PBRM1 Inactivation Promotes Upregulation of Human Endogenous Retroviruses in a HIF-Dependent Manner

Renal cell carcinoma is a common cancer with 76,080 new cases diagnosed and 13,780 deaths in the United States in 2021 (1). Clear cell renal cell carcinoma (ccRCC) is the most common histologic subtype and has been genomically characterized by multiple groups, including The Cancer Genome Atlas (TCGA; ref. 2). ccRCC is associated with several recurrent genomic alterations, including a high rate of inactivation of the von Hippel-Lindau tumor suppressor gene, VHL, as well as other genes located on chromosome 3p: PBRM1, SETD2, and BAP1 (2). PBRM1, the second-most frequently mutated gene in ccRCC (2), encodes a component of the polybromo BRG1-associated factor (PBAF) SWI/SNF chromatin remodeling complex, which plays key roles in nucleosome positioning and transcriptional regulation (3, 4).

It has been variably reported that PBRM1 mutations are associated with improved clinical outcome and response to immune checkpoint blockade (ICB) in patients with metastatic ccRCC. Whereas original observations and follow-up studies demonstrate a correlation between PBRM1 loss-of-function (LOF) mutations and second-line (VEGF TKI refractory), single-agent ICB response (5–7), other studies examining first-line, treatment-naïve cohorts treated with dual VEGF and ICB did not see this association (8, 9). In addition, evaluation of real-world datasets shows that although PBRM1 mutations associate with ICB response in ccRCC (10), this does not hold up across other cancer types (11). Therefore, a better understanding of the mechanisms underlying the potential association between PBRM1 mutations and ICB response might allow us to decipher these seemingly conflicting data.

Human endogenous retroviruses (hERV) are endogenous viral elements in the human genome derived from retroviruses (12). hERVs make up a substantial portion of the retroelements that comprise up to 8% of the human genome (13). It has been shown that hERV expression products can act as pathogen-associated molecular patterns (PAMP) to trigger innate immune responses (14) or provide antigenic epitopes stimulating adaptive immune responses (15). Multiple studies have identified altered expression of specific hERVs in many cancer types, including melanoma, breast cancer, and renal carcinomas (15–19). Recent studies also associate hERV expression with survival and response to ICB in ccRCC (15, 19). However, none of these studies examined the mechanisms by which hERVs are regulated in ccRCC.

We asked whether hERV expression is associated with ccRCC genomic alterations and found across multiple datasets that ccRCC tumors with PBRM1 mutations have significant alterations in hERV expression. In particular, members of the HERVERI family are upregulated in PBRM1-mutant tumors. In vitro knockdown (KD) of PBRM1 resulted in altered hERV expression in tumor cells, and these hERVs are enriched in HERVERI superfamily members. Moreover, the majority of PBRM1-induced changes in hERV expression are dependent on HIF1α and HIF2α. In aggregate, these results uncover a novel mechanism of hERV regulation in ccRCC and may underlie the clinical observations that PBRM1-mutant ccRCC have an improved prognosis and ICB response.

TCGA KIRC HERV expression matrix was requested from Smith and colleagues (15) Gene Mutation annotation was obtained from FireBrowse (http://firebrowse.org/#). RNA-sequencing (RNA-seq) data and mutation annotation of IMmotion150 cohort was downloaded from (EGAD00001004183) (20). RNA-seq from A704 cells was obtained from Gao and colleagues (21).

The alignment and quantification of HERV expression was processed using the hervQuant workflow (15) from RNA-seq FASTQ files. Raw expression matrices were normalized using RPM and log₂ transformed. hERVs with log₂ expression <0.1 were filtered out as low-expressed hERVs.

Differential expression analysis was performed using R package edgeR (v3.28.1) and limma (v3.42.2). Univariable logistic regression between PBRM1 mutation and hERV expression was performed using R package stats (v3.6.2). Enrichment of hERVs was tested using χ² test. PBRM1-truncated mutations indicate frameshift and nonsense mutations only. Differential hERV expression from UMRC2 cells was controlled for FDR using the Benjamini–Hochberg method.

All cells were short tandem repeat tested within 1 year of their use in these experiments. All cells were tested routinely for Mycoplasma every 2 months while in culture. All cells were cultured in DMEM (Sigma) supplemented with 10% FBS (Gibco) and maintained at 37°C at 5% CO₂. UMRC2 cells were obtained from Sigma Aldrich (catalog no. 08090511) in 2017. 293T cells were a gift from William G. Kaelin Jr. (Dana-Farber Cancer Institute, Harvard Medical School) in 2005. Cells with passage less than 50 passages were utilized for all experiments. PBRM1 KD was performed using an empty pLKO.1 vector (control) or pLKO.1 vector containing shRNA to human PBRM1 (TRCN0000015994, Thermo Fisher Scientific). Sequence CCGGCCGGAGTCTTTGATCTACAAACTCGAGTTTGTAGATCAAAGACTCCGGTTTT pLKO.1 empty or pLKO.1-shPBRM1 lentivirus was generated by transfection of 293T cells with the above plasmids using Fugene6 transfection reagent (Promega, catalog no. E2691). UMUC2 cells were transduced with unconcentrated lentivirual supernatant and selected for infection using 3 μg/mL of puromycin.

The siRNA sequences to HIF1A and HIF2A are as follows. All siRNA were obtained from Ambion.

• siHIF1–1 (Catalog: 4390824) Assay ID #s6539, human HIF1A

• Sense: CCAUAUAGAGAUACUCAAATT

• Antisense: UUUGAGUAUCUCUAUAUGGTG

• siHIF1–2 (Catalog: 4390824) Assay ID #s4699, human HIF1A

• Sense: CAAUAGCCCUGAAGACUAUTT

• Antisense: AUAGUCUUCAGGGCUAUUGGG 1–2

• siHIF2–1 (Catalog: 4390824) Assay ID #s4698, human EPAS1

• Sense: CACCUACUGUGAUGACAGATT

• Antisense: UCUGUCAUCACAGUAGGUGAA

• siHIF2–2 (Catalog: 4390824) Assay ID #s6541, human EPAS1

• Sense: CCUCAGUGUGGGUAUAAGATT

• Antisense: UCUUAUACCCACACUGAGGTT

200 pmol of each siRNA was used for transfected into a 6-cm plate of target cells using Lipofectamine RNAiMAX transfection reagent (catalog no. 13778075).

RNA was extracted using Qiagen RNeasy Plus Mini kit (catalog no. 74134). RNA-seq libraries were prepared using TruSeq Stranded mRNA Library Preparation Kit (Illumina, catalog no. 20020595) according to the manufacturer's protocol. 75b paired-end reads were sequenced on Nextseq 500 (Illumina). Quality control–passed reads were aligned, quantified, and normalized using hervQuant workflow to generate the hERV expression matrix and followed by differential expression analysis.

General linear model using the R stats package was used for all univariable regressions. Differential hERV expression was calculated using the R edgeR and limma packages. Categorical enrichment tests were compared using χ² test.

UMRC2 RNA-seq FASTQ files and hERV expression table are publicly available at GEO (Gene Expression Omnibus) under accession GSE176053.

We first asked whether hERV expression was associated with ccRCC genomic alterations. To this end, we obtained hERV expression from 449 TCGA KIRC samples using the hervQuant analysis workflow (15). After filtering out lowly expressed hERVs, 1,680 hERVs were evaluated for association between their expression and the mutation status of significantly mutated genes in ccRCC as defined by the TCGA KIRC project. Specifically, we performed univariable logistic regression of individual hERV expression by gene mutation status [wild-type (WT), mutated; Fig. 1A]. VHL, PBRM1, and BAP1 demonstrated strong positive association with the expression of a majority of hERVs. The positive association of PBRM1 mutation with hERV expression was maintained even when restricted to PBRM1 mutations predicted to cause LOF, that is, frameshift and nonsense (Fig. 1A). In contrast, mutations in SETD2 and MTOR were negatively correlated with hERV expression. To assess whether tumor purity accounted for the association between gene mutation status and hERV expression, we correlated ESTIMATE scores from TCGA as the measure of stromal content (22) to expression of each individual hERV (Supplementary Fig. S1A) and found that the majority of hERVs had weak correlation to tumor purity. In addition, we found no significant difference of tumor ESTIMATE score between PBRM1-mutated and WT samples (Supplementary Fig. S1B). In aggregate, these results suggest the associations between gene mutation and hERV expression is not confounded by tumor purity.

We next evaluated a second cohort of 262 primary RCC tumors from IMmotion150 (20), which investigated the clinical activity of atezolizumab with and without bevacizumab versus sunitinib in patients with ccRCC (20). RNA-seq data were used to quantify hERV expression by hervQuant (15) as described for the TCGA KIRC samples above. Application of the same filtering and transformation parameters resulted in 1,434 hERVs for downstream analysis. The expressed hERVs detected in the two datasets were similar, with 87% and 92% of the detected hERVs shared in the TCGA and IMmotion150 datasets, respectively (Supplementary Fig. S1C). Of the 9 significantly mutated genes detected in the TCGA study, the IMmotion150 dataset contained mutational analysis of 6 (PBRM1, VHL, SETD2, BAP1, MTOR, TP53). We found that mutations in VHL, PBRM1, and SETD2 were positively associated with hERV expression (Fig. 1A). Therefore, across both datasets, only PBRM1 and VHL demonstrated a consistent positive association between mutation status and hERV expression (Fig. 1B; Supplementary Fig. S1D). These data suggest that both VHL and PBRM1 function to suppress hERV expression.

We then asked whether the association between PBRM1 mutation and hERV expression was tissue-specific. We tested for the association of PBRM1 mutations and hERV expression in non-KIRC TCGA cancers with a greater than 5% PBRM1 alteration frequency and where hERV expression data was available (ref. 15; UCEC, CHOL, SKCM, MESO, and BLCA; Supplementary Fig. S2A). A consistent association between hERV expression and PBRM1 mutation (Supplementary Fig. S2B–S2D) was not identified. These data demonstrate that the regulation of hERV expression by PBRM1 is unique to ccRCC.

Although the taxonomy of hERVs remains to be standardized, a commonly used classification system assigns hERVs to 11 superfamilies (23). We hypothesized that PBRM1 inactivation may activate a specific subset of hERVs. We found that among hERVs that were upregulated in PBRM1-mutant tumors in both the TCGA and IMmotion150 datasets [generalized linear model (GLM) coefficiency>0, P < 0.05], the HERVERI superfamily was significantly enriched (Fig. 2A and B; Supplementary Fig. S3A). This enrichment was also seen when analyzing solely PBRM1 LOF mutations (χ²P = 2.645e-16, Supplementary Fig. S3A) in the TCGA KIRC dataset. The available IMmotion150 dataset regrettably does not include specifics about nucleotide or protein changes.

To validate that PBRM1 was functionally regulating hERV expression in a cell-autonomous manner, we silenced PBRM1 in the UMRC2 RCC cell line (Supplementary Fig. S4A) and assessed changes in hERV expression by RNA-seq. Differential expression analysis demonstrated that PBRM1 KD led to a change in a number of hERVs (log₂-FC>0 and FDR P < 0.05; Fig. 2C) and that the number of upregulated hERVs were significantly enriched in the HERVERI superfamily (Fig. 2D). Similar enrichment of increasing expression of HERVERI superfamily of hERVs was also detected when we analyzed published transcriptome data from isogenic PBRM1-intact and -deficient A704 ccRCC cells (Supplementary Fig. S4B; ref. 21). These findings in aggregate indicate that PBRM1 functionally and selectively regulates HERVERI expression in a cell-autonomous manner.

Because prior studies have demonstrated that PBRM1 inactivation amplifies HIF transcriptional activity (21) and that expression of HERV-E, a member of the HERVERI superfamily, is regulated in a HIF2α-dependent manner (24), we asked whether HIF1α and HIF2α were necessary for upregulation of HERVs following PBRM1 KD. To this end, we performed RNA-seq and HERV quantification after knockdown of HIF1A and HIF2A in the previously generated UMRC2-shPBRM1 cells (Supplementary Fig. S4A). A large portion of HERVs upregulated upon PBRM1 KD were downregulated by silencing of either HIF1α or HIF2α (Fig. 3A and B). Furthermore, HIF1α and HIF2α regulated the expression of a relatively similar set of hERVs upregulated upon PBRM1 KD (Supplementary Fig. S5) and to a similar degree (Fig. 3C). Restricting the analysis to only the HERVERI superfamily of hERVs demonstrated comparable findings (Fig. 3D–F). Nonetheless, there were hERVs upregulated upon PBRM1 KD that were not altered by HIF1A or HIF2A siRNA (Supplementary Fig. S5), suggesting that PBRM1 might regulate these hERVs through HIF-independent processes, such as solely through nucleosome repositioning. In aggregate, these results indicate that both HIF1α and HIF2α are necessary for the expression of a large proportion of hERVs downregulated by PBRM1.

hERV expression has been associated with both prognosis and ICB response in ccRCC (14, 15). To understand the underlying mechanism of hERV regulation in ccRCC, we asked whether genes mutated in ccRCC could influence hERV transcription. Herein we report that PBRM1 loss is associated with increased hERV expression, particularly the HERVERI superfamily. PRBM1's regulation of hERVs is specific to ccRCC and not seen in other TCGA tumor types. Moreover, we demonstrate that PBRM1 inactivation alters hERV expression in a cell-autonomous manner and that both HIF1α and HIF2α are necessary for upregulation of PBRM1-dependent hERVs.

Our study offers a potential explanation for the well-established observation that PBRM1 mutations are associated with a better prognosis relative to SETD2- and BAP1-mutant tumors, as well as the variably reported observation that PBRM1 mutations are associated with enhanced ICB response in ccRCC (5–7, 11, 20, 25–27). Because PBRM1 mutations appear to be an early event in ccRCC pathogenesis, often occurring as a truncal mutation (28), it is possible that the development of a PBRM1-deficient tumor necessitates a balance between immune surveillance and immune escape against highly immunogenic hERVs or through robust immune checkpoint upregulation and T-cell exhaustion. The addition of ICB could disrupt this balance to provoke a robust adaptive immune response, stimulating primed, anti-hERV T cells, resulting in favorable clinical outcomes. Indeed, prior work by Childs and colleagues supports this notion, as they saw an increase in HERVE tetramer–positive T cells in responding ccRCC patients following allogeneic stem cell transplant (29).

Choueiri and colleagues looked for association between SWI/SNF alterations and ICB response in two large datasets of ICB-treated solid tumor types (n = 7) (11) and saw that, in aggregate, mutations in SWI/SNF complex members did not associate with time to treatment failure (TTF) or overall survival (OS) in the cohort as a whole (11). However, when limiting the analysis to patients with ccRCC, tumors with SWI/SNF mutations showed significantly improved TTF and OS, an effect driven primarily by PBRM1 mutations. These results suggest that the combination of PBRM1 mutations and the genetics underlying ccRCC result in a unique phenotype. Our demonstration that PBRM1 regulates hERVs in a manner dependent on both HIF1α and HIF2α may explain this link, as the loss of VHL results in constitutively stabilized HIFα and HIF transcriptional activity, and VHL mutations are unique to ccRCC development. Our data demonstrate that tumors with dual inactivation of VHL and PBRM1, as expected, appear to have elevated hERV expression, and this may explain the consistent finding that PBRM1 mutations associate with improved ICB outcomes only in ccRCC.

It is interesting that mutations in other PBAF complex members unique to the PBAF complex (ARID2 and BRD7) are associated with improved clinical outcomes to ICB treatment in NRAS-mutant melanoma, even when controlled for tumor mutational burden, tumor purity, and treatment (30). Furthermore, PBRM1, ARID2, and BRD7 were found to be tumor cell–autonomous regulators of T-cell killing in murine melanoma cell lines (26). Along these lines, we had previously described a key role for HIF1A and HIF2A in BRAF-mutant melanoma, which, like ccRCC, is highly angiogenic (31). In addition, recently published work demonstrates that SETDB1 is a key H3K9 methyltransferase that is critical for silencing of transposable elements including hERVs (32, 33). In aggregate, these data suggest key roles for both nucleosome positioning and modification of histone tails in the repression of hERV expression and strongly argues for a special interaction between the PBAF complex and HIF in mediating ICB response.

PBRM1 inactivation has been shown to amplify HIF transcriptional activity and is therefore associated with higher angiogenic signaling (7, 21). This increased HIF activity and subsequent VEGF production might account for the inconsistent association of PBRM1 mutations with ICB benefit in ccRCC. VEGF is known to be highly immunosuppressive by preventing dendritic cell maturation, and the combination of VEGF inhibition and ICB has been successful in ccRCC. It is therefore notable that the absence of an association between PBRM1 mutation and ICB response in the IMmotion150, IMmotion151, and JAVELIN Renal 101 were all in the first line setting (8, 9, 20), where tumors have not had significant exposure to VEGF blockade.

Our work shows that PBRM1-regulated hERVs in both tumors and UMRC2 cells are enriched in the ERVERI superfamily. What mediates the selective upregulation of HERVERI hERVs remains unknown. Furthermore, although either HIF1α and HIF2α are necessary for upregulation of HERVERI hERVs, we do not know whether this represents a direct or indirect consequence of HIF activation. Finally, although we propose a model whereby upregulated hERV expression serves as a tumor-associated antigen in ccRCC, definitive evidence that ERVs serve as the primary tumor-associated antigens driving ICB response is still lacking. Future studies should focus on validation of our results in other model systems, as well as understanding additional modifiers of ICB response in PBRM1-deficient ccRCC.

I.J. Davis reports grants from NIH during the conduct of the study, as well as other support from Triangle Biotechnology, Inc. outside the submitted work. W.Y. Kim reports grants from NIH, AACR Kure It, Kidney Cancer Association, and University Cancer Research Fund during the conduct of the study; stock and other ownership interests in AbbVie, Abbott, Amgen, Arvinas, BeiGene, Bluebird Bio, Bristol Myers Squibb, Myovant, Natera, Oramed, Zentalis, and ACT (Advanced Chemotherapy Technologies); and research funding from Acerta, GeneCentric, and Merck. No disclosures were reported by the other authors.

M. Zhou: Data curation, formal analysis, validation, visualization, writing–original draft, writing–review and editing. J.Y. Leung: Formal analysis, investigation, writing–original draft. K.H. Gessner: Data curation, investigation. A.J. Hepperla: Formal analysis, investigation, writing–review and editing. J.M. Simon: Formal analysis, investigation, writing–review and editing. I.J. Davis: Supervision, investigation, writing–review and editing. W.Y. Kim: Conceptualization, supervision, writing–original draft, writing–review and editing.

We acknowledge the members of the Kim Lab for useful discussions. This work was supported by NIH R01 CA202053 (to W.Y. Kim); 2012 AACR-Kure It Grant for Kidney Cancer Research, grant number 12-60-36-KIM (to W.Y. Kim); the Kidney Cancer Association (to W.Y. Kim); and the University Cancer Research Fund (UCRF).