Fluctuations in tumor blood perfusion assessed by dynamic contrast-enhanced MRI

Authors

  • Kjetil G. Brurberg,

    1. Group of Radiation Biology and Tumor Physiology, Department of Radiation Biology, Institute for Cancer Research, The Norwegian Radium Hospital, Montebello, Oslo, Norway
    Search for more papers by this author
  • Ilana C. Benjaminsen,

    1. Group of Radiation Biology and Tumor Physiology, Department of Radiation Biology, Institute for Cancer Research, The Norwegian Radium Hospital, Montebello, Oslo, Norway
    Search for more papers by this author
  • Liv M.R. Dørum,

    1. Group of Radiation Biology and Tumor Physiology, Department of Radiation Biology, Institute for Cancer Research, The Norwegian Radium Hospital, Montebello, Oslo, Norway
    Search for more papers by this author
  • Einar K. Rofstad

    Corresponding author
    1. Group of Radiation Biology and Tumor Physiology, Department of Radiation Biology, Institute for Cancer Research, The Norwegian Radium Hospital, Montebello, Oslo, Norway
    • Department of Radiation Biology, Institute for Cancer Research, The Norwegian Radium Hospital, Montebello, N-0310 Oslo, Norway
    Search for more papers by this author

Abstract

Temporal heterogeneity in blood perfusion is a common phenomenon in tumors, but data characterizing the nature of the blood flow fluctuations are sparse. This study investigated the occurrence of blood flow fluctuations in A-07 melanoma xenografts by using gadopentetate dimeglumine (Gd-DTPA)-based dynamic contrast-enhanced MRI (DCE-MRI). Each tumor was subjected to two DCE-MRI acquisitions separated by 1 hour. The data were processed by Kety analysis and resulted in two E · F images (E is the initial extraction fraction of Gd-DTPA and F is the perfusion) and two λ images (λ is the partition coefficient of Gd-DTPA) for each tumor. The E · F images were used to determine the changes in blood perfusion arising in the time between the two imaging sequences. The λ images were used to control the reproducibility of the experimental procedure. The study showed that DCE-MRI with subsequent Kety analysis is a useful method for detection of blood flow fluctuations in A-07 tumors, and strongly suggested that the peripheral regions of A-07 tumors are more exposed to temporal changes in blood perfusion than are the central regions. Magn Reson Med 58:473–481, 2007. © 2007 Wiley-Liss, Inc.

Knowledge of the spatial and temporal heterogeneity in blood perfusion in tumor tissue is limited (1). However, there is a general consensus that tumors with low or heterogeneous blood perfusion are resistant to chemotherapy, immunotherapy, and other treatments that depend on adequate uptake of blood-borne therapeutic agents (2). Gadopentetate dimeglumine (Gd-DTPA)-based dynamic contrast-enhanced MRI (DCE-MRI) followed by Kety analysis is being used to obtain information on the blood perfusion of tumors (3). Kety analysis does not provide blood perfusion values directly, but yields values of the E · F (E is the initial extraction fraction of Gd-DTPA and F is the perfusion) product. Thus, correct assessment of tumor blood perfusion by the Kety method requires voxel-by-voxel determination of E. Theoretical and experimental studies have demonstrated that it is extremely difficult to obtain reliable E-values for Gd-DTPA by DCE-MRI (4, 5). However, many tumors have a hyperpermeable microvascular network and, therefore, show E-values for Gd-DTPA that are homogeneous and close to unity (2, 6). It has been suggested that the blood perfusion in these tumors is mirrored adequately by E · F images (7). By comparing E · F images assessed by DCE-MRI with tumor blood perfusion images assessed by invasive imaging, we have shown that E · F is an excellent parameter for blood perfusion in A-07 melanoma xenografts (8).

Insufficient blood perfusion may lead to the development of hypoxic regions in tumors (6, 9). Chronic hypoxia is a result of limitations in oxygen diffusion, whereas transient changes in blood perfusion or microvessel red blood cell flux may cause acute hypoxia (10–12). It has been shown that the oxygen tension (pO2) can vary significantly both spatially and temporally within tumors (13, 14). Furthermore, it is known that tumor hypoxia causes resistance to radiation therapy and some forms of chemotherapy, and may also promote malignant progression (11, 15) and the development of metastatic disease (16). Clinical investigations have shown that extensive hypoxia in the primary tumor is associated with locoregional treatment failure and poor disease-free and overall survival rates in several histologic types of cancer (16–18).

Diffusible dyes have been used to demonstrate tumor blood flow fluctuations with cycle times ranging from 30 min to several hours (19, 20). Baudelet et al. (21) used T2*-weighted gradient echo imaging to visualize that tumor blood flow may vary considerably within an interval of 1 hour, an observation which is consistent with studies in window chamber preparations showing red blood cell flux cycle times between 15 and 60 min (12, 22). Interestingly, the observed frequency range of tumor blood flow fluctuations corresponds well to the pO2 fluctuation frequencies measured in experimental tumors with recessed-tip microelectrodes (13) and OxyLite probes (23), suggesting that irregularities in tumor blood flow and pO2 fluctuations are related incidents.

Direct measurements of pO2 (14) and fraction of radiobiologically hypoxic cells (24) have suggested that pO2 fluctuations with cycle times of less than 1 hour occur frequently in A-07 melanoma xenografts. In the present study we used Gd-DTPA-based DCE-MRI to search for blood flow fluctuations in A-07 tumors, concentrating on the same frequency range as that identified previously for pO2 fluctuations. Thus, each tumor was subjected to two DCE-MRI mappings of E · F, 1 hour apart. The E · F images were processed to images showing the changes in blood perfusion occurring in the 1-hour-long time interval between the two DCE-MRI recordings. We hypothesized that these images would reveal whether blood flow fluctuations occur randomly throughout the entire volume of tumors or are restricted to particular tumor regions and, hence, would provide new and valuable knowledge regarding the nature of blood flow fluctuations in tumor tissue.

MATERIALS AND METHODS

Mice and Tumors

Adult (8–10 weeks of age) female BALB/c-nu/nu mice were used as host animals for xenografted tumors (24). Experiments were performed with tumors of the A-07 human melanoma line, established as described previously (25). The tumors were subjected to MR imaging when having volumes within the range of 300–800 mm3. Animal experiments were approved by the Institutional Committee on Research Animal Care and were performed according to the Interdisciplinary Principles and Guidelines for the Use of Animals in Research, Marketing, and Education (New York Academy of Sciences, New York, NY).

Anesthesia

DCE-MRI was carried out with anesthetized mice. A full dose of fentanyl citrate/fluanisone (Janssen Pharmaceutica, Beerse, Belgium) and midazolam (Hoffmann-La Roche, Basel, Switzerland) of 0.63/20 and 10 mg/kg of body weight, respectively, was administered intraperitoneally before the mice were placed in the MR scanner. A supplementary dose, 25% of the first dose, was administered 15 min before the second MRI acquisition was initiated. These doses did not alter the blood perfusion in A-07 tumors significantly as long as the body core temperature of the host mice was kept at 37–38°C, as verified by using the 89Rb uptake method to compare tumor blood perfusion in anesthetized and nonanesthetized BALB/c-nu/nu mice (26). The mice were breathing air freely during the experiments.

Contrast Agent

Gd-DTPA (Schering, Berlin, Germany), diluted in 0.9% saline to a final concentration of 0.06 M, was used as contrast agent. The contrast agent was administered in two separate doses, each of 5.0 mL/kg of body weight. The first dose was given after the mice had been placed in the MR scanner, whereas the second dose was administered 1 hour later (i.e., immediately before the second MR acquisition). Both doses were administered in the tail vein by using the same 24-G Neoflon.

DCE-MRI

A total of 12 mice were included in the study. Five mice served as controls and were subjected to imaging of the muscle in the right hindlimb, whereas seven mice were subjected to tumor imaging. The imaging was performed at a voxel size of 0.5 × 0.2 × 2.0 mm3 with a 1.5 T whole-body scanner (Signa, General Electric, Milwaukee, WI) in accordance with a procedure published earlier (7). Briefly, the tumors were imaged axially in a single section through the tumor center, whereas single coronal scans were used for muscle imaging. Two proton density (PD) images and three T1-weighted images were acquired before the contrast agent was administered. Following administration of Gd-DTPA, T1-weighted images were recorded every 14 sec for 15 min. Two PD images were acquired immediately after the last T1-weighted image. This imaging series (i.e., two precontrast PD images, three precontrast T1-weighted images, 57 postcontrast T1-weighted images, and two postcontrast PD images) was acquired twice. The second acquisition was started 1 hour after the first imaging series was initiated.

Arterial Input Function (AIF)

The present study involved two subsequent Gd-DTPA administrations separated by 1 hour. The blood concentration of Gd-DTPA following each administration was determined by analyzing blood samples in vitro. Blood samples of ≈50 μL were collected from a vein in the hindleg approximately once every minute and diluted to appropriate volumes in 0.9% saline before the Gd-DTPA concentration was determined by MRI (27). Twelve mice without tumors were used to determine the first AIF (7), whereas 20 samples from 11 tumor-free mice were analyzed to determine the second AIF.

Image Processing and Data Analysis

Gd-DTPA concentrations were calculated from signal intensities according to the method of Hittmair et al. (27). Plots of Gd-DTPA concentration versus time were generated and the Kety equation (28)

equation image

was fitted to the plots. In the equation, Ct(T) is the Gd-DTPA concentration in the tumor or muscle tissue at time T, E is the initial extraction fraction of Gd-DTPA, F is the perfusion per unit tumor weight, Ca(t) is the arterial input function, and λ is the partition coefficient of Gd-DTPA.

Numerical values of E · F and λ were calculated for each voxel from the best curve fit. The experimental uncertainties in E · F and λ due to noise were determined by subjecting the data to Monte Carlo analysis. E · F has been shown to mirror blood perfusion well in A-07 tumors (8), whereas λ is closely related to the extracellular volume fraction (ECVF) (7, 8) and the parameter νe in the Tofts-Larsson pharmacokinetic model (3). Therefore, E · F and λ were used as parameters for blood perfusion and ECVF in this study. T1-weighted images and images of E · F and λ were used to compute ΔT1-weighted, Δ(E · F), and Δλ images. Briefly, two subsequent E · F images were subtracted voxel by voxel and formed the basis for a Δ(E · F) image in which each voxel was subscribed a Δ(E · F) value according to the equation:

equation image

Δλ and ΔT1-weighted images were computed in the same manner and served as reproducibility controls.

RESULTS

AIFs were determined by collecting blood samples and measuring Gd-DTPA concentrations in vitro (Fig. 1). The open symbols refer to the first AIF, reported previously by Benjaminsen et al. (7), while the closed symbols refer to the second AIF. It should be noted that the dots at t = 0 do not refer to measured values, but to calculated values (i.e., the maximum Gd-DTPA concentrations that can be achieved, calculated from the bolus dose and the plasma volume of the mice). It should also be noted that E · F and λ values produced by Kety analysis are insensitive to minor variations in the Gd-DTPA concentration at t = 0. Ca(t) was determined by fitting a double exponential function to the blood sample data by regression analysis:

equation image
Figure 1.

AIFs (blood concentration of Gd-DTPA vs. time after Gd-DTPA administration) for BALB/c-nu/nu mice. Gd-DTPA (0.06 M) was administered intravenously in two separate bolus doses of 5.0 mL/kg of body weight, 1 hour apart. The open symbols refer to the Gd-DTPA concentration after the first dose, as reported by Benjaminsen et al. (7), and the closed symbols refer to the Gd-DTPA concentration after the second dose. The dots at t = 0 refer to theoretical values and represent the maximum concentrations that can be achieved immediately after Gd-DTPA administration. Due to accumulation of Gd-DTPA in the blood, the value at t = 0 is higher for the second than for the first Gd-DTPA dose. The curves represent the best fit of a double exponential function to the data.

The numerical values of the constants were determined to be: A = 2.55 ± 0.21 mM, B = 0.08 ± 0.02 s−1, C = 1.20 ± 0.11 mM, and D = 0.0010 ± 0.0002 s−1 for the first Gd-DTPA dose and A = 3.19 ± 0.22 mM, B = 0.07 ± 0.02 s−1, C = 0.82 ± 0.12 mM, and D = 0.0011 ± 0.0003 s−1 for the second Gd-DTPA dose.

Seven mice were scheduled for tumor imaging and five mice were subjected to imaging of the left hindlimb. Kety analysis at the voxel level showed that muscle tissue differed from tumor tissue in both median E · F and median λ (Fig. 2). Mean of median E · F was measured to be 0.07 mL/(g · min) (first imaging) and 0.08 mL/(g · min) (second imaging) in muscle tissue (Fig. 2a). The corresponding values for mean of median λ were 0.14 and 0.15 (Fig. 2b). In tumor tissue, mean of median E · F was 0.21 mL/(g · min) (first imaging) and 0.19 mL/(g · min) (second imaging) (Fig. 2a), while the corresponding values for mean of median λ were 0.50 and 0.55 (Fig. 2b). The Mann–Whitney Rank Sum Test showed that the median E · F values calculated from the first DCE-MRI recording were not significantly different from those determined from the second DCE-MRI recording, either for muscle tissue (P = 0.45) or for tumor tissue (P = 0.86). The median λ values did not differ significantly between the two DCE-MRI recordings either (P = 0.82 for muscle; P = 0.06 for tumor). However, the Mann–Whitney Rank Sum Test showed that both median E · F and median λ were significantly higher in tumors than in muscle tissue (P < 0.0001 for E · F; P < 0.0001 for λ).

Figure 2.

Seven A-07 tumors and five muscles in the hindlimb of BALB/c-nu/nu mice were subjected to two subsequent DCE-MRI recordings, 1 hour apart. a: Median E · F values derived from the first (open symbols) and the second (closed symbols) recording. b: Median λ values derived from the first (open symbols) and the second (closed symbols) recording.

Subsequent measurements of E · F in tumors may give different values, either because the perfusion of the tissue changes between the measurements or because of experimental uncertainty in the measurements. The present study depended on the ability to identify and exclude voxels where a change in E · F only reflected methodical uncertainty. E · F and λ were not expected to change significantly with time in resting muscle tissue and, therefore, muscle measurements were used to estimate the methodical uncertainty. Figure 3 illustrates the results of the Kety analyses of the DCE-MRI series of a representative muscle. The most important characteristics of the first λ image (Fig. 3a) were well retained in the λ image recorded 1 hour later (Fig. 3b). For example, both images showed high λ values within and close to regions with elevated E · F values. The λ distributions derived from the two imaging sequences differed only marginally, with median λ equaling 0.14 and 0.15 (Fig. 3c). The absolute value of Δλ was lower than 0.05 for 93% of the voxels lying within the muscular boundaries (Fig. 3d). Highly reproducible E · F images were also produced (Fig. 3e,f). A hot-spot, probably reflecting the femoral artery and vein, was observed to the right in the E · F images. This hot-spot occurred in the same location and showed the same shape and size in both images. The E · F distributions derived from the two imaging sequences were remarkably similar, with median E · F equaling 0.08 mL/(g · min) in both cases (Fig. 3g). Voxel-by-voxel subtraction revealed that 79% of the voxels showed Δ(E · F) values in the interval between –0.03 and +0.03 ml/(g · min) (Fig. 3h). On the basis of observations from studies of muscle tissue, illustrated by a typical example in Fig. 3, the following strategy was established for detecting fluctuations in blood perfusion in tumor tissue. First, voxels showing absolute values of Δλ exceeding 0.05 were excluded from the analysis because Δλ values in this range probably reflected methodical problems or biological changes in the tissue that were not related to fluctuations in blood flow. Second, absolute values of Δ(E · F) of less than 0.03 mL/(g · min) were regarded to be within the experimental uncertainty, and therefore, voxels with absolute values of Δλ < 0.05 and absolute values of Δ(E · F) < 0.03 mL/(g · min) were considered to reflect tumor regions without significant changes in blood perfusion. Consequently, only voxels with absolute values of Δλ < 0.05 and absolute values of Δ(E · F) > 0.03 mL/(g · min) were reckoned to represent tumor regions showing biologically significant blood flow fluctuations.

Figure 3.

Muscle tissue in the hindlimb of BALB/c-nu/nu mice was subjected to two subsequent DCE-MRI recordings, 1 hour apart. a,b: Color-coded λ images derived from the first and the second recording, respectively. c: λ frequency distributions derived from the first (blue line) and the second (red line) recording. d: The difference in λ[Δ(λ)] between the second and the first recording for all voxels lying within the boundary of the muscle, as found by voxel-by-voxel subtraction. e,f: Color-coded E · F images derived from the first and the second recording, respectively. g:E · F frequency distributions derived from the first (blue line) and the second (red line) recording. h: The difference in E · FE · F) between the second and the first recording for all voxels lying within the boundary of the muscle, as found by voxel-by-voxel subtraction.

In contrast to the muscle tissue, the A-07 tumors showed highly heterogeneous blood perfusion, with well-perfused regions in the periphery and poorly perfused regions centrally. Both E · F and λ varied over a wider range in the tumor tissue than in the muscle tissue. The E · F and λ images differed substantially among individual tumors, in agreement with previous DCE-MRI studies of A-07 tumors (7, 8). However, the E · F and λ images derived from the two recordings appeared similar in all tumors. The λ values were in general better preserved than were the E · F values. Despite these similarities, all tumors showed significant evidence of fluctuations in blood perfusion. Data pertaining to a representative tumor are presented in Fig. 4. Median E · F for this particular tumor decreased significantly from 0.105 to 0.092 mL/(g · min) (P = 0.0010, Mann–Whitney Rank Sum Test) between the two recordings (Fig. 4a–c), whereas median λ did not change significantly (median λ = 0.502 and 0.497; P = 0.99, Mann–Whitney Rank Sum Test; Fig. 4d–f). Thirty-one percent of the voxels had absolute values of Δλ < 0.05 (Fig. 4g). The Δ(E · F) values of these voxels are plotted in Fig. 4h, where the voxels with nonsignificant changes in E · F [i.e., absolute values of Δ(E · F) < 0.03 ml/(g · min)] appear in black and the voxels with significant changes in E · F [i.e., absolute values of Δ(E · F) > 0.03 ml/(g · min)] appear in color. The ΔE · F values of the voxels indicating significant fluctuations in blood perfusion [i.e., voxels with absolute values of Δλ < 0.05 and absolute values of Δ(E · F) > 0.03 ml/(g · min)] are shown in Fig. 4i.

Figure 4.

An A-07 tumor in BALB/c-nu/nu mice was subjected to two subsequent DCE-MRI recordings, 1 hour apart. a,b: Color-coded E · F images derived from the first and the second recording, respectively. c:E · F frequency distributions derived from the first (blue line) and the second (red line) recording. d,e: Color-coded λ images derived from the first and the second recording, respectively. f: λ frequency distributions derived from the first (blue line) and the second (red line) recording. g: Δλ image (i.e., image of the difference in λ between the second and the first recording, as found by voxel-by-voxel subtraction) highlighting all voxels with absolute values of Δλ < 0.05 (white symbols). Voxels with absolute values of Δλ > 0.05 (black symbols) were excluded from further analysis. h: Color-coded Δ(E · F) image (i.e., image of the difference in E · F between the second and the first recording, as found by voxel-by-voxel subtraction) highlighting all voxels having absolute values of Δλ < 0.05 and absolute values of Δ(E · F) > 0.03 mL/(g · min). Voxels having absolute values of Δλ > 0.05 or absolute values Δ(E · F) < 0.03 mL/(g · min) are coded in black. i: Histogram showing Δ(E · F) for all voxels indicating significant fluctuations in tumor blood perfusion [i.e., voxels having absolute values of Δλ < 0.05 and absolute values of Δ(E · F) > 0.03 mL/(g · min)].

The analysis of the data obtained for one of the tumors is described in detail in Fig. 4. The main results for the other six tumors are presented in Fig. 5, where Fig. 5a–f shows voxel plots of Δ(E · F) equivalent to that presented in Fig. 4h, and Fig. 5g–l shows histograms of Δ(E · F) equivalent to that presented in Fig. 4i. The seven tumors differed substantially in several characteristics. First, the fraction of voxels indicating significant fluctuations in blood perfusion ranged from 6% (Fig. 5b) to 51% (Fig. 5d). Second, three tumors were characterized by a general increase or a general decrease in E · F (Figs. 4i, 5h,j), whereas four tumors showed similar fractions of voxels with increased and decreased E · F values (Fig. 5g,i,k,l). Third, the spatial distribution of the voxels indicating significant fluctuations in blood perfusion varied from tumor to tumor. Interestingly, all tumors showed voxel clusters (n > 10 voxels) where E · F changed in the same direction. Such clusters were usually observed in the tumor periphery. The absolute changes in E · F differed among and within tumors, but rarely exceeded 0.2 mL/(g · min).

Figure 5.

A-07 tumors in BALB/c-nu/nu mice were subjected to two subsequent DCE-MRI recordings, 1 hour apart. a–f: Color-coded Δ(E · F) images (i.e., images of the difference in E · F between the second and the first recording, as found by voxel-by-voxel subtraction) for six tumors highlighting all voxels having absolute values of Δλ < 0.05 and absolute values of Δ(E · F) > 0.03 mL/(g · min). Voxels having absolute values of Δλ > 0.05 or absolute values of Δ(E · F) < 0.03 mL/(g · min) are coded in black. g–l: Histograms for six tumors showing the Δ(E · F) values for all voxels indicating significant fluctuations in tumor blood perfusion [i.e., voxels having absolute values of Δλ < 0.05 and absolute values of Δ(E · F) > 0.03 mL/(g · min)].

The criteria used to identify tumor regions showing biologically significant blood flow fluctuations [i.e., voxels with absolute values of Δλ < 0.05 and absolute values of Δ(E · F) > 0.03 ml/(g · min)] were derived from studies of muscle tissue. To examine the validity of these criteria for tumor tissue in more detail we subjected the DCE-MRI data from the tumor experiments to further analysis.

First, we investigated the possibility that our results were influenced significantly by noise. Thus, histograms of the residuals of the curve fits were established for both DCE-MRI recordings. The residuals of the second recording were not significantly different from those of the first one in any of the seven tumors included in the study (P > 0.05, Mann–Whitney Rank Sum Test). The uncertainties in E · F and λ due to noise were determined by subjecting the DCE-MRI data to Monte Carlo analysis using noise generated from the residual distributions. The standard deviation of E · F was found to be less than 0.005 mL/(g · min) for all values of E · F, whereas λ showed a standard deviation that increased from 0.007 to 0.012 with increasing E · F value. It is thus unlikely that noise could have led to absolute values of Δλ < 0.05 and Δ(E · F) > 0.03 mL/(g · min) in a significant number of voxels and, hence, to erroneous scoring of tumor blood flow fluctuations.

Second, we investigated the possibility that accumulation of contrast agent in poorly perfused or necrotic tumor regions during the interval between the two recordings could have led to erroneous conclusions. Regions with accumulated contrast agent were identified by subtracting the first baseline T1-weighted image from the second one. The ΔT1-weighted images showed signal intensities with high absolute values in the same regions as the Δλ images showed high absolute values of Δλ. One representative example referring to a large A-07 tumor with a massive central necrosis is presented in Fig. 6, showing the ΔT1-weighted image (Fig. 6a) and the Δλ image (Fig. 6b). Histological examination confirmed that the red-colored regions of the images in Fig. 6 corresponded to necrotic tumor regions. Comparisons of the ΔT1-weighted images and the corresponding Δλ images of the seven tumors included in the study revealed that the voxels showing evidence of contrast accumulation between the two recordings had absolute values of Δλ > 0.05 and thus were excluded from further analysis. Consequently, it is unlikely that the Δ(E · F) images in Figs. 4h and 5a–f were influenced significantly by the contrast accumulation in the tissue between the DCE-MRI recordings.

Figure 6.

An A-07 tumor in BALB/c-nu/nu mice was subjected to two subsequent DCE-MRI recordings, 1 hour apart. a: Color-coded ΔT1-weighted image (i.e., image of the difference in signal intensity between the second and the first baseline T1-weighted image, as found by voxel-by-voxel subtraction). b: Color-coded Δλ image (i.e., image of the difference in λ between the second and the first recording, as found by voxel-by-voxel subtraction).

DISCUSSION

The vascular network of many tumors is characterized by immature vessels and a chaotic organization (6) and, consequently, tumors frequently suffer from instabilities in the blood supply (29). Temporal variations in blood perfusion are important because the effects of many tumor treatments, for example, chemotherapy and immunotherapy, depend on an adequate blood supply at the time of drug administration (2). Moreover, fluctuations in blood flow and red blood cell flux may be involved in the development of acute hypoxia (12, 29). Many techniques used for continuous measurement of blood flow, for example, laser Doppler systems, offer a high temporal resolution in relatively small measurement volumes and, hence, allow an accurate description of the fluctuation kinetics in fixed spots. Information on the fluctuation kinetics in tumor tissue as a whole is sparse. Recently, however, T2*-weighted MRI series have been used to identify blood flow fluctuations in tumor cross sections (21). The method of Baudelet et al. (21) allows determination of blood flow fluctuations with high time resolution, but does not provide quantitative blood flow data. In the present study, A-07 melanoma xenografts were subjected to repeated DCE-MRI and Kety analysis to quantify the changes in perfusion arising within an interval of 1 hour.

The DCE-MRI acquisitions were performed over a period of 15 min with a time resolution of 14 sec and a voxel size of 0.5 × 0.2 × 2.0 mm3. Previous studies of A-07 tumors in our laboratory have shown that the signal-to-noise ratio at this voxel size is sufficiently high to provide well-defined Kety curves and accurate E · F and λ values (7, 30), and this was confirmed in the present study. The peak of the Kety curves, even for the voxels showing the highest E · F values, appears beyond 60 sec after the administration of contrast (8). This is later than observed for most experimental tumors in mice (3). The late appearance of the peak is not a result of limitations in Gd-DTPA diffusion across the blood vessel wall, but reflects rather that A-07 tumors have a particularly large extracellular volume fraction (25, 26). Two observations suggest that a time resolution of 14 sec is sufficient for studying blood perfusion in A-07 tumors. First, the initial rising part of the Kety curves is determined precisely at this time resolution, owing to the late appearance of the curve peak. Second, comparative studies of E · F images assessed by DCE-MRI at a time resolution of 14 sec and tumor blood perfusion images assessed by invasive imaging have shown that E · F is an excellent parameter for blood perfusion in A-07 tumors (8). The total acquisition time of 15 min of the DCE-MRI recordings, however, represents a major limitation in our study. Fluctuations in tumor blood flow may occur at a wide range of frequencies (6, 12, 20). The duration of the periods with high or low blood flow may range from less than a minute to several hours in experimental tumors (31). Consequently, fluctuations in tumor blood flow occurring in the high-frequency range cannot be detected by the experimental procedure used here.

Since the outcome of Kety analysis of DCE-MRI data is sensitive to the AIF, a significant effort was made to determine the AIF correctly. Counterintuitively, the blood concentration of Gd-DTPA was found to be slightly lower after the second than after the first injection. Because small amounts of Gd-DTPA may remain in the blood 1 hour after the first administration, one might expect increased blood levels of Gd-DTPA following the second administration, as was observed by Garcia-Martin et al. (32). In our study the two Gd-DTPA boluses were injected through the same Neoflon, and it was seen that up to 0.02 mL of blood diffused into the Neoflon in the time between the two injections. Consequently, the second dose of Gd-DTPA was slightly lower than the first one, explaining why decreased blood concentrations of Gd-DTPA were observed following the second injection. This difference in Gd-DTPA blood levels was corrected for by using two separate AIFs. Qualitative inspections of the E · F and λ images from muscles as well as quantitative comparisons of the E · F and λ histograms confirmed that the use of two separate AIFs led to better reproducibility than using the same AIF twice.

The DCE-MRI experiments consisted of a first imaging sequence of 15 min, an intermission of 45 min, and a second imaging sequence of 15 min. It was of utmost importance that the mice did not move during the experiments and that their physiological condition was the same during the two DCE-MRI recordings, as highly reproducible imaging is required to avoid detection of false-positive and false-negative fluctuations in E · F. The reproducibility of the imaging was investigated in separate experiments by subjecting muscle tissue to DCE-MRI. Although muscle tissue may not be optimal for this purpose, we chose to carry out the control experiments with hindlimb muscles because these muscles have adequate size and can be imaged easily without moving artifacts. Ideally, the muscle and tumor imaging should have been performed simultaneously in the same mice. However, since it was not possible to detect adequate muscle tissue in the axial tumor scans, we chose to perform the control experiments by using coronal scans in separate animals. The control experiments showed that the E · F and λ values derived for muscle tissue were not significantly different for the two recordings. This observation suggests that the results obtained in the present work were not influenced significantly by possible time-variations in the experimental conditions, implying that systemic factors like depth of anesthesia and body core temperature were controlled adequately during the time the mice were kept in the magnet.

The measurements in muscle tissue were also used to establish criteria for differentiating between changes in tumor E · F (i.e., Δ(E · F) = E · Finj2E · Finj1) caused by fluctuations in blood perfusion and changes in tumor E · F due to experimental uncertainty. The experiments showed that the absolute value of Δ(E · F) rarely exceeded 0.03 mL/(g · min) and the absolute value of Δλ rarely exceeded 0.05 in muscle tissue. Based on these observations, only tumor voxels showing absolute values of Δ(E · F) > 0.03 mL/(g · min) and absolute values of Δλ < 0.05 were considered to represent regions with biologically significant fluctuations in blood perfusion. These criteria imply that λ was used to control the reproducibility of the imaging procedure. The use of λ as a control parameter was based on the assumption that the true λ did not change during or between two DCE-MRI recordings. This assumption is probably valid since λ is closely related to the ECVF, and the ECVF of A-07 tumors is not expected to change significantly within the observation period used here. In all tumors included in the study, a significant fraction of the voxels showed absolute values of Δλ > 0.05 and these voxels were, therefore, excluded from further analysis. Several factors, unique to tumor tissue, may have contributed to the high fraction of voxels with large absolute values of Δλ. First, the calculation of E · F and λ was based on 15-min-long DCE-MRI sequences. Rapid changes in blood perfusion may occur within this time interval, and these changes will most likely result in noise and, hence, in inaccurate determination of E · F and λ. Second, Gd-DTPA may accumulate in poorly perfused tumor regions with high ECVF in the time between two contrast administrations, and accumulation of contrast can alter the distribution kinetics of the contrast agent and, hence, lead to poor reproducibility of E · F and λ. Third, the Kety model breaks down in tumor regions with extensive necrosis, as reported earlier (7). A-07 melanoma xenografts are free from necrosis as long as the tumor volume does not exceed 600 mm3 (24, 25, 30). However, two of the tumors used in this study were sufficiently large to contain necrotic regions. Detailed examinations of our data confirmed that the Δλ-based exclusion criterion successfully excluded voxels localized in necrotic tumor regions, voxels localized in tumor regions in which Gd-DTPA accumulated between the two contrast administrations, and voxels suffering from noisy Gd-DTPA concentration curves. Importantly, the voxels localized in tumor regions accumulating Gd-DTPA, identified by subtracting the first baseline T1-weighted image from the second one, consistently showed absolute values of Δλ > 0.05. Consequently, by using λ as a control parameter it was possible to identify the voxels for which the Kety method did not work properly and to exclude these voxels from the Δ(E · F) images. However, the use of λ as a control parameter may have led to exclusion of voxels with actual changes in E · F, implying that the present analysis may have resulted in an underestimation of the occurrence of blood flow fluctuations in A-07 tumors.

Seven A-07 tumors were included in the present study, and all tumors showed significant evidence of fluctuations in blood perfusion. However, the fraction of voxels indicating blood flow fluctuations differed substantially among the tumors, as did the fluctuation pattern. One of the tumors showed coordinated changes in blood perfusion involving the center as well as the periphery of the tumor (Fig. 5d). Changes of this type probably reflect vasomotor activity and altered blood flow in primary feeding arterioles (22). Two tumors showed coordinated changes in blood perfusion involving peripheral tumor regions only [Fig. 4h (decrease in E · F) and Fig. 5b (increase in E · F)], whereas in the other four tumors the fraction of voxels indicating increased blood perfusion was similar to that indicating decreased blood perfusion (Fig. 5a,c,e,f). A common feature of the seven tumors is that they showed voxel clusters (n > 10 voxels) where the blood perfusion changed in the same direction in all voxels, and these clusters occurred more frequently in the tumor periphery than in the tumor center. Coordinated changes in blood perfusion in small subvolumes of tumors may be caused by intermittent stasis or vasomotor activity in tumor arterioles (22). The changes observed here were most likely caused by vasomotor activity rather than intermittent stasis, as studies of tumors growing in window chamber preparations have demonstrated that the duration of periods with total stasis is usually on the order of seconds, whereas arteriolar diameters can change with cycle times ranging from 15 to 60 min (31). Consequently, the fluctuations in blood perfusion that could be detected in the present study of A-07 tumors were probably caused primarily by arteriolar vasomotion, and these fluctuations occurred more frequently in the tumor periphery than in the tumor center.

This conclusion is consistent with measurements of the intratumor heterogeneity in other microvascular parameters in A-07 tumors. A-07 tumors show significant radial heterogeneity in blood perfusion, with high perfusion in the periphery and low perfusion in the center, as revealed by DCE-MRI (7, 33) and Bioscope imaging of the uptake of freely diffusible radioactive blood flow tracers (8, 34). The radial heterogeneity in blood perfusion reflects the radial heterogeneity in rate of angiogenesis, microvascular density, and intracapillary HbO2 saturation (34–36). Histological examinations have shown that the vascular network of A-07 tumors consists of pathological microvessels, probably recruited by tumor-induced neovascularization, and normal microvessels, probably recruited from the preexisting network in the skin during tumor growth (25). The latter vessels include terminal arterioles invested in vascular smooth muscle, and these vessels are observed only in peripheral regions of A-07 tumors. It is thus possible that vasomotor activity in these supplying arterioles caused the fluctuations in blood perfusion observed here. This suggestion is consistent with data reported by Baudelet et al. (37), who showed that the frequency of spontaneous fluctuations in T2*-weighted gradient echo signal intensity in FSa II tumors is particularly high in tumor regions with a large proportion of mature vessels.

In contrast to our observations, studies using the double staining mismatch technique have indicated that transient perfusion is most likely to occur in central regions of experimental tumors (38). The design of these studies allowed detection of tumor regions suffering from complete vascular shutdown caused by intermittent stasis, but did not allow detection of fluctuations in blood perfusion caused by vasomotor activity. Studies in our laboratory using the double staining mismatch technique have shown that complete vascular shutdown does usually not occur in A-07 tumors (39).

Low-frequent fluctuations in tissue pO2 have been shown to be a characteristic feature of experimental tumors (13, 23, 29, 31). Radiobiological and immunohistochemical experiments have shown that acute hypoxia is a commonly occurring phenomenon in A-07 tumors (24). OxyLite fiberoptic probes have been used for direct measurements of pO2 fluctuations in A-07 tumors, and these measurements have revealed fluctuations consisting of high-frequency / low-amplitude oscillations (i.e., ≈1 cycle/min with amplitudes of 1–2 mmHg) superimposed on lower-frequency / higher-amplitude oscillations (i.e., ≈1 cycle/hr with amplitudes up to ≈10 mmHg) (14). A significant fraction of these pO2 fluctuations showed frequencies corresponding well to those of the blood flow fluctuations detected in the present work, suggesting a causal relationship between blood flow fluctuations and changes in pO2 in A-07 tumors. This suggestion is consistent with simultaneous measurements of blood flow and tissue pO2 in experimental tumors published recently by Lanzen et al. (40). It is thus possible that Δ(E · F) images may provide information on the extent of acute hypoxia and its spatial location in tumors. If so, the present study suggests that acute hypoxia may occur most frequently in the periphery in A-07 tumor, close to the surrounding normal tissue.

Fluctuations in blood perfusion and pO2 in tumors may lead to periods of transient hypoxia and, hence, reduced sensitivity to treatment with ionizing radiation and some chemotherapeutic agents, but also increased aggressiveness caused by hypoxia-induced changes in gene expression (15–18). There is an increasing interest in integrating biological information in radiotherapy treatment planning to target treatment-resistant and aggressive tumor regions. The results presented here suggest that it may be extremely difficult to target acutely hypoxic tumor regions efficiently during radiotherapy. First, fluctuations in blood perfusion and, hence, acute hypoxia are most likely to occur near the normal tissue, making it intractable to boost the tumor regions of interest without increasing the dose to the surrounding normal tissue. Moreover, the tumor regions experiencing changes in blood perfusion and tissue pO2 are often small, making it difficult to prepare radiation fields of satisfactory accuracy.

In summary, the present data suggest that Δ(E · F) maps can provide valuable information regarding the size and spatial distribution of tumor regions undergoing temporal changes in blood perfusion. Tumor blood perfusion may vary substantially within a period of 1 hour. In A-07 melanoma xenografts, blood flow fluctuations are most likely to occur in the tumor periphery, and are probably caused by vasomotor activity in supplying arterioles.

Ancillary