The von Hippel-Lindau Tumor Suppressor Gene: Implications and Therapeutic Opportunities

Abstract

The discovery of the von Hippel-Lindau (VHL) gene marked a milestone in our understanding of clear cell renal cell carcinoma (ccRCC) pathogenesis. VHL inactivation is not only a defining feature of ccRCC, but also the initiating event. Herein, we discuss canonical and noncanonical pVHL functions as well as subsequent breakthroughs shaping our understanding of ccRCC evolution and evolutionary subtypes. Current and evolving strategies to therapeutically exploit effector mechanisms downstream of pVHL are also discussed.

Introduction

Kidney cancer, which is largely made up of renal cell carcinoma (RCC), is among the top ten cancers in the United States, accounting for ~75,000 new cases per year¹. RCC is subclassified histologically, and ccRCC comprises ~75% of cases². ccRCC is resistant to chemotherapy, and until 2005, only one drug was approved by the FDA. However, there has been a surge in the number of drugs available for the treatment of ccRCC, initially with targeted drugs, and most recently, immune checkpoint inhibitors (ICIs). These advances, in particular the development of targeted therapies, were made possible through an improved understanding of ccRCC biology. The first breakthroughs arose from studying a rare inherited disorder, von Hippel-Lindau (VHL) syndrome. These findings propelled decades of research to elucidate the fundamental principles of ccRCC development and set the foundation for novel therapies.

Genetic linkage analysis locates the VHL gene to chromosome 3p

Von Hippel-Lindau syndrome is an autosomal dominant disease with age-dependent penetrance caused by germline mutations in the VHL gene³. This syndrome manifests with multiple malignancies including ccRCCs, hemangioblastomas of the central nervous system (CNS), pheochromocytomas, pancreatic neuroendocrine tumors, as well as renal and pancreatic cysts. Genetic linkage analysis of affected kindreds located the responsible gene to the chromosomal region 3p25-26 in the late 1980s^4,5. The VHL gene was identified by positional cloning in 1993⁶, and VHL mutations were subsequently identified in sporadic ccRCC⁷. Loss of heterozygosity (LOH) and inactivation of the second VHL allele were seen in both sporadic and hereditary cases implicating VHL as a classical two-hit tumor suppressor gene (TSG)^8,9. Consistent with VHL functioning as a TSG, reintroduction of VHL into the VHL-deficient ccRCC tumor cell line 786-O decreased tumor formation in mice¹⁰. More recent analyses have highlighted the pervasiveness of VHL inactivation in ccRCC. Inactivation of VHL, through intragenic mutation, epigenetic silencing, or large deletion, occurs in up to 91% of sporadic cases^11,12. In fact, VHL loss is the earliest event and a truncal finding in ccRCC development. However, contrary to initial suppositions that tumorigenesis is initiated by an intragenic mutation and is followed by 3p loss, recent evidence suggest that loss of 3p is the initiating event. Chromothripsis involving chromosomes 3p and 5q followed by mutation or epigenetic silencing of the second VHL allele is thought to initiate ccRCC development¹³.

The VHL gene product, pVHL, is a key regulator of the cellular response to hypoxia¹⁴. Biochemical studies revealed that pVHL forms a ternary complex with elongins B and C, which link pVHL to cullin-2 (CUL2) and RING-box protein 1 (RBX1)¹⁵.This E3 ubiquitin ligase complex binds hypoxia-inducible factor (HIF) α subunits after they have been modified at particular prolyl residues by prolyl hydroxylases (EglN1-3)¹⁶, resulting in its ubiquitination and degradation¹⁵. Interestingly, elongin C has also been found to be mutated in ccRCC, though to a significantly reduced extent (<5%) and mutations in the corresponding gene (TCEB1) are associated with LOH of chromosome 8, where TCEB1 resides¹⁷. In addition, mutations in CUL2 were reported in up to 1% of ccRCCs¹⁸. These mutations tend to be mutually exclusive and suggest that the oncogenic properties of the different genes result from disruption of the pVHL complex. While there are differences between ccRCC with mutations in TCEB1 and VHL^19,20, overall, inactivation of the pVHL complex may be regarded as a defining signature event in ccRCC.

The paradigmatic substrate of pVHL is HIF-α. The HIF family of transcription factors acts on hypoxia response elements (HREs) distributed throughout the genome controlling the expression of genes implicated, among other processes, in adaptation to hypoxia²¹. Hypoxia-inducible factor is a heterodimer complex consisting of a labile α subunit and a stable β subunit. Under normoxic conditions, the HIFα subunit is hydroxylated, leading to its recognition and ubiquitination by the pVHL complex²². Loss of VHL impairs HIF regulation, resulting in accumulation of HIFα and constitutive expression of HIF target genes. In humans, there are three HIFα proteins, HIF-1α, HIF-2α, and HIF-3α²¹. While there is substantial overlap between HIF-1α and HIF-2α target genes, some genes are exclusively regulated by HIF-1 or HIF-2^23-26. Furthermore, some evidence suggests that HIF-2 is the key driver of ccRCC, whereas HIF-1 may have inhibitory effects²⁷. In pre-clinical models, HIF-2α overexpression is both necessary and sufficient for tumor growth following VHL loss^23,28,29. In contrast, overexpression of HIF-1α in cell lines such 786-O, which don’t typically express HIF-1α, inhibits tumor growth in xenograft models³⁰. In addition 14q, which harbors the gene encoding HIF-1α, is lost in some ccRCC, and mutations of the remaining allele are occasionally seen^12,31,32. Conversely, genome wide association studies identified a single-nucleotide polymorphism mapping to EPAS1 (which encodes HIF-2α) in association with increased risk of ccRCC³³.

The VHL/HIF axis has implications well beyond cancer. Oxygen sensing is a fundamental element of pathophysiology, and plays a critical role in multiple processes. For their discovery and characterization of the cellular mechanism of oxygen sensing, the 2019 Nobel Prize in Physiology or Medicine was awarded to Drs. William G. Kaelin Jr., Sir Peter J. Ratcliffe and Gregg L. Semenza^34,35.

Identification of additional tumor suppressor genes on chromosome 3p

Whereas VHL inactivation is nearly universal in ccRCC, it is not sufficient to cause ccRCC. VHL loss has been observed in preneoplastic cysts, and mice with Vhl disruption in the kidneys do not develop ccRCC^36–38. This is the case even when both alleles of Vhl are inactivated in the appropriate compartment in the kidney^37,39. These data provide unequivocal evidence that VHL loss is not sufficient for ccRCC tumorigenesis, and that additional driver mutations are required.

The additional driver mutations required for ccRCC development remained elusive until the advent of massively parallel sequencing. This technology led to the discovery of three additional ccRCC TSGs located on chromosome 3p. The first to be discovered was the histone-lysine N-methyltransferase SET Domain Containing 2 (SETD2). Targeted sequencing of a ~3,500 gene panel revealed recurring truncating SETD2 mutations⁴⁰. Subsequent studies found that nonsilent SETD2 mutations occur in ~20% of ccRCC^12,17,18. After SETD2, the polybromo 1 (PBRM1) and the BRCA1 associated protein-1 (BAP1) genes were discovered. Varela et al., identified truncating mutations in PBRM1 through whole exome sequencing (WES) of 7 ccRCC⁴¹. PBRM1 mutations were subsequently identified in 88 (39.8%) of 221 ccRCCs sequenced, making it the second most common TSG in ccRCC⁴¹. A few months later, whole exome sequencing (WES) of paired tumor and patient-derived xenografts (also called tumorgrafts [TGs]) identified BAP1 mutations in ccRCC⁴². BAP1 targeted sequencing of 176 ccRCC tumors revealed 24 (14%) nonsilent mutations ⁴². These initial studies were confirmed and expanded upon by sequencing efforts that followed^12,17,18. Additional driver mutations and somatic copy number alterations (SCNAs) of some significance include: (i) activating mutations in the mTOR pathway^12,17,43, (ii) inactivating mutations in additional chromatin modifying genes^12,17,40,44, (iii) loss of TP53^12,17,45, and (iv) deletion of chromosomes 14 and 9^{12,17,32,46–48}.

SETD2, PBRM1, and BAP1, are all involved in epigenetic regulation of gene expression. SETD2 is a histone methyltransferase for histone 3 lysine 36 (H3K36), and the only known enzyme to trimethylates H3K36^49,50. PBRM1 encodes a component of the switching defective/sucrose nonfermenting (SWI/SNF) nucleosome remodeling complex (BAF180). Through its tandem bromo domains, BAF180 recruits the complex to specific nucleosomes characterized by particular histone tail acetylation patterns⁵⁰. The BAP1 protein is a deubiquitinase that acts on lysine 119 of histone H2A to reverse polycomb-mediated gene repression^50,51. Mutations in PBRM1 and BAP1 likely contribute to differences in gene expression we discovered between BAP1- and PBRM1-deficient tumors⁵².

Interestingly, BAP1, PBRM1, SETD2 are in physical proximity to VHL within a 50 Mb region on chromosome 3p^11,32,51. Notably, BAP1 and PBRM1 mutations tend to be mutually exclusive in ccRCC^12,42,52. The prevalence of tumors with combined mutations in BAP1 and PBRM1 is lower than expected by chance alone (~1-2% vs. ~5%) given the frequency of PBRM1 (~45%) and BAP1 (~12%) mutations^42,52–54. In most tumor types, mutual exclusivity typically indicates that the encoded proteins are in the same pathway (such as for pVHL and elongin C). Once the pathway is deregulated through mutation in one of the genes, there is limited added benefit to mutation of the other. However, inasmuch as ccRCC with loss of BAP1 and PBRM1 differ in grade and gene expression signatures, this is unlikely to be the case in ccRCC^42,52. Furthermore, BAP1 mutant tumors are associated with worse RCC-specific survival compared with tumors with loss of PBRM1 ^52,55–57. The differential association with outcomes between BAP1- and PBRM1-deficient tumors was subsequently extended to metastatic RCC^58,59.

These observations provided initial evidence of distinct molecular subgroups within ccRCC. PBRM1 and BAP1 tend to be mutually exclusive and are associated with unique tumor biology, ultimately resulting in divergent clinical outcomes. Multiregional sequencing of human tumors through the TRACERx consortium as well as studies in genetically-engineered mouse models expanded upon these findings.

Distinct evolutionary trajectories in ccRCC

The mouse has traditionally been the animal model of choice in oncology. Remarkably, dysregulation of relevant cancer genes in the corresponding organs often triggers tumor development and tumors reproduce those in humans. However, in contrast to humans, Vhl heterozygous mice were not predisposed to ccRCC^36,60. Furthermore, even targeting of two copies of Vhl in the mouse kidney failed to induce tumor development^37,61,62. While this may occur when the gene is inactivated in an irrelevant cell type, even when Vhl is disrupted in the appropriate cell type, Vhl alone was not sufficient for tumor development^37,61,62. As outlined above, the majority of ccRCCs in humans harbor not only VHL mutations, but also mutations in either PBRM1 (~50%) or BAP1 (~15%), and these mutations tend to be mutually exclusive. These data led us to hypothesize that ccRCC development in the mouse may require combined inactivation of Vhl together with Pbrm1 or Bap1. Indeed, simultaneous targeting of Vhl/Pbrm1 and Vhl/Bap1 in nephron progenitor cells induced ccRCC in the mouse^37,39,63,64. Interestingly, as in humans Vhl/ Pbrm1-deficient tumors were of low grade whereas Vhl/Bap1-deficient tumors were of high grade³⁹. In addition, Vhl/Bap1-deficient tumors developed with a shorter latency than Vhl/Pbrm1-deficient tumors. In contrast, Bap1 or Pbrm1 loss by themselves did not induce tumor formation^37,39. These data show that: (i) inactivation of VHL, PBRM1 or BAP1 alone is insufficient for ccRCC development; (ii) combined inactivation of VHL with either PBRM1 or BAP1 is necessary; and (iii) PBRM1 and BAP1 are determinants of tumor grade. Interestingly, Vhl/Pbrm1-deficient tumors could be transformed into tumors of high grade by targeting Tuberous Sclerosis Complex 1 (Tsc1)³⁹, which we previously discovered to be mutated in ccRCC⁴³. TSC1 forms a complex with Tuberous Sclerosis Complex 2 (TSC2) and is a critical regulator of mTOR complex 1. A third tumor suppressor gene on chromosome 3p is SETD2, but whether combined inactivation of Vhl and Setd2 is sufficient for ccRCC development remains unknown. In addition, unlike PBRM1 and BAP1, which are generally mutually exclusive and seem to bifurcate tumor development, SETD2 mutations tend to be enriched in PBRM1-deficient tumors⁵⁴. Parenthetically, whereas in the human VHL, PBRM1 and BAP1 are on the same chromosome, Vhl is on one chromosome, Bap1/Pbrm1 are on another, and Setd2 on a third chromosome in the mouse. This difference implies that whereas in the human a single event (loss of 3p) simultaneously inactivates one copy of all four TSGs, in the mouse, this would require deletions in three different chromosomes. This finding may explain the differential tumor predisposition across the two species.

Additional insight into the biology of ccRCC has been provided by the TRACERx consortium. The TRACERx consortium was established to assess the degree and impact of intratumoral heterogeneity on tumor biology and clinical outcomes, which early studies had revealed may be significant^65–67. Turajlic and colleagues performed deep sequencing of 110 oncogenes and TSG across 1,206 ccRCC tumor samples from 101 patients who were prospectively enrolled^68,69. This allowed the detection of clonal and subclonal somatic mutations providing an unprecedented view of the molecular diversity within a single tumor. While limited by gene density, the approach also offered a view of SCNAs, and in particular, LOH. After accounting for heterogeneity, the TRACERx consortium estimated the following mutation prevalence: PBRM1 (55%), SETD2 (25%), and BAP1 (19%)⁶⁹. In keeping with previous analyses^52,54, BAP1 and PBRM1 mutations, which were found, in some instances in the same tumor, were typically in spatially distinct regions⁶⁹.

The analyses of spatially distinct regions within the same tumor enabled an assessment of clonality and inferences about the timing of mutations. Thus, patterns of tumor evolution could be derived. Distinct evolutionary pathways were identified through rule-based clustering⁶⁹. 3p loss through chromothripsis was the most common initial event in ccRCC development, often occurring decades before clinical presentation¹³. Consistent with previous results⁶⁶, after inactivation of the second VHL allele, the most common truncal mutation was in PBRM1. Mutations in BAP1 were sometimes similarly truncal⁶⁹. As shown previously, the “VHL → BAP1” evolutionary subtype was associated with decreased disease free survival and overall survival^42,52,57,68. PBRM1 loss was often followed by other mutations which, when present, were associated with more aggressive disease and increased metastatic potential^68,69. Three main branches following PBRM1 loss were: (i) “PBRM1 → SETD2”, (ii) “PBRM1 → mTOR”, and (iii) “PBRM1 → Driver SCNAs”. Another subtype identified was the “multiple clonal drivers” subtype, which contained mutations in two or more of the following genes: BAP1, PBRM1, SETD2, or PTEN. In this group, the clonality of mutations could not be distinguished, indicating they occurred within a relatively short time span of each other. This subtype tended to be the most aggressive, and included the BAP1/PBRM1-deficient tumors which we and others noted to portend a poor prognosis previously^42,53,59.

In summary, strong evidence from genetically-engineered mouse models and patient tumors supports the conclusion that BAP1 and PBRM1 mutations represent distinct branches in ccRCC tumor evolution (Figure 1). In the setting of BAP1 mutations, tumors gain aggressive features and as a consequence have high fitness, reducing the need for additional driver mutations. As a result, these tumors have lower intratumoral heterogeneity, higher propensity to metastasize, and reduced overall survival^68,69. Conversely, PBRM1-mutant tumors tend to be associated with less aggressive features and are fertile ground for additional driver mutations which enhance tumor fitness. Perhaps the best supported of these are the “PBRM1 → mTOR pathway” and the “PBRM1 → SETD2” trajectories. The former is consistent with the finding that deletion of one allele of Tsc1 in Vhl/Pbrm1 deficient mice reduced tumor latency and increased tumor grade³⁹. The latter is consistent with the observation that mutations in SETD2 are found at increased frequency in PBRM1-deficient tumors suggesting cooperativity⁵⁴. Furthermore, in the TRACERx cohort, PBRM1 loss preceded SETD2 loss in in all the tumors where mutations in both were found⁶⁹, consistent with the notion that PBRM1 loss is an earlier event in ccRCC oncogenesis.

Figure 1. Integrated ccRCC development model.

ccRCC development often begins with a chromothripsis event involving chromosomes 3p and 5q, followed by inactivation of the remaining VHL allele. This is followed by loss of BAP1 or PBRM1. Tumors with mutations in VHL and BAP1 are more aggressive than those with VHL and PBRM1 mutations. Further acquisition of driver mutations in PBRM1-deficient tumors increases their aggressiveness. These include SETD2 mutations, activation of the mTOR pathway, or driver SCNAs. Though rare, simultaneous mutantion in BAP1/PBRM1 induce aggressive tumors and represent a distinct molecular subgroup. The trajectories depicted here likely account for ~60% of ccRCC. Remaining to be discovered are cooperative events in VHL-deficient tumors that are wildtype for PBRM1 and BAP1, as well as driving events in VHL wildtype tumors.

Several important questions remain, however. For instance, in the ~40% of tumors that are seemingly wild-type for both PBRM1 and BAP1, what other events cooperate with VHL loss to promote oncogenesis? Also, how do mutations in TCEB1 and CUL2, which are located on chromosome 8 and 10 respectively, lead to ccRCC formation, and what are the specific cooperating events? It is worth noting that in over a third of tumors in the TRACERx cohort, an evolutionary subtype could not be assigned⁶⁹, suggesting that while the majority of ccRCC tumors may follow the outlined trajectories, there are paths yet to be identified.

Targeting the VHL/HIF axis

Current therapies

An increased understanding of the VHL/HIF axis led to the development of targeted therapies in ccRCC. HIF regulates a number of genes controlling the cellular response to hypoxia, including VEGFA and PDGFB²¹. Notably, ccRCC appears to be the tumor with the highest levels of VEGFA expression⁷⁰. The first targeted therapies to receive FDA approval for the treatment of metastatic RCC were sorafenib and sunitinib^71,72. These tyrosine kinase inhibitors (TKIs) act primarily on the VEGF receptor (VEGFR1-3), but are also active against the related PDGF receptor (PDGFR)⁷³. The number of TKIs to gain FDA approval in RCC has expanded substantially and today includes pazopanib⁷⁴, axitinib⁷⁵, lenvatinib (used in combination with the mTOR complex 1 inhibitor everolimus)⁷⁶, and cabozantinib⁷⁷. While all converge on the VEGF pathway, lenvatinib targets the fibroblast growth factor receptor (FGFR), and cabozantinib has activity against the hepatocyte growth factor receptor (encoded by MET) and the tyrosine protein kinase receptor UFO (encoded by AXL)⁷³. As single agents in the frontline setting, TKIs result in objective responses in as many as 35% of patients with advanced ccRCC, with a median progression free survival between 8-11 months^78–82. Today, frontline therapy has been replaced with combinations of ICIs. These include combinations of ICIs targeting the PD-1 and CTLA-4 checkpoints or of ICI with TKIs, specifically axitinib^73,83–85. These combination therapies result in objective responses rates ranging from 40-60%, and a median progression free survival of 8 – 15 months^83–85. In addition, complete responses, which are rare with TKIs and can be durable, are observed in 10% of patients^83–85. It is worth noting that HIF also controls the transcription of several genes involved with immune evasion including CD274 (encoding PD-L1), NT5E (encoding CD73), ENTPD1 (encoding CD39)²¹, which provide one reason as to why ccRCC might be responsive to immunotherapies.

Targeting HIF-2

There are several advantages to targeting the pVHL pathway. First, VHL inactivation is the most common feature of ccRCC and thus, drugs targeting this pathway may be expected to have activity in a large number of patients. Second, VHL mutations are truncal, and thus drugs directed against this pathway may be less affected by intratumoral heterogeneity. Third, as the initiating event in ccRCC tumorigenesis, pVHL loss not only defines the context for future mutations, but likely engenders strong dependencies.

One promising approach is to directly target the HIF-2α/HIF-1β complex, which is immediately downstream of pVHL. Transcription factors such as HIF-2 have historically been considered undruggable as they generally lack pockets suitable for binding small molecules, such as those found in enzymes. The atomic structure of the HIF-2α PAS-B domain, however, revealed a highly structured pocket that could be bound by small molecule inhibitors^86,87. This led to the development of the first in-class HIF-2α inhibitor PT2385 by Peloton Therapeutics Inc., a small biotech company in the UT Southwestern Medical Center BioCenter^87,88 (the company was acquired by Merck in 2019). PT2385 is a potent and highly selective small molecule inhibitor that specifically dissociates HIF-2α/HIF-1β while leaving HIF-1α/HIF-1β intact⁸⁸. Preclinical testing of PT2399 (a close analogue of PT2385 and tool compound) demonstrated efficacy in VHL deficient RCC cell lines⁸⁹ and across an extensive platform of ccRCC tumorgrafts (TG). Overall, PT2399 decreased tumor growth by >80% in 10/22 (45%) independently derived ccRCC TGs, including in sunitinib resistant TGs²⁵. Overall, PT2399 had greater activity than sunitinib; TGs treated with PT2399 (n=96) had a mean tumor volume shrinkage of ~60% vs ~40% in sunitinib treated TGs (n=82, p=0.0126)²⁵. Interestingly, responses to PT2399 in VHL deficient ccRCC cell lines and tumorgrafts were variable, suggesting that VHL loss alone is not sufficient to predict responsiveness^25,89. Another possibility could be that pVHL may regulate other substrates, including Zinc finger homeoboxes 2 (ZHX2) and Scm-Like With Four Mbt Domain 1(SFMBT1)^90,91. Sensitivity did not appear to be associated with PBRM1 or BAP1 loss either. Overall, sensitive tumors exhibited higher levels of HIF-2α, providing the foundation for future biomarker studies²⁵. In TG analyses, PT2399 did not have activity against non-ccRCC tumors.

PT2385 was the first HIF-2 inhibitor evaluated in humans. In a phase I trial conducted on 51 patients with heavily pretreated metastatic ccRCC, PT2385 demonstrated a favorable safety profile and an efficacy signal; disease control lasting greater than 4 months was seen in 40% (21/51) of patients⁹². The leading grade 3/4 adverse events were anemia and hypoxia, both of which occurred in 5 (10%) patients (all grade 3). Interestingly, both anemia and hypoxia are on-target adverse events, mediated through suppression of erythropoietin (a HIF-2 regulated gene) and effects on carotid body biology respectively^93–95. Fortunately, severe hypoxia was rare (1/51 patients)⁹². Despite these encouraging results, a few patients who received the recommended phase II dose (800mg BID) had suboptimal circulating drug levels^92,96. This was due to extensive hepatic glucuronidation of PT2385 by the UDP-glucuronosyltransferase UGT2B17, which is differentially expressed among individuals^96,97. This led to the development of a second-generation inhibitor, PT2779 (MK-6482, after the purchase of Peloton Therapeutics by Merck), which demonstrated improved pharmacokinetics⁹⁶.

MK-6482 demonstrated a similar safety profile to PT2385 in an open label phase I/II study in advanced ccRCC^98,99. In this cohort of heavily pretreated patients, objective response rates were observed in 24% of patients (13/55) and disease control (defined as a best response of stable disease, partial response, or complete response) was achieved in 80% (44/55) of patients. After a median follow up of 13 months, the median duration of response was not reached, with 81% of patients having responses lasting over 6 months⁹⁹. An ongoing, open-label phase II study evaluating MK-6482 in patients with von Hippel-Lindau associated ccRCC demonstrated similar efficacy in ccRCC lesions, and responses were also observed in other VHL-syndrome associated lesions (CNS and pancreas)¹⁰⁰. This has led the Food and Drug Administration to grant MK-6482 breakthrough designation¹⁰¹. An international phase III trial comparing MK-6482 with everolimus in previously treated advanced ccRCC is currently enrolling (NCT04195750)¹⁰².

While HIF-2α inhibitors are promising, there is evidence of acquired resistance. Prolonged therapy with PT2399 led to the development of acquired resistance in TG models²⁵. Resistance arose from de novo point mutations which restored HIF-2α/HIF-1β dimerization²⁵. Two mutations were identified, one in HIF-2α (G323E), and another one in HIF-1β (F466L)²⁵. Structural and biochemical analyses of PT2385/HIF-2α/HIF-1β revealed that dissociation of the complex is mediated through allosteric modulation of HIF-2α and disruption of HIF-2α/HIF1β dimerization¹⁰³. Interestingly, the HIF-2α G323E mutation functions as a gatekeeper mutation, preventing PT2385 binding^25,103. HIF-1β F466L, on the other hand, is at the interface of HIF-2α and HIF-1β and is thought to increase binding affinity¹⁰³. The HIF-2α G323E mutation was subsequently found in two patients who participated in the phase I clinical trial of PT2385¹⁰⁴. One potential strategy to overcome resistance mediated by HIF-2α G323E mutation is to develop inhibitors which target different pockets. In fact, comprehensive structural analysis of HIF-2α revealed three additional pockets, potentially suitable for inhibitors^103,105.

Another potential strategy to target HIF-2α involves siRNA. While siRNA is an effective tool to inhibit gene expression in-vitro, challenges in the delivery of siRNA have been barriers to its effective use in patients¹⁰⁶. Arrowhead Pharmaceuticals has developed an siRNA delivery system (termed dynamic polyconjugates or DPC), which links siRNA with shielding moieties (which improve stability) and targeting ligands specific to RCC. The first generation siRNA exploits Arg/Gly/Asp (RGD) binding to the integrins αvβ3 and αvβ5, which are commonly expressed in ccRCC¹⁰⁷. In preclinical models, this DPC-RGD construct was able to inhibit tumor growth in xenografts. A second generation siRNA was developed using the TRIM™ platform, ARO-HIF2, which exhibits increased stability¹⁰⁸. A phase Ib trial of this agent in ccRCC has begun enrollment (NCT04169711). A phase Ib trial with a siRNA targeting HIF-1α (RO7070179) in advanced hepatocellular carcinoma was recently reported¹⁰⁹. The trial was terminated early due to failure to meet the primary endpoint of HIF-1α mRNA depletion following a single cycle (10mg IV dose), although one (of ten) patient demonstrated a partial response lasting over one year¹⁰⁹. In addition to potentially overcoming the HIF-2α resistance mutations highlighted above, the HIF-2α siRNA approach is attractive because it can reduce treatment-related adverse events by specifically targeting ccRCC cells.

Synthetic lethality with VHL loss in cancer

While HIF-2α/HIF-1β is a critical effector downstream of pVHL, evidence suggests that other pathways contribute to tumorigenesis. First, as observed across cell lines, TG and patients, not every tumor with a VHL mutation is sensitive to HIF-2α inhibitors. While pharmacokinetic and other factors may contribute, it is likely that not all VHL-deficient ccRCC tumors are HIF-2α dependent and some tumors, in fact, express low or undetectable levels of HIF-2α²⁵. In addition, pVHL has been shown to regulate other proteins besides the HIF2-α. pVHL substrates other than HIFα may account for different subtypes of VHL syndrome (Type 2A, 2B versus 2C)^110–112. Furthermore, Notch and nuclear factor (NF-κB) signaling pathways are constitutively active upon VHL loss in human ccRCC independent of HIF^113,114.

By performing an innovative genome-wide in vitro expression screen of ~17,000 proteins, recent studies identified ZHX2 and SFMBT1 as new pVHL substrates^90,91. ZHX2 was shown to promote NF-κB activation and SFMBT1 induced the expression of Sphingosine Kinase 1 (SPHK1), both of which acted in a HIF independent manner to promote ccRCC tumorigenesis^90,91. N-Myc Downstream-Regulated Gene 3 (NDRG3) was also identified as a potential pVHL substrate, which activates the RAF-ERK1/2 kinase pathway¹¹⁵. Other potential pVHL substrates have been previously described, including EPOR (erythropoietin receptor)¹¹⁶, transcription factor B-Myb¹¹⁷, FLNA (actin cross-linker filamin A)¹¹⁸, CEP68 (centrosomal protein 68)¹¹⁹, CERKL (ceramide kinase like protein)¹²⁰, hsRPB7 (human RNA polymerase II seventh subunit)¹²¹ and EHMT2 (euchromatic histone-lysine methyltransferase 2)¹²². Further research is needed to elucidate the precise conribution of these potential pVHL substrates and their role in tumorigenesis elicited by VHL loss.

In addition to the canonical role of pVHL as an important component of an E3 ubiquitin ligase, emerging research suggests that pVHL may function as an adaptor in other contexts. For example, Akt1 appears to be hydroxylated on Pro125 and Pro313 which are recognized by pVHL which in turn recruited phosphatase 2A (PP2A) promoting Akt1 dephosphorylation¹²³. A similar mechanism was observed with TANK Binding Kinase 1 (TBK1). TBK1 hydroxylation on Pro48 by EglN1 led to pVHL binding and recruitment of phosphatase PPM1B, ultimately dampening TBK1 activity¹²⁴. pVHL was also reported to associate with Card9 promoting its phosphorylation by CK2, which led to decreased NF-κB activity¹¹³. Finally, pVHL could bind BIM-EL, which inhibited BIM-EL phosphorylation by ERK on Ser69, therefore protecting it from proteasomal degradation¹²⁵. Thus, the function of pVHL is likely multi-faceted, and remains to be fully characterized ccRCC.

Effector Agnostic Approaches to pVHL Targeting

Thus, arguments can be made to target VHL loss in ccRCC beyond HIF-2. One strategy to exploit loss of function mutations is to leverage “synthetic lethality”, a scenario in which loss of a gene renders a second gene essential¹²⁶. Several groups have developed high-throughput screening platforms of chemical libraries capable of identifying compounds which exhibit selective killing of VHL-deficient cells. For example, initial reports identified the autophagy modulator STF-62247, and chemical inhibitors of GLUT-1 as selective inhibitors of VHL deficient RCC^127,128. Another study revealed that inhibitors of ROCK1 were toxic, another HIF target gene, were toxic to ccRCC with VHL loss¹²⁹. More recently, a report from the same group showed that inhibition of the Mevalonate pathway had anti-tumor effects in VHL deficient ccRCC cell-lines in a HIF-independent manner¹³⁰. In a high-throughput screen of ~13,000 small molecules, we identified homoharringtonine (HHT), an FDA-approved drug for chronic myeloid leukemia, as a potential VHL synthetic lethal compound. HHT was further shown to be efficacious in ~30% of tested tumorgrafts¹³¹. Finally, TBK1 was also found to be a synthetic lethal partner with VHL loss in ccRCC. Both sgRNA/shRNA mediated depletion and pharmacological inhibition of TBK1 preferentially killed VHL-deficient ccRCC cells¹²⁴. While synthetic lethal screens of chemical libraries have the advantage of producing hit compounds which can be refined via medicinal chemistry approaches, they target a small fraction of the proteome. Thus, these studies can be complemented by high throughput screens of short hairpin RNA (shRNA) libraries and more recently, small guide RNA (sgRNA)/CRISPR Cas-9.

One initial report utilizing a shRNA library directed against 88 kinases in VHL-deficient RCC cell lines identified CDK6, MET, and MAP2K1 as potential targets¹³². A follow up effort utilizing an expanded shRNA library targeting ~1,000 genes identified EZH1 as synthetically lethal with VHL¹³³. More recently, both in human and Drosophila, Nicholson and colleagues demonstrated synthetic lethality between CDK4/6 and pVHL¹³⁴. A genome-wide CRISPR Cas-9 screen identified the selenocysteine biosynthetic pathway as synthetic lethal with VHL^135,136. The extent to which these findings are influenced by other mutations in the cell line models is unknown. Synthetic lethality is highly context specific, and future efforts may leverage our greater understanding of ccRCC evolutionary pathways.

Conclusion

Our understanding of ccRCC tumorigenesis has improved substantially since the discovery of the VHL gene. Novel sequencing strategies and insightful experimental designs have refined our understanding of ccRCC biology. New agents targeting the “undruggable” HIF transcription factor have shown promise in the clinic, and strategies to exploit synthetic lethality with VHL loss have been identified. These exciting developments have the potential to build upon a decade which has seen a flurry of new treatments and improved clinical outcomes for ccRCC patients.