Abstract

The purpose of this study was to validate DCE-MRI parameters such as blood flow (), permeability surface area product (PS), fractional intravascular space (), and fractional extracellular extravascular space (), obtained using a standard two-compartment model against other established analysis methods and histological indices. DCE-MRI datasets of 28 mice implanted with various human cancer xenografts were acquired and analyzed. Statistically significant correlations were found between the parameters derived from the standard two-compartment model (, , , and PS) with the histological markers of intravascular and interstitial space and with the corresponding flow and permeability estimates obtained by the initial slope method and Patlak plot, respectively.

1. Introduction

It is important to validate the parameters derived from DCE-MRI against established techniques and to ascertain that the DCE-MRI parameters reflect the actual microcirculatory state and pathophysiology of the tumors imaged. Previous DCE-MRI validation studies [14] have attempted to correlate Generalized Kinetic (GK) model parameters, namely, the transfer constant and the fractional interstitial volume , with histological indices of the tumor microvasculature. However, according to GK model theory, incorporates both the effects of blood flow and vessel permeability [5] and a recent simulation study [6] has shown that has a significant positive correlation with blood flow, permeability, and blood volume, as well as a significant negative correlation with interstitial volume. With such ambiguities in interpretation, it might be difficult to identify appropriate markers for validation of .

The standard two-compartment model [710] describes tissue microcirculation with distinct parameters, namely, blood flow , vessel permeability surface area product (PS), fractional vascular volume (), and fractional interstitial volume (). These physiological parameters can be readily validated by comparison with appropriate histological markers or other established tracer techniques. More complex tracer kinetic models such as the distributed parameter models [10] are also capable of separately estimating blood flow and permeability. However, the distributed parameter models require rapid sampling of tissue tracer concentration curves and might not be suitable for use in the preclinical setting where the temporal resolution of the DCE-MRI scans is low compared to the rapid murine circulation.

In this study, we attempt to validate parameters of the standard two-compartment model derived from DCE-MRI of human cancer xenografts in mice by comparing and with histological markers; and by comparing and with the corresponding estimates obtained using a macromolecular tracer (Galbumin) with established tracer analysis techniques, namely, the initial slope method and Patlak plot [11].

2. Materials and Methods

2.1. Mice and Tumor Cell Line

This study was approved by the Institutional Animal Care and Use Committee and all mice were maintained according to the Guide for the Care and Use of Laboratory Animals published by the National Institutes of Health, USA. Male nonobese diabetic (NOD) mice (, 9-10 weeks old) with severe combined immunodeficiency (SCID, Animal Resources Centre, Canning Vale, Western Australia) were used as host animals for various types of human cancer xenografts. Details of the xenograft model and the formation of cancer cell lines can be found in [12]. Eight different types of human xenografts (Table 1) were implanted subcutaneously in the SCID mice. These tumors are known to differ in degree of angiogenesis.

2.2. DCE-MRI Protocol

DCE-MRI experiments were performed on a 7-T MR scanner (Bruker ClinScan, Bruker BioSpin MRI GmbH, Ettlingen, Germany) using a three-dimensional (3D) spoiled gradient recalled sequence (FLASH 3D) with the following parameters: repetition time TR = 3.04 ms, echo time TE = 1.23 ms, field-of-view FOV = 36 × 36 mm2, 128 × 128 matrix half-Fourier reconstructed to 256 × 256, 8 × 1 mm2 slices, final image resolution of 0.14 × 0.14 × 1 mm2, and acquisition time of 2 s.

Tissue precontrast and postcontrast values were estimated using the dual flip angle technique [13, 14]. To estimate the tissue native values, precontrast images were acquired with flip angles of 6° and 14°. For each flip angle, 5 repetitions were performed and averaged. One hundred and thirty postcontrast acquisitions were performed with flip angle 14° over a period of 260 s. 50 μL of gadoterate meglumine, INN (Dotarem, Guerbet S.A., Villepinte, France), was injected manually through the tail vein at the first postcontrast acquisition and was followed by a 50 μL saline flush.

2.3. Phantom Validation

At low concentrations, contrast agent concentration can be estimated by the difference in longitudinal relaxation rates where and are the postcontrast and precontrast (native) values, respectively, and denotes the longitudinal relaxivity. A phantom study was performed to validate the values estimated by the above variable flip angle protocol by comparing with those derived from inversion recovery measurements (inversion time TI = 50, 100, 200, 400, 600, 800, 1000, 1500, 2000, 4000 ms; repetition time TR = 5000 ms). The phantom consists of tubes of saline diluted with 0.05 g/L of copper sulphate to mimic value in human body tissue at 1.5 sec and filled with various concentration of Dotarem at 0.01, 0.05, and 0.1 mM.

2.4. Galbumin-Enhanced Imaging Protocol

Galbumin (BioPal, USA) is gadolinium-labeled bovine albumin with a molecular weight of ~74 kDa. Galbumin-enhanced scans commenced 15 min after the Gd-DOTA-enhanced scans, when the tissue enhancement due to Gd-DOTA has visibly faded. The Galbumin dynamic imaging protocol consists of two phases: an initial rapid imaging phase consisting of 15 acquisitions at 2 s temporal resolution using the 3D VIBE sequence mentioned above and a delayed phase of 15 acquisitions, with 58 s delay between each acquisition. A dose of 0.1 mL of Galbumin at 25 mg/mL was manually injected through the tail vein after the first set of dynamic images and followed by a 50 μL saline flush.

2.5. Data Processing

Postprocessing was performed offline on an Intel Core 2 Duo personal computer with Matlab (MathWorks, Natick, MA). For reduced inflow effects and wrap, only the 4 central slices from the imaging volume (of 8 slices) were selected for processing. Regions-of-interest (ROIs) corresponding to the lesions were manually outlined on the 4 central slices.

Galbumin concentration was estimated from the Galbumin-enhanced dynamic scans by the change in longitudinal relaxation rates after injection of Galbumin.

2.6. Tracer Kinetic Modeling

Consider a bicompartmental tissue system where the first compartment represents the vascular space and the second compartment represents the interstitial space. Assuming well-mixed compartments, tracer concentration in each compartment at time is given as follows [7, 8]: where denotes blood flow, and denote tracer concentration in the vascular and interstitial space, respectively, PS denotes the permeability surface area product, and denotes the tissue of density, which is set at 1 g/mL. is the fractional vascular volume, is the fractional interstitial volume, and denotes the arterial input concentration.

The operational equation for analysis of the DCE imaging data can be expressed as where denotes the tracer concentration in the tissue voxel and denotes the convolution operator. is the impulse residue response function, which describes the fractional amount of tracer remaining in the tissue owing to an impulse input at time . And is given by where and are the solutions of the following quadratic equation [15]:

2.7. Galbumin Kinetic Analysis

The Galbumin-enhanced scans were analyzed using more established methods to derive alternative estimates for blood flow and vessel permeability. Galbumin concentration-time data corresponding to the first rapid imaging phase was analyzed using the maximal slope method [16] to yield estimates for blood flow where is the arterial input function sampled from the Galbumin-enhanced scans.

Galbumin concentration-time data for the delayed imaging phase was analyzed using the Patlak plot method to derive the Galbumin influx rate constant , which is indicative of vessel permeability [17]: where is the ordinate intercept of the Patlak plot.

2.8. Histopathology
2.8.1. Tissue Sectioning

The tumor was manually dissected from the subcutaneous tissue. It was orientated in a plane similar to the imaging plane. The superior, inferior, left, and right borders were stained with different colors to allow orientation. The tumor was manually sectioned and subsequently fixed and embedded. A microtome section was obtained from the superior aspect of each section.

2.8.2. CD-31 Stain

Tumor cryosections (5 μm thick) were pretreated using optimized antigen retrieval methods and then immunohistochemical staining was performed. Vasculature was stained using a platelet/endothelial cell adhesion molecule/CD31 antibody and CD34 antibody (1 : 100 dilution; BD PharMingen, San Diego, CA). Visualization was done with automated stainer and biotin–avidin complex. The percentage of area stained for CD31 is used as a measure of the intravascular space.

2.8.3. Masson’s Trichrome

Interstitial collagen tissue was used as a marker of the interstitial space and was stained with the Masson’s trichrome stain. The digital image was threshold for interstitial collagen. The percentage of area stained for collagen is used as a measure of the interstitial space.

2.9. Microscope/Image Acquisition

Digital images of the tumor slices were created at a magnification of ×400 with a Nikon microscope (Nikon Instruments, Melville, NY). Image analysis for vessel counting was done using Nikon imaging software (NIS-Elements Basic Research 3.0, Nikon Instruments, Melville, NY). The software allowed the application of standardized computational algorithms as well as review and refinement of the results of those algorithms by human operators [18]. A camera resolution of 1300 × 1028 pixels would result in a pixel size of 0.240 μm/pixel. The pixel size of the DCE-MRI image was calculated by dividing the field-of-view with the resolution of the matrix. A grid sized to the DCE-MRI pixel is overlaid onto the digital histopathology image. Subsequent analysis (such as vessel count or percentage area of staining) is calculated with respect to each grid box (which correspond in size to the DCE-MRI pixel).

2.10. Statistical Analysis

Three hotspot areas were chosen in each tumor and the median value of each ROI was taken as its representative value. Histological sections were orientated according to the DCE-MRI images so that they will correspond to the same plane.

Correlation analysis using Pearson correlation coefficient was performed to compare the parameters derived from the standard two-compartment model (, , , and ) with the histological markers of intravascular and interstitial space and with the corresponding flow and permeability estimates obtained by the initial slope method and Patlak plot, respectively. All statistical analyses were performed using STATA v. 10 (StataCorp LP, TX), assuming a two-sided test at the conventional 0.05 level of significance.

3. Results

Parameters of DCE-MRI sequences used in human studies and its translated version in animal studies were given in Table 2. The sequences were based on a similar basic FLASH sequence. The values of TR and TE were set to be as similar as possible within the allowances given by the machine.

Comparison between values derived by the inversion recovery sequence with those estimated by the dual flip angle method was given in Figure 1. A linear correlation with slope of 1.22 and negligible ordinate intercept was obtained. Hence a reasonable precision of values estimated by the latter method was acquired and indicated the reliability of the estimated concentration values in the mouse tissue.

The various microcirculatory parameters estimated by the standard two-compartment model for each type of xenograft are presented in Table 3. Blood flow is highest in the renal xenografts, reaching 164.15 ± 1.05 mL/100 mL/min, and lowest in the HCC, GIST, and colorectal xenografts at around 25–30 mL/100 mL/min.

Correlations between the microcirculatory parameters derived by the standard two-compartment model with their respective immunohistochemistry staining were presented in Figure 2. Statistically significant correlations () were found for each parameter.

An example of a hepatocellular carcinoma xenograft is shown in Figure 3. Histological sections of the tumor with CD31 and Masson’s trichrome staining are shown in Figures 3(b) and 3(e). The hotspots of each staining are shown in Figures 3(c) and 3(f). The blood vessels and the interstitial collagen tissue are outlined in light green in the hotspot. The percentage of area of blood vessels is 3.3% and the interstitial volume is 18.55%. The values are similar to those obtained with DCE-MRI, as shown in Figures 3(a) and 3(d), with percentage of blood volume measured at 3.499% and interstitial volume at 18.435%.

The maps of maximum slope and influx rate constant derived from Galbumin-enhanced scans are shown in Figures 4(b) and 4(d). The maps of blood flow and permeability derived from DCE-MRI are shown in Figures 4(a) and 4(c).

4. Discussion

Validation studies of plasma flow in DCE-CT have been performed using microspheres. Stewart et al. [19] derived the hepatic arterial blood flow () from a dual-input version of the adiabatic tissue homogeneity model and compared it with the result obtained from radiolabeled microsphere studies. A strong correlation was observed between values from both techniques, with ().

Water PET has been used to validate plasma flow in DCE-CT. Bisdas et al. [20] used a distributed parameter (DP) model for tracer kinetic analysis in human brain studies to validate the derived perfusion values with PET scans and significant correlation was found with to 0.79 (-values ranged from 0.45 to 0.79) and . -PET has also been used to validate tumor blood flow derived from DCE-MRI. Nevertheless, was often used as a surrogate of blood flow and not directly [21]. Similarly, in another study using human melanoma xenografts, images of (where is the extraction fraction) were obtained by subjecting DCE-MR images to Kety analysis and compared to as a perfusion tracer [22].

Unlike in DCE-CT, there has not been a study that validates estimated by DCE-MRI directly, as previous studies used as a surrogate of blood flow. In this study we attempted to estimate vascular flow directly with a macromolecule Galbumin using initial slope method. The good correlation between and vascular flow derived from the initial slope method indicates the feasibility of using as an estimate of vascular flow.

has also been previously used to validate permeability. Ferrier et al. [1] used as an estimate of vascular permeability and compared them to estimates obtained using [14C]aminoisobutyric acid quantitative autoradiography ([14C]AIB QAR), which was an established method of evaluating blood-tumor barrier permeability. Significant correlation was found between and (, ).

Nevertheless, it is more difficult to validate vascular permeability in general. The blood-to-brain influx rate constant () as derived from the Patlak plot was the established estimate for vascular permeability [17]. The good correlation between and indicates the feasibility of using as an estimate of vascular permeability.

So, instead of using which can mean either flow or permeability, in this study we have shown the ability to measure flow and permeability separately and validate them independently.

Egeland et al. [23] compared the fractional volume of the extravascular extracellular space (EESF) derived by DCE-MRI with histology. They found that the numerical values of the DCE-MRI-derived parameters were not significantly different from the absolute values of tumor blood perfusion or fractional volume of the extravascular extracellular space in any of the tumor lines. Similarly, Benjaminsen et al. [24] investigated whether Gd-DTPA-based DCE-MRI can be used to assess the EESF of tumors using amelanotic human melanoma xenografts. Images of (the partition coefficient of Gd-DTPA) were obtained by Kety analysis of DCE-MRI data. Positive correlations were found between and EESF obtained by invasive imaging.

Gaustad et al. [25] investigated whether Gd-DTPA-based DCE-MRI can be a useful method for characterizing vascularity of tumors by comparing the images of as obtained by Kety analysis with Blood Supply Time (BST) images (i.e., images of the time from when arterial blood enters a tumor through the supplying artery until it reaches a vessel segment within the tumor) and morphologic images of the microvascular network which were produced by intravital microscopy. They found that the images mirrored the morphology (microvascular density) and the function (BST) of the microvascular networks well.

In this study, we compared both and directly with immunohistological chemistry stainings and not to invasive or morphological imaging.

This study suffered from the following limitations. As the parameters derived from DCE-MRI were compared with histology, there might be some distortions to the tissue during stain preparation. The orientation of scans might not exactly correspond with the histology slice as well. Difficulty in finding exact stains for flow and permeability also forced us to use a rather less than ideal substitute in the form of Galbumin (and its respective from Patlak plot).

5. Conclusion

We have shown that the parameters derived from the standard two-compartment model correlated with the respective values derived from histopathology.

It indicates the reliability of the proposed technique and potentials of applying this technique as a biomarker for preclinical drug developments, as it might allow monitoring of antiangiogenic therapy in murine models.

Conflict of Interests

The authors would like to declare the following conflict of interests. Septian Hartono, Choon Hua Thng, and Tong San Koh received research funding from Roche-Singapore Translational Medicine Hub. Laurent Martarello is employed by Roche-Singapore Translational Medicine Hub.