THE JEREMIAH METZGER LECTURE:VON HIPPEL-LINDAU DISEASE: INSIGHTS INTO OXYGEN SENSING, CANCER AND DRUGGING THE UNDRUGGABLE

Abstract

Germline VHL mutations predispose to hemangioblastomas of the retina, cerebellum, and spinal cord; clear cell renal cell carcinomas (ccRCCs); and paragangliomas. Consistent with the Knudson two-hit model, somatic biallelic VHL mutations are common in sporadic ccRCCs. The VHL gene product nucleates an ubiquitin ligase that targets the alpha subunits of the heterodimeric transcription factor HIF (hypoxia-inducible factor) for proteasomal degradation when oxygen is plentiful. The recognition of HIF↑ by pVHL requires that HIF↑ be hydroxylated on one (or both) of two conserved prolyl residues by the oxygen-dependent EglN (also called PHD) prolyl hydroxylases. HIF↑, bound to HIF↓ (also called ARNT), transcriptionally activates genes that promote adaptation to hypoxia such as VEGF and EPO. Deregulation of HIF, and particularly HIF2, drives pVHL-defective tumorigenesis. EglN inhibitors are being developed for the treatment of anemia and ischemic diseases, whereas HIF2 inhibitors are being developed for the treatment of pVHL-defective tumors. The thalidomide-like drugs (“IMiDs”) bind to cereblon, which is the substrate recognition subunit of another ubiquitin ligase that loosely resembles the pVHL ubiquitin ligase. The IMiDs kill multiple myeloma cells by reprogramming the cereblon ligase to earmark the transcription factors IKZF1 and IKZF3 for destruction. This discovery has galvanized interest in developing drugs that degrade otherwise undruggable proteins.

von Hippel-Lindau (VHL) Disease presents as a seemingly autosomal dominant hereditary cancer syndrome in which there is a predilection for various tumors including hemangioblastomas of the retina, cerebellum, and spinal cord; clear cell renal carcinomas; and paragangliomas (including intraadrenal paragangliomas, which are called pheochromocytomas). The first published report of this disease appeared in 1894 by the British geneticist Treacher Collins (1), although the syndrome derives its name from reports of the German ophthalmologist Eugen von Hippel in 1904 (2) and the Swedish pathologist Arvid Lindau in 1927 (3). In hindsight, some VHL families have clearly been affected for centuries.

Both the von Hippel and Lindau papers were published in German. This was advantageous in terms of readership when they were published, which has sometimes been referred to as the great descriptive era in medicine. Ironically, given circumstances today, the Collins paper might have been underappreciated because it was published in English. von Hippel also benefited from having a father who was in academic medicine and therefore in a position to publicize his son’s work.

VHL Disease affects about 1 in 35,000 people worldwide. At the molecular level, VHL disease is caused by germline (or less commonly, mosaic) loss of function mutations of the VHL tumor suppressor gene, which resides on chromosome 3p. Pathology develops when the remaining wild-type allele is spontaneously lost, mutated, or silenced in a susceptible cell such as in the eye, cerebellum, or kidney. The lifetime risk of such an event is extremely high, which accounts for the seemingly autosomal dominant pattern of inheritance. VHL Disease is not genetically heterogeneous insofar as patients with the classical stigmata of VHL Disease can be presumed to carry a VHL mutation. Nonetheless, there are strong genotype-phenotype correlations in VHL Disease that govern the spectrum of tumors (e.g., presence or absence of paraganglioma) observed in different families.

As would be predicted from the knowledge that germline VHL mutations predispose to clear cell renal cell carcinoma (ccRCC), biallelic VHL inactivation is also extremely common in sporadic ccRCCs. In this setting, the first “hit” is usually a spontaneous, somatic, intragenic VHL mutation, and the second “hit” is loss of the remaining VHL allele as the result of loss of chromosome 3p. Somatic VHL mutations have also been reported in sporadic hemangioblastomas and paragangliomas (4, 5). In short, the VHL gene conforms to the Knudson two-hit model.

ccRCCs and hemangioblastomas are notoriously rich in blood vessels, which has been linked to their ability to massively overproduce vascular endothelial growth factor (VEGF). These two tumors, together with paragangliomas, also occasionally cause paraneoplastic erythrocytosis due to their ectopic production of erythropoietin (EPO) (6). VEGF and EPO are both products of hypoxia-inducible genes. This was an important clue that the VHL gene product played a role in oxygen sensing.

Many hypoxia-inducible genes, including the aforementioned VEGF and EPO, are controlled by a DNA-binding heterodimeric transcription factor called hypoxia-inducible factor (HIF). HIF consists of a labile alpha subunit (HIFα) that is normally degraded when oxygen is plentiful and a stable beta subunit (HIFβ, also known as ARNT). When oxygen levels are low, HIFα accumulates, dimerizes with HIFβ, and transcriptionally activates genes such as VEGF and EPO that contain dedicated HIF-binding sites known as hypoxia-response elements.

The VHL gene encodes two protein isoforms because of alternative, in-frame, start codons. Both isoforms behave similarly in most of the biochemical and biological assays performed to date aimed at understanding their functions and, hence, are often referred to generically as pVHL. pVHL is the substrate recognition component of an ubiquitin ligase complex that also contains elongin B, elongin C, Cul2, and the ubiquitin conjugating enzyme Rbx1. When oxygen is plentiful, pVHL binds directly to HIFα subunits and targets them for proteasomal degradation. When oxygen is scarce, or pVHL is mutated or lost, HIFα accumulates, dimerizes with ARNT, and activates HIF-responsive genes such as VEGF.

The binding of pVHL to HIFα is oxygen-sensitive because binding only occurs if HIFα is hydroxylated on one (or both) of two conserved prolyl residues (7). Prolyl hydroxylation of HIFα is carried out by members of the EglN (also called PHD) family of 2-oxoglutarate-dependent dioxygenases (7). These enzymes have low oxygen affinities and hence respond to changes in oxygen over a physiologically relevant range of oxygen concentrations (7). The EglNs also required reduced iron and the cofactor 2-oxoglutarate (also called α-ketoglutarate), which is decarboxylated to succinate during the hydroxylation reaction.

The EglNs can be inhibited with orally available drug-like 2-oxoglutarate-competive antagonists. At least four such compounds have advanced to late-stage clinical trials for the treatment of anemia in the setting of chronic kidney disease (see Table 1) (8–10). In preclinical models and in patients, these compounds stabilize HIF, leading to the transcriptional activation of EPO and other HIF-responsive genes dedicated to erythropoiesis. Importantly, these compounds can reactivate hepatic EPO expression in preclinical models, which likely contributes to their efficacy in patients who are functionally anephric (11, 12). The most clinically advanced EglN inhibitor, Roxadustat, has been approved in a host of countries, including China, Japan, the United Kingdom, the European Union, South Korea, and Chile. The U.S. Food and Drug Administration (FDA) did not approve Roxadustat for the treatment of anemia because of potential safety concerns, including a possible increased risk of thrombosis. It remains to be seen whether this risk is real and, if so, can be mitigated by dose adjustments that result in a less rapid rise in red blood cell mass.

TABLE 1.

Late-Stage Prolyl Hydroxylase Inhibitors

Roxadustat

Vadadustat

Daprodustat

Molidustat

The pVHL-HIF-EglN oxygen-sensing pathway is conserved throughout metazoan evolution, presumably because it allows cells, tissues, and organisms to survive in changing oxygen environments. Many diseases of the developed world, such as myocardial infarctions and strokes, are linked to inadequate oxygen delivery. In preclinical models, inhibiting EglN, either genetically or pharmacologically, helps preserve tissue viability caused by experimentally induced ischemia of the brain, heart, or kidney.

Although HIF appears to acutely protect the heart against ischemia, sustained high-level activation of HIF in the heart, such as through genetic ablation of EglNs or transgenic expression of a stabilized version of HIF1α, causes dilated cardiomyopathy associated with loss of mitochondria, probably due to HIF-stimulated mitophagy mediated by BNIP3 (13,14). Atherosclerotic coronary artery disease is a major cause of ischemic dilated cardiomyopathy. In this setting, the degree of cardiac dysfunction is often out of proportion to the amount of heart muscle loss caused by prior myocardial infarctions. These new findings suggest that chronic HIF activation, per se, can contribute to cardiac dysfunction in the setting of ischemic cardiomyopathy.

Tissues that survive an ischemic insult are briefly partially protected from subsequent ischemic insults—a phenomenon known as “ischemic preconditioning.” Remarkably, ischemic tissues can also protect other tissues at a distance from ischemic insults. This phenomenon, called “remote ischemic preconditioning” (RIPC), has been demonstrated in a variety of animal models. Nonetheless, attempts to demonstrate and exploit RIPC in humans, such as by overinflating blood pressure cuffs to induce limb ischemia immediately prior to elective heart surgery, have produced mixed results, with two large randomized clinical trials failing to show a benefit of remote ischemia prior to surgery (15, 16).

We reasoned that elucidating the mechanism underlying RIPC might allow one to revisit whether RIPC could be harnessed in humans for therapeutic benefit. RIPC, at least as observed in laboratory models, is presumed to involve neural or hormonal factors. One predictable consequence of organ ischemia is to impair EglN activity, leading to HIF activation. We therefore made mice in which we could inactivate EglN specifically in skeletal muscle prior to subjecting them to experimental myocardial infarctions (17). Remarkably, skeletal inactivation of EglN1 was cardioprotective against ischemia-reperfusion injury. By conducting parabiosis and serum metabolomic experiments, we confirmed that this protection was mediated by a circulating factor, kynurenic acid, which had been shown previously to confer protection against ischemia in other preclinical models. We showed that EglN inactivation increases circulating 2-oxoglutarate, which then drives the formation of KynA by the liver, and that KynA was both necessary and sufficient for cardioprotection in our RIPC models (17).

KynA has a number of candidate receptors, although how it confers ischemic protection was unknown (18–24). We recently discovered that ischemic protection by KynA is mediated by the orphan G protein-coupled receptor GPR35 (25). Upon binding to KynA, GPR35 translocates from the cell membrane to mitochondria, whereupon it induces ATP synthase dimerization in an ATPIF1- and G protein-dependent manner. ATP synthase dimerization prevents its futile consumption of ATP under hypoxic conditions. These findings support the exploration of GPR35 agonists to prevent or treat organ ischemia.

In preclinical models, downregulation of HIF2α is both necessary and sufficient for suppression of ccRCC by pVHL, arguing that deregulation of HIF2α drives VHL-/- ccRCC tumorigenesis (26). Unexpectedly, HIF1α, the better and more widely studied HIF paralog, suppresses VHL-/- ccRCC in such preclinical models and is likely to be one of the targets of the large deletions of chromosome 14q that are typical of ccRCC (27–29).

By the 1990s, many pharmaceutical companies were developing inhibitors against the HIF-responsive gene product VEGF or its receptor KDR, which acts as a tyrosine kinase upon ligand binding. Such inhibitors have revolutionized the treatment of ccRCC, with eight such inhibitors FDA approved for this indication (Table 2) (26). Nonetheless, not all ccRCC patients respond to VEGF inhibitors, and those that do will invariably progress over months or years.

TABLE 1.

FDA Approved VEGF Inhibitors for Treating Kidney Cancer

Bevacizumab

Sunitinib

Sorafinib

Axitinib

Pazopanib

Cabozantib

Levantinib

Tivozanib

Based on first principles, targeting HIF2α itself should be more effective than targeting any one HIF-responsive gene product in isolation. However, the conventional wisdom was that HIF, as a DNA-binding transcription factor of the bHLH-PAS domain family, was undruggable. Fortunately, Rick Bruick and Kevin Gardiner, who were then at University of Texas–Southwestern, identified a potentially druggable pocket in the HIF2α PAS B domain and chemical matter that, upon binding to this pocket, induced an allosteric change in HIF2α such that it could no longer dimerize with ARNT (and hence no longer bind to DNA) (30–32). These initial chemicals were optimized further through medicinal chemistry efforts at Peloton Therapeutics, resulting in the tool compound PT2399 and the related clinical candidates PT2385 and PT2977 (Belzutifan). We and others showed that these compounds inhibited the growth of VHL-/- ccRCC in preclinical models in an on-target fashion (33–35). Belzutifan has advanced to Phase 3 testing for metastatic/advanced ccRCC based on promising Phase 1/2 data in ccRCC patients who had failed frontline agents such as VEGF inhibitors and immune checkpoint inhibitors (36).

Most cancer drugs work better in the frontline setting than in later lines of therapy. Partly for this reason, Belzutifan was tested in a clinical trial of 61 VHL patients who had measurable ccRCCs that were being monitored in surveillance programs in an attempt to delay or prevent the need for multiple renal surgeries and that had not been treated medically before (37, 38). Remarkably, almost all of the patients had measurable kidney tumor shrinkage, with a Response Evaluation Criteria in Solid Tumors (RECIST) response rate close to 60%. Gratifyingly, responses were also frequently observed in incidental non-renal tumors that could also be measured and followed in these patients, including many central nervous system hemangioblastomas and pancreatic neuroendocrine tumors. Moreover, the need for VHL Disease-associated surgeries, such as eye, spinal cord, and renal surgeries, decreased dramatically once Belzutifan was started (37). Based on these data, the FDA approved Belzutifan for the treatment of VHL Disease in August 2021.

Thalidomide was used in the mid-twentieth century to treat morning sickness and insomnia and was available without a prescription (i.e., was sold over the counter) in many countries around the world. Tragically, many children who were exposed to thalidomide in utero were born with severe limb defects (“thalidomide babies”), leading to its removal from the market. In the 1990s, thalidomide was resurrected when it was serendipitously found to be highly active against multiple myeloma. More potent thalidomide analogs, such as lenalidomide, are now mainstays of myeloma therapy and are collectively referred to as “IMiDs.”

How thalidomide killed myeloma cells, however, remained a complete mystery. The first clue came from Hiroshi Handa and coworkers in 2010 (39). They showed that thalidomide, coupled to a solid support, captured a ubiquitin ligase complex from cell extracts that included cereblon as its substrate recognition subunit. Moreover, they provided experimental evidence that thalidomide could inhibit cereblon ubiquitin ligase activity, at least in vitro.

Soon thereafter, however, several clinical reports showed that high cereblon levels in myeloma cells presaged their sensitivity to IMiDs, whereas some myeloma cells that had become resistant to IMiDs no longer expressed cereblon (40–44). This latter observation suggested to us, and to Ben Ebert’s group working in parallel, that the IMiDs were not killing myeloma cells by inhibiting cereblon, but instead required the cereblon protein to kill myeloma cells, perhaps by redirecting it to polyubiquitylate one or more proteins required for myeloma survival.

Our two groups used complementary approaches to show that, indeed, cereblon, once bound to an IMiD, acquires the ability to polyubiquitylate two transcription factors, IKZF1 and IKZF3 (45,46). We further showed that degradation of these proteins is both necessary and sufficient for the antimyeloma activity of the IMiDs. Subsequent studies by others showed that another neosubstrate, SALL4, accounts for the limb defects caused by in utero exposure to thalidomide (47, 48). Interestingly, loss of function SALL4 mutations cause Holt-Oram -Syndrome, which includes limb defects reminiscent of those observed in thalidomide babies. IKZF1 and IKZF2 would classically have been viewed as undruggable. These insights into the mechanism of action of IMiDs have galvanized interest in developing drugs that indirectly or directly degrade undruggable proteins linked to cancer and other diseases (49).

DISCUSSION

Goodenberger, St. Louis: That was a wonderful talk, thank you. Do you know if VEGF inhibitors have a salutary effect in hereditary hemorrhagic telangiectasia (Osler-Weber-Rendu disease)? Has your drug (Belzutifan) been tested in that setting, or are there any plans to do so?

Kaelin, Boston: That’s a great question. I don’t know if VEGF inhibitors or Belzutifan have been tried yet in Osler-Weber-Rendu disease, so I’m going to do a medical student trick and address a somewhat related question that I actually can answer. I worked with a company called TRACON that tested an antibody against CD105 as a potential anti-angiogenic agent. CD105 is a product of the gene that, when mutated, causes Osler-Weber-Rendu disease. The TRACON anti-CD105 antibody, as a pharmacodynamic effect of the drug, caused patients to develop the telangiectasias of Osler-Weber-Rendu disease.

Reiser, Chicago: Thank you for this very inspirational talk. I’m going to go back to the laboratory today! It seems like mitochondria would be an ideal site for oxygen sensing, and I think there were some data to support that idea, but then the field kind of stalled. So how does reactive oxygen species (ROS) production relate to HIF-1 alpha?

Kaelin, Boston: You are correct. In fact, I try to explain to students all the things that contributed to our winning the Nobel Prize. So many things are out of your control. Several things helped us—one was that there were so many competing models of how oxygen sensing would work that people were just completely lost. Sort of like the blind man and the elephant. The models were sometimes contradictory, and they certainly weren’t elegant. Then we stumbled upon this elegant, beautiful mechanism that everyone could appreciate. So, we benefited from the fact that the mechanism we discovered was elegant, which was a tribute to nature, not to us, and it really brought some clarity to this issue of oxygen sensing. Now as you state in your question, lots of other inputs affect oxygen sensing. For example, if you change mitochondria metabolism and ROS production, ROS can talk to the enzymes I described that prolyl hydroxylate hypoxia-inducible factor (HIF). That will then affect the upregulation of HIF, but the oxygen-dependent prolyl hydroxylation step seems to be at the core of the oxygen sensing mechanism. It is certainly, as I said, conserved throughout metazoan evolution. However, we can now reinterpret those other findings related to mitochondria and ROS production and see how they were influencing this prolyl hydroxylation mechanism.

Crowley, Boston: Terrific talk and I love talks from people who do their work on humans. So, I’ll start there because I think many of the clues are there (studying human diseases) all the time. There’s one theme in this meeting that I would ask about though. In oncology, there’s a certain tendency to look for known genes and known pathways. Once you have the first picture—sort of a boiler plate—you know how the paintings go from one to the next. There is a line of evidence, however, that’s bringing in entirely new and different genes that suggest there is a lot we don’t know. I am influenced greatly by the Broad Institute where genome-wide association studies (GWAS) are turning up a variety of loci and genes that don’t fall on known pathways, but which may be important when you’re trying to describe some of the heterogeneity of some people not responding and others responding. Is there any reason to take such genes and loci and map them to new pathways that we don’t know anything about? They might, in fact, be manifesting themselves through the resistance to therapy.

Kaelin, Boston: Yes, that’s a great question at multiple levels. Even though I gave you several examples of what some people might call translational research or applied research that was aimed at getting to a therapy, we are in a very dangerous time because I think we, and especially young people, are under too much pressure to translate. I think we should be in the knowledge generation business. We should do the fundamental work that frankly our colleagues in biotech and pharma can’t do because the timelines and the deliverables are too unpredictable. They rely on us to do that early, basic, fundamental work. They are understandably thinking, “We’re pretty good at that late-stage stuff, but we rely on you academics to be the knowledge creators.” So, I’m always a little sensitive about this talk because I don’t want young people to get the sense that you just do something when you’re on the five-yard line (with respect to something being applied), and you just have to run a couple of off tackle plays to get into the end zone. We should be in the knowledge generation business, and to your point, there are a lot of things we don’t know.

I agree with you with respect to only following known genes and pathways. If you do so, you’re like the drunk under the lamp post and you are losing all sorts of opportunities. I do think we have to marry going deep on certain things we think we know about to less biased approaches that bring in some of the things we don’t know. I know you mentioned the Broad Institute, where they do a lot of high throughput genomic screening. It’s not uncommon that some anonymous gene will suddenly score in such a screen. We do this too, and it’s actually quite fun when a gene that we know nothing about actually scores as being a modulator of one of these pathways that we thought we knew quite well. So, I kind of completely agree with you.

Tweardy, Houston: Fantastic talk on multiple levels. As a Nobel Prize winner, I think you become an oracle and are able to predict the future. I wonder if you could talk a little bit about where the direction of the proteolysis targeting chimeric (PROTAC) strategy is going and whether you think it will actually work.

Kaelin, Boston: Yes, that actually beautifully dovetails with the last question because a cottage industry has emerged almost overnight. Biotech and pharma are trying to make the next generation of degraders and are really running with this idea of how you’re going to degrade “undruggable” proteins with small molecule degraders, including PROTACs. That’s a wonderful thing. I think we should be symbiotic with the biotech and pharma companies or, put another way, I always ask myself, why are we doing this -experiment? For example, if we’re about to do something and it’s really obvious and it just needs to be scaled and to have resources, I’m not sure we should be doing it. Maybe our colleagues in pharma should do it because they do that kind of thing much better. At one point, Harvard was having a sort of “Come to Jesus moment” about how much drug discovery should be done at Harvard. Eric Lander, who was one of the advisors, stood up and said, “I think academics should only do the kinds of experiments where if you succeed it really messes with people’s heads.” So that’s one of my benchmarks. Once it’s turning the crank and is now an engineering exercise, that is better done in biotech or pharma. For PROTACS and other degraders to succeed, the main issues are now chemistry issues, not biology issues. Much elegant chemistry will need to be done to make degraders more drug-like, but I’m not sure that’s the best role for most academic labs and certainly not for my lab. I do think the degrader paradigm has been established.