How stereological analysis of vascular morphology can quantify the blood volume fraction as a marker for tumor vasculature: comparison with magnetic resonance imaging

Abstract

To assess angiogenesis noninvasively in a C6 rat brain tumor model, the rapid-steady-state-T₁ (RSST₁) magnetic resonance imaging (MRI) method was used for microvascular blood volume fraction (BVf) quantification with a novel contrast agent gadolinium per (3,6 anhydro) α-cyclodextrin (Gd-ACX). In brain tissue contralateral to the tumor, equal BVfs were obtained with Gd-ACX and the clinically approved gadoterate meglumine (Gd-DOTA). Contrary to Gd-DOTA, which leaks out of the tumor vasculature, Gd-ACX was shown to remain vascular in the tumor tissue allowing quantification of the tumor BVf. We sought to confirm the obtained tumor BVf using an independent method: instead of using a ‘standard' two-dimensional histologic method, we study here how vascular morphometry combined with a stereological technique can be used for three-dimensional assessment of the vascular volume fraction (V_V). The V_V is calculated from the vascular diameter and length density. First, the technique is evaluated on simulated data and the healthy rat brain vasculature and is then applied to the same C6 tumor vasculature previously quantified by RSST₁-MRI with Gd-ACX. The mean perfused V_V and the BVf obtained by MRI in tumor regions are practically equal and the technique confirms the spatial heterogeneity revealed by MRI.

Introduction

In neurooncology, it is well established that tumor angiogenesis is correlated with malignancy and clinical outcome. Therefore, objective assessment of tumor angiogenesis can guide the therapeutic strategy and helps in assessing its efficacy.

As malignant tumors are spatially heterogeneous, imaging modalities covering the whole tumor extension are required. Repetitive examinations to study disease evolution or response to treatment require noninvasive methods. Biopsies for histologic analysis neither cover the whole tumor extent nor can they be performed with the frequency of noninvasive imaging.

Magnetic resonance imaging (MRI) methods have been developed that are capable of mapping physiologic markers of angiogenesis qualitatively or quantitatively over a large volume of interest. Blood flow, blood volume fraction (BVf) (Aronen et al, 1994), vascular permeability (van der Sanden et al, 2000), and vessel size index (Dennie et al, 1998; Julien et al, 2004) have been used as surrogate markers for tumor angiogenesis and have been found to be related to the tumor grade (Aronen et al, 1994). Quantitative MRI methods that measure cerebral BVf all require the injection of a blood-pool contrast agent (CA). However, tumor vessels are often abnormally permeable to clinically approved low-molecular CAs. When quantifying the tumor BVf, CA leakage before or during acquisition can result in overstimulation or underestimation, depending on whether the imaging method is based on longitudinal or transverse relaxation, respectively.

We recently developed the rapid-steady-state T₁ (RSST₁) method for cerebral BVf mapping (Perles-Barbacaru and Lahrech, 2007) and validated it in healthy rat brain using Gd-DOTA, a clinically approved low-molecular-weight CA (≈0.56 kDa), and P760, a macro-molecular (≈5.3 kDa) experimental CA from Guerbet (Aulnay-Sous-Bois, France), which are both intravascular in cerebral tissue with an intact blood–brain barrier. For tumor BVf measurement with the RSST₁ method, we investigated the blood-pool properties of an experimental CA Gd-ACX composed of gadolinium linked to a modified α-cyclodextrin (hexakis(2-O-carboxymethyl-3,6-anhydro)-α-cyclodextrin) in a C6 brain tumor model in rats (Lahrech et al, 2008). The cerebral BVf measured in brain tissue contralateral to the tumor could be validated by a second measurement in the same animals 1 hour later using Gd-DOTA. Although from the MRI point of view, there is evidence that Gd-ACX remains confined to the vasculature in the C6 tumor model (Lahrech et al, 2008), we sought to confirm the obtained tumor BVf using an independent method.

Along with autoradiography, histologic assessment of angiogenesis is generally the standard reference method for validating measurements obtained by newly developed methods. Histologic techniques can reveal many morphologic (size, shape, architecture, density) and functional (vascular volume fraction (V_V), permeability) parameters that are often altered in the tumor vasculature. The standard histologic parameter for angiogenesis assessment is microvascular density, but several investigators point out that microvessel density is not a direct correlate of blood volume (Hawighorst et al, 1998). It does not reflect information about vessel diameter, length, and perfusion status. In particular, tumor vessels are characterized by a large diameter distribution, tortuosity, and incomplete perfusion (Bernsen et al, 1995). In stereology, which provides quantitative three-dimensional (3D) information of objects from measurements on planar sections through them, the vascular area fraction (A_A) is a two-dimensional surrogate of the V_V (principle of Delesse). More recently, stereological principles have been used to correct for slice thickness (Pathak et al, 2001) and derive a truly 3D vascular V_V, which can be better compared with the BVf obtained by MRI. Correlations between such histologic vascular V_V and MRI-derived BVf have been found in various studies, but the absolute values are often still different by a factor of two or more (Douma et al, 2010; Valable et al, 2008).

In this study, the V_V is computed from the mean vascular diameter obtained by digital morphometry and the vascular length density (L_V) obtained by a stereological method (Adair et al, 1994). To make sure we measure two corresponding physical quantities, both BVf and V_V were measured in the same animals by MRI and quantitative analysis, respectively. As the percentage of nonfunctional vessels can be relevant in many tumors (Jain, 2001) and the MRI methods based on exogenous CA quantify only circulating blood, a fluorescent dye was injected intravenously to mark the perfused vessels in the histologic sections (van der Sanden et al, 2000).

The first goal of this study was to investigate the feasibility and limitations of the vascular morphometric analysis on simulated data and on immunohistochemically stained rat vasculature, in particular to show whether vessel diameters can be accurately estimated. The second goal was to compare the BVf obtained by MRI in the C6 tumor model with the obtained histologic parameters (vascular A_A and V_V) in the same animals. The results were also compared with published values obtained by alternative techniques for this tumor model.

Materials and methods

Animals

All animal experiments strictly conformed to the Guidelines of the French Government (decree 87-848 of 19 October 1987, licenses 38 07 19, A 38 516 10004, and B 38 516 10003 from the French Ministry of Agriculture), and were approved by the Scientific Review Board of the Department of Functional and Metabolic Neuroimaging of the French National Institute for Health and Medical Research (INSERM, Institut National de la Santé et de la Recherche Médicale).

In Vivo Blood Volume Fraction Measurement

Four Wistar rats were unilaterally implanted with C6 tumor cells as described in the study by Lahrech et al (2008).

Three weeks later, in vivo BVf maps were obtained by MRI. Rats (weighing 310 to 400 g) were anesthetized with isoflurane, equipped with a venous and an arterial cannula, thermoregulated, and placed in the prone position for the imaging experiment (Lahrech et al, 2008).

Ex Vivo Vascular Volume Fraction Measurement

At the end of the MRI experiment, an intravenous injection of Hoechst 33342 dye (0.2 mL, 6 mg in normal saline solution, Sigma-Aldrich, Saint-Quentin-Fallavier, France), a fluorescent marker for perfused microvessels, was administered. The brains were immediately excised, frozen in liquid nitrogen, and stored at −80°C, to stop diffusion of the Hoechst dye into the perivascular tissue.

Tissue Processing

In all, 20 equally spaced coronal sections of 10-μm thickness located within the 2-mm thick MRI slice for BVf mapping were cut on a cryotome HM 560 (Microm International, Walldorf, Germany) at −18°C with minimum exposure to white light to avoid degradation of Hoechst fluorescent intensity. The distance from the olfactory bulb on transverse MR images was used to locate the coronal MRI plane through the tumor (expected accuracy within 0.5 mm). One additional section was cut for standard nonspecific histologic analysis of the tumor regions with hematoxylin–eosin staining.

All sections were processed with goat anti-collagen IV (ref 1340-01, Southern Biotechnology Associates., Birmingham, AL, USA) and a second anti-goat antibody labeled with Alexa Fluor 546 (ref A11056, Molecular Probes, Eugene, OR, USA). The staining procedure with the anti-collagen IV antibody (diluted 1/100 in phosphate-buffered saline) takes 12 hours in a humid chamber at 4°C. Slides are then rinsed in phosphate-buffered saline and exposed for 1 hour at room temperature and in the dark to the secondary antibody conjugated to Alexa 546 (diluted 1/100). After a last rinse with phosphate-buffered saline, the slides were refrozen at −90°C.

Type IV collagen is a component of the basement membrane, which is present in arterioles, capillaries, and venules and not specific for functional or nonfunctional vessels. The DNA-intercalating Hoechst dye stains the nuclei of endothelial cells in blood vessels, which are perfused at the moment of injection. In tumors, adjacent cells are also stained because the dye has a similar molecular weight (0.56 kDa) as Gd-DOTA and therefore diffuses across a ruptured blood–brain barrier into the tissue.

Image Acquisition

Magnetic Resonance Imaging

Cerebral BVf maps were obtained at a magnetic field strength of 2.35 T in a 40-cm diameter horizontal bore magnet (Bruker Spectrospin, Wissembourg, France). A homogeneous radio frequency (RF) saddle coil was used for emission, and a surface coil for reception. BVf maps were obtained in a coronal plane through the largest extension of the tumor with a spatial resolution of 0.75 × 0.75 × 2 mm³ (Lahrech et al, 2008) using the RSST₁-MRI method (Perles-Barbacaru and Lahrech, 2007). During MRI examination, each rat underwent two CA injections: Gd-ACX (administered at a dose of 0.05 mmol/kg in terms of gadolinium) was the CA under investigation and has been found to remain confined to the blood pool even in the tumor tissue, allowing quantification of tumor BVf in this tumor model (Lahrech et al, 2008). The Gd-DOTA was injected 1 hour later at a dose of 0.1 mmol/kg to show that the tumor vessels were permeable to this low-molecular CA, and to confirm the cerebral BVf in the contralateral tissue having vessels with an intact blood–brain barrier.

Microscopy

The microvasculature was analyzed by epifluorescent microscopy. The tissue sections were scanned using an Olympus DP50 Microscope Digital Camera System (ColorView CCD camera; Olympus, Rungis, France) mounted on a Nikon Microscope Eclipse E600 (Nikon, Champigny Sur Marne, France) and interfaced with digital image processing software (AnalySIS 5.0, Soft Imaging System GmbH, Münster, Germany). Two different filter blocks (Nikon) were used for the acquisition of the Hoechst dye (UV-2E/C (DAPI)) and the Alexa Flour 546 dye (G-2E/C ((TRITC)). A composite image (mosaic) was acquired twice, during the same session, at a × 10 magnification using the DAPI filter first, followed by the TRITC filter. The slides were moved by a high-precision motorized scanning stage controlled by the image processing software. This procedure avoids the necessity of image registration when images have to be matched (Rijken et al, 1995).

Owing to computer limitations, only a maximum array of 6 × 6 or 5 × 7 images could be scanned, representing a total area of 9,146 × 6,865 μm² or 10,664 × 5,727/7,628 × 8,002 μm² with a pixel size of 0.77 μm. The composite image was computed from individual images using the AnalySIS software. In most cases, this hardly covered the massively enlarged right hemisphere on the coronal slices, almost entirely occupied by the tumor. Little healthy-appearing surrounding tissue is included in these images.

The acquisition and image processing time for one composite image was 20 minutes with each filter, i.e., 14 hours for one brain. For this reason, only the tumor-bearing hemisphere has been processed.

Image Analysis

Magnetic Resonance Imaging

The BVf maps were computed by averaging over N_R=200 repetitions, according to , where S_prei are signals acquired before CA injection and S_posti signals acquired within 2.5 minutes after Gd-ACX or Gd-DOTA injection. S₀ is the signal in a proton-weighted acquisition (Perles-Barbacaru and Lahrech, 2007).

Microscopy

Intensities of fluorescent images were analyzed using ImageJ software (National Institute of Health, Bethesda, MD, USA). A custom-made semiautomatic macro was used for image segmentation and feature extraction. A flow chart of the macro is given as Supplementary Material at the JCBFM web site.

Red, green, blue (RGB) images were converted to 8-bit gray-scale images. Regions of interest (ROIs) were defined in the central part (minimum diameter of 1.5 mm) of the tumor and in its periphery (region enhanced by Gd-DOTA minus the central region). As much tumor tissue as possible was included in the ROIs, as had been done when analyzing the MRI data, but when tumor size exceeded the borders of the composite image, the ROIs were smaller than those drawn on MRI-based BVf-maps. Care was taken to exclude large vessels, such as branches of the anterior and middle cerebral arteries.

A plugin for background correction was systematically used. This plugin corrects an uneven image illumination and contrast enhances the resulting image. To avoid user bias, binary masks of the vessels were obtained by thresholding according to the nonparametric approach described by Otsu (1979) where possible, but a manual thresholding was sometimes necessary in tumor ROIs containing very few vessels because of the low contrast and variable background intensity.

After thresholding, profiles composed of at least five adjacent pixels were closed by dilatation, followed by erosion and the vessel lumina were filled in. The binarized image and the original image were compared to verify the binarization process.

Image processing as described above was performed separately on anti-collagen IV-stained images defining the vessel outlines and on Hoechst-stained images carrying information about perfusion status and permeability of the tumor vessels (Figure in Supplementary Material). The final processing step of the Hoechst-stained image was application of a Gaussian blur with σ=10. Binarized Hoechst and anti-collagen IV-stained images were overlaid to select the perfused vessel sections using a Boolean AND operation.

Digital Vascular Morphometry

Particle analysis was carried out using the ‘Analyse Particles' and the ‘Particles8 Plus' plugins in ImageJ. These plugins derive a number of morphologic parameters for each vessel profile. In addition, ellipses are fitted to each vessel profile, by equalizing the second-order central moments of the ellipse to those of the pixel distribution (Jain, 1989). For each vessel profile i in an ROI, the following parameters were derived:

• A_i: the area of the profile (lumen and vessel wall),

• ∅IC_i: the diameter of the inscribed circle,

• a_i and b_i: the major and minor axes of the corresponding ellipse, respectively,

• the height and width of the bounding box (smallest rectangle that encloses the profile),

• BDTH_i: the breadth which is the widest distance perpendicular to the longest distance (feret) between any two points along the profile boundary (as defined by G Landini in the Particles8 Plus plugin, http://www.dentistry.bham.ac.uk/landinig/software/software.html).

For each ROI j the following parameters were obtained:

• A_ROIj: the area of the ROI,

• N_j: the number of profiles in the ROI.

From these parameters we calculated:

• P_A: the number of profiles per unit area , which is commonly considered a measure of the vascular density,

• A_A: the area fraction ,

• L_V: the length density , in units of 1/μm², where is the ratio of the ellipse axes, which is used as a weighting factor based on the assumption that vessels can be approximated as curved and randomly oriented cylinders (Adair et al, 1994), and

• V_V, the volume fraction

  

  where 〈d〉 is the mean vessel diameter.

The mean edge-to-edge vascular diameters including the endothelial wall were estimated by one of the four parameters: diameter of inscribed circle (∅IC), minor axis of the ellipse (b), small side of the bounding box (SSBB), and the breadth (BDTH) (cf. Figure 1). To our knowledge, there is no established morphometric parameter that can be obtained from the two-dimensional shape of vessel cross-sections and that reliably represents the vessel diameter, in particular in irregular vessel cross-sections such as occurring in tumor tissue. Therefore, we evaluated the four different parameters (∅IC, b, SSBB, and BDTH) for proper estimation of the diameter and the resulting V_V using an idealized and a more realistic numeric model of brain vasculature, as well as on plane sections through healthy rat brain vasculature.

Figure 1.

Scheme of an irregular vessel section and the diameters measured. b=minor axis of ellipse, SSBB=bounding box, ∅IC=diameter of inscribed circle, BDTH=breadth.

Three-Dimensional Numerical Geometric Model

The simulated 512 × 512 × 512 voxel binary model (Verant et al, 2007) was composed of a number of straight cylinders with varying diameters and orientations (random distribution with equal probability). The number of cylinders, as well as the minimum and maximum diameter could be chosen to yield the desired total cylinder volume. Three such simulations with different parameters were carried out (cf. Figure 2A).

Figure 2.

Validation study and parameter definition on three different idealized numerical geometric models of vascular structure. Model parameters: model a: 40 cylinders, diameter 14 to 22 pixels, cylinder volume 3.4% model b: 30 cylinders, diameter 14 to 22 pixels, cylinder volume 2.5% model c: 60 cylinders, diameter 10 to 16 pixels, cylinder volume 2.9%. (A) 3D view of cylinder model a. (B) Measured number of cylinder profiles per unit area, cylinder length density, and area fraction as a function of the total number of analyzed slices. The average value obtained from a number (s) of equally spaced slices normalized by the average value obtained from the total of 512 slices is displayed: , X=P_A, L_V, A_A, respectively. The error bars are s.d. resulting from averaging over the measures on the three different cylinder models. The gray lines illustrate the ±5% error interval. (C) Measured mean cylinder diameter in pixels using the four morphologic parameters. As repeated morphometric analysis on the same cylinder model yielded identical results, the error bars represent the s.d. resulting from averaging 20 equally spaced slices. The distribution of cylinder diameters and their mean is shown in shades of gray in the background. The mean b and BDTH are not statistically different from the expected mean cylinder diameter. (D): Relative difference of the measured volume densities V_V (equation 1) from the real cylinder volume density V_Vreal. Compared are the performances of the different morphologic parameters for diameter estimation, as well as the area fraction as a diameter independent estimate of volume density. The lines represent the mean difference. P_A=profiles per unit area, L_V=length density, A_A=area fraction, ∅IC=diameter of inscribed circle, b=minor axis of ellipse, SSBB=small side of bounding box, BDTH=breadth, V_V=volume fraction. 3D, three dimensional. ^*Significantly (0.05) different from mean value.

Three-Dimensional Numerical Model of the Brain Vasculature

Although the numeric cylinder model was mainly used to determine the most accurate measure of the average diameter, it is a maximally simplified vasculature model. Therefore, we generated a second 512 × 512 × 512 voxel numeric model, which reflects the vascular architecture more appropriately (cf. Figure 3A). Three different models were generated from optic microscopy z-stack acquisitions of the mouse cortical vasculature (Verant et al, 2007) by binarization, closing of the profiles, and increasing the spatial resolution to achieve pixel sizes in the order of 1 μm, comparable to the spatial resolution of the histology sections of the rat cerebral and tumor vasculature analyzed in this study. Although this numeric model accurately reflects the irregular shape, length, and branching of vessels, we have no precise previous knowledge of the average vascular diameter.

Figure 3.

Validation study and parameter definition on three different numerical geometric models of the vascular structure. Model a: vascular volume 3.1% model b: vascular volume 4.3% model c: vascular volume 4.4%. (A) 3D view of vascular model a. (B) Measured number of vascular profiles per unit area, vascular length density, and area fraction as a function of the total number of analyzed slices. As in Figure 2B, the normalized average value is displayed. (C) Difference of the measured volume densities (equation 1) from the real vascular volume density (cf. Figure 2D for details and abbreviations). 3D, three dimensional.

The two numeric models were also used to determine the minimum number of plane sections required to extrapolate reliable stereological quantities for rather complex structures. Therefore, up to 20 equally spaced entire slices were examined for each of the 3 cylinder models.

Identical image analysis as for the brain slices was performed omitting background correction and thresholding.

For the cylinder and vascular numeric models, the real total volumes V_V were determined as the fraction of white voxels over black voxels after analyzing all 512 slices.

Healthy Rat Brain Vasculature

The stereological algorithm and vascular morphometry were also applied to eight anti-collagen IV-stained sections of Wistar rat brains (n=4). The section thickness was 5 μm and the digitized image had a pixel size of 1.38 μm. Gray matter ROIs in the parietal cortex and in the striatum and a white matter ROI in the corpus callosum were analyzed.

C6 Tumor Vasculature

Vessel analysis as described above was applied on the anti-collagen IV-stained images and then repeated on the selected perfused vessels (Figure in Supplementary Material).

Statistical Analysis

Statistical comparisons were performed using GraphPad Prism version 3.02 for Windows (GraphPad Software, San Diego, CA, USA; http://www.graphpad.com). The mean cylinder diameters were compared with the hypothetical mean diameter using a one-sample t-test. The measured vascular parameters were compared using paired t-tests. The significance level was set at 0.05. Values are reported as mean±one s.d.

Results

Validation Studies

Three-Dimensional Numerical Geometric Model

The pilot study on the cylinder models showed that the measured or calculated parameters such as L_V, P_A (number of cylinder profiles per unit area), and mean cylinder diameter reach a constant value with an accuracy within ±5% when the number of test slices used for analysis exceeds ∼7 or 8. The convergence is shown in Figure 2B for L_V, P_A, and the cylinder A_A. To compare these parameters for the three different cylinder models, the values are normalized with regard to the average value obtained from all 512 slices. The measured anisotropy L_V/P_A averaged over the three models was 1.74±0.08.

The bar chart in Figure 2C compares the performance of four different morphologic parameters used to estimate the mean cylinder diameter. All four parameters resulted in mean values that were within the range of cylinder diameters (illustrated by the gray box), but the ∅IC underestimates the mean cylinder diameter and the SSBB tends to overestimate it. The methods using the b and the BDTH yield cylinder diameters that are not significantly different from the hypothetical mean diameter (18 pixels for the first 2 models and 13 pixels for the third illustrated by the dashed line in Figure 2C).

Length density correlated well with the number of cylinders in each model. Using the measured length density and the measured mean cylinder diameter, the total cylinder volume was calculated. The relative difference from the given cylinder volume is plotted in Figure 2D. When calculating the cylinder volume from L_V using the average b or the average BDTH as an estimate for cylinder diameter, the maximum error was within 16%.

Three-Dimensional Numerical Model of the Brain Vasculature

The same tendencies were found with the vascular geometric model. Figure 3B shows convergence for L_V, P_A, and the cylinder A_A starting at about six slices. Figure 3C shows how the V_V is underestimated and overestimated using the ∅IC and SSBB parameters for diameter estimation, respectively. The V_V calculated from the parameter b still tends to underestimate the real V_V, with a maximum error of −18% for the vascular geometric models.

From these studies, we conclude that BDTH is the parameter best suited to estimate vessel diameter.

Healthy Rat Brain Vasculature

Table 1 shows that the diameters of the healthy rat brain vasculature estimated by BDTH are in the range of those reported in the literature: 7 to 8 μm (Deane and Lantos, 1981; Pathak et al, 2001). No significant regional differences in this parameter were found in accordance with the literature (Schlageter et al, 1999; Weiss et al, 1982).

Table 1. Vascular parameters obtained in a healthy rat brain (n=4).

	BDTH (μm)	Lv (1/mm2)	PA (1/mm2)	VV (BDTH) (%)	AA (%)
Cortex	7.9±0.2	1173±147	425.7±35.9	5.80±0.56	4.60±0.54
Corpus callosum	7.5±0.4	490±197	170.6±56.8	2.10±0.71	1.68±0.55
Striatum	7.7±0.2	799±235	301.4±54.9	3.69±1.06	2.97±0.72

BDTH, breadth; L_V, length density; P_A, profiles per unit area; V_V, volume fraction; A_A, area fraction.

The vascular A_A (a measure independent of the estimation of the vessel diameters) yielded a significantly lower value. For in vivo studies, the results obtained with ∅IC, b, and SSBB are omitted because for these parameters the same tendencies as for the numerical geometric models were observed yielding even unphysiologically small and large vessel diameters for ∅IC and SSBB, respectively.

Table 1 also shows the vascular P_A and L_V, both being highest in the cortex and lowest in the corpus callosum in accordance with the literature.

C6 Tumor Vasculature

Microscopy

Histologic examination of the hematoxylin–eosin-stained C6 tumor-bearing sections revealed large areas of necrosis, predominantly in the central part of the tumor, containing very few and deformed vessels, cystic formations, and a narrow band of viable pleomorphic proliferating cells in the tumor periphery. In some animals, the histologically defined tumor size exceeded the one deduced from T₂-weighted and contrast-enhanced T₁-weighted images.

Magnetic Resonance Imaging

This study was carried out on very large advanced-stage tumors: in the slice used for BVf mapping by MRI, the area of contrast enhancement after Gd-DOTA injection covered ∼50% of the whole brain section (Lahrech et al, 2008) and a substantial shift of the median line was observed in all brains.

The PaCO₂ in tumor-bearing rats measured before both CA administrations was slightly higher (mean of 47.0±6.8 mm Hg) than usually observed in healthy rats under these experimental conditions. Magnetic resonance imaging-derived BVfs averaged for all eight tumor-bearing rats were (Lahrech et al, 2008): 1.32%±0.40% in the tumor periphery, 0.34%±0.28% in the tumor center, and 0.94%±0.16% in the contralateral hemisphere.

The BVf obtained by MRI in the four animals that underwent vascular morphometry was 1.38%±0.46% in the tumor periphery and 0.46%±0.29% in the tumor center. In the contralateral hemisphere, similar BVf values were obtained with both CA: 1.02%±0.25% and 1.09%±0.33% with Gd-ACX and Gd-DOTA, respectively. In Figure 4, images of a representative tumor-bearing rat brain are shown. The area enhanced by Gd-DOTA on the T₁-weighted image (Figure 4A) is outlined. The corresponding BVf map is given in Figure 4B together with a corresponding histologic Hoechst-stained image covering a large part of the tumor (Figure 4C). This advanced-stage C6 tumor model is characterized by a high BVf heterogeneity. The BVf map exhibits low blood volume in the center of the tumor ROI (Figure 4B) and some areas characterized by a higher blood volume in the periphery. Hoechst staining confirms a low vessel density in this ROI (Figure 4C). Only few vessels that are highly permeable to the Hoechst dye are visible. Interestingly, in this example, vasculature permeable to the Hoechst dye was observed well beyond the region enhanced by Gd-DOTA, showing that the tumor extent might be misinterpreted using contrast-enhanced MRI only.

Figure 4.

Coronal section through a representative C6 glioma. Images showing the extension of a typical C6 glioma of a rat. (A) T₁-weighted MRI acquisition 5 after intravenous Gd-DOTA administration. The contrast enhancement is used to define the outline of the tumor in the right hemisphere. The brain is outlined in white. (B) Corresponding BVf map (expressed in %) obtained by RSST₁-MRI revealing a high BVf heterogeneity in the tumor region. (C) Hoechst-stained coronal section at the same level (≈bregma) showing a few enlarged vessels in the tumor periphery and sparse-perfused and Hoechst-permeable vessels in the tumor center. Such vessels are also observed beyond the tumor outline deduced from Gd-DOTA signal enhancement. In this particular case, acquisition of the fluorescent image does not cover the dorsal part of the tumor. BVf, blood volume fraction; MRI, magnetic resonance imaging; RSST₁-MRI, rapid-steady-state-T₁-MRI.

Figure 5 shows overlays of Hoechst- and collagen IV-stained microvasculature in the periphery (Figure 5A) and in the center (Figure 5B) of a C6 glioma showing extravasation of the Hoechst dye with diffusion into the perivascular tissue. Figure 5C corresponds to the view in Figure 5B and illustrates the segmented vessels with the perfused vessels in red and the nonperfused vessels in gray. Figure 5 also shows the collagen IV-stained microvasculature in healthy cortical gray matter (Figure 5D) and in healthy callosal white matter (Figure 5E) for comparison. No Hoechst extravasation was observed in the contralateral brain tissue.

Figure 5.

Microvasculature in C6 tumor tissue and healthy brain. Immunofluorescent images comparing the microvascular structure in tumor and brain tissue. (A and B) Microvasculature in the C6 tumor periphery (panel A) and center (panel B). The anti-collagen IV (red) and Hoechst staining (blue) are overlaid to illustrate the partial perfusion and vessel permeability. (C) Segmented perfused (red) and nonperfused (white) vessels for the tumor center shown in panel B. The profiles have breadths in the range of 2.74 to 45 μm. (D and E) Anti-collagen IV staining of microvasculature in healthy white matter of the corpus callosum (panel D) and in healthy cortical gray matter (panel E). The vessel distribution and shapes are much more regular than in the tumor regions.

The bar charts in Figures 6A and 6B compare the V_V obtained by vascular morphometry using BDTH to estimate the mean vessel diameter and the vascular A_A with the BVf obtained by MRI with Gd-ACX as blood-pool CA.

Figure 6.

Comparison between stereologically obtained vascular volume fractions and blood volume fractions obtained by MRI. (A and B) The V_V calculated from BDTH, as well as the A_A for perfused vessels (anti-collagen IV staining+Hoechst overlay) only (panel A) and for all anti-collagen IV-stained vessels (panel B) are compared with the BVf measured by MRI with Gd-ACX in the C6 tumor periphery and center. The volume fractions obtained by vascular morphometry for the perfused vasculature are not statistically different from the BVf obtained by MRI, whereas using anti-collagen IV staining alone leads to an overestimation. (C) Comparison of vascular parameters obtained from all microvessels and perfused microvessels in C6 tumor regions. The vascular V_V is not only determined by the vascular density reflected by the P_A but in particular also by the vessel diameter. V_V=volume fraction, A_A=area fraction, BVf=blood volume fraction, b=minor axis of ellipse, BDTH=breadth, P_A=profiles per unit area. MRI, magnetic resonance imaging. ^*Significantly (0.05) different.

Compared with the perfused vasculature (Figure 6A), V_V and A_A of all vessels (Figure 6B) were by a factor of ∼2 and 2.6 greater, respectively. As shown in Figure 6C, this is rather attributed to a smaller diameter of the perfused vessels than to a partial perfusion (reflected by a slightly decreased vascular P_A).

Discussion

To date, MRI is the only noninvasive brain imaging method that can measure the BVf regionally regardless of the tumor location and depth, but blood-pool CAs are required for quantification. In malignant brain tumors, the vascular endothelium is often permeable to low-molecular-weight CAs, such as Gd-DOTA, precluding BVf quantification. In our previous study (Lahrech et al, 2008), we used the RSST₁ technique (Perles-Barbacaru and Lahrech, 2007) with Gd-ACX, providing evidence for the vascular confinement of this new CA in a C6 rat brain tumor model enabling the quantification of the tumor BVf. In this study, we measured the tumor BVf in the same animals by MRI with Gd-ACX and by vascular morphometric analysis combined with a stereological technique.

The stereological technique derives the V_V from the mean vessel diameter (power of two!; cf. equation 1) and is therefore very sensitive to the correct estimation of this parameter. Consequently, the first aim of this study was to establish the most appropriate measure of vessel diameter. This validation study was necessary because there is no consensus on which is the best measure for the ‘diameter' of an irregularly shaped profile. It was carried out on 3D numerical models of straight cylinders and cortical vasculature, as well as on histology sections of the rat brain vasculature. Although in the cylinder model, all four studied morphologic parameters used for diameter estimation yielded results within the given range of cylinder diameters, the four morphologic parameters yielded fairly different results for the more irregularly shaped vascular profiles in the brain tissue. Our studies on the numerical geometric models show that when using b or BDTH as diameter, the estimation of the total V_V is accurate within 18%. The mean vessel diameters in the rat brain obtained from BDTH are in accordance with the literature (Deane and Lantos, 1981; Farrell et al, 1991; Pathak et al, 2001; Schlageter et al, 1999). However, because determination of the diameter is method dependent, it is difficult to compare vessel diameters measured by different investigators. In addition, some authors average the vessel diameters only over a particular range, defining, e.g., the microvasculature as vessels with diameters <12 μm (Weiss et al, 1982). The vascular A_A, a two-dimensional surrogate for the vascular V_V or BVf, although more sensitive to the Holmes effect (projected size of the object greater than or equal to real size (Hennig, 1969)), is independent of the estimation of the vessel diameters, and yields values in the order of the stereologically obtained V_V, supporting the stereological technique.

Vascular V_V values obtained in healthy rat brain tissue in our study are in the range of published values (Table 2). Morphometric methods generally report vascular V_V or A_A between 1% and 2% (Dunn et al, 2004; Pathak et al, 2001, 2003), although some authors found values up to 3% (Schlageter et al, 1999) or 5% (Weiss et al, 1982), depending on the stereological technique used. There also is evidence that not all vessels are perfused in the healthy brain tissue (Shockley and LaManna, 1988; Weiss et al, 1982).

Table 2. Regional cerebral blood volume (healthy regions) in normocapnic, normothermic, and anesthetized rats.

ROI	CBV value	Unit	Technique	References
Whole braina	2.51	mL/100 g	Autoradiographyb	Todd et al (1992)
Whole braina	2.96±0.57	mL/100 g	Autoradiographyb	Todd et al (1993)
Whole brainc	2.77±0.24	mL/100 g	Autoradiographyb	Todd and Weeks (1996)
Whole brain	1.3±0.1	mL/100 g	Autoradiographyd	Bereczki et al (1992)
Cortex	3.4	mL/100 g	Optical bolus tracking method	Shockley and LaManna (1988)
Whole braine	2.40±0.34	%	3D SST1-MRI	Lin et al (1997)
Whole braine	2.96±0.82	%	3D SST1-MRI	Lin et al (1999)
Whole brainc	3.14±0.32	%	SST2-MRI	Dunn et al (2004)
Cortexc	1.63±0.18	mL/100 g	SST1-MRI	Schwarzbauer et al (1997)
Corpus callosumc	1.22±0.25	mL/100 g	SST1-MRI	Schwarzbauer et al (1997)
Thalamusc	3.03±0.36	mL/100 g	SST1-MRI	Schwarzbauer et al (1997)
Whole brainc	3.14±0.32	%	SSΔR2*-MRI	Dunn et al (2004)
Cortexf	4.3±0.7	%	SSΔR2*-MRI	Tropres et al (2004)
Striatumg	3.1±0.7	mL/100 g	SSΔR2*-MRI	Julien-Dolbec et al (2002)
Striatumh	2.2±0.6	mL/100 g	SSΔR2*-MRI	Julien et al (2004)
Cortexa	3.01±0.43	%	SSΔR2*-MRI	Payen et al (2000)
Striatuma	2.94±0.49	%	SSΔR2*-MRI	Payen et al (2000)
Cortexa	4.07	%	SSΔR2*-MRI	Tropres et al (2001)
Striatuma	2.87	%	SSΔR2*-MRI	Tropres et al (2001)
Whole brain	1.89±0.39	%	Morphometryi	Pathak et al (2001)
Whole brain without MVj	1.92±0.32	mL/100 g	SRQCT	Adam et al (2003)
Whole brain with MVj	4.18±1.06	mL/100 g	SRQCT	Adam et al (2003)
Cortexj	2.27	mL/100 g	SRQCT	Adam et al (2003)
Striatumj	2.01	mL/100 g	SRQCT	Adam et al (2003)
Striatumk	5.6	mL/100 g	SRQCT	Adam et al (2005)

CBV, cerebral blood volume; MV, macroscopic vessels; ROI, region of interest; SSΔR₂^*−MRI, steady-state ΔR₂^* magnetic resonance imaging; SST₁−MRI, steady-state T₁-weighted magnetic resonance imaging; SRQCT, synchrotron radiation quantitative computed tomography; 3D, three dimensional.

Reported values for the regional cerebral blood volume obtained using various imaging techniques. Decimal places and s.d. are given as reported in the original study.

^aRats anesthetized with halothane.

^b14C-dextran labeled plasma and ^99mTc-labeled red blood cells.

^cRats anesthetized with isoflurane.

^d125I- labeled serum albumin and ⁵⁵Fe labeled red blood cells.

^eRats anesthetized with intraperitoneal pentobarbital.

^fContralateral to C6 glioma, under moderate hypoxia, rats anesthetized with halothane.

^gRats anesthetized with intraperitoneal thiopental.

^hContralateral to C6 glioma, rats anesthetized with intraperitoneal thiopental.

ⁱWith stereo correction for slice thickness, contralateral to 9L tumor.

^jAnesthetized by intraperitoneal infusion of chloral hydrate.

^kContralateral to F98 glioma (n=1).

However, the major methodological impact on cerebral BVf measurement seems to originate from whether the BVf is measured in vivo or post mortem. From Table 2, it can be seen that with few exceptions, in vivo BVf measurements yield values in the range of 2% to 4%, whereas lower BVf values were reported in postmortem studies (Bereczki et al, 1992).

Comparing in vivo MRI with an invasive method such as histology is technically challenging because the data (the brain sections) are not in the same format and have different spatial resolutions. Each technique has additional limitations.

Although we took care to define similar ROIs on the MRI and histologic slices when possible, the nonparametric correlation between MRI BVf and stereological vascular V_V in the tumor tissue was not statistically significant (r_S=0.32). It is generally easier to show a correlation between two techniques when data from a healthy brain are used, rather than data from heterogeneous tumor tissues (Pathak et al, 2001). Although Pathak et al found a significant correlation between morphologic estimates and MRI-based BVf in healthy tissues, the correlation was poor for the tumor microvasculature. Unfortunately, contralateral vascular V_V for correlation with MRI data was not available in this experiment. Recent studies carried out in the laboratory (Valable et al, 2008) compare the BVf obtained in a C6 rat brain tumor model by the steady-state ΔR₂^* method using a superparamagnetic CA with the vascular V_V computed according to Pathak's stereological method for slice thickness correction (Pathak et al, 2001). Although only anti-collagen IV staining was used for microvessel delineation in this study, a significant correlation was observed between the histologic and MRI data, but the histologic vascular V_V was only about one-third of the BVf measured by MRI. For the C6 tumor tissue in our study, there was a good numerical agreement between the BVf measured by RSST₁-MRI and vascular V_V obtained by the stereological technique. This might be attributed to the lower slice thickness or the high number of slices used in our study.

Blood water detected by the RSST₁ technique or any other MRI technique using CAs for BVf measurement is the one contained in perfused vessels. Our results confirm that, when comparing the BVf measured by MRI in certain tumors with a histologic technique, care should be taken to restrict vascular analysis to perfused vessels. Numerous tumor specific features, such as irregular vessel architecture, abnormal vascular barrier, and elevated interstitial fluid pressure, result in reduced blood flow in parts of the tumor (Jain, 1988).

The size difference between perfused and unperfused vessels might be explained by vessel compression outside the plane leading to stasis and enlarged vessels. The diameter of the perfused vessels may also be decreased in the postmortem study owing to vessel collapse. The mean diameters of the tumor vessel profiles were in accordance with published values for the C6 tumor model (Farrell et al, 1991; Tropres et al, 2004). In our study, the diameters of perfused tumor vessels were not significantly different from those in the healthy rat brain, although enlarged tumor vessels with regard to the contralateral cortical tissue were observed in some other studies carried out on the C6 tumor model (Dennie et al, 1998; Tropres et al, 2004). As reported in the study by Lahrech et al (2008), the vasculature in brain tissue contralateral to the tumor might not be representative of the healthy brain vasculature owing to compression, edema, or inflammation caused by the tumor and affecting the vessel size. As shown in our study, another potential reason for this discrepancy might be that the ‘diameter' of irregular vessel cross-sections depends very much on the morphologic parameter used. For example, Tropres et al (2004) (p. 536 to 537) calculated for every intravascular pixel the shortest distance to the vessel wall and used the greatest value among all intravascular pixels as an estimate for the vessel diameter. Although in almost elliptical cross-sections, this parameter would result in the same measure as the breadth, it is likely that it differs from the breadth in irregular vessel cross-sections. In addition, when reporting the average vessel diameter in tumors, these authors also included nonfunctional vessels.

The ratio of perfused V_v to total V_v was ∼0.4 to 0.6 in this late-stage glioma model. This is in accordance with observations made by Bernsen et al (1995) who found that the perfused (area) fraction of tumor vessels ranged from 0.20 to 0.85 in subcutaneous tumors in mice. The perfusion fraction was in the same range or significantly lower in intracerebrally implanted human glioma xenografts in nude rats (Bernsen et al, 1999).

Owing to the characteristics and combined limits of the immunohistological, digital vascular morphometry, and stereological techniques, including image quality, thresholding, choice of morphometric parameters, etc., the vascular V_V evaluated in this study is only an estimation and no uncertainty interval can be given. Uncertainties also exist with the MRI technique: Although with the RSST₁-MRI technique only the intravascular water should contribute to the signal, the water exchange (Perles-Barbacaru and Lahrech, 2007) probably extends the origin of the acquired signal to the vessel wall or beyond.

Nevertheless, the stereological estimation of V_V of the perfused tumor vasculature is similar to BVf obtained by MRI supporting the vascular confinement of the CA Gd-ACX, despite the presence of the vascular endothelium permeable to Gd-DOTA.

Conclusions

There is no real gold standard for quantifying the BVf by histologic vascular analysis. Here, a robust technique for vascular volume fraction measurement is proposed, which requires the measurement of the vascular diameters. There is no gold standard for measuring the vascular diameters neither. In this study, we therefore compare the vascular diameters obtained in four different ways and we conclude that the BDTH is the most appropriate parameter. This conclusion is based on the criterion that the estimated diameters are in accordance with the literature and that resulting vascular volume densities correlate best with the vascular area fraction, which is a generally accepted histologic surrogate for BVf measurement.

The vascular volume fraction obtained by quantitative vascular morphometry combined with a stereological algorithm was compared with the tumor blood volume measured using the quantitative RSST₁-MRI method in conjunction with Gd-ACX, a new CA with blood-pool properties in the tumor vasculature. Despite some uncertainties regarding the morphometric analysis, an almost quantitative equivalence could be established for the tumor blood volume between the two independent techniques. The fact that the in vivo BVf measured by MRI is not significantly greater than the vascular V_V, confirms that Gd-ACX remains confined to the blood pool even in the vasculature of this advanced-stage C6 tumor that is mostly permeable to Gd-DOTA. The BVf measurement using a blood-pool CA such as Gd-ACX is therefore reliable and noninvasive and suitable for angiogenesis assessment.

Morphometric analyses also revealed one of the reasons for the low tumor blood volume, which is that the perfused vasculature makes up only half of the total vascular volume fraction. This partial perfusion is much less apparent when only vascular density is considered.

The authors declare no conflict of interest.