Anti-VEGF agents in oncology
The mechanism, and where the toxicity comes from
These products interrupt vascular endothelial growth factor signalling, with the aim of limiting the blood supply a tumour builds for itself. The point that governs nonclinical design is that the same signalling maintains normal vasculature and supports several physiological processes that involve new vessel formation. The toxicity of the class follows from the mechanism working as intended in tissue other than the tumour, which is a different situation from an off-target effect.
Approved products, and three different ways of blocking the pathway
Bevacizumab
Bevacizumab was approved in February 2004.[S1] It is a monoclonal antibody that binds the ligand, so it removes the signal rather than blocking the receptor that receives it.
Ramucirumab
Ramucirumab was approved in April 2014.[S2] It binds the receptor rather than the ligand. The distinction matters for nonclinical work: an antibody against a human receptor may not recognize the animal receptor, whereas an antibody against a soluble ligand faces the same question about the animal ligand. Either way, cross-reactivity has to be established experimentally before the toxicology species is fixed.
Ziv-aflibercept
Ziv-aflibercept is not an antibody. Its prescribing information describes it as a recombinant fusion protein consisting of the VEGF-binding portions from the extracellular domains of human VEGF receptors 1 and 2, fused to the Fc portion of human immunoglobulin G1.[S3] It works as a decoy receptor, capturing ligand before it reaches the cell surface. Because the binding domain is derived from the human receptor rather than raised against it, the species cross-reactivity profile is set by receptor biology rather than by an immunization campaign, and the analytical methods used to measure it in serum differ from those used for a conventional antibody.
The problems specific to this class
Compensatory pathways
Resistance to antiangiogenic therapy is described in the literature as something that emerges over time.[S4] Blocking one pathway can leave others available, which means a short study in a responsive model may not surface the problem at all. The design implication is about duration: a study that ends before the compensatory response would have developed cannot report on it, and study length in this class is a scientific decision rather than a scheduling one.
Dependence on the pathway varies
How much a given tumour depends on this pathway differs between tumours, and no biomarker reliably identifies the ones that do. For nonclinical work this means the vasculature of the model has to be characterized rather than assumed. Two models with similar growth rates can have quite different vessel densities, and a result obtained in a heavily vascularized model does not transfer to a poorly vascularized one.
Effects on normal tissue
The pathway maintains normal vasculature as well as tumour vasculature, and it supports processes such as wound healing that depend on forming new vessels. These effects are not avoidable by improving selectivity for the target, because the target is the same molecule in both settings. What nonclinical work can do is characterize them, and the approved labels record what was actually found.
What the monkey studies actually found
Two findings from the approved labels are worth planning around. Female cynomolgus monkeys given bevacizumab at 0.4 to 20 times the recommended human dose showed arrested follicular development or absent corpora lutea.[S1] With ramucirumab, adverse effects in the kidney, specifically glomerulonephritis, occurred in monkeys at 16 to 50 mg/kg.[S2] Both are consistent with the mechanism: ovarian follicle development and glomerular capillary maintenance both involve the pathway being blocked. For a developer this sets expectations for the toxicology protocol. Reproductive organ histopathology and renal function with urinalysis for protein are not optional add-ons in this class; they are where the class effects appear, and a protocol that does not include them will not detect what the approved products were shown to do.
The nonclinical work
Models where the vasculature is the readout
One consideration applies to every model choice here. In a xenograft, the tumour cells are human but the blood vessels growing into them are the host animal's. The target the product must engage to work is therefore on animal endothelium, not human, and an antibody selective for the human protein may have no effect in that setting for reasons unrelated to its quality. Establishing what the product does to the animal ortholog is a prerequisite to interpreting any of these models.
Measuring the vascular effect
Vessel density measured by quantitative pathology is the direct readout, and it should be reported with the method stated, since counts depend heavily on how vessels were identified and which regions were sampled. Imaging that reports perfusion adds something density does not: a vessel that is present but not carrying blood counts in one measure and not the other. Free ligand concentration in serum indicates whether the product engaged its target systemically.
Pharmacokinetics
A circularity is worth noting in this class. The product acts on the vasculature, and the vasculature is what delivers the product to the tissue. Reducing vessel density or permeability can therefore reduce the delivery of the next dose, so exposure measured early in a study may not describe exposure later in it. Sampling at more than one timepoint, and measuring free rather than only total ligand, is what makes this visible.
Safety assessment
Blood pressure, coagulation parameters, renal function including urine protein, and reproductive organ histopathology are the areas where this class produces findings, based on what the approved labels record.[S1][S2] Anti-drug antibody formation is measured because it changes exposure within the study and can make later timepoints uninterpretable. Where the product is a fusion protein rather than an antibody, the immunogenicity assay has to be developed for that construct, and an assay format validated for antibodies does not carry over.[S3]
What to establish with the laboratory
Four things are worth establishing before award. How the laboratory quantifies vessel density, including the marker used and the sampling scheme, since this is the primary readout and methods differ. Whether it can measure free as well as total ligand in serum. Whether its repeat-dose protocols in this area include reproductive organ histopathology and urine protein, given where the class findings appear.[S1][S2] And, for a fusion protein rather than an antibody, whether it has developed immunogenicity assays for that construct type.[S3]
About this article
This is an independent editorial article for people who commission nonclinical work in the United States. Product descriptions and nonclinical findings are taken from the prescribing information as published, and approvals are recorded as facts; this site does not discuss the efficacy of any product or compare products. Findings in animals are reported as animal findings and are not statements about people. This article covers oncology products only; it does not address intravitreal use. It does not recommend any laboratory. Last reviewed: September 3, 2026.
References
- FDA — AVASTIN (bevacizumab) full prescribing information (Nonclinical Toxicology) and approval records (BLA 125085, approved February 26, 2004). https://www.accessdata.fda.gov/scripts/cder/daf/ (accessed 2026-09-03)
- FDA — CYRAMZA (ramucirumab) full prescribing information (Nonclinical Toxicology) and approval records (BLA 125477, approved April 21, 2014). https://www.accessdata.fda.gov/scripts/cder/daf/ (accessed 2026-09-03)
- FDA — ZALTRAP (ziv-aflibercept) full prescribing information (Description, Nonclinical Toxicology) and approval records. https://www.accessdata.fda.gov/scripts/cder/daf/ (accessed 2026-09-03)
- Breaking the barrier: a guidelines-based review of antiangiogenesis drug resistance. https://pubmed.ncbi.nlm.nih.gov/ (accessed 2026-09-03)
Japanese
