2.1.

Xenograft Tumor Model

Seven-week-old male nude mice (BalB/cAnNCriBgi-nu; Charles River Japan, Yokohama, Japan) were used in all of the experiments. In experiments with dynamic fluorescence imaging (DyFI), mice received subcutaneous injections of 1 × 10⁷ U251MG cells into the right lower flank under ketamine/xylazine anesthesia. The animal care and experimental procedures were performed with the approval of the Animal Care Committee of KAIST (Daejeon, South Korea).

2.2.

Dynamic Fluorescence Imaging

The system for DyFI was built to allow serial acquisition of ICG fluorescence emission with high sensitivity and included a charge-coupled device (CCD) camera (PIXIS1024; Princeton Instruments, Princeton, New Jersey), two 760-nm light-emitting diode (LED) arrays (SMC760; Marubeni America, Santa Clara, California) to excite the ICG dye, a near-infrared (NIR)-sensitive right-angled mirror (W47-308; Edmund Optics, Singapore), two 760 ± 20-nm excitation bandpass filters (Asahi Spectra USA, Torrance, California), and an 830 ± 20-nm emission bandpass filter (Asahi Spectra USA), as shown in Fig. 1a. The LED arrays were adjusted to gain a uniform illumination field, and the mirror was introduced to visualize both dorsal and lateral sides of the animal simultaneously [Fig. 1b]. The optical setup was shielded with a light-tight box to ensure high signal-to-noise ratios.

Fig. 1

DyFI. (a) Schematic diagram of the optical setup for DyFI. (b) A bright-field image of a tumor xenograft mouse and a temporal sequence of fluorescence images after the intravenous injection of ICG at 0 s. The box with the dashed line indicates the region of interest in the fluorescence images. The three colored dots (red, green, and blue) indicate the region of interest in (d) with the corresponding colors. Scale bar, 1 cm. Five representative time points are shown. (c) Modeling for ICG pharmacokinetics. Three main determinants consisting of elimination, perfusion, and permeability are considered in the model. For elimination, the constant fractional elimination of ICG from plasma by the liver and the kidney is represented by τ (elimination time constant). For perfusion, flow rate (F) is introduced to describe the transport of blood from the arterial compartment to the capillary and venous compartments. For permeability, bidirectional transport of ICG between plasma and the extravascular space is described by two parameters, each representing extravasation (K _ext) and intravasation (K _int). (d) Regression analysis of time-series ICG fluorescence to the model. Temporal ICG dynamics of the three regions indicated in (b) with colored dots (red, green, and blue), their fitting curves with the corresponding colors, and a table for the extracted parameters are shown [Spearman correlation: r _S = 0.999 (P < 0.0001) for red, r _S = 0.961 (P < 0.0001) for green, and r _S = 0.995 (P < 0.0001) for blue]. r.f.u., relative fluorescence unit; V _A, volume fraction of the arterial compartment; V _CV, volume fraction of the capillary and venous compartment; V _T, volume fraction of the tissue compartment.

DyFI was performed when the tumor size reached approximately 1 cm³. For time-series ICG imaging, mice were anesthetized with ketamine/xylazine, laid on the imaging system, and held with tape. Under the serial acquisition mode of a CCD camera at 3-s intervals, ICG (0.1 ml of 400 μmol/L ICG; Sigma Chemical, St. Louis, Missouri) was injected intravenously through the tail vein. The time-series image acquisition of the fluorescence emission lasted for 10 min after the ICG injection. Following the serial imaging, a bright-field image of the mouse was taken to aid in the selection of regions of interest.

Image processing and regression analyses were performed using matlab software (MathWorks, Natick, Massachusetts) functions, including nonlinear least-squares fitting (lsqnonlin). After loading 200 successive images (0 to 10 min at 3-s intervals), the average background noise was subtracted in each image. Using the lsqnonlin function on the images obtained from 1 to 4 min, the half-life of ICG was determined in the manually selected region where the ICG dynamics were almost completely exponential. Since the ICG dynamics contain dense information at early time points, the images were sampled sparsely at later time points with the experimentally determined variable sampling rate to ensure the best fit; usually, 1 out of 1 image until 1 min, 1 out of 3 to 5 images until 5 min, and 1 out of 5 to 10 images until 10 min were used. For each pixel, excluding the background, the temporal ICG dynamics were regressed to the devised mathematical model [Eq. 4] using the lsqnonlin function. The regression analysis in a pixel generates a set of parameters including volume fractions for each compartment, perfusion, permeability, and vascular density. To obtain each parameter map, the spatial distributions of the corresponding parameters were displayed using the imagesc function. For merged parameter maps, each parameter was normalized, mapped to red, green, or blue, superimposed, and displayed using the imagesc function.

2.3.

Anti-Angiogenesis Study

VEGF-Trap (25 mg/kg) was injected intraperitoneally when the subcutaneous tumor size became approximately 1 cm³, and DyFI was performed immediately before and 2 days after the treatment.

2.4.

Statistical Analysis

The data are expressed as means ±standard error of mean (SEM). Statistical differences were determined using the t-test or analysis of variance (ANOVA) as indicated, with P < 0.05 considered statistically significant.

3.1.

Development of Dynamic Fluorescence Imaging

To trace blood dynamics in vivo, we used the nontoxic tricarbonate fluorophore ICG. The action spectrum of ICG is in the near-infrared region and it has greater tissue transmittance and lower scattering compared to visible light. Using ICG, it is possible to visualize relatively deep-lying structures in vivo.¹³ When administered intravenously, ICG predominantly binds to serum proteins, especially albumin. Accordingly, the probe is suitable for studying macromolecular permeability, as well as the perfusion of tumor vasculature. The ICG–albumin complex is segregated in the liver and excreted by the hepatic or renal pathway. Since this elimination process is an exponential decay with a half-life of 3 to 5 min, the arterial input function, which is indispensable for pharmacokinetic analysis, can easily be determined with high accuracy.^{12, 14}

To perform time-series fluorescence imaging, we developed a reflectance NIR fluorescence imaging system as described in Sec. 2 [Fig. 1a]. After tail vein injection of ICG into an anesthetized tumor-bearing mouse, fluorescence emission was acquired with 3-s intervals for 10 min. The ICG profile in the tumor was distinctive with high but uneven fluorescence due to aberrant vessel formation. No apparent retention of ICG was observed in tumors or normal tissues, which might interfere with repetitive imaging [Fig. 1b].

The spatiotemporal dynamics of ICG were modeled mathematically using differential equations [Fig. 1c]. Three main determinants for ICG pharmacokinetics consisting of elimination, perfusion, and permeability were considered in the model. The arterial input function (C _A) was determined by the elimination of ICG in the liver and was expressed as an exponential decay function with the time constant τ. Although some studies assumed a biexponential model to consider the long- and short-term decay of ICG, since the long-term decay is on the order of hours, ICG decay for 10 min can be approximated to a single exponential Eq. 1. The half-life of ICG decay ranged from 3 to 5 min. Therefore, ICG concentrations in an artery (C _A) with time can be expressed as Eq. 1 (t is time; C ₀ is the initial concentration for ICG).

Transport of ICG through the vascular lumen, specifically between the arterial compartment to the compartments of the capillaries and veins (C _CV), is indicative of perfusion and was represented using Fick's law with the flow rate F [Eq. 2]. Permeability, indicating bidirectional transport of ICG between intravascular and extravascular compartments (C _T), was modeled with two parameters of extravasation (K _e) and intravasation (K _i) [Eq. 3].

To determine whether the model is relevant, we tested the fitness of the devised model with experimentally obtained time-series ICG fluorescence data. In both tumors and normal tissues, the curve fits matched the experimentally obtained ICG dynamic data, reflecting the relevance of the model [Fig. 1d].

Regression analysis of time-series ICG dynamics at a pixel generates a set of pharmacokinetic parameters comprising vascular density, perfusion, and permeability. By applying regression analysis in a pixel-by-pixel manner, we could map the spatial distribution of multiple pharmacokinetic parameters with high resolution [Figs. 2a, 2b, 2c]. We called this method dynamic fluorescence imaging.

Fig. 2

Spatial profile of vascular parameters in tumors. (a)–(c) Color-coded maps (left panels) and histogram analysis (right panels) for vascular density, perfusion, and permeability in the tumor. Since the vascular density map is dependent on excitation light intensity, vascular density in the shaded area (arrow) was underestimated. The regions in (a) demarcated by the dashed and solid lines indicate the tumor and normal tissue used for the histogram analysis, respectively. a.u., arbitrary unit.

3.2.

Measuring Vascular Parameters in Tumors Using DyFI

In multiparametric pharmacokinetic maps, tumors showed distinctive spatial profiles of functional vascular density, perfusion, and permeability compared to the normal counterparts. Vascular density was markedly enhanced but unevenly distributed, suggesting spatial heterogeneity of tumor-induced vessel formation [Fig. 2a]. Unexpectedly, high vascularity did not always couple to better perfusion in the tumors. While perfusion increased in some regions, it decreased in other regions, demonstrative of functionally inefficient angiogenesis [Fig. 2b]. Nevertheless, perfusion was usually enhanced at the periphery of tumors and decreased near the center of the tumors, which contained several poorly perfused foci. These ischemic foci often led to macroscopic necrosis at the surface, reflecting the relevance of measuring perfusion by DyFI and its diagnostic potential. Highly permeable vasculature was observed only in the restricted area of tumors [Fig. 2c]. The permeability values for tumor vasculature were comparable to those measured using other optical methods^{11, 12} and DCE-MRI using albumin-Gd-diethylenetriamine penta-acetic acid (DTPA).¹⁵

While vascular density and permeability were significantly increased in tumors compared to normal tissues, no significant change was observed in the mean perfusion rate. Since a broadly distributed shape was consistently observed in the histograms of tumors, we evaluated whether differences in spatial heterogeneity existed between tumors and normal tissues. Spatial heterogeneity was defined as the standard deviation of a pharmacokinetic parameter in the region of interest. For all of the pharmacokinetic parameters measured, tumors had significantly higher spatial heterogeneity compared to normal tissues (Fig. 3).

Fig. 3

Comparison of vascular parameters between tumors and normal tissues. (a)–(c) In tumors, vascular density and permeability were significantly elevated compared to normal tissues while perfusion was not significantly different (n = 6, Wilcoxon test). (d)–(f) spatial heterogeneity for each vascular parameter. Spatial heterogeneity is defined as the standard deviation of the parameter in the region of interest. For all pharmacokinetic parameters shown, tumors had significantly higher spatial heterogeneity than normal counterparts (n = 6, Wilcoxon test).

The spatially resolving, multimodal nature of DyFI allowed co-registration of the acquired pharmacokinetic maps without the need for correcting movement artifacts. Therefore, we studied whether spatial correlations exist among different pharmacokinetic parameters in tumors. Although high permeability and regional ischemia have already been reported as tumor characteristics,¹⁶ the spatial relationships between perfusion and permeability have not been elucidated. In the merged maps of perfusion and permeability, we found that the highly permeable regions co-registered with the ischemic regions [Fig. 4a]. The negative spatial correlation between perfusion and permeability was clearly demonstrated in tumors using regional analysis. When a tumor was divided into three concentric regions with the ischemic focus at the center, permeability increased significantly nearer the center, while perfusion decreased in this region [Figs. 4b and 4c]. This pattern was observed in five of six ischemic foci in five tumors.

Fig. 4

Negative spatial correlation between perfusion and permeability. (a) A merged map for perfusion and permeability. Green and blue indicate perfusion and permeability, respectively. Note that highly permeable regions co-registered with regions of low perfusion. (b) and (c) Spatial correlation analysis of perfusion and permeability. The tumor was divided into three concentric regions [indicated in the left picture in (b); red for region I, yellow for region II, and green for region III] and analyzed separately [Kruskal–Wallis ANOVA (compared to region I): *, P < 0.001; #, P < 0.001].

3.3.

Application of DyFI for Preclinical Study of Anti-Angiogenic Therapy

Since DyFI was successful for measuring tumor functional vascular parameters, we addressed whether DyFI might be a useful tool for a preclinical study of anti-angiogenic therapies. We designed a preclinical study to test the short-term effects of vascular endothelial growth factor (VEGF)-Trap, a soluble decoy receptor for VEGF.¹⁷ DyFI was performed twice in each animal, and the changes in perfusion and permeability were evaluated immediately before and 2 days after a single treatment with VEGF-Trap (n = 4). Although the tumors showed no apparent changes, the spatiotemporal ICG dynamics were notably altered by treatment with the VEGF-Trap, suggesting rapid remodeling of the tumor vasculature [Fig. 5a]. The most noticeable change was permeability. For all of the treated mice, highly permeable regions in tumors almost completely disappeared within 2 days of treatment. Mean permeability levels of the treated tumors became comparable to normal tissues [Figs. 5b and 6a]. This rapid normalization in permeability was consistent with a previous study that used a multi-tyrosine kinase inhibitor to target the same VEGF signaling pathway.⁸ No significant change was detected in mean tumor perfusion. However, significant decreases occurred in the spatial heterogeneity of perfusion (n = 4, P = 0.022, paired t-test), reflecting redistribution of blood flow and a noticeable, but nonsignificant, decrease in the ischemic area of all tumors [Figs. 6b, 6c, 6d].

Fig. 5

Rapid vascular normalization effect of VEGF-Trap on perfusion and permeability. DyFIs were performed at day 0 and day 2. VEGF-Trap was administered immediately after the first imaging. (a) Bright-field images and representative ICG dynamics. r.f.u., relative fluorescence unit. (b) Changes in the spatial profile of perfusion and permeability. Note that poorly perfused regions at day 0 became normalized after the VEGF-Trap treatment (arrows).

Fig. 6

Quantitative analysis of the rapid vascular normalization effect of VEGF-Trap. VEGF-Trap was administered immediately after the first imaging at day 0. (a) Quantitative analysis of permeability. (b) Quantitative analysis of perfusion. (c) Quantitative analysis of spatial heterogeneity of perfusion (n = 4, paired t-test). The spatial heterogeneity of perfusion was defined as the standard deviation of the perfusion parameters in the region of interest. (d) Quantitative analysis of the ischemic area. The ischemic area was defined using three criteria (regions with perfusion rates lower than 100%, 70%, or 50% of normal counterparts).