by showing its sensitivity in quantitating tumor angiogenesis and therapeutic modulation using an anti–vascular endothelial growth factor antibody. Furthermore, we show the feasibility of simultaneous multichannel measurements of distinct biological phenomena such as receptor tyrosine kinase expression and angiogenesis. FMT measurements can be done serially, with short imaging times and within the same live animal. The described method should be valuable for rapidly profiling biological phenomena in vivo.
A number of high-resolution microscopic imaging techniques have recently been developed to study molecular events in vivo. In particular, intravital fluorescence microscopy ( 1– 3), confocal laser scanning microscopy ( 4– 6), multiphoton laser scanning microscopy ( 7, 8), and in situ scanning force microscopy ( 9) have recently been introduced. Collectively, these methods have allowed striking insights into molecular targets and physiology. Despite their exceptional spatial resolution, these methods lack the millimeter/centimeter depth penetration and capability of surveying pathologies in their entirety. This is particularly important for cancers as they have been shown to be spatially and temporally heterogeneous and microscopic snapshots may not suffice to capture molecular events occurring throughout an entire tumor. Global quantitative measurements would be particularly important for assessing treatment efficacy and for studying complex phenomena such as angiogenesis, apoptosis, and gene regulation.
For the above reasons, there has been much interest in extending optical measurements to the whole body in experimental animals ( 10) and even humans ( 11). It has previously been shown that photonic measurements can be made using near IR light for improved tissue penetration and applying sophisticated reconstruction algorithm ( 12– 14). Whereas most reports to date have relied on the absorption of near IR light (e.g., by oxyhemoglobin and deoxyhemoglobin), exogenous near IR fluorochromes have been applied more recently. One particularly enticing approach has been the use of quenched near IR fluorochromes (e.g., protease activatable near IR probes), which minimizes signal contribution from nontarget tissues ( 15, 16) and simplifies reconstructions. However, given the large number of nonenzyme protein targets, fluorescently labeled antibodies ( 17– 19), and fluorescent proteins ( 20, 21), there is a clear need to develop tomographic methods that would allow quantization of continuously bright fluorochromes in vivo.
A number of previous reports have described fluorescence tomographic imaging systems of various sensitivities and resolutions ( 22, 23), whereas a few have focused on high spatial resolution imaging systems for their ultimate use in mouse models ( 10). Most of the described reports to date have dealt with system optimization ( 13, 24) and/or characterization of imaging systems using phantoms ( 25). However, there is a lack of biological validation studies, to assure that observed measurements indeed correlate with biological processes under consideration. The goal of this study was therefore to validate a new high-resolution, multislice fluorescence-mediated tomographic (FMT) imaging system designed to sense affinity tagged nonquenched fluorochromes (e.g., tagged antibodies) in live mice. Compared with previous reports ( 10, 26), the current system has improved spatial resolution, dual wavelength capability and is highly accurate in quantifying brightly fluorescent molecules. The method has the advantage of centimeter (rather than micrometer) depth penetration, submillimeter spatial resolution, and minute acquisition time for entire data sets. Together with the emergence of an array of imaging probes ( 17– 19, 27) and fluorescent proteins ( 5, 20, 21), the technique can now be used to rapidly quantitate variables in organs and tumors at time scales previously not possible. In this report, we validate the method and apply it to measurements of angiogenesis and a receptor tyrosine kinase (HER-2/neu) in murine models. These variables were chosen because of their proven clinical importance in the outcome of patients with breast cancer ( 28).
Materials and Methods
Fluorescence-mediated tomographic imaging system. All fluorescence imaging was acquired using a modular home-built scanner capable of acquiring transillumination, reflectance, and absorption data. Components of this device had previously been described ( 10) and used to measure a unique set of protease activatable imaging probes such as cathepsin reporters ( 16). Image data sets were reconstructed using a normalized Born forward model ( 24). Details of the algorithm have been published before ( 26).
Excitation laser diode sources included a 672-nm laser and a 748-nm laser (BW Tek, Newark, DE). The excitation system consisted of 46 fibers spread over a 20 × 20 matrix in a slab geometry imaging chamber. For each fiber, four different sets of data were acquired (intrinsic fluorescence, extrinsic fluorescence, and intrinsic and fluorescence noise) via an ultra-low noise cooled CCD camera (Model 7471, Roper Scientific, Trenton, NJ). Image calibration was done with phantoms of capillary glass containing known concentration (100-4,000 μmol/L) of Cy5.5 and AlexaFluor 750 (AF750, Molecular Probes, Eugene, OR). Images were acquired using a scattering media consisting of 1% Intralipid and 0.5% ink. The imaging probes used in this study were a long circulating dextranated magnetofluorescent nanoparticle (CLIO) containing Cy 5.5 (CLIO-Cy5.5; ref. 29) or near IR fluorochrome labeled long circulating synthetic graft copolymer (Angiosense-750, VisEn Medical, Inc., Woburn, MA) for use as a vascular marker, and fluorescently labeled herceptin (Herceptin-Cy5.5; Trastuzumab, Genentech, San Francisco, CA). The latter was synthesized by reacting column purified herceptin with Cy 5.5 (Molecular Probes) at a 4 fluorochrome/antibody molar ratio and purifying the reaction product on a Sephadex G50 column. The final conjugation ratio of fluorochrome to antibody was 1.05.
Correlative in vivo studies. Magnetic resonance imaging of tumor vasculature was done on a 4.7-T Bruker imaging system (Pharmascan, Karlsruhe, Germany). T2-weighed sequences were obtained before and after i.v. injection of CLIO-Cy5.5 (10 mg Fe/kg), a long circulating dextranated nanoparticle ( 29). The MR sequences were acquired with the following variables: TR = 2,000 ms, 16 different TE values from 6.5 to 104 ms, flip angle = 90 degrees, matrix size = 128 × 64, number of average = 4, field of view = 4.24 × 2.12 cm, slice thickness = 0.8 mm. Vascular volume fractions were obtained from T2* calculations using CMIR-Image as described elsewhere ( 30) and normalized to muscle.
Single photon emission computed tomography (SPECT) data were acquired on a combined high resolution SPECT-CT scanner (XSPECT, Gamma-Medica, Northridge, CA), equipped with a dual-gamma camera SPECT system as well as X-ray fan beam CT scanner. Radiolabeled RBC were prepared using a commercially available kit (UltraTag RBC, Mallinckrodt, Hazelwood, MO) used clinically and injected systemically via the tail vein (200 μCi/mouse). Photons from the tracer were collimated with a 1-mm pinhole collimator on each camera, resulting in a submillimeter resolution. After the SPECT acquisition (ROR 5 cm, 32 projections, 60 seconds per projection) a CT scan was acquired (256 projections, 50 kV, 500 mA) and coregistered with the SPECT data set for image fusion and exact three-dimensional anatomic localization of the tracer signal. SPECT data analysis was done by placing regions of interest on the tumor and muscles and evaluating the number of counts per region.
Intravital confocal microscopy was done using a Radiance 2100 system (Bio-Rad, Richmond, CA) equipped with four lasers, water immersion lenses, and custom-built heated stages for intravital imaging. 9L-GFP tumors were implanted in a dorsal skin chamber, 1 week before imaging session ( 2). Immediately before imaging, gas anesthetized animals received an i.v. injection of CLIO-Cy5.5 (10 mg/kg) through the tail vein. Images of microvasculature were obtained.
Animal models. The primary cell lines used in this study were obtained from the American Tissue Culture Collection (Manassas, VA). 9L is a rodent gliosarcoma and MDA-MB-468 is a human breast cancer cell line. 9L cells were stably transfected with EGFP for correlative intravital confocal measurements. Both cells lines were cultured in DMEM with 2 mmol/L l-glutamine and 10% heat-inactivated fetal bovine serum at 37°C in a humidified 5% CO2 atmosphere. One million cells were implanted into the mammary fat pad in athymic female nude mice that weighed 25 to 30 g. Cohorts of tumor bearing mice were treated with an anti–vascular endothelial growth factor (VEGF) antibody (Avastin, Bevacizumab, Genentech) twice weekly by i.p. injection of 0, 1, 3, 5, and 7 mg/kg. Each group consisted of three animals. All in vivo experiments were done on anesthetized [ketamine/xylazine 80/12 mg/kg i.p. or gas anesthesia (isoflurane 2%)] mice and all studies were approved by the ethics committee of the Institution's review board and are in accordance with NIH guidelines.
Histology. Tumors were excised, snap frozen, and cut into 8-μm sections. Air-dried slides were used directly for fluorescence microscopy of Herceptin-Cy5.5 distribution. H&E staining was done in all tumors. To identify vessels unequivocally, sections were also stained with a primary anti-CD31 antibody (Santa Cruz Biotechnology, Santa Cruz, CA) and revealed with a biotinylated secondary antibody (Abcam, Inc., Cambridge, MA) and developed for 8 minutes using the Vectastain Elite avidin-biotin complex kit (Vector Laboratories, Burlingame, CA). Control sections were processed identically, with the exception of omitting the incubation with the primary antibody. Immunohistochemical analysis was done with Image J to automate vessels counting. Three representative slices per tumor were analyzed. Results are given as mean vessel number ± SD per high power field (20×, HPF).
Imaging system and performance. Figure 1 summarizes the components of the imaging system, acquisition, reconstruction, and display. In brief, 46 excitation sources are dispersed over a 20 × 20 mm imaging area and each of them is illuminated at a given time. For each experiment, imaging data is acquired at the excitation (intrinsic) and emission wavelengths and recorded on a 512 × 512 element ultra-low noise cooled CCD camera. Noise measurements are also collected for noise subtraction. The acquisition time is ∼5 seconds per source yielding total acquisition times of ∼5 to 10 minutes per imaging session. The data sets are reconstructed using a normalized Born forward model ( 24) to yield 21 z-direction resolved data (analogous to X-ray tomosynthesis). Reconstructed voxel dimensions were 570 × 570 × 620 μm. Typical reconstruction times were <3 minutes for an entire three-dimensional FMT data set using a standard PC. The tomographic method employed here, the normalized Born approximation, is a technique that minimizes the sensitivity of the reconstructions to background optical heterogeneity. In absolute terms, the accuracy of the methods is such that the error is <20%. Better accuracy is also expected for relative changes on the same animal or animal to animal variability as similar optical heterogeneity is expected from animals of the same species ( 31).
A set of initial phantom experiments concluded that the system had a detection threshold for Cy5.5 (excitation, 675 nm; emission, 700 nm) of less than 500 fmol in the center of the imaging chamber, a linear and quantitative dose-response to extrinsic fluorochromes (r = 0.99, error <5% over 100-1,000 nmol/L) and the capacity to resolve fluorescent point sources at 600 μm apart in all three dimensions ( 10). Similar measurements were obtained with AlexaFluor 750 (excitation, 720 nm; emission, 750 nm) with spectral overlap of <1%. Importantly, the system was capable of resolving fluorescence measurements in the presence of strong absorbers such as melanin and hemoglobin. In one set of experiments, we showed that a 4-fold increase in hemoglobin concentration essentially had no effect on fluorochrome estimations at nanomolar concentrations.
Quantifying angiogenesis. Angiogenesis can be characterized by the functional (i.e., perfused) vascular volume per unit area ( 32). To determine, whether fluorescent tracers could be reconstructed in live animals, we injected nude mice (n = 9) with a dual-modality nanoparticle, CLIO-Cy5.5, which had previously been validated as an intravascular marker by confocal microscopy ( 29) and which can be used for simultaneous magnetic and fluorescent measurements ( 30). Using tracer amounts (2 nmol of fluorochrome/mouse), we were able to operate at 102 dB signal/noise levels and show that this was sufficient for reconstructions throughout entire 9L-GFP tumors ( Fig. 1). Because each tomographic slice contains quantitative information, entire tumor volumes and fluorochrome concentrations can be reconstructed three dimensionally.
Angiogenesis measurements are typically done microscopically by quantitating vessel area in two-dimensional projections ( 33), by radiotracer measurements of plasma volume or RBC ( 34, 35) or through the use of circulating markers such as fluorescent dextrans ( 7, 30). Using a 9L-GFP glioma tumor model, we compared the optical measurements to these different gold standards. Figure 2 summarizes correlations on nine different tumors between the optical measurements and CD31 microvascular density measurements (r = 0.88) and magnetic resonance imaging of vascular volume fraction (r = 0.95). The best correlation was observed between sensitive radiotracer measurements of tagged RBC and optical measurements (r = 0.97).
Therapeutic modulation of angiogenesis. To determine if global tumoral measurements could serve as early end points for measuring antiangiogenic effects, we tested an anti-VEGF antibody. Gliosarcomas were grown in mice until tumors reached 8 to 10 mm in diameter. Different groups of animals were then dosed with the anti-VEGF antibody ranging from 0 to 7 mg/kg i.p. twice weekly for 1 week. At the end of the treatment course, animals underwent optical measurements. As can be seen in Fig. 3 , there was a dose-response curve with an ED50 at 3.5 mg/kg, similar as reported previously using different efficacy end points ( 36, 37). At doses higher than 5 mg/kg, no incremental effect could be seen, a finding which was also corroborated by MVD measurements (542 ± 13 vessels/HPF, 487 ± 12 and 492 ± 13 for animals treated with either 0, 5, or 7 mg/kg, respectively). To further corroborate these findings, we also did confocal microscopy using chamber-implanted 9L-GFP tumors undergoing the same therapeutic regimen ( Fig. 3). Results from these qualitative studies mirrored those of FMT imaging (i.e., a reduction of FMT signal corresponds to a reduction of the number of vessels seen by confocal imaging). Finally, we imaged angiogenesis serially in a cohort of animals (n = 24) before and after inhibitor treatment, showing similar results.
Dual measurements. The above data indicates that FMT was sensitive, fast, and accurate in measuring angiogenesis with a perfusion marker. To expand on the number of simultaneous physiologic measurements, we hypothesized that it should be feasible to acquire multiple data sets of near IR reporters in different channels, similar as is done in multicolor confocal microscopy. We therefore prepared fluorescently labeled herceptin-Cy5.5 for visualization of HER-2/neu and coinjected it with the Angiosense-750 labeled vascular probe into two different xenograft models (9L as above and breast cancer MDA-MB-468; n = 5 for each group). Results from these experiments ( Fig. 4 ) and validation studies ( Fig. 5 ) show that both tumor types were vascularized (9L: VVF = 1.8 ± 0.3 %, MDA-MB-468: VVF = 3.1 ± 0.5 %). Conversely, the breast cancer model showed significant tumoral uptake of the fluorescently labeled herceptin whereas the glioma was essentially negative (MDA-MB-468: 14.7 ± 1.3 pmol versus 9L: 0.9 ± 0.3 pmol; P < 0.001). Histology and fluorescence-activated cell sorting analysis corroborated the imaging findings ( Fig. 5).
The examples provided show the utility of the volumetric fluorescence imaging technique. When combined with powerful new fluorescent probes and mouse models, the technique should prove particularly useful to rapidly study molecular and physiologic variables in whole animals. Because the technique uses three-dimensional reconstructions, it is superior to reflectance imaging that depends nonlinearly on depth and optical properties ( 10). We show that the technique can be used to study at least two different physiologic variables (angiogenesis and expression of HER-2/neu) simultaneously. We envision that the technique could be easily adapted to other measurements such as receptor profiling ( 38), proteolysis ( 39, 40), metastases formation ( 41), or protein-protein interactions ( 42), for example. Three-dimensional resolution and whole animal depth offer the unique ability to study tumors orthotopically, in transgenic models and in their entirety. Given fast image acquisition and reconstruction times, the method should be particularly useful as a means for higher throughput in vivo drug screens. As new antitumor treatments against multiple targets are developed, the possibility to measure early treatment response by FMT becomes very attractive.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
We thank Sarah Rhee and Eric Swart for assistance with magnetic resonance imaging, Pratik Patel for confocal acquisitions, Gordon Turner and Antoine Soubret for assistance with fluorescence experiments, and Fred Reynolds for synthesis of imaging probes.
- Received February 3, 2005.
- Revision received April 7, 2005.
- Accepted April 29, 2005.
- ©2005 American Association for Cancer Research.