The potential of DCE-MRI to provide clinically valuable information on the tumor microenvironment has been investigated in numerous studies, both preclinical and clinical. On the clinical side, significant correlations between DCE-MRI-derived parameters and microvascular density, IFP, and oxygen tension have been reported [14, 15, 27], as well as promising findings on the ability of semi-quantitative and quantitative DCE-MRI measures to predict response to therapy [13, 28, 29]. Nevertheless, correlations have in general been relatively weak, and semi-quantitative DCE-MRI analyses that are hard to standardize remain to be preferred over more complex and labor intensive quantitative analyses in clinical practice.
A major advantage of preclinical studies compared to clinical studies in imaging-based research, is that the imaging conditions can be controlled more easily. High-quality preclinical investigations may therefore be a prerequisite for the true value of imaging-based approaches to be explored. Moreover, as tumors of the same genetic basis can be established in several animals, multiple copies of a single patient tumor can be examined in the same experiment. These tumor copies will inevitably differ somewhat in vascular density, interstitial hypertension, and the level of hypoxia, as a result of the stochastic nature of angiogenesis and differences in lesion size . They may therefore, collectively, depict some of the intratumor heterogeneity characteristic of many human tumors.
Previous DCE-MRI studies of human cervical cancer xenografts in mice have revealed robust correlations between pharmacokinetic parameters and IFP or the extent of hypoxia, i.e. the fraction of radiobiologically hypoxic cells or the fraction of hypoxic tissue as determined histologically [9, 10, 12]. A possible drawback of these encouraging preclinical investigations is, nonetheless, the use of xenografted tumors initiated from previously established cell lines. Increasing evidence suggests that cell line-derived tumor xenografts may be suboptimal models of human cancer, primarily due to their limited capability of recapitulating the biological characteristics and diversity of patient tumors, as well as their lack of success in predicting clinical treatment outcome . In this regard, PDX models, established by transplanting samples of patient tumors directly into host animals and maintaining the tumor tissue exclusively in vivo, have shown promise [16, 17, 31]. Our group has developed a panel of four PDX models of squamous cell carcinoma of the uterine cervix, differing in molecular and biological properties. Despite being intramuscular and thus ectopic tumor models, previous validation of the BK-12, ED-15, HL-16, and LA-19 tumors has confirmed that essential features of the donor patients’ tumors are retained after xenotransplantation [18, 19]. Furthermore, the highly different MVDs, IFPs, and fractions of hypoxic tissue found among these tumor xenografts make them suitable models for evaluating the hypothesis of investigation in the present study. It should be noted, however, that since the stroma and hematopoietic system in xenografted tumors is of murine origin, the vascular density, IFP, and supply of oxygen may deviate from that in human tumors .
The Tofts pharmacokinetic model was used to analyze the imaging data acquired in this study. Among other frequently used pharmacokinetic models, e.g. the Brix model and the shutter-speed model, the standard Tofts model has been reported to be preferable in analysis of clinical DCE-MRI data . Previous investigations in our group have shown that good fits to experimental data are obtained using this model [20, 33], as well as highly reproducible parametric images . Important prerequisites for the Tofts model, such as insignificant effects of water exchange, uniform concentration of the contrast agent within each voxel, and negligible contribution of intravascular contrast agent to the total tumor concentration, were assumed to be adequately fulfilled also in the present study. Although variation in arterial input function between animals was not considered, this pharmacokinetic model appeared to be well suited for analyzing the current DCE-MRI data. Nevertheless, in all four PDX models, a sizeable fraction of the voxels was assigned unphysiological v
e values, i.e. a fractional distribution volume of the contrast agent higher than unity. These voxels are likely to represent necrotic or fibrotic tissue, for which the Tofts pharmacokinetic model breaks down . Consequently, voxels with v
e > 1.0 were excluded from the subsequent analysis. Exclusion of voxels with unphysiologically low v
e values was also considered, but since removal of these voxels only had a minor impact on the tumor median K
trans and v
e values, simplicity of the analysis was chosen over the introduction of more exclusion criteria. In the present investigation, positive correlations were detected between the fraction of unphysiological voxels and the extent of hypoxia, in accordance with the well-established fact that severe and prolonged hypoxia may lead to necrosis .
When the basic assumptions of the Tofts pharmacokinetic model are valid, the volume transfer constant K
trans is determined by the blood perfusion and the permeability surface area product of the vessel wall in varying proportions . Considering the relatively high microvascular permeability in many cancerous tissues, as well as the low molecular weight of Gd-DOTA, K
trans is thought to be largely dependent on blood perfusion in our DCE-MRI examinations of cervical cancer. Both on the PDX model level and the single tumor level, we found significant correlations between K
trans and hypoxic fraction. Furthermore, the data were well described by common exponential decay curves, and comprised both early and late generation tumor xenografts. The indication that DCE-MRI can provide reliable information on the extent of hypoxia therefore holds true across varying transplantation conditions and several patient-derived xenograft models differing in biological features.
Hypoxia is caused by an imbalance between oxygen supply and oxygen consumption . The oxygen supply is governed by the blood perfusion, whereas the oxygen consumption depends on the respiratory activity of the tissue and, consequently, the density of cells. Hypoxic regions are therefore expected to coincide with areas characterized by poor blood perfusion and/or low extracellular volume fraction. No association was found between v
e and the hypoxic fraction in this study, suggesting that the level of hypoxia is determined primarily by the blood perfusion, i.e. the extent to which a functional vascular network is present, in our cervical cancer models. Also, an abnormal microvasculature, with a varying fraction of non-perfused tumor vessels, can help explain why no correlation was seen between K
trans and MVD in these experiments. K
trans quantifies the functioning of perfused vessels, while the CD31 analysis incorporates all microvessels, both perfused and non-perfused.
As mentioned introductorily, elevated IFP is a common feature of malignant lesions . Fluid is forced out of tortuous and leaky tumor vessels by the hydrostatic microvascular pressure, and accumulates in the tumor interstitium due to impaired lymphatic drainage . In the present investigation, none of the DCE-MRI parameters were related to IFP, implying that neither the blood supply nor the extravascular extracellular volume fraction is determinative for the interstitial hypertension in these tumors.
Former characterization of BK-12, ED-15, HL-16, and LA-19 tumors has revealed fairly large fractions of stroma (∼ 20–35%) in these PDX models . A densely structured extracellular matrix could possibly represent a barrier to the transvascular and interstitial transport of molecules , and one could therefore speculate whether the underlying assumptions of the Tofts pharmacokinetic model are violated when subjecting our cervical cancer models to DCE-MRI. However, the results reported here on the inverse association between K
trans and hypoxic fraction are similar to previous findings on melanoma xenografts in our group [42, 43]. The PDX models of current interest differ significantly in histological appearance and stromal content from these melanoma xenografts, and it is thus possible that K
trans may relate to tumor hypoxia in a qualitatively similar way for tumors showing substantial variation in extracellular composition.