Can We Identify Predictive Biomarkers for Antiangiogenic Therapy of Cancer Using Mathematical Modeling?

Angiogenesis (the term referring generically to new blood vessel formation) is a hallmark of all solid tumors (1), and vascular endothelial growth factor (VEGF) is the most prevalent and potent angiogenic growth factor in these tumors (2,3). For this reason, there have been intense efforts to develop therapies that target the VEGF pathway. Currently, there are nine antiangiogenic drugs approved by the US Food and Drug Administration and one pending approval (Table 1) (4). These agents are antibodies that target VEGF itself (bevacizumab and aflibercept) or its receptor VEGFR2 (ramucirumab), or are tyrosine kinase inhibitors that interfere with VEGFR2 signaling as well as other receptor and cellular kinases. Introduction of these new drugs over the last decade has established antiangiogenic therapy as a novel therapeutic modality, but their implementation has raised several important questions. Why do they work in some cancers and not others? Is the mechanism of action similar when targeting the ligand versus the receptors, or when using specific versus multitargeted drugs? Can any of them completely block the VEGF pathway? What mediates the inevitable escape from therapy? Can we find biomarkers to identify patients who will benefit from these agents or pathways that must be targeted when tumors become refractory to a given antiangiogenic agent?

Table 1.

Overview of successful phase III trials of antiangiogenic agents*

Drug	Indication	Improvement in RR (%)	Improvement in PFS (months)	Improvement in OS (months)
Bevacizumab	Metastatic colorectal cancer (with chemotherapy)	10	4.4	4.7
		0	1.4	1.4
		7.8	2.8	2.5
		14.1	2.6	2.1
	Metastatic nonsquamous NSCLC (with chemotherapy)	20	1.7	2.0
		10.3–14.0	0.4–0.6	NS
	Metastatic breast cancer (with chemotherapy)	15.7	5.9	NS
		9–18	0.8–1.9	NS
		11.8–13.4	1.2–2.9	NS
		9.9	2.1	NS
	Recurrent GBM (monotherapy)	Currently only phase II data reported
	Metastatic RCC (with IFN-α)	18	4.8	NS
		12.4	3.3	NS
Sunitinib	Metastatic RCC	35	6.0	4.6
	GIST	6.8	4.5	NS
	PNET	9.3	4.8	?
Sorafenib	Metastatic RCC	8	2.7	NS
	Unresectable HCC	1	NS	2.8
	Unresectable HCC	2	1.4	2.3
Pazopanib	Metastatic RCC	27	5.0	N/A
	Advanced soft tissue sarcoma	6.0	3.0	NS
Vandetanib	Advanced medullary thyroid cancer	43	6.2	N/A
Axitinib	Advanced RCC	10	2.0	N/A
Regorafenib	Chemo-refractory metastatic colorectal cancer	0.6	0.2	1.4
Aflibercept	Chemo-refractory metastatic colorectal cancer	8.7	2.2	1.4
Cabozantinib	Advanced medullary thyroid cancer	25	7.2	NS
Ramucirumab	Metastatic gastric and gastroesophageal junction cancers†	0.8	0.8	1.4

* GBM = glioblastoma; GIST = gastrointestinal stromal tumor; HCC = hepatocellular carcinoma; IFN-α = interferon alpha; NSCLC = non–small cell lung cancer; OS = overall survival; PFS = progression-free survival; PNET = pancreatic neuroendocrine tumor; RCC = renal cell carcinoma.

Source: Updated from Jain, 2013 (4).

† Not currently FDA approved.

Most cancer cell–targeted drugs target identifiable oncogenic pathways known to drive tumorigenesis and have corresponding biomarkers related to those pathways that help guide treatment. However, the theoretical advantage of antiangiogenic therapy over these cancer cell–targeted treatments is that it can be applied more broadly, irrespective of the genetic makeup of each cancer. Unfortunately, unlike in the development of anticancer targeted agents—where target is often an identifiable oncogenic signal known to drive tumorigenesis—the clinical development of antiangiogenic agents has not been biomarker-based. This, coupled with the fact that the exact mechanism(s) of benefit of anti-VEGF drugs remain unclear, has resulted in a complete lack of biomarker-based selection of patients for antiangiogenic therapy with anti-VEGF drugs (5,6). Clearly, the approach of indiscriminately using antiangiogenic drugs in all patients with an approved indication limits their overall efficacy, as some patients show inherent resistance.

This raises important concerns for the use of antiangiogenic therapy and makes the identification of mechanistic biomarkers of response a priority. Such a biomarker would allow selection of patients who would experience survival benefits in excess of the 2–5 months seen in the overall population with antiangiogenic therapies. This would likely equal or even surpass the benefits of cancer cell–targeted drugs, which are given in selected populations (eg, patients with cKIT, EGFR, or BRAF mutations or HER2 amplification) (4). Moreover, it would rekindle the interest for developing antiangiogenic therapies for diseases in which current antiangiogenic drugs have failed in unselected populations (eg, prostate, pancreatic, brain, or breast cancer). Finally, a predictive biomarker would allow exclusion of patients who do not benefit from antiangiogenic therapy and spare them the toxicities and cost of these drugs.

The importance of these issues has stimulated active research in these areas, producing hypothesis-generating results that could potentially impact the treatment outcome. In an article in this issue of the Journal, Finley and Popel (7) use a computational model to address this problem. They model VEGF kinetics in the relevant physiologic compartments to investigate the distribution in the body of two major soluble VEGF isoforms (VEGF₁₂₁ and VEGF₁₆₅). The model predicts changes in VEGF concentrations in each compartment after intravenous injection of a VEGF blocker. One interesting conclusion is that the concentration of free VEGF is higher in the tumor interstitium than the circulating plasma and is predominantly in the form of VEGF₁₂₁. The model also predicts that tumor VEGF levels can either increase or decrease after anti-VEGF treatment depending on the tumor microenvironment. Collectively, these intriguing results suggest that the rate of VEGF secretion by tumor cells is the major determinant of response and could serve as a predictive biomarker for anti-VEGF drugs.

Limited clinical data lend support to this model. VEGF expression level has been the natural candidate as a biomarker for anti-VEGF drugs, but data have been inconsistent (5). Some studies suggest that circulating VEGF may predict response to anti-VEGF therapy in hepatocellular carcinoma and breast cancer (8,9). Other studies, including many randomized phase III trials, found no association between circulating VEGF and response to the anti-VEGF antibody bevacizumab (10–12). Some studies have measured the levels of purely soluble isoforms (eg, VEGF₁₂₁, not bound to tumor matrix), and have found intriguing associations with outcome (13). Others have not detected any statistically significant association for VEGF₁₂₁ (6,14). Finally, the limited available data on the changes in tumor VEGF levels after antiangiogenic therapy suggest a decrease in VEGF expression by cancer cells (15).

Thus, there is still controversy regarding the value of tumor or circulating VEGF as predictive biomarkers in the clinic, and mathematical models such as presented by Finley and Popel (7) could be valuable tools to help sort out this complex issue. These models can be easily adapted to reflect new empirical information to fine-tune the use of VEGF isoforms as predictive biomarkers. For example, additional factors such as host cells (eg, tumor-associated fibroblasts or myeloid cells) that secrete VEGF could be included (16,17). Also, modeling could be performed at higher resolution to investigate how spatiotemporal variations in VEGF levels might be produced by the microenvironment of each compartment or whether endogenous production of VEGF blockers—such as soluble (s) VEGFR1 (also known as sFLT1)—should be considered (18–23). Finally, modeling might determine whether resistance from antiangiogenic therapy is due to the incomplete blockade of VEGF signaling by analyzing ligand-targeted versus receptor-targeted agents: receptor-targeted agents do not directly interact with VEGF ligand, so their biological effects should not be affected by changes in VEGF secretion of specific isoforms.

A potentially critical factor is the existence of endogenous blockers of VEGFs such as sVEGFR1. In 2009, we proposed circulating plasma sVEGFR1 as a potential biomarker that predicts inherent resistance to anti-VEGF therapies in cancer (22). Indeed, we found in five single-arm phase II studies that cancer patients with high levels of circulating sVEGFR1 had a poor outcome after anti-VEGF therapies with antibodies as well as tyrosine kinase inhibitors (18–23). In some of these studies we found that these patients also experienced fewer side effects, further supporting the notion that high sVEGFR1 levels lower biological activity of anti-VEGF agents irrespective of their mode of action (18,21,23). Moreover, specific single-nucleotide polymorphisms in the FLT1 gene that are associated with higher VEGFR1 expression are also associated with poor outcomes in phase III studies of bevacizumab (24). It is unclear if the predictions from the currently available computational models support the use of sVEGFR1 or ratio of sVEGFR1 to VEGF as a biomarker.

In summary, the mathematical model of Finley and Popel (7) is a step forward in our understanding of the potential roles of different VEGF isoforms in the tumor and in blood circulation. It also raises many exciting questions about potential biomarkers for anti-VEGF therapies, and emphasizes once again the need for prospective clinical studies to specifically address them.

Funding

NIH grants P01-CA080124 (to RKJ, DGD, LLM), R01-CA115767, R01-CA085140, and R01-CA126642 (RKJ), R21-CA139168 and R01-CA159258 (to DGD), R01-CA149285 (to LLM), and Federal Share/NCI Proton Beam Program Income Grants (to RKJ, DGD, LLM); Department of Defense Innovator Award W81XWH-10–1-0016 (to RKJ); American Cancer Society Grant RSG-11–073-01-TBG (to DGD); and National Foundation for Cancer Research Grant (to RKJ). RKJ received research funding from Roche and MedImmune.