Abstract
Background: Quantitative analysis of tumour angiogenesis is an indispensable prerequisite for the development of novel antivascular treatment strategies. The aim of this study was to develop an automated analysis technique enabling quantitative assessment of complete tumour vascular networks non-invasively in vivo. Materials and Methods: Experiments were performed in A-Mel-3 implanted in transparent dorsal skinfold chambers of the Syrian Golden hamster. Tumours were imaged in two dimensions by means of a computer controlled scanning-table combined with intravital microscopy. Functional vessel density was automatically quantified online with a digital imaging application package. Results: Data for functional vessel density correlated excellently with data obtained by off-line video-based frame-by-frame analysis (R(S)=0.947, r2=0.959, n=48). Conclusion: This new technique enables automated, quantitative assessment of complete tumour vascular networks and therefore provides a promising tool to characterise changes of the entire tumour vascularity in response to antivascular treatment regimes.
Angiogenesis, the formation of new blood vessels from the endothelium of the existing vasculature, plays a key role in tumour growth, progression and metastasis (1). The growing vascular tumour network guarantees adequate supply of tumour cells with oxygen and nutrients and provides efficient drainage of metabolites. Quantitative analysis of the tumour microvascular architecture is of great importance for the development of vascular targeting treatment strategies (2) as well as for antiangiogenic approaches (3). However, methods for quantitative analysis of microcirculatory parameters in vivo are rare.
Different techniques for imaging tumour microvascularity have been developed, including dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) (4, 5), microbubble contrast-enhanced sonography (6), as well as intravital microscopy in different chamber-preparations (7) and animal models (8, 9). The strength of in vivo microscopy is its ability to provide a direct and continuous approach to visualise and quantify tumour microvasculature (10). Microvascular parameters such as functional microvessel density, blood flow and diameter of the microvessels can be measured but are very heterogeneous in different tumour entities and depend on the location within these tumours. In fast proliferating tumours there is a high proliferation of new blood vessels in the tumour rim in contrast to the tumour centre where, due to changes in the surrounding milieu, an increasing number of tumour cells are necrotic (11, 12). This leads to trial results that are highly examiner-dependent if heterogeneously vascularised regions of the tumour are measured. Reproducibility of results, for example for functional vessel density, therefore is low, if only a small number of regions of interest are investigated. Even if tumours are screened almost completely, the vascular network of the tumour and its complexity can hardly be visualised.
The aim of the present study was to develop an automated analysis technique for the quantitative assessment of entire tumour vascular networks in vivo. The approach was validated by correlation of the results with data obtained by in vivo fluorescence microscopy and off-line video based analysis.
Materials and Methods
Animal and tumour model. Experiments were carried out using male Syrian golden hamsters of 40-50 g body weight (Charles River, Sulzbach, Germany) in accordance with UKCCCR guidelines (13). The animals were housed in single cages and received tap water and standard laboratory food (ssniff Spezialdiaeten GmbH, Soest, Germany) ad libitum. To permit quantitative analysis of microvasculature in vivo, a dorsal skinfold chamber, consisting of two symmetric titanium frames, was surgically implanted into the dorsal skin of the animals as described previously (14, 15). After a recovery period of 24 h from anaesthesia and microsurgery, chamber preparations were inoculated with 2 μl of dense tumour cell suspension (~2×105 cells) of the amelanotic melanoma of the hamster A-Mel-3 (16) (Figure 1). For the administration of flouroscence markers, permanently indwelling fine polyethylene catheters (PE10; inner diameter, 0.28 mm) were implanted into the right jugular vein. All surgical procedures were performed under anaesthesia with ketamine (100 mg/kg body weight i.p., Ketavet®; Parke-Davis, Berlin, Germany) and xylazine (10 mg/kg body weight i.p., Rompun®; Bayer, Leverkusen, Germany).
In vivo microscopy. For in vivo microscopy, the awake and chamber-bearing hamster was immobilised in a Perspex tube on a specially designed stage (Effenberger, Munich, Germany) under a modified Leitz microscope (Orthoplan; Leitz, Munich, Germany). After setting the starting point and the number of images that should be taken in x and y directions, the tumour was scanned in two dimensions and imaged by means of a computer controlled scanning table. Images had a 10% overlap in x and y directions. Blue light transillumination technique with a magnification of 630-fold and automated focusing was used to assure good quality of the acquired images.
Image analysis. Analysis of the acquired images was performed with a digital imaging application package (KS400, Zeiss, Germany) on a personal computer (Scenic, Siemens Fujitsu, Munich, Germany). The algorithm for the calculation of functional vessel density was programmed using the KS400 macro editor function.
To minimise the loss of quality due to irregular illumination and inhomogeneous sensitivity of the camera sensors, a shading correction procedure was used (shadcorr). The best correction in terms of maximal contrast of the images could be reached with a multiplicative correction (Appendix). In the algorithm, the acquired images had to be transferred into binary images for further processing (Figure 2). This was achieved by a multiple step procedure. In the first step, each pixel of the image was scaled along a grayscale between 0 and 255 (normalize). To amplify the contrast between the light gray background and the darker vessels, the grayscale values of each pixel in the image were multiplied with themselves (multiply) in the second step. Therefore, pixels with low grayscale values such as those in the background stayed low, while pixels with greater values such as those within a vessel were intensified. In the next step, a smooth-function was used to reduce maxima and to abrade sharp edges (median). In the fourth step, each pixel of each image was inverted; for example, black was inverted to white and light gray was inverted to dark gray (binnot). The final step was a gray value segmentation to result in a binary image (disdyn). The code for the multiple step procedure is given in the Appendix. Finally the vessels, now white in the binary images, were reduced to single lines along their smallest diameter by a thinning procedure (binthinning) developed by Arcelli et al. (17) (Appendix).
The length of all lines/vessels within a region of interest (ROI) was calculated and divided through the area of the ROI. The result was saved on the computer as a functional vessel density map together with the ID number of the ROI. The whole procedure was performed online by means of the aforementioned self-written algorithm. To visualise differences in functional vessel density among different areas of the tumour, false colour rendering and 3D-mesh-graphs were used (Figures 3, 4, 5 and 6).
Fluorescence microscopy. For the validation of this new technique, the data were correlated with the results acquired by in vivo fluorescence microscopy and off-line video analysis. For this purpose, FITC-labelled dextran (MW 500000; 0.05-0.1 ml of a 5% solution in 0.9% NaCl; Sigma, Deisenhofen, Germany) was injected intravenously to visualise tumour microcirculation. Selective observation of FITC-labelled plasma was carried out using epiillumination with a 100 W mercury lamp attached to a Leitz Ploemopak illuminator with a Leitz I2/3 filter block (excitation wavelength: 450-490 nm, emission wavelength ≥515 nm). In vivo fluorescence microscopy was performed immediately after the scanning procedure of the tumour. The scanning table was set back to randomly selected ROIs where images were acquired by a SIT video camera (C2400-08; Hamamatsu, Herrsching, Germany) and recorded on a S-VHS video tape.
Off-line analysis of the S-VHS video tapes was performed by an image analysis system (Cap Image; Zeintl, Heidelberg, Germany) (18). This system allowed the measurement of functional vessel density as a parameter for angiogenic activity (19).
Study protocol. For data acquisition, three animals with constant growing tumours and chamber preparations without irritations were used (20). Twenty-four hours after surgical chamber preparation, tumour cells were implanted and jugular vein catheters were inserted. The scanning procedure was performed on day 6 after tumour cell inoculation. Fluorescent markers were injected 30 min before measurement.
Results
The three tumours were completely scanned with up to 286 ROIs (13×22 ROIs in x and y directions) depending on tumour size. After cutting off the normal tissue vasculature, the whole tumour vascular network was visualised in blue-light transillumination for further processing (Figure 3, left). By running the new automated algorithm, binary images of the entire tumour vasculature were created. These images were used for the quantification of functional vessel density (fvd) in each ROI (Figure 3, middle). To visualise differences in fvd in different parts of the tumour, false colour rendering (Figure 3, right) and 3D-mesh-graphs were used (Figure 4, 5 and 6).
Differences in fvd between the fast proliferating tumour rim and the tumour centre were detected in two of the three tumours. Tumour #1 had an fvd of 217.1 cm−1 in the tumour rim and an fvd of 81.0 cm−1 in the tumour centre (Figure 4). The analysis of tumour #2 resulted in an fvd of 215.1 cm−1 in the tumour rim and an fvd of 80.1 cm−1 in the tumour centre (Figure 5). In tumour #3 it was not possible to differentiate between the rim and the centre; the fvd of the whole tumour was 252.6 cm−1 (Figure 6).
To validate this new approach, 48 randomly selected ROIs out of the three tumours were also examined by fluorescence microscopy and off-line video based frame-by-frame analysis. Correlation of data of both techniques revealed a Spearman's correlation coefficient of R(s)=0.947 (r2=0.959, n=48) for functional vessel density (Figure 7).
The scanning procedure of the entire vascular network was also used in combination with fluorescence microscopy to get reliable data of the actually perfused parts in the tumour (Figure 8).
Discussion
Intravital microscopy is one of the leading techniques to study tumour microcirculation (10). However, it has not yet been possible to visualise the complexity of whole tumour microvascular networks. Therefore in this study, an examiner-independent automated procedure was developed to measure tumour vascularity in terms of fvd. The strength of this new technique is its ability to visualise and analyse the complex microvascular structure of entire tumours without examiner-dependent bias. The results correlated well with data obtained by fluorescence microscopy and off-line video based frame-by-frame analysis. Previous studies showed comparable but slightly higher results for fvd in untreated A-Mel-3 tumours (21). This overestimation may be due to the fact that only a small number of ROIs was measured in that study (20).
This new technique using blue-light transillumination microscopy, however, failed to distinguish between perfused and non-perfused microvessels because it was not possible to set a threshold for separating one against the other. Nevertheless, measurement of fvd with this new approach and fluorescence microscopy showed comparable results in the A-Mel-3 tumour. This indicates a high functional efficacy of the newly formed blood vessels.
Attempts to combine the newly developed algorithm with fluorescence microscopy for a single measurement of perfused vessels failed because the contrast between FITC-labelled plasma and the extravascular background was too low to reach a sufficient transformation into binary images. However, the generation of clear binary images is a prerequisite for a correct calculation of vessel density. To reach this goal a precise illumination and shading correction of the whole ROI especially at the edges is necessary when acquiring the initial images. This may be difficult in long-term observations when the tumour is growing in three dimensions and transillumination decreases. A 10% overlap of the images in x and y directions can reduce this problem.
In comparison to other techniques for the measurement of tumour microcirculation such as DCE-MRI and microbubble contrast-enhanced sonography, the present technique gives no information about the functionality of the tumour vessels. Hyperperfusion or stasis in different parts of the tumour cannot be detected as well as the formation of thrombosis within the vessels. In contrast, DCE-MRI can be used to quantify tumour perfusion and tumour vessel permeability and to characterise the intra- and extravascular compartments. Unfortunately, there is a number of different tracer kinetic models with a variety of microvascular parameters, making the comparability among different DCE-MRI studies very difficult. Furthermore, there are no standardised measurement protocols to compare studies from different institutions (22, 23). Similar problems occur when using microbubble contrast-enhanced sonography. Despite its low cost, its availability and its first promising results in monitoring antivascular therapy there is also a lack in standardisation so far (24, 25).
The results of the present study demonstrated the heterogeneity of functional vessel density between the fast proliferating tumour rim and the increasingly necrotic tumour centre which could be seen in two of the three tumours. These findings were due to the decreasing nutrient and oxygen supply to the tumour centre normally seen around day 4 to 5 after tumour cell inoculation (14). Tumour #3 was smaller in diameter and therefore could have had a better supply of oxygen and nutrients for the central tumour parts where no necrotic areas were observed.
Graphical illustration by 3D meshes and false colour rendering enabled the visualisation of functional vessel density within the tumours and can be used to characterize changes of the entire tumour vascular network in response to antivascular treatment regimes.
The new algorithm is not only an application for the skinfold chamber in the hamster, but can also be used for other preparations and tumour models; for example in the cranial window preparation of rats or mice (26). The only limitation is that the acquired images must have enough contrast between the vessels and the background to achieve a good transformation into binary images.
In this study, an automated examiner-independent technique was developed for the assessment of tumour microvasculature. Further studies will focus on the improvement of this imaging technique to make other microcirculatory parameters such as vessel diameter and red blood cell velocity accessible for automated imaging and quantification algorithms.
Appendix - Processing procedures in KS400 environment
Shading correction
imgdelete “*”
imgload “ROI1”,1
imgload “shading_correction.img”,4
shadcorr 1,4,2,2,0
Multistep image transfer
normalize 1,2,0
multiply 2,2,3,200
median 3,4,7
binnot 4,5
disdyn 5,6,65,-20,1
Vessel thinning
binthinning 1,2,1,0
- Received March 30, 2010.
- Revision received May 19, 2010.
- Accepted May 26, 2010.
- Copyright© 2010 International Institute of Anticancer Research (Dr. John G. Delinassios), All rights reserved