Judah Folkman spent thirty years on this idea. Bevacizumab proved he was right about the biology — and showed the cure was far more complicated than the logic.
The logic was as clean as logic gets in oncology. Solid tumors cannot grow beyond a few millimeters without recruiting new blood vessels. Those vessels depend on a protein called Vascular Endothelial Growth Factor — VEGF. Block VEGF, the vessels cannot form, the tumor is starved of oxygen and nutrients, and the patient lives. Judah Folkman spent thirty years defending that idea when most of the field thought it was wrong. Bevacizumab, a humanized monoclonal antibody that sequesters circulating VEGF-A, vindicated him. It was approved. It extended survival in multiple cancer types. And then the great cure that was supposed to follow never arrived. The drug worked — but not the way Folkman described. What actually happens inside a tumor when you block VEGF is stranger, more interesting, and more instructive than starvation.
Metastatic colorectal cancer. The gain was real, reproducible, and the basis for approval across six-plus tumor types. The side-effect profile mapped directly to what VEGF does in healthy tissue.
The pivotal metastatic colorectal cancer trial told the story with numbers. Median overall survival in the control arm was fifteen point six months. In the bevacizumab arm it was twenty point three. Roughly five months. That is not nothing — in metastatic colorectal cancer, five months is a clinically meaningful difference, and the benefit was real, reproducible across trials, and the basis for regulatory approval. Approvals followed in non-small-cell lung cancer, breast cancer, renal cell carcinoma, glioblastoma, and ovarian cancer. The side effect profile made biological sense: hypertension because VEGF regulates vessel tone in normal tissue; proteinuria because VEGF maintains the glomerular filtration barrier in the kidney; impaired wound healing because VEGF drives the angiogenesis that repairs tissue. Every toxicity was a readout of VEGF's normal jobs. The drug was doing exactly what it was designed to do. And yet the five-month gain in colorectal cancer was as good as it got. No tumor type was cured.
Multi-kinase coverage addresses angiogenic redundancy — but also multiplies toxicity profiles.
The pharmaceutical response to bevacizumab's limitations was to build more tools attacking the same pathway from different angles. Aflibercept is a fusion protein — a decoy receptor constructed from VEGFR-1 and VEGFR-2 domains — that acts as a trap, sequestering VEGF-A, VEGF-B, and placental growth factor before they reach endothelial cells. Ramucirumab takes the opposite approach: it binds VEGFR-2 directly and blocks ligand docking from the receptor side rather than the ligand side. Then came the small molecules. Tyrosine kinase inhibitors cross cell membranes and compete with ATP at the kinase domain of VEGF receptors, blocking the signaling that follows ligand binding. Sunitinib is the archetype of the multi-kinase approach: it inhibits VEGFR-1, VEGFR-2, and VEGFR-3, plus PDGFR, c-KIT, FLT3, and RET. The breadth is a strategy — different tumors upregulate different pro-angiogenic signals, and blocking multiple kinases simultaneously limits the escape routes.
VHL-deleted renal cell carcinoma became the proving ground for anti-angiogenic tyrosine kinase inhibitors. VHL loss constitutively activates HIF and drives extreme VEGF overexpression, making these tumors genuinely dependent on continuous angiogenic signaling. Sunitinib produced response rates around thirty percent in this setting, with progression-free survival around eleven months compared to roughly five months on interferon alfa, which had been the prior standard. That was a genuine advance, and sunitinib became a standard first-line therapy. The agents that followed — sorafenib, which added RAF kinase inhibition to VEGFR blockade; pazopanib; axitinib; cabozantinib, which added MET and AXL to cover additional escape routes; lenvatinib; and regorafenib — each carry different kinase profiles and different toxicity patterns. Hand-foot skin reaction, a painful palmar-plantar desquamation, is the characteristic toxicity of inhibitors that hit PDGFR and c-KIT in skin, and it distinguishes TKI-class toxicity from bevacizumab-class toxicity.
The normalization window: drug delivery improves, chemo penetrates, radiation works — hypoxia falls, radiosensitivity increases. Timing is the critical variable.
The vascular normalization hypothesis reframed what anti-VEGF drugs actually do. The mechanism is not starvation. It is pruning of the most dysfunctional vessels combined with transient structural repair of the surviving ones. During what Jain called the normalization window, drug delivery to the tumor improves — chemotherapy can penetrate farther into the tumor interior. Radiation works better because oxygenation improves, and hypoxic tumors are radioresistant while well-oxygenated ones are not. This explains why combinations of anti-angiogenic therapy with chemotherapy and radiation outperform either alone: the anti-angiogenic drug is making the tumor temporarily more accessible. But the window has a threshold. Continued or excessive VEGF blockade tips from transient normalization into over-pruning — severe hypoxia, reduced drug delivery, and, critically, selection pressure for cancer cell clones that can invade the surrounding tissue rather than wait for a blood supply. Timing is not just important. It is the entire variable.
The glioblastoma experience made the distinction between imaging response and clinical benefit unavoidable. When bevacizumab was added to glioblastoma treatment, the gadolinium-enhancing lesion on MRI shrank dramatically. Gadolinium enhancement reflects blood-brain barrier breakdown — leaky vessels accumulate the contrast agent. Tightening the vessels through VEGF blockade reduces enhancement. The tumor appeared to shrink on the scan. Response rates were impressive. Progression-free survival improved in some analyses. Overall survival did not. This is pseudoresponse: an imaging improvement driven by vascular normalization rather than actual cancer cell kill. The lesson from this is not obscure — it is fundamental to how clinical trials should be designed and interpreted. Response rate is the weakest endpoint. Progression-free survival is better. Overall survival is the endpoint that cannot be fooled by a biological effect on vessels. When response rate and overall survival point in different directions, overall survival wins. Always.
Scientific reframings earn credibility by generating predictions that turn out to be right. This one did.
What makes the vascular normalization model more than a redescription of the data is that it generated predictions that could be tested and that turned out to be correct. The prediction was: normalize vessels, reduce hypoxia, and immune cell trafficking into the tumor will improve. Reduce hypoxia and block VEGF's direct immunosuppressive signaling, and checkpoint inhibitors will work better inside the normalized tumor. Add a checkpoint inhibitor during the normalization window, and you will see additive or synergistic benefit on a hard endpoint — overall survival — not just on imaging. That prediction was made before the combination trials ran. The combination trials confirmed it. Scientific frameworks earn credibility by generating correct predictions. The starvation framework predicted bevacizumab would be curative in some patients. It was not. The normalization framework predicted the combination with immunotherapy would improve overall survival. It did. The reframe is not semantic. It is the difference between knowing why a drug works and knowing when and with what to combine it.
The thesis of this chapter can be stated in three lines. First: the mechanism of anti-angiogenic therapy is not starvation. It is vascular normalization — pruning of the most dysfunctional vessels and transient structural improvement of the survivors. Second: the clinical strategy that follows from normalization is combination, not monotherapy. The drug's most important job is making chemotherapy penetrate, making radiation more lethal, and making immunotherapy work by lifting VEGF's suppression of immune trafficking. Third: because vascular normalization creates a transient window that imaging can mimic without actual tumor cell kill, endpoint hierarchy matters. The drug that improves gadolinium enhancement on an MRI scan may be doing something biologically real that does not translate to the patient living longer. Knowing which endpoint to trust is not a statistical nicety. It is the difference between a drug that works and a drug that appears to work — and in oncology that difference is the patient.
Cancer Medicine · Chapter 2 · Anti-Angiogenic Therapy: Promise and Reality
That is the framework: not starvation as a goal, normalization as the mechanism, and combination as the strategy that follows from understanding what normalization actually does. The open questions are genuinely open. How do you detect the normalization window in a living patient — not in a mouse with a dorsal window chamber, but in a person, on a clinical schedule? Which tumor types are truly VEGF-addicted enough that anti-angiogenic monotherapy produces durable benefit rather than transient response? And when you combine anti-angiogenic therapy with immunotherapy, what is the optimal sequencing and dosing to sustain the normalization window rather than collapse it into over-pruning? Folkman was right about the biology. The vessels matter. The question his successors are still answering is how to use that insight precisely enough that the right patients, in the right tumors, get the right combination at the right moment in the window.