1.

Introduction

In vivo observation and imaging of living subjects using fluorescence is one of the most rapidly evolving fields in clinical and experimental biology. Fluorescent probes offer a number of practical advantages in that they are inexpensive, present straightforward synthetic and conjugation chemistries, and emit detectable, nonionizing photons on excitation. A variety of strategies have been developed to image specific cellular and biochemical events in vivo using targeted and∕or activatable fluorochromes conjugated to biocompatible delivery vehicles suitable for systemic injection. Imaging of cell-surface receptors and antigens has been accomplished in animal models using fluorescently labeled antibodies and antibody fragments,^{1, 2} proteins that bind cell-surface moieties,³ and peptides that bind to cell-surface receptors.^{4, 5} Self-quenched fluorescent probes activated by specific proteolytic cleavage⁶ or specific nucleic acid binding⁷ have also been developed for in vivo use, allowing functional biochemical imaging of living subjects. Imaging of gene expression via fluorescence has been achieved using genetically encoded fluorescent reporters such as green fluorescent protein (GFP).⁸

With advances in fluorescent probe technology, corresponding improvements in fluorescence imaging have also been achieved. Fluorescence has historically been used as a means of contrast in microscopy,^{9, 10, 11, 12} a technique that has also been applied to living subjects through the use of intravital windows.^{13, 14} Photographic techniques such as fluorescence reflectance imaging (FRI) have been applied toward imaging macroscopic fluorescent signatures deeper in tissues.^{5, 8, 15, 16, 17, 18} Such methods have offered a simple means of imaging the distribution and∕or activation of fluorescent probes in living subjects, and can provide semiquantitative results when the sizes and tissue depths of the imaging targets are properly controlled. However, as FRI is a single projection technique, it is unable to provide fully quantitative results because of nonlinear depth-dependent photon absorption in the subject.^{19, 20, 21} To quantitatively image fluorescence in this situation, tomographic methods have been adopted. Hardware and software for in vivo fluorescence imaging have been developed, including means for localizing fluorophores in diffuse media^{22, 23} as well as tomographically resolving fluorophore distributions,^{24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37} as demonstrated with simulated and phantom data. Recently, these techniques have been applied to the observation of protease activity³⁸ and of chemotherapeutic action²¹ in mouse models. These efforts have resulted in the emergence of fluorescence-mediated tomography (FMT) as a technique for quantitative in vivo molecular imaging using fluorescent agents.

Phantom experiments using known fluorochrome concentrations have demonstrated the accuracy of FMT employing a normalized Born reconstruction strategy in measuring fluorochrome concentrations in homogeneous media.^{34, 38} However, it has not been established that the quantitative performance of fluorescence tomography observed in vitro extends to the in vivo situation. This validation includes not only evaluation of the quantitative accuracy of FMT for fluorescent objects in living subjects, but also determining how factors such as background fluorescence and local optical properties affect this accuracy. These questions are directly applicable to biological situations involving quantitation of a fluorescent target, such as a tumor that has been labeled through accumulation and∕or activation of a fluorescent probe. In such an experiment, one wishes to absolutely determine the fluorochrome concentration and hence probe accumulation∕activation within the tumor. This measurement should ideally be independent of low levels of background fluorescence in surrounding tissue as well as the optical properties of the tumor itself, which are in part determined by its blood content and vascularity.

The extent to which FMT satisfies these criteria was investigated in this study by imaging tumor-like phantoms containing known concentrations of fluorescent dye before and after subcutaneous implantation in nude mice. Comparisons were made between reconstructions of FMT data acquired before and after implantation of the fluorescent tubes, allowing estimation of the sensitivity of FMT to the optical heterogeneity of the surrounding mouse tissues. In addition, identical tubes were imaged in mice after intravenous injection of a fluorescent probe targeting phosphatidylserine, a cell surface moiety implicated in cellular apoptosis. The selection of this probe represents a realistic case of a nonactivatable probe that typically yields background fluorescence signals due to nonspecific biodistribution. This case allowed observation of the effects of background, nonspecific fluorescence on the quantitation of a fluorescent focus. Finally, the influence of local optical heterogeneities on quantitation accuracy were assessed by imaging implanted fluorescent tubes with varying ink concentrations. From these experiments, it was observed that large dataset fluorescence tomography using a normalized Born reconstruction methodology can offer reliable quantification performance in vivo.

2.

Methods and Materials

2.1.

Fluorescent Tube Construction and Implantation

A turbid fluorescent medium simulating fluorochromes embedded in tissue was created by adding Cy 5.5 fluorescent dye (peak excitation $670\mathrm{nm}$ , peak emission $710\mathrm{nm}$ ) to a solution of 1% intralipid and black india ink. The exact concentrations of Cy 5.5 and ink varied based on the experiment and are described in detail later. Translucent plastic tubes of inner diameter 1.5, wall thickness $0.25\mathrm{mm}$ , and length $8\mathrm{mm}$ , as shown in Fig. 1 , were sealed at one end and filled with $6\mu \mathrm{L}$ of the fluorescent medium. The end of the tube was then sealed by melting the plastic at the opening. All animals were treated in accordance with the regulations of Massachusetts General Hospital. Nude mice (nu∕nu, COX-7) were prepared for fluorescent tube placement by administration of general anesthesia through an intraperitoneal injection of ketamine $(80\mathrm{mg}∕\mathrm{kg})$ and xylazine $(12\mathrm{mg}∕\mathrm{kg})$ . A $5\text{‐}\mathrm{mm}$ incision was made in the skin at the superior aspect of the ventral mammary fat pad, and a subcutaneous pocket approximately $10\mathrm{mm}$ long was made by separating the skin from the underlying muscle. A fluorescent tube was completely inserted into the subcutaneous pocket, after which the mouse was imaged with the system and protocol discussed in the next section. The tube was then removed and the implantation and imaging procedure was repeated using a new tube, depending on the experiment as detailed next. A single dose of anesthesia for a mouse was sufficient for fluorescent tube implantation and the acquisition of all subsequent data. After completion of the imaging battery and while still under anesthesia, the mice were euthanized by an overdose of halothane.

Fig. 1

The experimental setup used for the in vivo measurements. A small plastic tube filled with a turbid intralipid∕ink∕Cy 5.5 solution is implanted in a nude mouse, which is then placed in an imaging chamber and bathed in an intralipid and ink optical matching fluid. A $670\text{‐}\mathrm{nm}$ wavelength laser diode, suitable for the excitation of Cy 5.5 fluorescent dye, is routed into a two-channel optical switch. Light is then routed either into a lamp adjacent to the detection apparatus, facilitating FRI, or into a multichannel optical switch that controls the excitation sources used for FMT. Photon detection is accomplished with a CCD camera focused on a window implanted in the imaging chamber. Appropriate bandpass filters are used to selectively detect fluorescent excitation or emission photons.

3.

Results

A total of 39 FRI∕FMT datasets were acquired from six nude mice over the course of the experiments described. A single FRI∕FMT examination, including insertion of the fluorescent tube, positioning of the mouse within the imaging chamber, and acquisition of a single FRI and FMT dataset, required approximately $10\min$ . Reconstruction of raw FMT data into a 3-D fluorochrome distribution was completed in approximately $5\min$ per dataset.

3.2.

Influence of Background Fluorescence

A slice from the FMT reconstruction of a $500\mathrm{nM}$ fluorescent tube implanted in mice preinjected with Cy 5.5-labeled Annexin V are shown in Fig. 3(b) , while a corresponding FMT slice acquired from a mouse that was not injected with a probe is shown in Fig. 3(a). The slices were taken from the same locations as those presented in Figs. 2(b) and 2(c). The reconstructed volumes in these images have dimensions of 2.5 by $2\mathrm{mm}$ , and are more spherical than those obtained with the $1000\mathrm{nM}$ phantom in Fig. 2. It is immediately apparent that in addition to the fluorescent tube, background heterogeneities are present in the reconstructions of data for which the mouse was initially injected with a fluorescent agent. The effects of these heterogeneities on quantitation of the fluorescent tubes are shown graphically in Fig. 3(d), in which average fluorochrome concentrations measured by FMT are plotted as a function of the tube Cy 5.5 concentration. The trace representing acquisitions made in a mouse with no circulating probe (circles) closely follows the actual fluorochrome concentrations, shown as a dotted line. The reconstructions of data acquired from a mouse injected with Cy 5.5-labeled Annexin V (squares), however, demonstrate a pronounced deviation from the actual values at lower tube fluorochrome concentrations $\left(0\text{to}500\mathrm{nM}\right)$ . For the $1000\mathrm{nM}$ fluorescent tube experiment, the traces for the two acquisitions converge.

Fig. 2

FMT of fluorescent tubes before and after subcutaneous implantation in a nude mouse. (a) A series of coronal slices from a reconstructed FMT dataset acquired from a fluorescent tube containing $1000\text{‐}\mathrm{nM}$ Cy 5.5 implanted subcutaneously in a nude mouse. Slice locations are given as distances from the glass imaging window, with the source plane being at a depth of $13\mathrm{mm}$ . The central slices contained no significant signals and are omitted. The scale bar in (a) applies to (b) and (c). (b) A $1.6\text{‐}\mathrm{mm}$ -deep reconstructed FMT slice acquired from a fluorescent tube containing $1000\text{‐}\mathrm{nM}$ Cy 5.5 immersed in an intralipid and ink solution. The FMT slice is shown overlaid on a white light image of the tube. (c) A corresponding reconstructed FMT overlay from a dataset acquired after implanting the same fluorescent tube subcutaneously in a nude mouse. (d) Average fluorochrome concentrations measured from fluorescent tubes containing 0-, 250-, 500-, and $1000\text{‐}\mathrm{nM}$ Cy 5.5 plotted against actual fluorochrome concentration. Measurements obtained with the tube immersed in a homogeneous medium are represented as circles, while measurements made after implantation of the tube in a living subject are shown as triangles. Error bars represent the standard deviation of three independent measurements.

A simple correction for background fluorescence was applied by subtracting $20\text{counts}∕\mathrm{s}$ from all fluorescence measurements prior to solving the inverse problem. Fluorochrome concentrations reconstructed from these corrected FMT datasets are shown in the red trace in Fig. 3(d). The result of the correction is that reconstructed FMT fluorochrome concentrations approach those obtained from the same fluorescent tubes imaged in mice without background fluorescence. The effect of this correction on the reconstructed image is shown in Fig. 3(c), which shows the FMT slice from Fig. 3(b) reconstructed after dc subtraction. The background heterogeneities in the image are reduced, and the reconstructed fluorochrome concentrations in the focus approach the values measured from the fluorescent tube in a mouse with no background fluorescence.

3.3.

Influence of Target Absorption

The effect of varying local absorption on FMT measurements was assessed by imaging fluorescent tubes containing $500\mathrm{nM}$ Cy 5.5 and 25, 50, or $100\mathrm{ppm}$ black india ink, as shown in Figs. 4(a), 4(b), 4(c) . The reconstructed volumes in these images have dimensions 2.5 by $2\mathrm{mm}$ , and appear similar in shape to those observed in Fig. 3. As depicted in the bar graph in Fig. 4(d), the average reconstructed FMT fluorochrome concentrations are approximately equal for the tubes containing 25 and $50\mathrm{ppm}$ ink, while the average fluorochrome concentration for the tube containing $100\text{‐}\mathrm{ppm}$ ink was approximately 20% larger. This difference is on the order of the measurement variability, as given by the error bars in Fig. 4(d). In contrast, a 40% decrease in photon counts was measured between FRI images collected from the fluorescent tubes containing 25 and $100\mathrm{ppm}$ ink.

Fig. 3

FMT of fluorescent tubes in nude mice with and without previous intravenous injection of Cy 5.5-labeled Annexin V. (a) A $1.6\text{‐}\mathrm{mm}$ -deep coronal reconstructed FMT slice of a fluorescent tube containing $500\text{‐}\mathrm{nM}$ Cy 5.5 subcutaneously implanted in a mouse with no injected fluorescent probe, shown overlaid on the corresponding white light image. (b) A corresponding reconstructed FMT slice acquired from $500\text{‐}\mathrm{nM}$ Cy 5.5 phantom subcutaneously implanted in a mouse injected with $2\mathrm{nM}$ of Annexin V Cy 5.5 $1.5\mathrm{h}$ before imaging. (c) An FMT slice reconstructed from the raw data in (b) after applying a correction for background fluorescence. The scale bar shown applies to (a), (b), and (c). (d) Fluorochrome concentrations measured from FMT reconstructions of fluorescent tubes in mice with and without background fluorescence and with and without background correction plotted as a function of the actual tube fluorochrome concentrations. Error bars represent the standard deviation of three independent measurements.

4.

Discussion

The quantitative accuracy of fluorescence tomography has been investigated previously.^{32, 34} However, these studies involved titrations of known amounts of fluorescent dye into a translucent tube immersed in a turbid solution of intralipid and ink. This idealized experiment utilizing a homogeneous medium may not reflect the performance of FMT in vivo, where biological optical heterogeneity and background fluorescence may complicate the measurement of fluorochrome concentrations. In the present study, we sought to generate a model system in which we could probe the quantitative performance of FMT in living subjects. By creating small clear plastic vessels that were filled with known concentrations of fluorescent dye, we were able to produce a controlled phantom that could be inserted subcutaneously to mimic a subcutaneous tumor model containing a known amount of fluorochrome. Furthermore, by adding intralipid and ink to the fluorochrome solution in the tube, we simulated the biological situation of fluorochromes embedded in a turbid, absorbing tissue. This experimental system allowed the evaluation of the response of FMT to biological heterogeneity, background fluorescence, and local tissue absorption.

FMT reconstructions of measurements taken of the same fluorescent tube both before and after implantation in a nude mouse demonstrate a linear relationship between measured and actual fluorochrome concentration. FMT reconstructions of fluorescent tubes imaged while in a mouse were observed to give larger fluorochrome concentration measurements than when the tubes are imaged in a homogeneous intralipid-ink medium, as seen in Fig. 2. The slope of the regression between measured and actual fluorochrome concentrations in the in vivo case was 24.7% greater than that in the ex vivo case. This effect can be attributed to the optical heterogeneity of the animal that typically yields a higher net absorption coefficient for photon paths passing through the middle of the torso, because of the heat and other highly absorbing structures, relative to photon paths through boundaries that generally contain low-absorbing adipose tissue. This quantification bias in the presence of increased absorption was also observed in the results of Fig. 4, and is discussed further below. This discrepancy could be addressed by incorporating independent reconstructions of background absorption, yielding more accurate forward models for fluorescence tomography. An alternative approach is to employ an experiment-based calibration factor derived from animal measurements to improve in vivo quantification. It is important to note that, similar to observations obtained in simpler phantoms, normalized Born methods yield robust reconstruction of subcutaneous fluorescent foci even in the presence of biological heterogeneity.

Fig. 4

FMT of fluorescent tubes with varying ink concentrations in nude mice. Coronal slices located $1.6\mathrm{mm}$ beneath the imaging window taken from FMT reconstructions of fluorescent tubes containing $500\text{‐}\mathrm{nM}$ Cy 5.5 and (a) 25-, (b) 50-, and (c) $100\text{‐}\mathrm{ppm}$ ink implanted in a nude mouse overlaid on white light images, with a scale bar that applies to each image. (d) A bar graph of the reconstructed fluorochrome concentration versus the actual concentration, with error bars representing the standard deviation of three independent measurements.

One of the most exciting possibilities for fluorescence tomography is its application in the detection of molecular-specific fluorescent probes in vivo, permitting noninvasive visualization of a range of biochemical and physiologic activity. In these applications, the sensitivity of FMT to background fluorescence due to nonspecific probe accumulation in tissues within the field of view becomes critical. In this study, we have addressed this issue by injecting mice with implanted subcutaneous fluorescent tubes with a standard dose of a representative fluorescent molecular probe, Cy 5.5 Annexin V. As the animal model does not contain a focus of apoptotic activity in which accumulation of this probe might be expected, the agent follows its normal biodistribution, resulting in distributed, diffuse fluorescent signals throughout the subject. As observed in Figs. 3(a) and 3(b), FMT measurements of this model to identify the presence of the subcutaneous fluorescent tube. However, there are also low levels of fluorochrome reconstructed in the background, which are not observed in reconstructions of measurements in which the mouse was not injected with a probe. Background fluorescence clearly affects measurement of focal fluorochrome concentrations in a structure of interest, as evidenced by Fig. 3(b), where a $500\mathrm{nM}$ target was measured to have a fluorochrome concentration of $600\mathrm{nM}$ . With increased target fluorochrome concentration, the effect of background fluorescence becomes less noticeable, as observed in Fig. 3(d).

To advance FMT as an imaging modality capable of quantitatively measuring flurochrome concentrations in heterogeneous, in vivo settings, the influence of background fluorescence on imaging results must be addressed. As a preliminary approach, we applied a correction in which a dc value of $20\text{counts}∕\mathrm{s}$ was subtracted from all fluorescence field measurements prior to image reconstruction. This procedure makes the approximation that the contribution of the background fluorochrome distribution adds $20\text{counts}∕\mathrm{s}$ for each source-detector pair. While this correction proved to be quite effective in this study for generating measurements that are unaffected by background fluorescence, as seen in Fig. 3(d), it is not a robust means of dealing with background signals. Further work is being conducted to develop adaptive techniques for measuring background fluorescence and subtracting its contribution to the measured fluorescence photon field. Effort is also being directed to develop more sophisticated treatments of photon propagation from small concentrations of distributed fluorochromes, and to develop to more accurately correct for the effects of a background of low intensity fluorescence in in vivo FMT datasets.

As discussed before, tomographic measurements of fluorescence can be greatly influenced by the optical properties of the surrounding medium. However, for measurements of tumors and tumor models targeted by fluorescent contrast agents, the local optical properties of the fluorescent object itself can also be of significance. Different tumor types can exhibit a wide variety of vascular profiles and therefore blood contents. As hemoglobin is a strong photon absorber in the near-infrared band, this can result in different tumors possessing distinct photon absorption properties. We simulated this phenomenon by constructing fluorescent tubes with varying concentrations of black india ink mixed with the intralipid and Cy 5.5 solution. Spectrophotometer measurements verified that the corresponding absorption coefficients of the resulting solutions ranged from $0.15{\mathrm{cm}}^{-1}$ ( $25\mathrm{ppm}$ ink) to $0.6{\mathrm{cm}}^{-1}$ ( $100\mathrm{ppm}$ ink). As seen in Fig. 4(d), our data suggest that the effect of varying local absorption on fluorochrome concentrations measured by FMT is limited. Tubes with ink concentrations of 25 and $50\mathrm{ppm}$ ( ${\mu }_a=0.15$ and $0.3{\mathrm{cm}}^{-1}$ ) reconstructed to average fluorochrome concentrations of 489 and $492\mathrm{nM}$ , respectively, while a tube with $100\text{‐}\mathrm{ppm}$ ink $({\mu }_a=0.6{\mathrm{cm}}^{-1})$ reconstructed to an average fluorochrome concentration of $582\mathrm{nM}$ . It is encouraging that given a four-fold change in the absorption coefficient of the fluorescent tube, the fluorochrome concentration measured by FMT changed only by approximately 20%. Corrections involving measured or a priori information on absorption heterogeneities in the subject may further reduce this variation and maximize the accuracy of quantitation for FMT. FRI, however, is unable to distinguish changes in absorption from changes in fluorochrome concentrations in the target, as evident by the 40% decrease in photon counts measured from the phantoms with absorption coefficients of 0.15 and $0.6{\mathrm{cm}}^{-1}$ . This benefit of FMT relative to FRI has recently been demonstrated by Ntziachristos by imaging tumors with different degrees of vascularity.²¹

The results shown herein were obtained with a second-generation FMT prototype employing 31 light sources and an array of $19\times 13$ detectors sampled from the CCD images acquired. These data were reconstructed using a homogeneous slab solution to the diffusion equation after the normalized Born formulation. This approach demonstrated reduced sensitivity on varying medium optical properties and is experimentally simple to implement, as it is independent of the source and detector gain and several coupling issues between tissue and the optical apparatus. Current research in acquisition hardware and reconstruction techniques for FMT can be expected to further refine the quantitative accuracy of fluorescence tomography in vivo. In particular, optimization studies are shedding light on the source and detector arrays that most effectively balance dataset size and information content for small animal imaging.^{41, 42} Also, progress has been made toward more accurately modeling the photon propagation boundaries encountered in FMT,^{36, 43} permitting fluorescence tomography without the assumption of a slab geometry, and therefore without bathing the subject in an optical matching fluid. Other groups have also had success in combining high resolution anatomic imaging techniques such as magnetic resonance imaging (MRI) with diffuse optical tomography, permitting consideration of tissue optical heterogeneity in the tomographic reconstruction process.⁴⁴ Finally, in contrast to the diffusion equation-based solutions used herein, the use of forward models based on the transport equation may be used to improve the fluorescence imaging problem,⁴⁵ especially in systems with small source-detector separations (less than 5 to 10 photon mean-free paths) and when imaging through void regions. More advanced theoretical approaches to the problem of photon propagation in tissue can improve the imaging accuracy, especially when imaging absolute quantities. These developments can further improve FMT robustness and image fidelity, and can be used to specifically detect and measure fluorescence in live tissues.

5.

Conclusion

We demonstrate the ability of a fluorescent molecular tomography system to measure fluorochrome concentrations in vivo. Furthermore, we show that the quantitative accuracy of FMT is robust in the presence of varying local photon absorption coefficients. The presence of background fluorescence, modeled here by a previous intravenous injection of a fluorescent-labeled molecular probe in a mouse bearing a fluorescent tube, is seen to offset fluorochrome concentrations measured by FMT for tubes with Cy 5.5 concentrations from $0\text{to}500\mathrm{nM}$ Cy 5.5. This effect is reduced by applying a simple correction for background fluorescence to the measured fluorescent field. Advances in FMT hardware and software will further reduce the sensitivity of the technique to background fluorescence and to local absorption. These findings encourage the use of fluorescence tomography as a quantitative technique for measuring molecular-specific fluorescent probes in vivo, supporting the development of this emerging branch of molecular imaging.