We have developed a molecularly-detailed model of VEGF transport in the body. This is the first human model to include experimental measurements of receptor densities as well as receptors on parenchymal cells, and thus represents a significant advance compared to previous models. By incorporating in vivo and in vitro quantifications of VEGF receptor expression on endothelial and parenchymal cells and experimentally-based values for VEGF degradation in the tissue compartments and VEGF secretion by tumors, we have created a predictive tool that reflects physiological elements of VEGF-mediated angiogenesis. Using the model, we are able to predict how systemic properties, drug design parameters, and properties of the tumor microenvironment influence the response to anti-VEGF treatment. Specifically, we have predicted the fold-change in free VEGF following intravenous injection of an anti-VEGF agent. Importantly, the model predicted that the outcome of the anti-VEGF treatment (e.g., the level of free VEGF in the tumor interstitium) is dependent on the tumor microenvironment (e.g., receptor expression, internalization of neuropilins, and VEGF isoform ratio) and therefore may vary from patient to patient or between groups of patients. The model predicted that transport and binding parameters of the anti-VEGF can be fine-tuned such that the drug acts to deplete free VEGF in the tumor.
Our model predicted that tight binding between the anti-VEGF and VEGF results in a strong therapeutic effect. Although free VEGF was reduced to nearly zero in all compartments in the few hours following the treatment, and the desired therapeutic effect is observed, tight binding between VEGF and the anti-VEGF agent may also induce adverse effects. The work of Gerber et al.
] shows a positive correlation between tight binding of the anti-VEGF to VEGF and the inhibition of tumor growth; however, tighter association between anti-VEGF and VEGF results in increased toxicity. Specifically, the authors identify renal changes that result from increased antibody affinity, including glomerulosclerosis and anti-VEGF deposition in glomeruli, as well as hypoalbuminemia and ascites formation [53
]. Further investigation into the mechanism of glomerular injury reveals that inhibition of VEGF in noncancerous tissues such as the kidney results in a reduction of glomerular VEGF, which is required for the health and integrity of the adjacent microvasculature [54
]. As VEGF is required for cell maintenance and tissue homeostasis, inhibiting endogenous VEGF signaling leads to additional side effects [55
]. Thus, there is an optimal value for the binding affinity where the anti-VEGF binds VEGF tightly enough to inhibit tumor growth while limiting toxicity, and this drug design parameter must be carefully balanced with the occurrence of adverse events.
The occurrence of the desired therapeutic effect is sensitive to the expression of VEGF receptors on tumor cells. Our current model incorporates in vivo experimental data for the density of VEGFR1 and VEGFR2 located on the normal and tumor endothelia, muscle fibers (myocytes), and tumor cells. In contrast, the number of neuropilin receptors on various cell types has yet to be quantified in vivo. Our analysis shows that the occurrence of a therapeutic effect is dependent on the density of VEGFRs and co-receptors. Therefore, accurate estimates of neuropilin expression and the relative density of NRP1 compared to NRP2 are required, as the response to anti-VEGF therapy is sensitive to this property of the tumor microenvironment.
In addition to the density of neuropilin expression on tumor cells, the availability of these receptors also influences the response to anti-VEGF treatment. Since neuropilins are co-receptors involved in angiogenesis and are estimated to be present in large numbers on parenchymal cells, it is possible that the internalization of these receptors may be dysregulated in pathological conditions. Our model predicts that a low rate of internalization of neuropilins leads to a more drastic reduction of free VEGF in the tumor. A previous model was used to compare therapeutic approaches of targeting NRP1 and shows that blocking NRP1 expression does not result in persistent inhibition of VEGF signaling [56
]. Our results support this finding, and in fact show that prolonged expression of NRPs (inhibiting NRP internalization) will improve the therapeutic effect of anti-VEGF treatment. Teesalu and coworkers recently characterized the amino acid motif of NRP1 that promotes its internalization and found that blocking interaction at that site inhibits internalization of the receptor [57
]. Their work suggests that it is possible to specifically target neuropilin internalization.
The rate at which VEGF is secreted and the VEGF isoform secretion ratio VEGF165:VEGF121 in the tumor have a significant impact on the response to anti-VEGF treatment and are crucial for prediction of the therapeutic effect. Specifically, the rate of VEGF secretion in the tumor can be used to tune the steady-state level of free VEGF in the tumor and influences whether an anti-VEGF agent works to deplete tumor VEGF. Our model predicts that the steady-state concentration of tumor free VEGF prior to treatment influences whether the anti-VEGF has a therapeutic effect. This underscores the need for isolation of tumor interstitial fluid and measurement of its VEGF concentration, which may be used as a predictive biomarker for administration of specific anti-angiogenic drugs and serve in stratification of patients who would best respond to a specific therapy.
Similarly, there is a need for quantitative measurements of the relative secretion of VEGF isoforms. These data would aid in refining the model and would lead to a better understanding of the effects of drugs that target VEGF-mediated angiogenesis. It is interesting to note that the VEGF isoform ratio in tumor tissues is in the range required for an anti-VEGF agent to have a therapeutic effect, as compared to the ratio observed in other tissues such as muscle. These results bring into question whether the VEGF isoform expression may be a useful biomarker to predict a therapeutic response to anti-VEGF treatment.
We have shown that the properties of the tumor should be taken into account when developing treatment strategies. It would be of interest to predict the effect of anti-VEGF treatment when using receptor quantification and the relative secretion of VEGF isoforms for specific types of tumors. This type of analysis may provide insight as to why certain tumors respond to anti-VEGF treatment better than others. Additionally, the incorporation of tumor-specific properties is required to develop personalized medicine and identify the patient population that is best-suited for VEGF-targeted therapies [58
Our model of VEGF distribution is used to investigate the effect of anti-VEGF agents in targeting VEGF and inhibiting angiogenesis. An alternative or complementary view is that the anti-VEGF therapy works through vascular normalization, i.e., repairing tumor vasculature to resemble normal vessels, leading to increased pericyte coverage and increased blood perfusion, reduced interstitial pressure, and tightened endothelial cell junctions [51
]. Clinically, metastasis is reduced and the efficacy of chemo-, radiation- and immune-therapies is improved upon vessel normalization [7
]. Interestingly, our model predicts that when tumor lymphatics become functional, perhaps due to a reduction in interstitial pressure following normalization, the anti-VEGF has a more potent therapeutic effect. Similarly, it would be of interest to incorporate the dynamic effects of vascular normalization on macromolecular permeability, which is reduced following normalization, in order to understand how this influences the response to anti-VEGF therapy. Vascular normalization is a transient response, characterized by an optimal time window after which the normalized features of the tumor vasculature are lost, possibly due to prolonged anti-VEGF treatment or development of a resistance to treatment [7
]. Our current model predicts that tumor VEGF is immediately reduced following anti-VEGF treatment and then rebounds to a new pseudo-steady state below pre-treatment levels within 7 days. The time it takes for tumor VEGF to rebound may correspond to the normalization window. Therefore, our model can be used to explore how the duration of the normalization window depends on systemic and drug parameters and tumor properties.
Several assumptions influence the model predictions and should be re-evaluated as quantitative experimental data become available. We have assumed equal distribution of receptors on the luminal and abluminal surface of endothelial cells. Previous work predicts that quantification of luminal and abluminal receptors influences VEGF distribution in the body [34
]. We have assumed that the total number of receptors is conserved; however receptor expression is a dynamic process and the cell-surface receptor density might depend on VEGF concentration among other factors [31
]. Additionally, the administration of an anti-VEGF agent may affect receptor expression over days and weeks, the time scale investigated in the present study. Therefore, incorporating receptor dynamics and quantifying the effect of anti-VEGF therapy on receptor density would better reflect biological conditions. Anti-VEGF therapies influence the tumor vasculature and may also have anti-tumor effects leading to tumor growth inhibition. However, we have assumed that the size of the tumor remains constant throughout the model simulation. It would be of interest to incorporate a function for tumor growth and/or regression as a function of time. Lastly, the effects of platelet content or the ability of platelets to secrete and sequester angiogenic proteins has not been addressed. Since degranulation of platelets is a source of VEGF and platelets have been shown to sequester the VEGF antibody bevacizumab [60
], it is important to add platelets to the model.