However, as discussed in more detail below, these MRI biomarkers typically each depend on a variety of physiological factors that may all be influenced by tumor therapy, thereby complicating the interpretation of the biomarker changes. The sensitivity of a particular biomarker may vary greatly depending on the particular tumor phenotype and what parts of the tumor physiology are being effected by a given tumor therapy. In addition, the models used to relate the biomarkers to the relevant physiology may be inadequate or poorly characterized further complicating interpretation of biomarker responses. Here we examine in detail the sensitivity of these MRI tumor biomarkers to treatment with cediranib (AstraZeneca Pharmaceuticals), a potent inhibitor of vascular endothelial growth factor receptors (VEGFR) [19]. The MRI biomarker responses are compared with both ex vivo histology and in vivo optical microscopy studies performed previously [20] in the same mouse tumor model to more directly link the biomarkers to the relevant tumor physiology and validate their sensitivity to anti-angiogenic therapy. The results reported here provide insight into which of the MRI biomarkers are most sensitive to changes in tumor morphology and predictive of tumor response to anti-angiogenic therapy.

The VCI is increasingly being used in both clinical and animal model studies as a biomarker of vessel normalization with anti-angiogenic therapy [4], [6]–[8], [21]–[23]. The VCI is defined as the ratio between the blood volume weighted for large diameter vessels (CBV, cerebral blood volume) to the blood volume weighted for small diameter vessels (MBV, microvascular blood volume) and is proportional to the average blood vessel diameter [4], [5]. Angiogenesis driven tumor growth is typically associated with increased vessel density and caliber and a disordered and tortuous vasculature structure [24]–[26]. An increased VCI with therapy has therefore typically been interpreted as a lack of a therapeutic response. However, an accurate assessment of therapeutic efficacy and interpretation of changes in the tumor vasculature is difficult to obtain from changes in the average vessel diameter alone. For example, if both the CBV and MBV decrease, suggesting a positive therapeutic response, but the MBV decreases more than the CBV then an increased VCI will be observed. Here we examine the sensitivity of the VCI to changes induced by cediranib therapy. The changes in VCI with therapy are compared with histological and intravital optical microscopy (IVM) measurements of the average tumor blood vessel diameter and CBV.

The water ADC has previously been shown to be correlated with tumor cell density [27]–[29] and has been used to monitor response to chemotherapy, where increased cell death (decreased cellular volume fraction) was associated with increased ADC [30]–[34]. However, anti-angiogenic therapies may have only anti-edema effects and the role of ADC as a sensitive biomarker of tumor edema is less well established. The ADC depends on a number of factors including the intra- and extra-cellular water diffusion coefficients and transverse relaxation times, the cellular volume fraction, and the tortuosity of the extra-cellular space [35]. Since anti-angiogenic therapies may alter a number of these parameters it is difficult to predict how the ADC will change. In addition, some anti-angiogenic therapies may have both anti-edema and anti-tumor effects that could lead to offsetting ADC responses thereby masking changes in edema. Finally, ADC may also correlate with glioma invasion [36], further complicating its interpretation over time. Given all of these possibilities, it would be helpful to explore the relationship of changes in ADC acutely after anti-VEGF therapy to changes in tumor edema. The transverse relaxation time, T2, has also been used as a biomarker of edema [11], [12]. Increased protein concentrations associated with decreased edema typically lead to decreased T2 relaxation times. However, T2 is influenced by many other factors, such as water compartmentalization and diffusion, hemorrhage, and necrosis. These factors could all lead to therapy induced T2 changes independent of changes in edema. Here we examine the sensitivity of both ADC and T2 to cediranib therapy and validate the sensitivity of these tumor edema biomarkers with ex vivo measurements of tumor water content determined from wet and dry tumor weights.

Similarly, an increased rVCI was observed on day 2–3 for both cediranib (33.2±1.9%) and control (33.8±3.1%) animals with no significant difference between treatment groups (Figure 1, right). This result is in good agreement with previous histology measures of the average tumor blood vessel diameter [20], which saw no differences between control (9.4±0.3 µm) and cediranib (9.5±0.4 µm) treated animals in the average vessel diameter of the tumor core after 2 days of treatment.

While the rVCI increased equally for cediranib and control groups, the increase occurred for different reasons. As shown in Figure 2 (left panel), cediranib treatment lead to a decrease in both rCBV (−16.5±2.2%) and rMBV (−39.2±5.8%). The larger drop in rMBV compared to rCBV lead to the increased rVCI observed in Figure 1. In contrast, the control group observed an increase in rCBV (+14.1±2.3%) and a decrease in rMBV (−20.6±2.1%), again resulting an increased rVCI. A statistically significant difference between control and cediranib groups was observed in both the change over time of the rCBV (p<0.01) and of the rMBV (p<0.05). As demonstrated in Figure 2 (right panel), there were no statistically significant differences between MRI and IVM measurements of the change over time in rCBV for either control or cediranib treated animals.

While there was a statistically significant (p=0.01) decrease in ADC (−6.2±2.5%) with cediranib treatment, the ADC also decreased slightly for control animals (−2.4±3.1%) with no statistically significant difference between cediranib and control groups (Figure 3, left). In contrast, a significant (p<0.01) difference in the change over time of T2 (ΔT2) was observed between control (1.6±1.3%) and cediranib (−8.6±1.4%) treated animals (Figure 3, right). The decreased T2 for cediranib treated animals is consistent with the decreased water content measured ex vivo from the wet and dry tumor weights on day 2 of treatment, where a 6.3±1.9% decrease in tumor water content was observed between cediranib and control animals [20]. This is in good agreement with the 8.5±0.2% difference in T2 observed on day 2 by MRI for control (T2=52.8 ms) and cediranib (T2=48.3 ms) treated animals.

A statistically significant difference between cediranib and control groups was observed in both the change over time of K^trans (p=0.04) and ν_e (p=0.03). A 34.3±10.1% decrease in K^trans (±SEM) was observed over time for cediranib treated animals, while a 6.3±15.8% increase in K^trans was observed for control animals (Figure 4, left). Similarly, changes over time in ν_e (±SEM) of −31.6±4.3% and −2.8±13.6% were observed for cediranib and control groups, respectively (Figure 4, middle). The changes in vascular permeability to Gd-DTPA (K^trans) observed by MRI are in good agreement with the changes in permeability to tetramethylrhodamine-labeled bovine serum albumen (TMR-BSA) observed previously [20] by IVM (Figure 4, right), where a decreased permeability (ΔP=−62.3±19.2%) for the cediranib group and an increased permeability (ΔP=8.8±3.5%) for the control group were observed on day 2.

Recent studies have shown that the vascular models used to derive the relationship between the average blood vessel diameter and the MRI measured CBV (ΔR2*) and MBV (ΔR2) may be inadequate for modeling the very abnormal tumor vasculature [43], [44]. The vasculature is typically modeled as a random uniformly distributed collection of perfect cylinders [45], [46]. Not only may this vascular model be inadequate for tumors, but also it is unclear how anti-angiogenic tumor therapies that normalize the tumor vasculature will affect the appropriateness of such a fixed vascular model. Here we find that the increased rVCI observed for both control and cediranib treated animals (Figure 1) is consistent with histology measurements of the average vessel diameter [20], which observed no difference in vessel diameter between control and cediranib treated animals after 2 days of treatment. This suggests that despite the simplistic static vascular model used, the VCI does accurately reflect changes in the average blood vessel diameter.

Cediranib treated mice have been shown to have a significantly increased survival rate compared to controls [20]. Evaluation of the rVCI alone therefore might mistakenly suggest that cediranib has no therapeutic benefit as the VCI increased significantly for both cediranib and control animals. However, analysis of changes in the rCBV and rMBV indicates that the increased rVCI for control and cediranib groups occurred for quite different reasons (Figure 2). While the rCBV and rMBV both decreased significantly for cediranib treated animals, the rMBV decreased more than the rCBV resulting in an increased rVCI. In contrast, for control animals the rCBV increased while the rMBV decreased, again resulting in an increased rVCI. The larger decrease over time in the MBV compared to the CBV observed for cediranib treated animals suggests that cediranib is preferentially pruning smaller caliber, less mature tumor blood vessels and has a smaller, but still significant, effect on the larger blood vessels. In contrast, the increased CBV and decreased MBV observed over time for control mice would be consistent with an increasingly avascular tumor with vessel regression in the tumor core and fewer, but larger, blood vessels. Such progression to an avascular phenotype for the core of large tumors is not uncommon and has, for example, been observed previously in a rat glioma model [47].

The change in rCBV measured by MRI is in excellent agreement with that measured by IVM (Figure 2). In addition, the strong response of the rCBV and rMBV to cediranib therapy is consistent with a previous clinical study where cediranib treatment lead to a significant decrease in both rCBV and rMBV [7]. While the clinical study did observe a transient decrease in rVCI that was not observed in our mouse brain tumor model, this might simply reflect somewhat different relative responses of the rMBV and rCBV in the clinical subjects, where a greater decrease is observed for the rCBV than rMBV. Finally, in contrast to the VCI, the decreased rMBV measured after only one treatment in the clinical study was strongly correlated with the duration of overall and/or progression-free survival [42]. These results therefore suggest that the CBV and MBV may be better gauges of therapeutic response than the VCI.

Changes in the T2-weighted signal intensity of tumors are frequently taken as evidence of changes in tumor edema [11]. Here we quantified tumor T2 relaxation times and compared the changes in T2 with changes in tumor water content measured ex vivo. A significant difference between cediranib and control groups is observed in the T2 response (Figure 3). In particular, cediranib treated animals had an 8.5±0.2% lower T2 on day 2 than control animals, which is in good agreement with the 6.3±1.9% decreased tumor water content measured ex vivo after 2 days of cediranib therapy [20]. This suggests that T2 is a sensitive and quantitative biomarker of changes in tumor edema. However, care must be taken when analyzing tumors with regions of necrosis and hemorrhage. While the U87 tumor model studied here displayed no sign of necrosis, hypointense regions consistent with hemorrhage were evident in some cases, particularly along the periphery of the tumor. It is therefore important to define tumor regions-of-interest that do not contain hemorrhage as this will result in significantly decreased T2 values. In addition, therapies that induce large changes in water compartmentalization (i.e. due to necrosis and/or changes in the extravascular-extracellular space) and diffusion could also complicate interpretation of T2 changes. Changes in water compartmentalization and water diffusion would, however, be reflected in the ADC. For the U87 tumor model studied here, only very small changes in ADC were observed with therapy (Figure 3).

The ADC is also frequently used as a biomarker of tumor edema. While the ADC did decrease significantly over time for cediranib treated animals, the ADC also decreased slightly for control animals leading to no significant differences in the ADC response to treatment between cediranib and control groups (Figure 3). In contrast, tumor edema determined from tumor T2 and from ex vivo wet-dry tumor weights [20] showed a significant decrease in tumor edema over time for cediranib treated animals compared to controls. The ADC depends on a large number of factors including the intra- and extra-cellular water diffusion coefficients and transverse relaxation times, the cellular volume fraction, and the tortuosity of the extra-cellular space [35]. The significant decrease in ADC for cediranib treated animals likely resulted from a combination of decreased edema and decreased extravascular-extracellular space (increased cellular volume fraction), observed by DCE MRI (Figure 4). In contrast, the slight decrease in ADC for control animals likely resulted from an increase in tumor edema, which would lead to an increased ADC, being offset by the decrease in the tumor extravascular-extracellular space (Figure 4), which leads to a decreased ADC. The decreased extravascular-extracellular space for control animals may be a result of the increased tumor blood volume (Figure 2) and an increased tumor cell volume induced by increased intra-cellular water content. This illustrates that offsetting responses in different parts of the tumor physiology can compromise the sensitivity of the ADC to changes in tumor edema. Thus, for this tumor model, T2 is a more sensitive gauge of changes in tumor edema than ADC. This is in agreement with a previous clinical study, which found T2 to be more sensitive and “useful” than ADC for differentiating contrast-enhancing tumor and immediate edema regions [11], though it conflicts with clinical experience with cediranib where early changes were seen on ADC before they were seen on T2-weighted images [7]. The dependence of the MRI biomarkers on a number of physiological factors points to the need to consider the biomarker changes in relation to one another to properly interpret the therapy induced changes. Only by considering the ADC, T2, and ν_e responses together, for example, can insight be obtained into the likely changes in tumor physiology that are occurring for cediranib and control groups. In general, the sensitivity of a particular MRI biomarker may vary greatly depending on which physiological factors are being altered most by a given therapy for a particular tumor.

Finally, DCE experiments using low molecular weight Gd-based contrast agents, such as Gd-DTPA, are routinely performed for assessing changes in vascular permeability (K^trans) in response to anti-angiogenic therapy. However, accurate measurement of K^trans requires the use of an accurate arterial input function (AIF) to properly model the tracer kinetics. Obtaining an accurate AIF is complicated by many factors, including partial volume and contrast agent induced T2* distortion of the measured AIF and lack of an artery in the field of view from which to measure the AIF. In particular, for mouse brain images obtained with a surface RF coil, no arteries are typically visible from which to measure an AIF. Using a reference tissue, such as scalp tissue, to calibrate the tumor DCE curves can in principle allow the tracer kinetic parameters to be determined without the need for a direct AIF measurement [48]. However, in practice using a reference tissue calibration is complicated by B₁ field inhomogeneities associated with surface coils and B₀ field inhomogeneity present at air-tissue interfaces, such as the scalp. Recent studies suggest that given the large potential errors in measurement of the AIF, it is better to use an assumed, fixed AIF model for all subjects [49]. The validity of such an approach, however, remains unclear. The AIF is sensitive to blood flow velocity (linked to body weight), blood pressure (dependent on anesthesia dose and body temperature), and the amount of contrast agent injected (variable due to manual injection method typically used for animal studies), all parameters that are difficult to assess with great accuracy and likely vary from subject to subject. While in theory an accurate measure of AIF could help reduce all of these sources of subject-to-subject variability, in practice measurement of the AIF may itself add uncertainty, rather than reduce it, and thus these uncertainties in many of the kinetic tracer model input parameters may lead to large uncertainties in the assessment of the vessel permeability parameters K^trans and ν_e. A previous study of a subcutaneous flank tumor model in rats, where an AIF was directly measured and compared with a variety of AIF models, suggests that errors introduced by using a fixed bi-exponential AIF model are less than 5% [40]. However, others have suggested that the use of a fixed AIF model can lead to large systematic errors in the determination of permeability parameters [13], [50]. The agreement between permeability parameters extracted using experimentally measured and fixed models is likely to be highly variable depending on the tumor model and reproducibility of the experimental techniques used (i.e. anesthesia, contrast agent injection technique, etc.). The decreased permeability to Gd-DTPA induced by cediranib observed in this study (Figure 4), as quantified by K^trans, is consistent with the decreased permeability to BSA observed previously by IVM [20]. The IVM measurement of permeability involves a much simpler and more straightforward analysis of the fluorescent signal intensity in the vascular and extravascular spaces, with no complicated kinetic modeling or need for an accurate AIF. The good agreement between DCE and IVM permeability measurements therefore suggests that the simple, fixed bi-exponential AIF model can be adequate for assessing changes in vascular permeability. The strong response of K^trans to cediranib therapy is again in agreement with a previous clinical study, which saw a significant decrease in K^trans for cediranib treated subjects [7] that was correlated with the duration of overall and/or progression-free survival [42].