Correlation of MRI Biomarkers with Tumor Necrosis in Hras5 Tumor Xenograft in Athymic Rats

Abstract

Magnetic resonance imaging (MRI) can measure the effects of therapies targeting the tumor vasculature and has demonstrated that vascular-damaging agents (VDA) induce acute vascular shutdown in tumors in human and animal models. However, at subtherapeutic doses, blood flow may recover before the induction of significant levels of necrosis. We present the relationship between changes in MRI biomarkers and tumor necrosis. Multiple MRI measurements were taken at 4.7 T in athymic rats (n = 24) bearing 1.94 ± 0.2-cm³ subcutaneous Hras5 tumors (ATCC 41000) before and 24 hours after clinically relevant doses of the VDA, ZD6126 (0–10 mg/kg, i.v.). We measured effective transverse relaxation rate (R₂*), initial area under the gadolinium concentration-time curve (IAUGC_60/150), equivalent enhancing fractions (EHF_60/150), time constant (K^trans), proportion of hypoperfused voxels as estimated from fit failures in K^trans analysis, and signal intensity (SI) in T₂-weighted MRI (T₂W). ZD6126 treatment induced > 90% dose-dependent tumor necrosis at 10 mg/kg; correspondingly, SI changes were evident from T₂W MRI. Although R₂* did not correlate, other MRI biomarkers significantly correlated with necrosis at doses of ≥ 5 mg/kg ZD6126. These data on Hras5 tumors suggest that the quantification of hypoperfused voxels might provide a useful biomarker of tumor necrosis.

Introduction

The growth of solid tumors requires a functional vascular network to supply oxygen and nutrients and to remove waste products [1–3]. In contrast to antiangiogenic approaches, vascular-damaging agents (VDAs) such as ZD6126 selectively aim to destroy existing tumor blood vessels, leaving normal blood vessels relatively unaffected [4]. ZD6126 is a phosphate prodrug that, in vivo, is rapidly converted by phosphatases in serum to ZD6126 phenol. ZD6126 phenol binds to tubulin, inhibiting microtubule assembly, thereby destabilizing the intracellular microtubule network. Disruption of the microtubule network selectively affects tumor blood vessels because immature proliferating endothelial cells are thought to rely, at least in part, on a microtubule cytoskeleton to maintain cell shape and function [5]. In contrast, mature endothelial cells have a well-developed actin cytoskeleton that provides structural support [6]. Therefore, ZD6126 treatment leads to selective rounding up of immature endothelial cells, disrupting the endothelial cell monolayer lining the tumor vasculature, reducing blood flow [7], and inducing blood vessel occlusion, with subsequent tumor cell death due to nutrient deprivation [8]. In preclinical models, VDAs have been shown to induce extensive necrosis throughout the tumor, usually within 24 hours [9–12]. Characteristically, a rim of viable tumor survives treatment, as these cells are sustained by blood vessels in surrounding nontumor tissues, which remain unaffected by VDA treatment. For a description of VDAs and the principal actions of ZD6126, see Siemann et al. [13]. Due to rapid tumor regrowth from this viable rim, the antitumor activity of single doses of VDAs, as assessed by changes in tumor size, usually appears modest despite the induction of extensive tumor cell necrosis [9]. Therefore, monitoring tumor size alone is unlikely to provide a reliable biomarker for the underlying biologic activity or antitumor effects of single doses of VDAs. Consequently, other methods to measure the antivascular effects of VDAs, such as pharmacodynamic magnetic resonance imaging (MRI) biomarkers, have been investigated both clinically and preclinically [14–17]. In contrast to single doses, multiple doses of VDAs, or a combination with therapeutic approaches targeting the surviving tumor rim, have demonstrated significant tumor growth delay or tumor shrinkage in preclinical models [5,9,18,19].

Dynamic contrast-enhanced (DCE) MRI permits an indirect and noninvasive measurement of tumor function in both human and animal models. Different approaches can be taken toward the analysis of DCE-MRI data. For example, initial area under the gadolinium contrast agent concentration-time curve (IAUGC) data from a region of interest (ROI) or from single voxels can be used as a biomarker of tumor physiology. More complex compartmental modeling has attempted to derive other biomarkers of tumor physiology, such as K^trans [the volume transfer constant for a contrast agent into the extravascular extracellular space (EES)], V_e (the volume of the EES), and k_ep (the rate constant for the backflux from the EES to the intravascular space) [20–22]. For an extensive review of different MR approaches for measuring tumor angiogenesis and its pharmacological modulation, see Leach et al. [20], drup-Link et al. [23], Jeswani and Padhani [24], and Kiessling et al. [25].

DCE-MRI has been used to measure the effects of VDAs on humans and has shown decreases in both tumor K^trans and IAUGC [15,16]. For example, after a single dose of ZD6126, Evelhoch et al. [15] showed an acute decrease in the IAUGC, with only partial recovery at later times. Similarly, Galbraith et al. [16] demonstrated that a single dose of combretastatin A4 phosphate (CA4P) caused a reduction in IAUGC showing partial recovery over 24 hours and that multiple CA4P doses in one patient maintained a reduction in K^trans from baseline. Similar changes in DCE-MRI biomarkers have also been described in either ZD6126-treated or CA4P-treated animal xenograft tumor models. However, Evelhoch et al. [15], Galbraith et al. [16], and Maxwell et al. [26] were unable to clearly correlate drug-induced changes in tumor DCE-MRI biomarkers with induced tumor necrosis. Robinson et al. [17] demonstrated a dose-dependent reduction in tumor IAUGC and R₂* 24 hours after ZD6126 administration, which was associated with increased tumor necrosis. Emerging MR techniques that measure vascular maturation [27], size [28–30], and function [27,31] are also of interest to the research fields of vascular physiology and therapeutic modulation.

MRI uniquely permits the acquisition of multiple contrasts in one sitting. However, the underlying biologic process dictates that contrast in humans is often not accessible and that, if biopsies are available, they are limited in their frequency and tumoral location. The current study was undertaken to explore the relationships, at clinically relevant doses, between ZD6126-induced tumor necrosis and MRI biomarkers in a tumor xenograft model to identify potential acute biomarkers for ZD6126-induced tumor necrosis in patients. The Hras5 transformed mouse NIH 3T3 fibroblast tumor line was selected because it has a reproducibly low level of background tumor necrosis. MRI protocol included multislice T₂-weighted MRI (T₂W), R₂*, enhancing fraction (EHF), IAUGC, and K^trans measurements, all performed within the same experiment. This imaging protocol exploited those imaging biomarkers most used in phase I clinical investigations of abdominal tumors exposed to vascular-modulating therapeutic regimes [15,16,32] and included more investigational biomarkers. In addition, multislice hematoxylin-eosin (H&E) histology was used to measure tumor necrosis. In parallel studies, it was shown that a single dose of 2.5 to 12.5 mg/kg ZD6126 in rats produced plasma levels of ZD6126 phenol broadly equivalent to those achieved in humans at single doses of 40 to 112 mg/m² (data not presented here). Therefore, in this study, we focused on exploring the effects of ZD6126 within this clinically relevant dose range (2.5–10 mg/kg).

Materials and Methods

Rats were given food and water ad libitum. All animal procedures were performed in full compliance with licenses issued under the UK Animals (Scientific Procedures) Act following local ethical committee review and with United Kingdom Coordinating Committee on Cancer Research guidelines.

Tumor Development and Scheduling

Male athymic rats (V 200 g) bearing subcutaneous hindflank Hras5 tumors were established (mean tumor volume ± SEM, 1.94 ± 0.2 cm³). The Hras5 cell line was isolated from mouse NIH 3T3 fibroblasts transfected with pT24C3 plasmid containing a mutant, the transforming form of the HRAS1 gene (ATCC 41000) [8].

Three groups of 15 animals implanted at 10-day intervals. The animals were randomized by tumor size 3 days before their first imaging session, and 10 animals per implant group were recruited into the imaging study. Animals underwent two imaging sessions: 24 hours before and 24 hours after ZD6126 (2.5, 5, 7.5, or 10 mg/kg) or vehicle administration intravenously. ZD6126 was prepared as described elsewhere [5].

Animal Preparation

Restraint during MRI was achieved by anesthesia with an isoflurane/air mixture (1.5% isoflurane at 2 l/min air). Each animal was prepared and imaged in a purpose-built polymethylmethacrylate-lidded bed. The tail was warmed, a vein was catheterized using a 26-gauge catheter, and a cannula containing gadopentetate dimeglumine (0.3 mmol/kg Gd-DTPA, “Magnevist”; Schering,Berlin,Germany) and heparinized saline was connected. Forepaws were then connected to “wrap-around” silver electrodes connected to a respiration device and an electrocardiogram monitoring device (SA Instruments, Stony Brook, NY). Once in the magnet, the temperature was monitored and maintained at 37°C by a home-built thermocouple system and a flow of warm air.

MRI and Tumor Histology

MRI was carried out in a 400-mm-diameter horizontalbore 4.7-T magnet (Varian, Palo Alto, CA) using a 63-mm quadrature birdcage transmit/receive radiofrequency coil.

Each animal underwent a fast gradient-echo coronal pilot scan to confirm positioning within the radiofrequency coil. A fast spin-echo (SE) sagittal sequence then provided anatomic images to determine the coordinates of tumor slices. Except in multigradient echo (MGRE) and rapid acquisition with relaxation enhancement (RARE) scans, subsequent acquisitions employed a field of view of 80 x 60 mm with four contiguous 2-mm sagittal slices through the tumor interleaved with a fifth transverse 2-mm slice through dorsal spinal muscle. Firstly, R₂* measurements were taken using a four-slice MGRE technique (matrix = 128 x 64; average = 12; repetition time T_R = 11 milliseconds; echo time T_E = 5, 10, 15, 20, 25, 30, 35, and 40 milliseconds). Secondly, a fat-suppressed fast four-slice RARE sequence provided a heavily T₂-weighted image to permit good tumor delineation and qualitative morphologic interpretation (matrix = 128 x 128; echo train length = 8; echo spacing = 10 milliseconds; average = 2; T_R = 1 second; effective T_E = 40 milliseconds). Thirdly, a five-slice SE sequence (T_R = 0.5, 2, and 10 seconds; matrix = 128 x 64; average = 2; T_E = 9 milliseconds) provided a precontrast T₁ map. Finally, a precontrast set of five five-slice T₁-weighted SE images (T_R = 0.12 seconds; matrix = 128 x 64; average = 2; T_E = 9 milliseconds) was acquired followed by the commencement of DCE-MRI scan in the same five-slice locations for a further 10 minutes in which the manual injection of contrast agent over 3 seconds occurred.

After the second imaging time point at 24 hours after ZD6126 treatment, each animal was killed using a technique specified in Schedule 1 of the UK Animals (Scientific Procedures) Act 1986, and the tumor was excised and placed in formalin fixative for H&E processing and necrosis measurements. For each tumor, necrosis was determined from five tumor slices with 200 µm of separation using operator-guided image analysis software. The area of necrosis within the whole-tumor section was determined visually, and the proportion of necrotic nonviable tumor over the whole section was calculated using image analysis software written for the Zeiss KS400 version 3.0 image analyzer (Carl Zeiss Vission GmbH, Hallbergmoos, Germany). For each group, the mean percentage of necrosis and standard error were calculated. The results are presented as the mean tumor necrosis (%) for all tumors (five slices per each tumor) in each treatment group.

In the current investigation, 30 animals were entered, and 24 complete data sets were analyzable. Data were excluded if they did not satisfy predetermined inclusion criteria, including injection failure, poor tumor implant as determined from T₂W image, or experimental/acquisition failure. There was no detectable change in muscle values after ZD6126 treatment.

MRI Data Analysis

Whole-tumor ROI were manually drawn over T₂-weighted high-spatial-resolution images and used to segment the tumor and to extract the biomarkers described below.

Tumor R₂* (sec^-1) maps were generated for each slice, using all eight gradient-echo images and fitting an exponential model voxelwise [17]. For each slice, R₂* was determined for an ROI encompassing the whole tumor, but excluding muscle and skin. R₂*, the effective transverse relaxation rate, is the sum of the homogeneous component R₂ and the inhomogeneous component R₂′. An increase in R₂* may arise from an increase in the intracellular compartment tissue content of paramagnetic iron species, primarily deoxyhemoglobin (dHb), consistent with: 1) a reduction in perfusion; 2) an increase in metabolic activity uncoupled to a vascular response (increase in the concentration of intracellular dHb); 3) congestion of erythrocytes after ZD6126 thrombosis resulting in erythrocyte deoxygenation (increase in intracellular dHb); or 4) intratumoral bleeding, all of which can ultimately affect the overall heterogeneity of the tissue under investigation. A decrease in R₂* may arise from a decrease in R₂′, reflecting an increase in tissue water associated with edema and/or necrosis or vessel collapse.

T₁ values were measured immediately before the injection of contrast agent using multi-T_R T₁W scan. Changes in T₁ values during DCE-MRI acquisition were estimated from changes in signal intensity (SI) obtained for each voxel, given baseline T₁ values. From T₁W images, voxelwise IAUGC, and EHF_60/150 and K^trans biomarkers were calculated. For each tumor, a median IAUGC of the whole tumor was determined over the first 60 and 150 seconds (mM sec) from the four imaging slices. Tumor IAUGC voxel values greater than the muscle median IAUGC were defined as enhancing voxels. EHF is the fraction of tumor voxels defined as, thus, enhancing. Group values were then presented as the mean of individual animals' voxel medians. Muscle ROI were drawn on T₁W precontrast agent scans.

K^trans (sec^-1) was obtained by pharmacokinetic modeling using the model of Tofts and Kermode (Eq. (1)), which was implemented using a software written in this laboratory. The Nelder-Mead Simplex Method was used when fitting the data using a program developed in-house using MATLAB software (MathWorks, Natick, MA) [33]. Concentrations of Gd-DTPA were derived using the known relaxivity of gadopentetate from the time course of T₁ values in plasma and tumor determined from DCE-MRI series. Vascular input function (VIF) was modeled by:

$${[\text{Gd}]}_{\text{t}}^{\text{plasma}}=D[a_1\exp (-m_{1}t)+a_2\exp (-m_{2}t)],$$

where D = 0. 26388 mmol/kg, the bolus dose of Gd-DTPA; a₁ and m₁ reflect the equilibration of Gd-DTPA between plasma and extracellular space; a₂ and m₂ represent the kidney clearance of Gd-DTPA; and (a₁ + a₂)^-1 (l/kg) represents the plasma volume of the rat per unit of body weight. The values a₁ = 17, a₂ = 7, m₁ = 0.03, and m₂ = 0.001 were obtained from unpublished studies in athymic rats in this laboratory and agree with earlier reports [34].

Not all voxels provide valid values of K^trans. For some voxels, the Nelder-Mead Simplex Method fit fails to converge, whereas some give values that would be classified as nonphysiological (K^trans < 0 or > 0.01 sec^-1). These fit failures were set to zero, and the data, both excluding and including fit failures, are presented.

Mean whole-tumor values from the four imaging slices for each tumor from each animal were calculated before and after treatment. Significance testing used one-tailed unpaired Student's t test: the biomarker compared was the change in IAUGC, K^trans, or EHF in the ZD6126-treated group compared to that in the vehicle-treated group. Data are presented as mean ± SEM, unless otherwise stated.

Results

ZD6126 Induces Extensive Tumor Necrosis in Hras5 Xenografts in Nude Rats

Tumor volume, as determined by caliper measurements, was 1.94 ± 0.2 cm³ before ZD6126 treatment. Significant tumor growth inhibition would not be anticipated after only one dose of ZD6126 at any dose [9]. Consistent with previous reports [8], ZD6126 induced extensive (> 90%) central tumor necrosis in the Hras5 model 24 hours after drug treatment, with viable tumor cells observed only at the periphery of the tumor at a dose of 10 mg/kg (Figures 1/and 2a). In contrast, untreated tumors showed a low background level of necrosis (Figures 1i and 2a). A dose response was observed, with a significant increase in tumor necrosis at ZD6126 doses of ≥ 5.0 mg/kg (Figure 2a), with some regions of heterogeneous staining on H&E sections at intermediate doses of 5.0 and 7.5 mg/kg (Figure 1k).

Figure 1.

T₂W images of Hras5 tumors before and after ZD6126 treatment. T₂W tumor images 24 hours before (a–d) or 24 hours after (e–h) treatment with vehicle (a and e) or ZD6126 at 5.0 mg/kg (b and f), 7.5 mg/kg (c and g), or 10.0 mg/kg (d and h). Visual inspection revealed a broadly homogenous signal across the tumor mass in pretreated and vehicle-treated animals. After ZD6126 treatment, heterogeneity in SI within tumors was seen (f–h). One hour after the second MRI, tumors were excised. Representative H&E- stained tumor sections are shown for animals' treatment with vehicle (i) or ZD6126 at a dose of 5.0 mg/kg (j), 7.5 mg/kg (k and m), or 10.0 mg/kg (l and n), respectively. Necrotic (N) and viable (V) regions of the tumor are identified. Vehicle-treated animals showed a low level of spontaneous tumor necrosis (i), whereas higher doses of ZD6126 resulted in extensive central tumor necrosis with smaller regions of viable tissue toward the periphery (k and l). After 10 mg/kg ZD6126, there is a clear boundary between the necrotic core and the viable tumor rim (n), whereas at lower doses, there was evidence of heterogeneity across the tumor (m).

Figure 2.

Analysis of necrosis and MRI biomarkers in Hras5 tumors following ZD6126 treatment. (a) Necrosis (mean ± SEM; %) was assessed in H&E-stained sections. ZD6126 at doses of ≥ 5.0 mg/kg produced significant increases in tumor necrosis 24 hours after treatment. Whole-tumor R₂* (mean ± SEM; msec^-1), IAUGC₆₀ (mean ± SEM; mmol/sec), and EHF (mean ± SEM) are shown in (b)–(d), respectively. Pretreatment (open bars) and posttreatment (hatched bars) values are shown for each dose group. *P < .05, **P < .01; one-tailed unpaired t-test compared with vehicle.

T₂W Tumor Images Reveal Varied Tumor Heterogeneity in Ascending Doses of ZD6126

Figure 1 illustrates T2W data for four tumors from different animals before (Figure 1, a–d) and 24 hours after (Figure 1, e–h) a single dose of 0, 5, 7.5, or 10 mg/kg ZD6126. Changes in T₂W SI are seen following ZD6126 treatment. However, a heterogeneous response in signal changes over the whole tumor is seen after different doses of ZD6126. For example, a hypointense core is seen in the tumor of animals treated with 10 mg/kg, corresponding to a region of necrosis (Figure 1h, l, and n for corresponding tumor H&E sections from the same animals). In contrast, the animals treated with 5.0 or 7.5 mg/kg ZD6126 showed a heterogeneous T₂W SI within the tumor core, consistent with the heterogeneity seen on H&E sections (Figure 1, f, g, j, k, and m, for corresponding tumor H&E sections from the same animals). Such changes in T₂ SI may reflect different necrotic states (e.g., acute or chronic necrosis and/or hypoxia or varied types of tissue integrity as a result of different doses of ZD6126). Interestingly, “penumbral”-like regions are evident in K^trans maps illustrated in Figure 3. This observation appeared consistent within the different dose groups (data not shown).

Figure 3.

K^trans in Hras5 tumors following ZD6126 treatment. (a) Single-slice K^trans maps are illustrated 24 hours before and 24 hours after ZD6126 administration. K^trans maps illustrate characteristic VDA effects. In both pretreatment and vehicle-treated tumors, there are some regions of low K^trans, possibly corresponding with spontaneous necrosis. In contrast, 24 hours after treatment with ZD6126, the number of fitted K^trans values in the tumors became progressively lower until < 50% of tumor voxel values fitted the model. Whole-tumor K^trans measurements (mean ± SEM; sec^-1) before (open columns) or after (hatched columns) treatment arshown either including (b) or excluding (c) zero fit-failure values. *P < .05, one-tailed unpaired t-test compared with vehicle.

R₂* Values Significantly Decrease 24 Hours after ZD6126 Treatment

Figure 2b illustrates changes in whole-tumor mean R₂* values 24 hours before and after treatment with vehicle or ZD6126. Compared with vehicle-treated controls, a significant decrease in R₂* was observed after ZD6126 treatment at doses of > 5 mg/kg. This observation agrees with earlier reports showing a decrease in R₂* after administration [17].

EHF and IAUGC_60/150 Changes after ZD6126 Administration

Figure 2c summarizes the mean change in tumor IAUGC₆₀ in each treatment group 24 hours after ZD6126 administration. Tumor IAUGC₆₀ 24 hours after 5, 7.5, and 10 mg/kg ZD6126 administration was lower than those in vehicle-treated controls (changes in pre to post ZD6126 in each group: -52 ± 13%, P = .02; -70 ± 25%, P = .05; -77 ± 19%, P = .1). Although the decrease after 10 mg/kg ZD6126 did not reach statistical significance, this treatment group had a mean IAUGC value before drug treatment lower (IAUGC₆₀ = 4.22 ± 1.23 sec^-1) than those of vehicle controls (IAUGC₆₀ = 5.82 ± 0.87 sec^-1) or other ZD6126 treatment groups (IAUGC₆₀ range = 7.67 ± 0.33 to 9.04 ± 1.19 sec^-1), which may have limited the capability to detect a statistically significant decrease due to VDA therapy. Analysis of IAUGC₁₅₀ produced similar results (data not shown).

Tumor IAUGC values greater than the muscle median IAUGC were defined as highly enhancing pixels. This threshold was not chosen to compare muscle and tumor, but rather to identify the proportion of the tumor that enhanced significantly (EHF). Figure 2d summarizes the mean changes in EHF₆₀ observed 24 hours after ZD6126 treatment. After 5, 7.5, or 10 mg/kg ZD6126, there was a reduction in EHF₆₀ compared with vehicle-treated controls (changes in pre to post ZD6126 in each group: -35 ± 2.8%, P = .05; -61 ± 19%, P = .058; -63 ± 9%, P = .002). Although the animals treated with 7.5 mg/kg ZD6126 did not reach statistical significance, there was a strong trend indicating a decrease. Analysis of EHF₁₅₀ showed similar results (data not shown).

Analysis of K^trans and K^trans Fit Failures Illustrates Spatial Changes across the Tumor after ZD6126 Treatment

Figure 3 summarizes the mean K^trans results from the whole tumor, including and excluding zeroed K^trans fit failures before and after ZD6126 administration. Vessel occlusion will drastically compromise the inflow of contrast agent and will result in poorly enhancing regions. Where K^trans is < 0 or > 0.01 sec^-1, voxels were classified as fit failures. Separate analyses were carried out either: 1) excluding fit-failure voxels, or 2) setting the value of fit-failure voxels to zero and including these in the analysis. The decrease in mean tumor K^trans is significantly greater than that in controls 24 hours after treatment with either 5 or 7.5 mg/kg ZD6126 whether the analysis included or excluded fit failures (Figure 3, b and c). The failure to detect significant changes 24 hours after 10 mg/kg ZD6126 may be associated with very few voxels fitting after treatment, but those that did largely skewed the data. Overall, such results suggest that the perfusion status of the rim was also affected by ZD6126 treatment. Nevertheless, H&E results indicated that the rim contained viable tumor cells (Figure 1, i–l).

K^trans maps illustrated the effects of ZD6126 across whole Hras5 tumors (Figure 3a). Vehicle-treated and pre-ZD6126-treatment tumors showed voxels with low K^trans values with small regions where K^trans has failed to fit (dark blue) into the core. The latter may correspond to areas of background (non-drug-induced) tumor necrosis, which have been observed, albeit at a low level, in the Hras5 tumor model [9]. After 5, 7.5, and 10 mg/kg ZD6126, the number of fitted K^trans voxels in the tumor decreased until, in some examples, < 35% of tumor voxels had K^trans values that satisfied inclusion criteria (Figure 4). As the dose of ZD6126 increased, fitted K^trans voxels were more limited to the rim of the tumor, with fit failures becoming more prevalent in core regions (Figure 3a). At intermediate does of ZD6126, qualitative observations recognized a distinct penumbral region between necrotic and viable tumor regions discernable on T₂W images and H&E slides. Similarly, K^trans maps suggested that these regions, although poorly perfused, were not lacking functional vascularity and maintaining a viable microenvironment (as determined by the viable rim from H&E slides) and that these poorly perfused regions may even extend to the rim of the Hras5 tumor 24 hours after ZD6126 treatment. Thus, even though fitted voxels in the viable rim of the tumor retain a degree of contrast agent flux into the EES within a given volume, there is a significant decrease in K^trans of the tumor periphery after 5 and 7.5 mg/kg (Figure 3, b and c).

Figure 4.

K^trans fit failures in the Hras-5 tumor following ZD6126 treatment. The proportion (mean ± SEM; %) of voxels that failed to fit the Tofts and Kermode model was calculated before (open columns) and after (hatched columns) ZD6126 treatment. Mean tumor necrosis is also plotted for reference (π). In pretreated tumors, the proportion of voxels failing to fit to the model was low, which suggests that the Hras5 tumor is well perfused with contrast agent. However, following ZD6126 treatment at doses of ≥ 5 mg/kg, the proportion of fit failures increases markedly, indicating inaccessibility for the contrast agent.

Figure 4 shows the extent of necrosis and the proportion of voxels that were classified as K^trans fit failures within each treatment group before and after ZD6126 treatment. This analysis confirmed the qualitative assessment of K^trans maps in Figure 3a. Indeed, before ZD6126 administration and at zero and low doses, few or no voxels could not be fitted with the K^trans algorithm, suggesting that there was significant contrast agent flux within the tumor. In contrast, after ZD6126 administration at doses of ≥ 5 mg/kg, the number of voxels failing to fit greatly increased. Overlaying the degree of necrosis on this plot illustrates the relationship between histology measure and that using DCE-MRI.

Finally, ZD6126-induced changes in MRI measurements taken 24 hours after drug treatment were related to the percentage of tumor necrosis on an animal-by-animal basis. R₂* showed a poor correlation with the degree of necrosis (R₂ < 0.1, P > .1). K^trans fit failures, EHF_60/150, IAUGC_60/150, and K^trans, including fit failures set to zero, all displayed a significant correlation with tumor necrosis, with the latter being largely affected by the fit failures being set to zero (K^trans fit failures: R₂ = 0.69, P < .001; EHF_60/150: R₂ = 0.61, P < .001; IAUGC_60/150: R₂ = 0.32, P < .002; K^trans including fit failures: R₂ = 0.545, P < .001). Figure 5 illustrates the correlation between K^trans fit failures and tumor necrosis.

Figure 5.

A strong and significant correlation between fit failures and tumor necrosis in the Hras5 tumor model. The proportion of fit failures (%) and tumor necrosis 24 hours after vehicle or ZD6126 treatment was calculated on an animal-by-animal basis (n = 24).

Discussion

Acute events associated with VDA therapy include tumor and endothelial cell apoptosis, vascular collapse and thrombus formation, tumor hypoxia, and fluctuations in extracellular pH and oxygenation. Subsequent events may include vascular remodeling, tumor and endothelial cellular ischemia and necrosis, hemorrhage, and interstitial fluid pressure changes [5,9,35–38]. Although DCE-MRI has been utilized in clinical studies to monitor the effects of VDAs on the tumor vasculature, there has been little work to date to assess other clinically important biologic endpoints such as drug-induced tumor cell necrosis and hypoxia in these patients. In addition, there has been little preclinical evidence correlating VDA-induced changes in tumor MRI biomarkers with tumor necrosis at clinically relevant drug exposure levels.

In this study, we have measured multiple noninvasive MRI biomarkers (T₂W, R₂*, IAUGC_60/150, EHF_60/150, and K^trans) in Hras5 tumors in nude rats 24 hours before and after treatment with ZD6126. We looked for correlations between these imaging endpoints and ZD6126-induced tumor necrosis 24 hours after treatment. We selected the Hras5 tumor model for these studies as it has been reported to have a low fraction of background necrosis in untreated animals, even in established tumors [8,9]. In this study, background necrosis was < 10%, confirming earlier reports. Doses of ZD6126 that were ≥ 5.0 mg/kg produced significant increases in tumor necrosis, although 2.5 mg/kg ZD6126 had no significant effect. The marked difference between the 2.5-mg/kg and the 5.0-mg/kg dose groups suggests that the antitumor effects of ZD6126 may have a very steep dose response, or perhaps a threshold dose, below which tumor necrosis is not induced [8,16]. More detailed studies within the dose range 2.5 to 5.0 mg/kg are required to investigate this further.

Clinical studies investigating ZD6126 as a single agent once every 3 weeks have shown over the dose range studied (5–112 mg/m²) that plasma ZD6126 phenol levels generally increased in a linear fashion, reaching an AUC of 3894 ng hr/ml at 112 mg/m² [39,40]. In parallel pharmacokinetic studies, we have shown that doses of up to 10 mg/kg produced a similar plasma exposure of ZD6126 phenol in nude rats [41], supporting the concept that the dose range examined in these studies is clinically relevant.

In this study, T₂W images were acquired primarily to provide tumor delineation for R₂* and DCE-MRI measurements. Although it is accepted that T₂W SI can depend on many factors, these rapidly acquired and mapping independent images of endogenous T₂ of Hras5 tumors showed a qualitative difference 24 hours after treatment with ZD6126. Images presented from a previous study illustrate a degree of heterogeneity in the T₂W images of the tumor after CA4P administration [42], but not to the degree observed in the present experimental conditions. Additionally, in the present study, heterogeneous penumbral areas were apparent on both T₂W and H&E images, particularly at intermediate doses, possibly representing viable—but morphologically appearing as—hypoxic areas of tumor. However, immunohistologic confirmation of this would require further studies (e.g., employing a probe specific for hypoxic conditions) [43].

The whole-tumor R₂* data in the Hras5 model showed a significant decrease 24 hours after ≥ 5 mg/kg ZD6126, compared with controls. Robinson et al. [17] hypothesized that R₂* would increase after ZD6126 therapy as a response to a relative increase in dHb under hypoxic conditions as a result of reduced tumor perfusion. However, that study found that R₂* decreased 24 hours after ZD6126 treatment in two tumor models. A further study demonstrated that R₂* increased shortly after ZD6126 dosing (within 35 minutes), consistent with the original hypothesis, but confirmed that R₂* significantly decreased 24 hours later [44]. The biologic process(es) underlying the paradoxical decrease in R₂* is unknown, but may involve denaturation of dHb within necrotic tumor tissues [44]. Nevertheless, this parameter represented the weakest correlation with necrosis.

IAUGC is an MRI biomarker that has been commonly applied in both preclinical and clinical DCE-MRI examinations. Although measurements of tumor IAUGC have the advantage of being relatively robust and straightforward to calculate, this biomarker has a complex link to underlying tissue physiology and contrast agent kinetics. Nevertheless, IAUGC data can be a precedent indicator to more complex modeling results. Indeed, some literature has presented a positive correlation between IAUGC and the more complexly derived K^trans values in colorectal liver metastases in humans [45]. In agreement, the current IAUGC and K^trans show a strong and significant correlation at baseline (R² = 0.91, P > .001; data not presented). In the present study, we have demonstrated significant decreases in IAUGC_60/150 following ZD6126. Evelhoch et al. [15] have performed DCE-MRI measurements on mice bearing murine C38 tumors immediately before and 24 hours after ZD6126 administration. Extensive necrosis was observed 24 hours after ZD6126 administration, and this was associated with a reduction in the median tumor IAUGC. However, the ZD6126 doses used in these studies would be expected to produce plasma ZD6126 phenol levels substantially higher than those yet achieved in the clinic [5,8,39,40]. Two studies have investigated DCE-MRI changes after single doses of CA4P in rats bearing P22 carcinosarcomas at 1, 6, or 24 hours after drug administration [16,26]. Together, these studies demonstrated a prolonged reduction in IAUGC at a high CA4P dose, but with full or partial recovery after 24 hours at more clinically relevant doses.

EHF is a ratio measure using the median muscle IAUGC signal change as a threshold for tumor signal changes [17]. Using muscle or normal tissue as a reference can permit a subject-controlled calculation of IAUGC without the requirement for vascular input function. In the current study, IAUGC from the muscle was not significantly different either day to day, irrespective of treatment, or between animals. At doses of ≥ 5 mg/kg ZD6126, there were significant decreases in EHF_60/150 in Hras5 tumors. However, although the group treated with 7.5 mg/kg did not reach statistical significance (P = .058), the data obtained from each animal suggest a trend to decrease 24 hours after treatment. In an earlier study, Robinson et al. [17] demonstrated that ZD6126 induced a dose-dependent decrease in EHF in the GH3 prolactinoma tumor model in Wistar-Furth rats, which appeared to correlate with increased tumor necrosis 24 hours after ZD6126 treatment. Compared with the study reported here, higher doses of ZD6126 (≥ 25 mg/kg) were necessary to demonstrate statistically significant increases in tumor necrosis and decreases in EHF. At a lower dose (12.5 mg/kg), IAUGC was reduced, but only in restricted areas of the tumor. Pharmacokinetic studies used to determine the relationship between ZD6126 dosing and plasma ZD6126 phenol levels were not reported in these earlier studies, so it is uncertain whether the doses used were clinically relevant [17]. In addition, the GH3 model has higher and more variable levels of background tumor necrosis when compared with Hras5, perhaps necessitating higher ZD6126 doses to demonstrate statistically significant changes both in necrosis and IAUGC. Alternatively, the intrinsic sensitivity of the tumor vasculature to the effects of VDAs may have some level of animal strain or tumor model dependency [36]. Nonetheless, previous work has demonstrated consistent antitumor effects of ZD6126 in a histologically diverse panel of tumor xenografts [9].

Quantitative pharmacokinetic modeling techniques can be applied to changes in contrast agent concentrations in a tissue or pathology of interest. Using pharmacokinetic modeling, vascular kinetic biomarkers associated with underlying physiological processes can be derived and monitored in response to treatments. The K^trans data presented here were derived from the Tofts and Kermode model using a set of inhouse modeled VIF parameters. In the current study, the mean K^trans for the whole Hras5 tumor was significantly reduced 24 hours after treatment with doses of 5 and 7.5 mg/kg ZD6126. Furthermore, when only fitted voxels (i.e., those that were not set to zero) were evaluated, treatment effects remained. Earlier results have described the retention of a perfused and histologically viable rim of tumor as a characteristic response to VDAs. This phenomenon is thought to be due to the diffusion of oxygen and nutrients from adjacent vasculature within neighboring nontumor tissues. Nevertheless, the current results suggest that not only does ZD6126 cause massive central necrosis but also causes a prolonged disturbance of contrast flux of the tumor rim because H&E histology shows a rim occupied by viable tumor cells, but MRI data suggest that this region is functionally poorly perfused, even 24 hours after drug treatment. In support of this interpretation, Prise et al. [37] showed that following treatment with a clinically relevant dose of CA4P, blood flow was shut down across the whole tumor, including the rim. Blood flow gradually returned, recovering more quickly in the rim compared with central regions, but reaching pretreatment levels only 48 to 96 hours after drug dosing.

In the present study, the final R₂* measurement across the whole tumor did not correlate well with necrosis. Indeed, R₂* has been shown to both increase and decrease after VDA treatment, possibly reflecting both vascular effects and underlying pathological changes within the tumor [17,44]. Furthermore, R₂* alone may not be able to differentiate, for example, between a phase of deoxygenation or an underlying hemorrhagic pool resulting from loss of tissue integrity.

IAUGC is a relatively simple biomarker that has been used in the clinic as an early measure of therapeutic effects on tumor vasculature for both VDA and antiangiogenic therapies [15,16,46]. By using an internal reference, the EHF measure takes into account a degree of systemic change associated with contrast agent delivery. In the present study, we have shown a good correlation between tumor necrosis and EHF. Thus, these very poorly perfused values appear to closely match with the degree of necrosis induced. Further analysis may permit the resolution of regions associated with drug-induced hypoxia. If a suitable reference tissue could be identified within the imaging plane of the tumor, EHF measurement may be applicable in certain clinical situations.

K^trans, excluding fit failures set to zero, showed no correlation with tumor necrosis. This is most likely because this measurement principally determines K^trans at the viable tumor rim because nonfitting K^trans voxels are associated with necrosis in the tumor core and have been excluded from this measurement. Different factors can affect the success of a fitting algorithm. However, in the current experiments, pretreatment groups show a mean 2.2% fit failure across the whole sampled tumor. The proportion of voxels that failed to fit using the Nelder-Mead Simplex Method algorithm showed the strongest and most significant correlation with tumor necrosis. This provides preclinical support for the suggestion of Galbraith et al. [16], who proposed that non-enhancing pixels after VDA therapy might represent areas of tumor necrosis.

Sagittal MRI planes that would traverse the central region of the tumor were acquired. On excision, the tumor was embedded and then sectioned in the plane, matching that of MR acquisition. Ensuring accurate spatial registration of the tumor from in vivo to ex vivo image analysis is often problematic. Soft tissues, in particular subcutaneous tumors, 1) do not pose obvious landmarks or stereotactic coordinates (unlike the brain with bregma), and 2) are not encased in a stable framework (unlike the brain with the cranium, or the cartilage fixed to the bone surface). One approach is to mark excised and embedded samples to orientate the sectioning process [47].

Clinical studies have routinely employed noninvasive DCE-MRI to discern the biologic and antitumor effects of VDAs on phase I cancer patients. However, the temporal and spatial complexity of events following VDA treatment may necessitate multiple imaging endpoints before any conclusion can be drawn about potential biologic effects (e.g., diffusion-weighted imaging) [42,44,48]. For example, at early time points following VDA therapy, acute changes in vessel perfusion and permeability are likely to predominate, whereas at later time points, MRI measurements may be more weighted by vascular volume changes or necrosis.

In conclusion, we have examined the relationship between VDA-induced necrosis and multiple MRI biomarkers in the Hras5 tumor model in rats 24 hours after treatment with clinically relevant doses of ZD6126 and have found a good correlation for two imaging biomarkers (EHF and nonenhancing/nonfitted voxels) with tumor necrosis. This study provides a preclinical rational for investigating nonenhancing voxels as an imaging biomarker within clinical studies with potential for noninvasive assessment of tumor necrosis.