Proteostasis Modulation in Germline Missense von Hippel Lindau Disease

von Hippel-Lindau disease (VHL) is an autosomal dominant neoplasia syndrome (incidence of 1/40,000 live births) caused by germline VHL mutations on 3p25-26 (1, 2). Patients with VHL develop tumors of the central nervous system (CNS), kidneys, adrenal glands, and pancreas. VHL-associated CNS tumors include hemangioblastomas (HB) and endolymphatic sac tumors (3). HBs are the most common VHL-associated tumor and surgical resection remains the definitive treatment of choice for symptomatic lesions (3, 4). Because of the multiplicity of HBs, patients with VHL often undergo multiple CNS surgeries that are associated with morbidity and mortality (5). Preclinical (6, 7) and clinical studies (8, 9) have demonstrated the role of beta-adrenergic receptor antagonist propranolol as a tumoristatic agent for HBs. Propranolol is now approved for the treatment of VHL disease in Europe (10). Recently, hypoxia-inducible factor-2 alpha (HIF2α) inhibitor belzutifan (11) was approved for the treatment of VHL disease in the United States (11). However, clinical trial data and early reports confirm that continued growth may occur in some VHL tumors despite best medical treatment. Consequently, there is a critical need for complementary nonoperative treatments for VHL-associated HBs.

VHL protein (pVHL) is an E3-ligase that binds and degrades HIF1α and HIF2α through the ubiquitin proteasome pathway (12). LOH at the VHL locus (13) and loss of pVHL function causes amplified HIF signaling, driving oncogenic cell proliferation/survival, metabolism, motility, and angiogenesis (14). Missense mutations are the most common germline VHL mutation (up to 40% of patients with VHL; refs. 15, 16). While missense mutant pVHL retains intrinsic E3-ligase function, it undergoes proteasomal degradation after translation resulting in loss of function and tumor formation (17). Proteostasis modulation has been shown in the laboratory as a powerful technique to refunctionalize missense mutated proteins in human disease (18, 19). We and others have shown that proteostasis modulation can rescue missense mutant pVHL from degradation and arrest VHL-associated tumor initiation and/or growth (17, 20). However, despite the strong theoretical underpinnings of this mechanism (21), clinical applications for proteostasis modulation have been limited to lysosomal storage disorders (22).

On the basis of our prior findings of cytosolic rescue of missense mutant pVHL by an orally available histone deacetylase inhibitor (HDACi) vorinostat (17), we conducted a window-of-opportunity study in patients with germline missense VHL. Patients with germline missense VHL and symptomatic HBs were treated with oral vorinostat prior to planned surgery and tumor tissue analyzed. Targeted gene expression, protein, and single-cell transcriptomic comparisons with untreated tumors from the same patients confirmed the rescue of missense mutated pVHL in CNS HB. We further validated this treatment paradigm by investigating the mechanistic pathways with in vitro recapitulation of patient-specific VHL mutations.

We designed a phase 0 open-label clinical trial (clinicaltrials.gov identifier NCT02108002) with a priori power analysis suggesting a minimum 6 patients to detect a difference in tumor protein (Protocol). We included 7 adult patients (age 18 years or greater) with known VHL disease caused by a missense mutation (Supplementary Table S1). The patients were screened at the NIH Clinical Center, a natural history of VHL clinical trial (NCT00005902). These patients demonstrated clinical progression of a CNS HB requiring surgical intervention (eligibility criteria). The surgical interventions were conducted through a separate clinical trial (NCT00060541). We excluded patients with a prior history of treatment with vorinostat (or similar agents), severe medical illnesses, concurrent unrelated cancer, and pregnancy/breastfeeding (Supplementary Fig. S1). Patients were administered 400 mg of vorinostat orally daily for the 7 days immediately before resection (on study day 8) in an inpatient setting. A 7-day treatment course was decided following consultations with the Institutional Review Board (IRB), based on the need for early surgical intervention in patients with symptomatic HBs, and based on established precedents for “window-of-opportunity” trials (23, 24). Adverse events were monitored and recorded per Common Terminology Criteria for Adverse Events (v4.0) grading. HB tissue from enrolled patients obtained from prior surgical resection without vorinostat treatment was used for comparative analysis (April 2014 through September 2018). One patient received stereotactic radiosurgery (SRS) prior to resection of HB. Specifically, a single fraction of 25 Gy was administered approximately 1 year before surgical excision of vorinostat-treated HB. HB tissue from 5 additional patients (not treated with vorinostat) with VHL germline missense mutations that underwent resection at the NIH was also used for comparative analysis. Clinical, imaging, and laboratory assessment was performed.

We analyzed tumor tissue for changes in pVHL (primary outcome) and downstream signaling (secondary outcome) compared with untreated HBs. Formalin-fixed paraffin-embedded (FFPE) sections were used to assess pVHL levels in tumor samples and stained using polyclonal rabbit anti-human VHL antibody (1:300; GTX101087, GeneTex) and 3,3-diaminobenzene (DAB) chromogen. pVHL expression was quantified using mean pVHL stain saturation in ImageJ software (NIH, Bethesda, MD). Stromal cells and vessels wall were sampled for mean intensity of pVHL stain (five sections each). Mature capillary network pVHL was used to normalize stromal tumor pVHL. Light microscopy pVHL expression was quantified using ImageJ software (25).

Multiplex fluorescence IHC was performed on 5-μm-thick paraffin sections of tumor tissue samples using select antibody panels targeting relevant biomarkers of pituitary adenoma signaling pathways. Briefly, sections were first deparaffinized and treated using a standard antigen unmasking step in 10 mmol/L Tris/ethylenediaminetertaacetic acid buffer pH 9.0. Sections were then blocked with Human BD Fc Blocking solution (BD Biosciences) and treated with True Black Reagent (Biotium) to quench intrinsic tissue autofluorescence. Next, sections were immunoreacted for 1 hour at room temperature using 1 μg/mL cocktail mixture of immunocompatible antibodies targeting HIF1α (Thermo Fisher Scientific), HIF2α (Santa Cruz Biotechnology), pVHL (Thermo Fisher Scientific), VEGF-A (R&D), GLUT-1 (LSBio), Hsp90 (Enzo Life Sciences) which were either directly conjugated or indirectly labeled using secondary antibodies, using the following spectrally compatible fluorophores: Alexa Fluor 546, Alexa Fluor 594, Alexa Fluor 647, PerCP, Pacific Orange, and IRDye 800CW. After washing off excess antibodies, sections were counterstained using 1 μg/mL 4′,6-diamidino-2-phenylindole (DAPI; Thermo Fisher Scientific) for visualization of cell nuclei. Slides were then cover slipped using Immu-Mount medium (Thermo Fisher Scientific) and imaged using an Axio Imager.Z2 slide scanning epifluorescence microscope (Zeiss) equipped with a 20X/0.8 Plan-Apochromat (Phase-II) non-immersion objective (Zeiss), a high-resolution ORCA-Flash4.0 sCMOS digital camera (Hamamatsu), a 200W X-Cite 200DC broad band lamp source (Excelitas Technologies), and seven filter sets customized to detect the aforementioned fluorophores with minimal spectral cross-talk (Semrock). Image tiles (600×600 μm viewing area) were individually captured at 0.325 μm/pixel spatial resolution, and tiles seamlessly stitched into whole specimen images using the ZEN 2 image acquisition and analysis software program (Zeiss). Pseudocolored stitched images were then exported to Adobe Photoshop and overlaid as individual layers to create multicolored merged composites.

For protein extraction, cell pellet or frozen HB tissue (mechanically disrupted) were collected in RIPA buffer containing a protease inhibitor cocktail. After incubation on ice for 15 minutes, samples were centrifuged at 12,000 rpm for 15 minutes at 4°C. The supernatant was considered as cell lysate and protein concentration was estimated by Pierce BCA Protein Assay Kit (Thermo Fisher Scientific). Proteins were electrophoretically separated on 4%–12% NuPAGE Bis-Tris gels (Invitrogen) and subsequently electroblotted onto polyvinylidene difluoride membranes with the Trans-Blot Transfer Turbo System (Bio-Rad). Membranes were blocked for 1 hour in 5% nonfat dry milk in TBS with tween 20 and incubated overnight with primary antibodies; the primary antibodies targeting pVHL, Hif1α, Hif2α, VEGF-A, acetylated Hsp90, GAPDH, and vinculin were purchased from Santa Cruz Biotechnology; Hsp90 and GLUT1 from Cell Signaling Technology; Flag-tagged antibody from Origene. Finally, the membrane was incubated with horseradish peroxidase–conjugated secondary antibodies (mouse or rabbit; Amersham) for 1 hour at room temperature and Super Signal West Femto (Thermo Fisher Scientific) was used to induce immunoreactive signal, which was detected by ChemiDoc MP Imaging System (Bio-Rad). Band intensity was quantified by densitometry via Image Lab Software (Bio-Rad) and normalized to Vinculin or GAPDH as endogenous controls.

Total RNA was extracted from frozen human HB samples and cultured cells using RNeasy Mini Kit (Qiagen) and reverse transcribed to cDNA with Superscript III qRT-PCR Supermix (Life Technologies). qPCR was performed using target-specific primers of the selected genes. SYBR Select Master Mix on the Illumina Eco Real-Time PCR System (Illumina) was employed. The oligonucleotides used for targeted genes were: VHL (For: 5′-AGAATTACAGGAGACTGGAC-3′, Rev: 5′- AAAGCTGAGATGAAACAGTG-3′); HIF1Α (For: 5′-GCTGGCCCCAGCCGCTGGAG- 3′ Rev: 5′-GAGTGCAGGGTCAGCACTAC-3′); EPAS1 (For: 5′-CTGTGTCTGAGAAGAGTAACTTCC

Rev: 5′-TTGCCATAGGCTGAGGACTCCT); SCLA21 (For: 5′-GCAACGGCTTAGACTTCGAC-3′ Rev: 5′-CCAAATCGGCATCATTCTCAT-3′), VEGFA (For: 5′-ACCCATGGCAGAAGGAG GAG -3′, Rev: 5′-ACGCGAGTCTGTGTTTTTGC-3′). Relative gene expression values were calculated using the comparative CT method with beta-actin as an internal control.

The sequence encoding the pathogenic VHL variants were cloned into a pLenti-C-Myc-DDK-IRES-Puro plasmid (Origene) between the MluI/AsisI sites. Recombinant VHL lentiviral constructs with point mutations in p.L89P, p.W117C, p.W117R, p.Q164R, p.R167Q, p.L169P, and p.L188P were generated using a site-directed mutagenesis kit (Agilent). A wild-type (WT) construct was also generated and introduced into 786-O cell cultures. The sequences of VHL gene were verified by analyzing the entire coding regions through Sanger sequencing.

The commercially available, VHL-deficient cell line 786-O (ATCC) was cultured in DMEM with 10% FBS and following the standard protocol, cells were transfected with WT and mutant pVHL lentivectors using the JetOPTIMUS transfection reagent (Polyplus).

786-O cells were plated on chambered slides and transiently transfected with WT or mutant pVHL vectors. After 48 hours of transfection, cells were washed with ice-cold PBS and fixed with 4% paraformaldehyde (Sigma-Aldrich) for 10 minutes at room temperature. Cell permeabilization was done using 0.1% Triton X-100 in PBS for 10 minutes at room temperature. In situ proximity ligation assay (PLA) was performed with Duolink In-Situ Detection Reagent RED (Sigma-Aldrich) as per manufacturer's recommendations.

For the FFPE tissue sections, deparaffinized step was used before permeabilization. During this step, we incubated and washed tissue sections with Xylene and graded ethanol (100%–70%) each for 10 minutes and finally in water for 1 minute. Rest of the procedures for PLA was same as mentioned for transiently transfected fixed cells. Coverslips were mounted on glass slides with Duolink In Situ Mounting Medium with DAPI (Sigma-Aldrich) for subsequent observation at fluorescence microscope (Zeiss LSM 880 AiryScan Confocal Microscope).

To better quantify the degree of cellular colocalization signal of targeted molecules, we employed QuPath (version 0.3.1), a publicly available software platform for analysis of histopathological tissue sections. The utilized pipeline was adopted as described previously (26). Briefly, image files (.czi) were loaded onto QuPath platform and an optimized cell detection algorithm was then applied on the blue channel (DAPI) to identify cells based on nuclear size and fluorescence intensity. Output file was subsequently exported to ImageJ suite, where the optimized “FindMaxima” algorithm was used to detect dimer signals in the red channel (Alexa Fluor 594-positive foci).

Protein stability measurement was done via pulse-chase analysis. Briefly, 786-O cells were transfected with lentiviral constructs of recombinant pVHL for 48 hours as discussed previously. Cells were next incubated with cycloheximide (CHX; 20 mg/mL) or CHX (20 mg/mL) with 3 μmol/L vorinostat at 37°C in a CO₂ incubator; protein synthesis was chased hourly up to 4 hours. At desired timepoint maintaining cold condition, cell plates were washed once with cold PBS and RIPA lysis buffer was added to collect lysates for further processing.

Optimal cutting temperature compound (OCT)-embedded HB samples were removed from −80°C storage and placed in a cryostat. For maximal removal of surrounding OCT medium, a clean razor blade was used to cut around the tissue. Once a sample was stripped to satisfaction, it was placed in a 1.5 mL cryotube in dry ice before being moved to −80°C storage. In between samples, the blade was thoroughly rinsed and wiped clean using 70% ethanol.

Nuclei were isolated from the frozen tissues using an adaptation of a previously established protocol (27). Briefly, each sample was placed on a clean petri dish and washed with 800 μL of Detergent-Lysis Buffer (0.1% Triton-X). A no. 22 blade was used to dissociate tissue and then transferred into a 7 mL Dounce Homogenizer. Tissue was homogenized with eight strokes of the loose Pestle, followed by 12 strokes of the tight Pestle. The crude nuclei homogenate was then passed through a 40 μm strainer and spun down at 3,200 × g for 5 minutes at 4°C in a swing-bucket centrifuge. The nuclei pellet was resuspended in 1 mL of low-sucrose buffer and then gently layered above 4 mL of high sucrose buffer without disrupting the density gradient. The gradient was then spun down at 3,200 × g for 20 minutes at 4°C, and the ensuing pellet was resuspended in 25–150 μL of Nuclei Resuspension Buffer (10X Genomics). An aliquot was taken to be stained with Acridine Orange dye (1:10 dilution) and counted/imaged using a Luna-FL Dual-Fluorescence Cell Counter (Logos Biosystems). Following manufacturer's recommendations, the nuclei suspensions were next processed via the Single Cell Multiome ATAC+Gene Expression Assay (10X Genomics). Nuclei were loaded onto a Next GEM Chip J (10X Genomics) targeting a yield of 3,000 nuclei.

For all single-nuclei experiments, the Single Cell Multiome ATAC+Gene Expression kit (10X Genomics) was used according to the manufacturer's protocol recommendations. Library preparation was performed according to the manufacturer's instructions. Libraries were pooled and sequenced on the NextSeq 500 System (Illumina) using High-Output Reagent Kit v2.5 (150 cycles; Illumina).

Sequenced reads from human HBs were demultiplexed and aligned to the GRCh38 reference genome (28) using CellRanger ARC (v2.0.1; 10X Genomics) mkfastq function with default settings and subsequent counts were generated using CellRanger ARC count function. Datasets were analyzed using the NIH BioWulf High Performance Computing Platform.

Individual datasets were integrated using the CCA integration method as described by Hao and colleagues (29). In summary, the Seurat v4 R-package integration workflow was utilized to remove batch effects while preserving biological variation in a shared space. We subset the nuclei by removing those with fewer than 300 or greater than 6,000 unique molecular identifiers per cell, as well as those with high expression of mitochondrial genes (>10%). Cell-cycle scoring was performed using Seurat workflow (regevlab list). Integration anchors were found using “FindIntegationAnchors()” with the number of dimensions set to 30, and each sample's gene expression matrix was integrated into one object via “IntegrateData().” The integrated object was then scaled, run through a principal component analysis, and clustering was performed through Uniform Manifold Approximation and Projection (UMAP; ref. 30). Clusters were annotated to one of several cell types [macrophages, T cells, natural killer (NK) cells, B cells, oligodendrocytes, pericytes, HB-stromal, HB-endothelial, and HB-neuronal] through per-cluster detection of several established marker genes. Differentially expressed genes per cluster were identified using the FindMarkers() function using default settings (log₂-fold change >0.25, and gene must be expressed in >25% of cells in that cluster).

We identified copy-number variants (CNV) in our single-nucleus RNA sequencing (snRNA-seq) datasets using the R-package InferCNV (31). The package was run according to the developer's recommended workflow, and cells that were presumed to be non-neoplastic were used as the “reference” cells (e.g., T cells, NK cells, macrophages, oligodendrocytes, B cells). The CONICSmat package (32) was used to produce UMAP plots that describe the inferred chromosomal expression.

Clusters were annotated by cell type both through manual investigation of cell-type specific markers and cross-referencing to previously described single-cell datasets using the package “SingleR.” (33) SingleR was used to validate the common immune and circulating blood cell types that were manually annotated with suggested labels. The HB cells could not be properly called by the SingleR algorithm included cells/clusters that we named neoplastic stromal cells. Bioinformatics pipelines and data processing was performed using the High Performance Computing platform (BioWulf) at the NIH.

Canonical pathway exploration was performed with Ingenuity Pathway Analysis (IPA; Qiagen) software version 01-20-04 on differentially expressed genes within the stromal compartment of HB patient samples, as identified following single-cell sequencing.

For comparative analysis of generated data, unpaired and paired Student t test or two-way ANOVA, where appropriate, were performed using Prism v9.0 software (GraphPad). A P value <0.05 was used to determine statistical significance.

The Combined Neuroscience IRB of the NIH approved the clinical trial. Written informed consent was received prior to participation.

The human data generated in this study are not publicly available due to patient privacy requirements but are available upon reasonable request from the corresponding author. Other data generated in this study are available within the article and its Supplemental Data.

We recruited 7 patients (1 female, 6 male) with germline missense VHL (Supplementary Fig. S1) that completed the 7-day course of oral vorinostat and underwent surgical HB resection (Fig. 1A and B; Table 1). Mean age at enrollment was 46.0 ± 14.5 years (range, 25–61 years). No patients developed dose-limiting toxicity (Supplementary Table S2). Two patients developed transient acute renal injury (Table 1) that was treated with intravenous fluids. One patient had insufficient tissue for analysis. Previously excised, untreated HBs from study patients and HBs from 5 additional patients with germline missense VHL were used for comparison with the vorinostat-treated group (Supplementary Table S3). Radiographically, we did not detect a change in the volume of the symptomatic HB or the associated brain edema and/or cyst within 4 days of drug administration (Supplementary Fig. S2A). We found no additional adverse events over the 42.0 ± 17.5 months (range, 17–64 months) follow-up (Supplementary Table S2). Taken together, these findings suggest that short-term vorinostat is tolerated without serious adverse events by patients with germline missense VHL.

We first asked whether vorinostat crossed the blood–tumor barrier. We found elevated acetylated α-tubulin, a cytosolic HDAC6 substrate (ref. 35; Supplementary Fig. S2B and S2C) with vorinostat treatment compared with paired, untreated HBs from the same patients. pVHL was stable in the mature HB vessel wall (without VHL LOH; ref. 26) compared with neoplastic stromal tumor cells with IHC assays (Fig. 1C). We found no difference in pVHL expression in the mature vessel wall with vorinostat treatment (absorbance value 0.3 ± 0.03 vs. 0.3 ± 0.02; P = 0.1; Fig. 1D). In contrast, we found significantly elevated pVHL expression in neoplastic stromal cells in vorinostat treated (absorbance values 0.2 ± 0.02 vs. 0.05 ± 0.003; P = 0.04; Fig. 1D and E). After normalizing to vessel wall pVHL, we found that pVHL was significantly elevated within neoplastic stromal cells in treated HBs compared with paired, untreated HBs (absorbance values 0.4 ± 0.06 vs. 0.2 ± 0.01; P = 0.0009; Fig. 1E). Tissue subtypes (neoplastic stromal cell vs. mature capillary walls; 81%; P < 0.0001) and vorinostat treatment (6%; P = 0.008) accounted for the majority of variability in pVHL expression within tumors (two-way ANOVA). We also found elevated pVHL by Western immunoblotting in resected vorinostat-treated HBs compared with paired, archival tissue (Supplementary Fig. S2D) except in one tumor that had received presurgical radiotherapy (Pt 4; Supplementary Table S2). Using multiplexed IHC on paired HBs (Pt1, Pt2, Pt4, and Pt6), we confirmed that vorinostat induced potent expression of pVHL with concomitant suppression of downstream effectors including HIF1α, HIF2α, GLUT-1, and VEGF-A (Fig. 1E; Supplementary Fig. S3; Supplementary Table S4). These findings indicated that short-term vorinostat treatment rescued pVHL in patients with missense VHL disease.

Consistent with prior in vitro data (17), no apparent change in VHL mRNA expression was noted in surgical tumors following vorinostat treatment (ΔC_t; P = 0.74; Fig. 2A). pVHL reduction leads to transcriptional upregulation of HIF target genes (including VEGFA) via HIF activation (35). Correspondingly, we found unchanged HIF gene expression (P = 0.05) but suppressed downstream VEGFA expression (P = 0.02) with vorinostat treatment (Fig. 2B and C). The vorinostat effect on pVHL rescue was not detected in patient tumor samples with germline partially deleted VHL (Supplementary Fig. S2E; Supplementary Table S3), suggesting a missense mutation–specific effect of vorinostat. The location of missense mutations has functional and phenotypic consequences (36) in VHL disease. In our prior report (17), we compared the effects of vorinostat on mutations at p.Ser68Trp and p.Tyr112Asn. Here, we explored the location-specific effects of VHL missense mutations in our patient cohort (Fig. 2D). To mechanistically explore the effects of location of missense mutations, we first introduced WT VHL in a VHL^−/− 786-O cell line and found moderate elevation of pVHL with vorinostat treatment (Supplementary Fig. S2F). We then tested a cognate model using patient-derived VHL renal cell cancer cells (p.Trp117Arg) and compared the effects of vorinostat treatment on VHL^−/− 786-O cells induced with flagged, mutated VHL lentiviral constructs. Vorinostat triggered a robust elevation of flagged pVHL with reductions in HIF1α, HIF2α, VEGF-A, and GLUT-1 expression (Supplementary Fig. S4A) in both models. Similarly, qRT-PCR revealed comparable downregulation of downstream genes including VEGFA and SCLC2A1 (Supplementary Fig. S4B and S4C). We then asked whether missense mutated pVHL interacted with HIF1α, a key marker of pVHL function. We found detectable pVHL–HIF1α interactions in 786O cells transfected with WT and missense mutated VHL (Supplementary Fig. S4D). We found an increase in PLA signal with vorinostat treatment in missense VHL, but not in WT VHL transfections. These data supported the use of patient-specific missense constructs to explore the effects of vorinostat. We then introduced lentiviral constructs with patient-specific VHL mutations (p.Leu89Pro, p.Trp117Cys, p.Gln164Arg, p.Arg167Gln, and p.Leu188Pro) into 786-O cells. We found that vorinostat (3 mmol/L, 24 hours) reduced HDAC activity (Supplementary Fig. S4E) and rescued pVHL expression across the spectrum of patient mutations. We found consistent suppression of VEGF-A, and GLUT-1 protein expression (Fig. 2E and F). Half-life measurements following cycloheximide treatment confirmed the broad efficacy of vorinostat in rescuing missense mutated pVHL across the patient mutation spectrum (P = 0.0008, two-way ANOVA; Fig. 2G and H). As expected, we found inconsistent transcriptional effects on VHL and HIF gene expression, but a robust suppression of downstream VEGFA expression (P = 0.0001; Figs. 2I–M) in vorinostat treated conditions. These results suggested a broad, location unrestricted effect of vorinostat in rescuing missense mutated pVHL.

Quality control of defective pVHL is maintained via Hsp90 recruitment and proteasomal degradation (37). We postulated decreased pVHL–Hsp90 interaction (38) across the spectrum of missense mutations underlying pVHL stabilization upon vorinostat exposure (Fig. 3A). PLA confirmed a robust reduction in pVHL–Hsp90 interactions in available tumor samples with 7-day vorinostat treatment compared with paired, untreated controls (across a mean of 764.5 cells detected per tumor sample; P = 0.02; Fig. 3B and C). We then expanded PLA analysis to account for all patient-specific mutations using transfected 786-O cells. We found a consistent reduction in the PLA signal in the tested pairs (64.4 mean cell detections per construct/condition; mean of differences = −3.97, SD = 3.53, t[5] = 2.75, P = 0.04; Fig. 3D–F).

To deconvolute the cell-specific effects of vorinostat treatment in HIF signaling, we analyzed single-nucleus transcriptomic profiles of paired (prior untreated vs. vorinostat treated) HBs from 2 patients (Pt 4 and Pt 6; 10X Genomics; Table 1). Fresh-frozen tumors were processed and nuclei isolated to generate transcriptomic data from 1,690 nuclei. Initial inspection of UMAP embedding (30) following batch correction (29) revealed no major clustering effects of patients or treatment conditions (Fig. 4A; refs. 27, 29). We then employed reference-based single-cell (33) annotation and cluster-level annotation using canonical cell type marker genes for each sample separately (Supplementary Fig S5A and S5B). We used consensus canonical cell class overlaps (cell vs. cluster) to transcriptionally identify immune cell populations including macrophages, as well as endothelial cells and pericytes (Fig. 4B; Supplementary Fig. S5C and S5D; Supplementary Table S5; ref. 23). We found that exposure to radiation also led to elevated pVHL (Supplementary Fig. S2D, Pt#4) but unchanged UMAP clustering (Supplementary Fig. S6A) and a distinct pattern of gene expression compared with untreated HBs (Supplementary Fig. S6B; Supplementary Table S6). We confirmed that VEGFA expression (21) was restricted to neoplastic stromal cells (Fig. 4C), whereas HIF1A, EPAS1 (HIF2A), and SLC2A1 (GLUT1) were widely expressed across all cell classes (Fig. 4D; Supplementary Fig. S6C and S6D). We then analyzed the effects of vorinostat treatment on gene expression in canonical and noncanonical HIF signaling pathways [gene set enrichment analysis (GSEA), v7.5.1; Broad Institute]. We found robust downregulation of downstream HIF targets including VEGFA, hexokinase 2 (HK2) and insulin-like growth factor binding protein 3 (IGFBP3) in neoplastic stromal cells following vorinostat treatment (Fig. 4E). The effects of vorinostat on other cell classes (Fig. 4E; Supplementary Table S7) and the effects of radiation on all cell classes (Supplementary Fig. S6E) were variable and less robust. We identified the “second-hit” (39) at the VHL locus (chromosome 3p) exclusively in neoplastic stromal and neuronal-like cells across all samples with inferred genomic CNV analysis (Fig. 4F; ref. 31). We then isolated the transcriptional signature of the neoplastic stromal cells and examined the effect of 7-day vorinostat treatment with pathway enrichment analysis (IPA; Qiagen). Differentially expressed gene sets (132 genes, 42 downregulated and 90 upregulated; Supplementary Table S6) with statistical significance annotated to the neoplastic stromal cell cluster were studied. Network-level exploration indicated potent repression (P = 0.03) of the nuclear HIF1α transduction pathway (Fig. 4G). Induction of the Integrin (P = 0.01), G-Protein Coupled Receptor (P = 0.02), and Autophagy (P = 0.03) signaling pathways was also observed, while significant inhibition of the IL8 (P = 0.001), Production of Nitric Oxide and Reactive Oxygen Species in Macrophages (P = 0.005), and Apelin Endothelial Signaling (0.009) networks was noted. These findings suggest targeted effects of vorinostat on neoplastic stromal cells within germline missense VHL-related HBs.

Multiple CNS HBs are found in over 80% of patients with VHL and constitute the most frequent cause of disease-associated loss of function and death (40). Because these tumors exhibit an unpredictable and saltatory growth pattern, optimal management includes observation until symptom onset, at which point surgical resection becomes the standard treatment. Because of the protean nature and multiplicity of these tumors in VHL, patients frequently require resection of multiple HBs over their lifetime with associated morbidity and mortality (41). Recently, belzutifan, a HIF2α inhibitor was approved for use as an oral agent in patients with VHL-related HBs and renal cell carcinomas (11). However, the study data demonstrated some CNS HBs remained stable or continued to grow during therapy. Therefore, effective, complementary nonsurgical therapies are needed (42, 43).

Missense mutant pVHL maintains intrinsic function, but its immediate posttranslational degradation drives the pVHL loss-of-function in VHL-associated tumors through upregulation of HIF target genes (17). HDACi modulation of proteostasis in missense mutant pVHL could provide a new therapeutic option (21, 44). Currently, vorinostat is approved for the treatment of the cutaneous manifestations of cutaneous T-cell lymphoma (45) and leads to lymphoma regression (46). Prolonged use for more than 2 years has been well tolerated with rare adverse events (47). Here, we demonstrated that short-term daily administration of vorinostat is safe in patients with CNS HBs. Vorinostat was bioactive at the tumor level and induced a robust expression of pVHL in the neoplastic stromal cell compartment of HB with subsequent suppression of key HIF pathway genes.

Two patients harbored the hotspot mutation p.R167Q, which has been tested for stability and shown to produce reduced pVHL levels in vitro (20, 48). To characterize pVHL stability based on mutational status and ascertain the effects of vorinostat in broader pVHL variants, we transfected 786-O cells with missense pVHL lentiviral vectors. We found that vorinostat led to a consistent and robust elevation of pVHL and subsequent reduction of key protumorigenic factors across the spectrum of missense mutations, suggesting that the pVHL complex retains sufficient ligase activity to prevent tumorigenesis. Mechanistically, vorinostat modulates the proteasomal pathway by limiting Hsp90-pVHL assembly and blocking degradation of mutant pVHL as demonstrated previously (17). Similarly, we found that short-term oral vorinostat had a potent biologic effect on missense mutant pVHL levels ex vivo. Within surgically acquired HBs, vorinostat potently disrupted Hsp90 interaction with pVHL in situ and this was recapitulated in vitro.

HDACis may suppress hypoxia signaling via additional mechanisms (49). Because HDACis have wide pleiotropic effects on the cellular transcriptome (50, 51), we characterized the transcriptional profile of HB following vorinostat treatment at the single-nucleus resolution in human surgical tumor samples. snRNA-seq identified clustering by tumor-resident cell types with minor distribution of resident immune cell types. We identified the stromal tumor cells via transcriptional profiling and consensus cell annotation. Stromal cells, the putative protean tumor drivers in HB (52), demonstrated cell-specific VEGFA expression. In VHL, the “second hit” discovered in tumors is often a highly variably sized deletion of the WT allele (53). Using inferred CNV analysis, we corroborated a loss of genomic material within regions of chromosome 3p (second hit) that was restricted to the neoplastic stromal cells. We then found that vorinostat treatment induced a potent repression of the nuclear HIF1α signaling pathway in neoplastic stromal cells using gene ontology analysis.

Here, we demonstrated that short-term daily administration of vorinostat is safe in patients with missense VHL with CNS HBs. Chronic (or multiple intermittent) use of vorinostat may be needed for management in patients with VHL as increased levels of pVHL may prevent tumor progression and initiation. However, treatment may only be necessary during times of maximal tumor initiation and/or progression, as determined by serial surveillance. Future studies to define the long-term effect on VHL tumor control (i.e., HBs, renal cell carcinoma, pheochromocytomas, and pancreatic neuroendocrine tumors) will be important as the same biologic mechanisms underlie VHL-associated tumor initiation and progression.

Oral vorinostat treatment in patients with germline missense VHL mutations has a potent biologic effect that warrants further clinical study. Our results provide the first biologic evidence to support the use of proteostasis modulation for the clinical treatment of syndromic solid tumors mediated by protein misfolding.

R.R. Lonser reports other support from NIH during the conduct of the study as well as personal fees from Merck, Biogen, and Uniqure outside the submitted work. No disclosures were reported by the other authors.

P. Chittiboina: Conceptualization, resources, formal analysis, supervision, methodology, writing–original draft, project administration, writing–review and editing. D. Mandal: Formal analysis, writing–review and editing. A. Bugarini: Software, formal analysis, writing–original draft. D.T. Asuzu: Data curation, formal analysis, visualization. D. Mullaney: Visualization. P. Mastorakos: Data curation, investigation, methodology. S. Stoica: Formal analysis, investigation. R. Alvarez: Formal analysis, investigation. G. Scott: Resources, data curation. D. Maric: Visualization. A. Elkahloun: Formal analysis. Z. Zhuang: Conceptualization. E.Y. Chew: Conceptualization, resources. C. Yang: Conceptualization, formal analysis, methodology. M. Linehan: Conceptualization, resources. R.R. Lonser: Conceptualization, supervision.

This study was funded by the Intramural Research Programs of the National Institute of Neurological Diseases and Stroke, NCI, and the National Eye Institute of the NIH, Bethesda, MD. The study was supported by grant ZIA NS003053-16 to P. Chittiboina.

The authors are clinicians committed to taking care of patients with VHL disease. The authors thank the patients for volunteering their trust and time to take part in this study.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.