Distinct endogenous retroviruses are expressed in mutational subtypes of clear cell renal cell carcinoma and are linked to improved clinical outcomes

SUMMARY

Distinct mutations in chromatin regulators and aberrant expression of transposable elements (TEs), have been associated with clinical benefit to immunotherapy (IO) in specific clear cell renal cell carcinoma (ccRCC) clinical contexts. However, the relationship between mutations in chromatin regulators and TE expression, and their effect on clinical outcomes, are incompletely understood. Here, we identified TEs expressed in distinct mutational subtypes of ccRCC, with endogenous retroviruses (ERVs) comprising the majority of TEs observed. Of these, ERVs 544 and 2014 were upregulated in PBRM1 mutant samples. Patients with high expression of these ERVs and somatic PBRM1 mutations had improved progression-free survival with IO monotherapy, but not targeted therapy, and their upregulation associated with expression of innate immune pathways. Chromatin accessibility increased at ERV 544 and 2014 loci in PBRM1-deficient ccRCC cells, and ERV 544 and 2014 were upregulated upon in vitro PBRM1 knockout in ccRCC cell line clones. Broadly, our study supports a link between PBRM1 mutations, subsequent chromatin accessibility changes, and aberrant but immunoresponsive ERVs in ccRCC.

Graphical abstract

In brief

Meli et al. conduct a comprehensive characterization of transposable element expression in clear cell renal cell carcinoma (ccRCC) mutational subtypes. They identify two specific endogenous retroviruses that are upregulated in PBRM1-mutant ccRCC and whose expression may contribute to selective immune checkpoint blockade response in distinct clinical settings.

INTRODUCTION

Renal cell carcinoma (RCC) is the most frequent type of primary renal neoplasm and represents approximately 2% of global cancer diagnoses.¹ In the US alone, there were over 80,000 estimated cases of RCC in 2023.^2,3 Clear cell renal cell carcinoma (ccRCC) is the most common histological subtype of RCC, accounting for 75% of RCC diagnoses.⁴ ccRCC is characterized by somatic loss of chromosome 3p and the VHL tumor suppressor, followed by mutations in chromatin modulators including PBRM1, SETD2, BAP1, KDM5C, and PI3K-MTOR pathway members such as PTEN and MTOR.⁵ VHL, PBRM1, SETD2, and BAP1 are also located on 3p, leading to biallelic inactivation in a subset of patients with mutations in these genes.

Despite having a low tumor mutational burden, ccRCC is considered an immunogenic cancer.^6,7 Treatment paradigms for advanced-stage ccRCC currently consist of immune checkpoint blockade therapy, combinations of immunotherapy (IO) and antiangiogenic targeted agents (primarily targeting the VEGF receptor, and to a lesser extent, the mTOR pathway), and HIF2ɑ inhibition.^8,9 Specific genomic alterations, such as PBRM1 mutations, have been associated with favorable response to IO or anti-angiogenic treatment in this cancer type in specific clinical contexts.^10–12 Other alterations, such as BAP1 mutations, mark subsets of ccRCC patients with poor prognosis.¹³ These and other genomic alterations characterize well-defined tumor evolutionary trajectories which associate with distinct clinical and biological features.¹⁴

Apart from somatic mutations, ccRCC has also been shown to have aberrant expression of endogenous retroviruses (ERVs), a type of transposable element (TE).¹⁵ TEs are generally silenced through epigenetic mechanisms and many of them have accumulated mutations and rearrangements, losing open reading frames.^16–18 However, a subset of ERVs retain similarities to their proviral sequence of origin.¹⁹ In addition, expression of a few specific ERVs, such as HERV-E or hERV4700, has been shown to associate with clinical benefit in ccRCC patients.^20,21 ERV derepression can lead to activation of innate and adaptive immune responses, possibly synergizing with IO therapy, and also may provide a source for tumor-associated antigens.^22–26 For instance, ERV-derived antigens have been detected in a long-term ccRCC responder to allogeneic hematopoietic stem cell transplantation.^27,28 Recently, PBRM1 inactivation has been associated with upregulation of a specific ERV superfamily.²⁹ However, the extent of TE expression generally (and ERV expression in particular) in distinct ccRCC mutational subtypes and its potential association with clinical outcomes remains incompletely understood. Moreover, the cell type-specific patterns of ERV expression in ccRCC and the potential epigenetic regulation at these loci are unknown. While initial reports have utilized bulk sequencing datasets to determine ERV expression, probing single-cell chromatin accessibility of ERVs through assay for transposase-accessible chromatin (scATAC-seq) could shed light on the cell type-specific patterns of ERV expression and the potential epigenetic regulation at these loci.

In this study, we quantified TE expression in tumors from IO-treated and untreated ccRCC patients, as well as cell line ccRCC RNA-sequencing (RNA-seq) datasets, and defined the mutational subtype-specific patterns of TE expression. We additionally identified specific ERVs that are differentially expressed in mutational subtypes, and observed that expression of two specific ERVs, in conjunction with PBRM1 inactivation, associated with selective response to IO in specific clinical settings. Moreover, increased expression of distinct ERVs associated with upregulation of distinct innate immune pathways. Utilizing scATAC-seq data, we found greater accessibility in PBRM1-deficient ccRCC cells at these ERV loci. Lastly, we experimentally validated that PBRM1 loss is sufficient to induce upregulation of ERVs. Overall, these results provide support for the mutational subtype-specific expression patterns of ERVs, their association with clinical outcomes in conjunction with distinct mutations, and the possible mechanistic link between chromatin modulation and aberrant ERV expression in ccRCC.

RESULTS

Global TE expression in ccRCC clinical cohorts and cell lines

To first determine the landscape of TE expression patterns in representative ccRCC settings, we characterized TE expression in three complementary cohorts: (i) an immunotherapy-treated ccRCC dataset (IO; includes CheckMate 009, 010, 025 and Miao et al., 2018),^10,30–32 (ii) an untreated ccRCC dataset (TCGA-KIRC),⁵ and (iii) ccRCC cell lines from the Cancer Cell Line Encyclopedia (CCLE) (Figure 1A; Methods).³³ In all three datasets, we quantified expression of long terminal repeat (LTR), long interspersed nuclear elements (LINEs), short interspersed nuclear elements (SINEs) and DNA transposon (DNA) elements. The relative mean expression of these TEs was similar across the clinical cohorts and cell lines when stratified by either TE types or TE families (Figures 1B; S1A), and further confirmed using orthogonal quantification methods (Figure S1C). Within the individual cohorts, LINE and SINE elements showed higher expression relative to LTR and DNA elements. Moreover, at the locus level, the relative expression across the four major TE types was comparable.

Figure 1. Global TE expression in IO-treated ccRCC (n = 107), TCGA-KIRC (n = 301) and CCLE ccRCC (n = 8) samples.

(A) Schematic of the experimental design of this study summarizing the datasets, computational pipelines and downstream analyses.

(B) Average TE expression in the three ccRCC cohorts. Each dot represents a TE of a specific class on the x axis, with the y axis values representing DESeq2-normalized expression.

(C) Average TE expression in the TCGA-KIRC cohort, stratified by pathologic stage (n = 8 I, n = 134 II, n = 117 III, n = 41 IV); Kruskall-Wallis test p values displayed.

(D) Average TE expression in responders (R, n = 42) and non-responders (NR, n = 65) in the IO cohort; unpaired Wilcoxon rank-sum test p values displayed.

We next analyzed whether global TE expression distributions varied in distinct clinical contexts, including across tumor stage and grade, as well as in the setting of selective clinical response to IO. Overall, TE expression did not differ appreciably when clinical cohorts were stratified by pathologic stage (Figure 1C, Kruskal-Wallis p value >0.05), primary tumor stage, tumor grade (Figure S1B, Kruskal-Wallis p value >0.05), or clinical benefit to IO (Figure 1D, unpaired Wilcoxon rank-sum p value >0.05). These results indicate that global TE expression was, on average, consistent across ccRCC clinical cohorts and cell lines, and that overall TE expression distributions did not exhibit distinct shifts across ccRCC clinical subsets.

TE expression in ccRCC mutational subtypes

Whereas we did not observe global changes in TE expression across cohorts or contexts, whether TE expression varied across the major mutational subtypes of ccRCC was uncertain and formed the basis for our subsequent investigation. Apart from the tumor suppressor VHL, multiple chromatin modulators are among the most frequently mutated genes in ccRCC, including PBRM1, SETD2, KDM5C, and BAP1.⁵ Yet, average TE expression was not significantly different across the clinical cohorts and cell lines when stratified by PBRM1 mutation status (Figure 2A, unpaired Wilcoxon rank-sum p value >0.05). Similar results were observed in the case of SETD2, KDM5C, and BAP1 mutational subtypes (Figures S2A–S2C, unpaired Wilcoxon rank-sum p value >0.05).

Figure 2. TE expression in IO-treated ccRCC (n = 107), TCGA-KIRC (n = 301), CCLE ccRCC (n = 8) and A704 PBRM1 KO (n = 4) mutational subtypes.

(A) Average TE expression stratified by PBRM1 mutation status across IO (top), TCGA-KIRC (middle) and CCLE (bottom) cohorts; unpaired Wilcoxon rank-sum test p values are displayed.

(B) PBRM1 mutant (log₂fc > 0) vs. wild-type (log₂fc < 0) differential gene and TE expression across datasets (padj ≤0.05). Representative TE subfamilies, and number of differentially expressed TEs, labeled.

(C) Frequency plot of the TE classes of differentially expressed TEs (padj ≤0.05) found in two or more mutational subtypes.

Even though TE average expression distributions were similar within these mutational subtypes, specific TE subfamilies could be differentially expressed upon epigenetic dysregulation. Thus, we examined whether specific TE subfamilies were differentially expressed based on PBRM1 mutant versus wild-type status, and we further complemented this analysis with examination of TE expression in an isogenic PBRM1 null and wild-type ccRCC cell line, namely A704.³⁴ Several TE subfamilies were differentially expressed in PBRM1 mutant compared to wild-type ccRCC across clinical and cell line datasets, including LTR16D, THE1-int, and the hAT-Tip100 transposon subfamily (Figure 2B; adjusted p value ≤0.05; Methods). In particular, HERVE was also differentially expressed in PBRM1 mutant CCLE samples (Figure 2B; adjusted p value ≤0.05). This ERV has been previously shown to encode a potential tumor-associated antigen in ccRCC.²⁷ We were similarly able to identify differentially expressed TEs in SETD2, KDM5C, and BAP1 subtypes (Figures S3A–S3C; adjusted p value ≤0.05).

We next asked whether LTR, LINE, SINE, or DNA transposon elements were enriched among the differentially expressed TEs. Across the mutational subtypes in patient and cell line datasets, 29 out of 795 TE subfamilies were differentially expressed in two or more subtypes (Figure S2D; adjusted p value ≤0.05; Methods). The majority of these TEs (22/29; 76%) belonged to the LTR class, while the rest were composed of LINE (4), SINE (1), and DNA (2) elements (Figure 2C). Further, ERV1 family members comprised the majority of differentially expressed TEs (Figure S2E). These results were consistent with prior literature reporting that ERVs (the most abundant LTR elements in the human genome) are aberrantly expressed in ccRCC, and taken together with our findings suggest that LTR elements generally—and ERVs in particular—are the major TE type that is differentially regulated by epigenetic mechanisms in this cancer type.^20,21,29,35

Differential ERV expression in mutational subtypes

Our preceding finding that LTRs (inclusive of ERVs) may be differentially regulated by epigenetic modulators prompted the subsequent goal of nominating specific ERVs differentially expressed in ccRCC mutational subtypes, as multiple ERV elements have been reported to be aberrantly expressed in ccRCC.^{20,21,27,28,36,37} Indeed, the mutational subtype-specific expression patterns of specific ERVs in ccRCC are unknown, though these properties cannot be robustly identified using global TE analysis algorithms (inclusive of those employed in this study). We thus utilized a complementary computational strategy to quantify specific ERV expression patterns in our ccRCC clinical cohorts (IO treated and TCGA) and cell lines, inclusive of an additional isogenic PBRM1 null cell line (A704; Methods; Figure 1A). Then, we stratified differential ERV expression analyses across the four most frequent mutational subtypes in ccRCC, namely PBRM1, SETD2, KDM5C, and BAP1. Notably, we identified multiple ERVs that were differentially expressed in PBRM1 mutant versus wild-type ccRCC across multiple cohorts leveraged herein (Figure 3A; adjusted p value ≤0.05; Methods). In particular, ERVs 2256 (HERV-E) and 4700, which were previously reported to be upregulated in ccRCC^21,27 generally, were differentially expressed in a PBRM1 subtype-specific manner. Similarly, we identified additional ERVs that were differentially expressed in SETD2, KDM5C, and BAP1 mutational subtypes (Figure S4). We also asked whether any of these subtype-specific ERVs were recurrently differentially expressed in more than one mutational subtype. We found that 52 out of the 3,173 reference ERV elements (1.6%) were differentially expressed across multiple mutational subtypes and ccRCC data-sets (Figure S5A). Thus, distinct ERV expression patterns were observed in a mutational subtype-specific manner and suggest that ERV expression may be regulated by epigenetic mechanisms in ccRCC.

Figure 3. Differential ERV expression in IO-treated ccRCC (n = 123), TCGA-KIRC (n = 355), CCLE ccRCC (n = 8) and A704 PBRM1 KO (n = 4) and association with survival.

(A) PBRM1 mutated (right) vs. wild-type (left) differential ERV expression.

(B) ERV expression association with OS and PFS in the IO cohort. Association was determined by a multivariable CoxPH model adjusted for age, sex, and purity, with ERV expression as a continuous variable. An association was deemed significant if the ERV variable nominal p value ≤0.05. Labeled ERVs in blue were also differentially expressed. Purple dashed lines indicate the approximate FDR q value ≤0.05. No ERVs had an FDR q value ≤0.05 for OS.

Mutation subtype-specific ERV expression and association with survival

Expression of specific ERVs has also been previously associated with distinct clinical outcomes in ccRCC in the context of immunotherapy using observational or small trial cohorts.^20,21,38 Thus, we next assessed whether ERVs were associated with overall survival (OS) or progression-free survival (PFS) in the IO cohort (Methods). Multiple ERVs were nominally associated with either of these outcomes, or both (Figure 3B; Table S1; multivariable CoxPH p ≤ 0.05; Methods). Importantly, the ERVs associating with OS or PFS included nine of the 52 ERVs that were differentially expressed in multiple mutational subtypes. Of these nine ERVs that were (i) differentially expressed in multiple mutational subtypes and (ii) associated with improved PFS or OS in the IO cohort, only ERVs 544 and 2014 had previously passed all prior minimum expression thresholds in all datasets utilized in this study (Figure S5B; multivariable CoxPH p = 0.03 and 0.02, respectively; Methods). Both ERV 544 and 2014 are HERV-Hs; ERV 544 contains polymerase and protease sequence motifs, while ERV 2014 contains polymerase motifs only, as previously described.¹⁹ ERV544 is located ~100 kb upstream of its nearest gene, A2M, which has potentially been associated with vascularization and immune infiltration in ccRCC, and H3K27Ac marks can be detected in cell lines near ERV544, suggesting a potentially important role for this particular region of chromosome 2.³⁹ ERV2014 is located upstream and antisense to SCGB3A2, a gene that is largely expressed in lung tissue and downstream of the lung lineage transcription factor TTF-1, expression of which is lost during lung adenocarcinoma progression.^39,40 These results suggested that mutational subtype-specific ERVs may also be associated with distinct survival patterns in ccRCC.

In ccRCC, PBRM1 loss-of-function mutations were previously associated with improved survival in the context of IO after tyro-sine kinase inhibitor (TKI) treatment, whereas BAP1 mutant ccRCC have been associated with an aggressive, treatment refractory clinical course.^10,11 Thus, we hypothesized that if ERVs 544 and 2014 have a potential role in their association with clinical outcomes, they should be upregulated in favorable mutational subtypes, such as PBRM1, and downregulated in aggressive ccRCC, such as BAP1 mutant. Accordingly, ERVs 544 and 2014 were upregulated in PBRM1 mutant samples compared to wild-type in the IO and TCGA cohorts (Figure 4A; ERV 544 unpaired Wilcoxon rank-sum p value ≤0.05 in TCGA; ERV 2014 p value ≤0.05 in IO, TCGA). Conversely, these two ERVs were downregulated in BAP1 mutant ccRCC (Figure 4B; ERV 544 unpaired Wilcoxon rank-sum p value ≤0.05 in TCGA; ERV 2014 p value ≤0.05 in IO).

Figure 4. Expression of ERV 544 and 2014 in ccRCC mutational subtypes and association with survival in IO cohort.

(A) Expression of ERV 544 and 2014 in IO cohort and TCGA, stratified by PBRM1 mutations.

(B) Same as in A, but stratified by BAP1 mutations. Statistical significance was determined with an unpaired Wilcoxon rank-sum test.

(C) Kaplan-Meier plots of PFS probability in the IO cohort, stratified by ERV expression (greater or less than mean) and PBRM1 mutations. p value represents the statistical significance determined by a log rank test. The CoxPH model p value is also included. Asterisks represent significance level: *p ≤ 0.05, **p ≤ 0.005, ***p ≤ 0.0005, ****p ≤ 0.00005.

We also found concordant mutational subtype expression patterns of these ERVs in PBRM1 mutant versus wild-type ccRCC cell lines (Figure S6A; ERV 544 and ERV 2256 unpaired Wilcoxon rank-sum p value ≤0.05). In fact, ERVs 544 and 2014 exhibited very similar expression patterns as ERV 2256, previously reported to be translated and presented as a ccRCC-associated antigen,²⁷ and ERV 4700, determined to be ccRCC-specific and possibly translated and recognized by anti-tumor T cells.²¹ While not all these comparisons achieved statistical significance, the consistent expression patterns suggested a shared underlying biology of expression of these ERVs in PBRM1 and BAP1 mutant ccRCC.

We further validated our findings in two additional, external ccRCC datasets, the everolimus (MTOR inhibitor; non-IO treated) arm of the CM025 trial, and IMMotion 150, a trial comprising atezolizumab (IO), atezolizumab and bevacizumab (IO and VEGF) and sunitinib (VEGF) arms (Methods). In both these cohorts, we observed consistent results with our previous analyses. Namely, ERVs 544 and 2014 exhibited higher expression across PBRM1 mutant samples compared to wild-type, and lower expression in BAP1 mutant samples (Figures S6B and S6C; ERV 544 unpaired Wilcoxon rank-sum p value ≤0.05 [IMMotion150 PBRM1, CM025 everolimus BAP1]; ERV 2014 p value ≤0.05 [IMMotion150 PBRM1 and BAP1]). These mutations occurred at different frequencies in ccRCC, with PBRM1 mutations observed in 40%–50% and BAP1 mutations in 10%–15% of tumors, and thus their reported exclusivity with each other cannot fully account for the opposite expression patterns observed.⁵

Given the prior evidence on PBRM1 mutation association with IO outcomes in ccRCC pre-treated with VEGF TKIs, we investigated whether aberrant ERV expression confers additional survival benefit beyond PBRM1 mutation status in this patient population. In the IO cohort, elevated expression of ERVs 544 and 2014 in conjunction with PBRM1 mutations conferred improved PFS compared to other sample subsets (Figure 4C; log rank p value = 0.039 and 0.025, respectively). In the everolimus arm of CM025, ERV expression and PBRM1 mutation status did not associate with PFS (Figure S7; log rank p value = 0.86 for ERV 544 and 0.58 for ERV 2014). In the IO arm of IMMotion 150, there was a modest association of ERV 544 with PFS, but no significant association in the case of ERV 2014 (Figure S7; log rank p value = 0.09 and 0.78, respectively). These conclusions were similarly consistent in a multivariable survival analysis, whereby high ERV 544 or 2014 expression and PBRM1 mutations had a hazard ratio of less than 1 in the IO cohort, but not in the everolimus arm of CM025 (Figure S8; CoxPH p value = 0.008 and 0.009, respectively, in IO cohort; 0.7 and 0.1, respectively, in CM025 everolimus cohort). These results suggest that aberrant ERV 544 and 2014 expression, in conjunction with PBRM1 mutations, may associate with improved PFS in IO-treated, TKI-refractory, ccRCC patients.

To investigate a potential biological process underlying the association between upregulation of ERVs 544 and 2014 with selective response to IO treatment, we stratified samples from patients treated with IO in the CM025 cohort by expression of hallmark pathways using ssGSEA (Methods).¹¹ Innate immune pathways, including TLR, RIG-I, JAK/STAT, and IRF dependent programs, were significantly enriched in ERV-high samples compared to ERV-low samples (Figure S9). We did not observe enrichment of adaptive immune responses in these samples, suggesting that the potential mechanism associated with response to IO relies on an inflammatory response and upregulation of type I IFN genes, sensitizing tumor cells to IO. Moreover, key oncogenic, non-immune related pathways such as MAPK signaling were not enriched in ERV-high vs. low samples, providing further support for our proposed mechanism underlying the clinical responses.

ERV chromatin accessibility and expression evaluation

TE expression and chromatin accessibility can differ across cell types.⁴¹ In addition, TE sequences can serve as regulatory binding sites, leading to distinct transcriptional outputs.⁴² Furthermore, one study suggested that, in ccRCC, PBRM1 mutations led to increased chromatin accessibility, whereas BAP1 mutations lead to decreased chromatin accessibility.⁴³ Following our observations that ERV 544 and 2014 exhibited increased expression in PBRM1 mutant ccRCC and have relevance in selective immunotherapy response, we hypothesized that chromatin accessibility at these ERV loci would accordingly change in these mutational subtypes in a cell-type specific manner. To this end, we utilized scATAC-seq data from the aforementioned ccRCC study⁴⁴ (n = 23 patient samples), and investigated the frequency of Tn5 transposase insertion events at the ERV 544 and 2014 locuses.

Chromatin accessibility at the ERV 544 and 2014 loci was increased in malignant cells from patients with putative LOF PBRM1 mutations compared to patients with wild-type PBRM1 (Figure 5A, Methods). To test the hypothesis that ERV expression from the bulk RNA-seq analyses was tumor-specific, we also investigated the chromatin accessibility profiles at the same ERV loci in CD45⁺ cells derived from the same ccRCC tissue samples. ERV 2014 accessibility was lower in CD45⁺ cells, whereas ERV 544 was also accessible in CD45⁺ cells. Coverage profiles at an immune cell-specific gene, GZMB, were markedly CD45⁺ cell-specific as expected, further corroborating the difference in accessibility observed at the ERV 2014 locus in the malignant cell population across PBRM1 mutational subtypes (Figure S10A). The majority of the remaining seven ERVs identified as mutational subtype-specific and associated with clinical outcomes were also exclusively expressed in malignant cells, compared to CD45⁺ cells (Figure S11). To better understand the impact of PBRM1 mutations on TEs at single-cell resolution, we also quantified TE accessibility in the scATACseq dataset using scTE (Methods).^40,45 We observed that patterns of TE activity varied substantially by cell type, with several TEs specifically expressed in tumor cells or immune cells (Table S2). Groups of cells identified by unsupervised clustering tended to be composed of cells from just one cell type (Figure S12A).

Figure 5. ERV 544 and 2014 single-cell chromatin accessibility and RT-qPCR expression in isogenic cell line.

(A) Coverage plots of chromatin accessibility in scATAC-seq data of ccRCC.

(B) RT-qPCR relative expression of ERV 544 and 2014, along with ERV 2256 (HERV-E), ERV 2637 (ERV3.2) and ERV 4700. Error bars represent standard deviation. Statistical significance of expression differences was calculated using an unpaired t test with Welch correction and q values are displayed.

We also characterized TE activity patterns within the malignant cell compartment. Through unsupervised clustering, we identified sets of tumor cells with distinct TE activity patterns. Two of these clusters contained cells from multiple samples, while the remaining four were predominantly sample-specific. We did not observe a strong interaction between PBRM1 mutation status and these unsupervised clusters (Figure S12B). However, through a supervised analysis of PBRM1 mutated tumor cells and PBRM1 wild type tumor cells, we identified several TEs whose accessibility is significantly associated with PBRM1 mutation status (Table S3). Several MERs (medium reiterated frequency repeats), which can derive from potentially all other TE types,³⁵ were found as differentially expressed and accessible in PBRM1 mutant or wild-type samples, though we did not find specific LTRs of the same families as the ERVs we identified.

Finally, we sought to experimentally test whether expression of ERVs 544 and 2014 was upregulated in ccRCC cells upon loss of PBRM1. We performed CRISPR-based knockout of PBRM1 in single-cell clones of 786O ccRCC cells and tested ERV expression via RT-qPCR (Figure S10B, Methods). We observed upregulation of these two ERVs upon loss of PBRM1 in three out of four single-cell clones (Figure 5B). In accordance with our previous analyses, ERV 4700 expression patterns were similar to those of ERVs 544 and 2014, though ERVE-4 (ERV 2256) was not similarly upregulated, suggesting other potential mechanisms that may govern individual ERV expression beyond loss of PBRM1. In addition, ERV 2637, which may also be expressed in immune cells and not ccRCC-specific based on prior literature,³⁷ did not show a consistent association between expression and PBRM1 loss, as we had determined in the bulk RNA-seq CCLE dataset. Taken together, these results supported a mechanistic link between PBRM1 mutation status and expression of select ERVs. In this model, chromatin at distinct ERV loci is less accessible in the presence of functional PBRM1, and conversely, more accessible in PBRM1-deficient ccRCC, leading to aberrant ERV expression. Their expression, coupled with PBRM1 loss, may associate with selective immunoresponsiveness in distinct clinical contexts in ccRCC.

DISCUSSION

In this study, we sought to characterize the mutational subtype-specific patterns of TE expression in ccRCC, as well as their potential association with relevant clinical outcomes in this patient population. To pursue this goal, we utilized patient RNA-seq data from IO-treated and untreated ccRCC, as well as ccRCC cell lines. Our findings revealed that TEs were similarly expressed across clinical cohorts and cell lines. In addition, average expression of TEs was not significantly different across clinical subsets of patients, suggesting the absence of a global change in the expression patterns of these elements. Within specific mutational subtypes, global TE expression was not significantly different, suggesting that while specific elements might be differentially expressed, the global TE expression distribution was not significantly altered. Indeed, we found that 29 out of approximately 800 TE subfamilies were recurrently differentially expressed across ccRCC mutational subtypes and datasets in this study. The majority of these TEs belonged to the LTR class, supporting previous literature that distinct ERVs are upregulated in ccRCC,^11,15,20,21 and suggesting that individual ERVs (the most common type of LTR elements in the human genome) may be contributors to anti-tumor immune responses in ccRCC.

Given these findings, we next aimed to determine whether specific ERVs are differentially expressed in ccRCC mutational subtypes, and whether their expression associates with clinical outcomes. We showed that multiple ERVs (52 out of 3,173 reference ERVs) were differentially expressed across mutational subtypes in clinical and cell line datasets of ccRCC. As continuous variables, expression of ERVs 544 and 2014 associated with PFS in the IO-treated ccRCC cohort. ERVs 544 and 2014 were upregulated in PBRM1 mutant ccRCC, across clinical cohorts and cell lines, and they were downregulated in BAP1 mutant samples. This pattern was validated in two external ccRCC cohorts, the everolimus arm of CM025 and IMMotion150. Given the reported association of PBRM1 mutations with benefit to IO monotherapy after TKI exposure, we tested whether expression of these ERVs synergizes with PBRM1 mutations to confer a survival benefit. Patients with high ERV 544 or 2014 expression, and PBRM1 mutations, had higher PFS probability compared to patients with low ERV expression, with or without PBRM1 mutations, in the IO cohort. This association was significant both through a univariable and a multivariable analysis adjusting for sex, age and tumor purity. Stratification by ERV expression and PBRM1 mutations was not predictive of PFS in the MTOR-inhibitor arm of CM025 and the metastatic IO-treated arm of IMMotion150, and ERV-high samples exhibited upregulation of innate immune pathways. These results suggest that ERV 544 and 2014 expression, in conjunction with PBRM1 mutation status, may associate with improved PFS post IO treatment in TKI refractory patients. However, in treatment-naive metastatic ccRCC, novel predictive biomarkers of response to IO are needed.

Given the expression upregulation of ERVs 544 and 2014 in PBRM1 mutant ccRCC, we also investigated whether there were concordant changes in chromatin accessibility specific to tumor cells using a scATAC-seq dataset of ccRCC. In putative malignant cells, chromatin accessibility was increased at these ERV loci in PBRM1 null cells compared to wild-type PBRM1 cells. Moreover, ERV 2014 exhibited decreased chromatin accessibility in CD45⁺ cells, suggesting that this ERV is more tumor restricted. Finally, to validate whether PBRM1 loss is a sufficient event for ERV 544 and 2014 upregulation in malignant cells, we performed CRISPR-Cas9 mediated knockout of PBRM1 in single-cell clones of the 786O ccRCC cell line. We observed an upregulation trend, albeit not statistically significant in a small sample size, of ERVs 544 and 2014 upon PBRM1 loss. This pattern also applied to the previously reported ERV4700, but not ERVs 2256 and 2637. Overall, these results support a potential mechanistic link between PBRM1 loss, subsequent increase in chromatin accessibility, and aberrant expression of specific ERVs in ccRCC cells, consistent with prior studies.²⁹

Limitations of the study

This study had several limitations. First, despite methodologic efforts to overcome these issues, TE expression can be confounded by read multimapping. While one method to circumvent this limitation is to add reads that multimap as pertaining to one TE subfamily, as is the case of the TEtranscripts tool, we lose locus-level information and thus cannot pinpoint which specific ERV loci are expressed. However, we did not observe LTR elements being differentially expressed in PBRM1 mutational subtypes of CCLE ccRCC samples at the locus-level, suggesting potential limitations when comparing distinct locus-level TE quantification tools such as hervQuant and SQUIRE. Moreover, some TEs have been shown to be expressed in immune cells, in addition to tumor cells.³⁷ While we accounted for tumor purity in our analyses of bulk RNAseq data and also investigated relevant single cell data-sets for cell-type specific inference, it is possible that some ERV reads in bulk samples come from cell types other than malignant cells. In addition, the variable sizes of the RNA-seq datasets in this study could have contributed to specific TEs/ERVs deemed statistically differentially expressed but having small effect sizes. When validating our bulk RNA-seq findings, the only accessible IO-treated ccRCC cohort, in addition to our discovery cohort, was IMMotion150. However, our discovery cohort represented advanced, targeted treatment-refractory ccRCC, whereas IMMotion150 participants had treatment-naive metastatic RCC. These two distinct clinical presentations may entail different underlying biologies and thus may account for the observed difference in ERV expression and PBRM1 mutation association with PFS. Indeed, PBRM1 mutations were not predictive of response to IO in IMMotion150.⁴⁶ In addition, it is possible that the locus-level RNA-seq analysis through which we identified distinct ERVs associated with immune checkpoint blockade in this study may not fully address ERVs that have a role in gene regulation, which may be better captured by family-level tools such as TEtranscripts and scTE, potentially explaining the fact that we did not observe HERV-Hs specifically differentially accessible in PBRM1 mutant versus wild-type ccRCC cells in our scATAC analyses. Last, we have shown that specific ERVs can be detected as expressed in ccRCC mutational subtypes, but additional experimental work is needed to show possible ERV-derived epitope presentation, and individual single-cell clone variability may account for potential ERV expression differences across clones. For example, a recent study showed that a distinct set of HIF-regulated ERVs are also translated and serve as a source of tumor-associated antigens through complementary functional approaches, which complements our study demonstrating that there may be distinct molecular differences within PBRM1-mutant ccRCC that could result in differential expression and translation of specific ERVs.

RESOURCE AVAILABILITY

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Eliezer M. Van Allen (eliezerm_vanallen@dfci.harvard.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• The IO cohort refers to the CheckMate 009, 010 and 025, and Miao et al. cohorts.^10,30–32 TCGA-KIRC RNA-seq fastqs were downloaded from the Genomic Data Commons (GDC).⁵ CCLE and A704 cell line RNA-seq fastqs were downloaded from the Sequence Read Archive (SRA).³³ CM025 everolimus data are part of the CM025 cohort, and IMMotion 150 genomic data were previously published.⁴⁶ scATAC-seq FASTQs from the ccRCC CPTAC cohort were obtained through the Cancer Data Service after dbGaP approval phs001287.v16.p6.⁴³

• The computational tools utilized in this study have been previously published and are publicly available. The STAR and Salmon docker images are available at quay.io/biocontainers/star and quay.io/biocontainers/salmon, respectively. The utilized samtools docker image is available at quay.io/biocontainers/samtools. The TEtranscripts docker image is available at quay.io/biocontainers/tetranscripts. Detailed information on these pipelines, including command line parameters, can be found in the following sections. This study does not report original code.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Cell lines

786-O cells were purchased from ATCC (CRL-1932) and cultured in Roswell Park Memorial Institute (RPMI) 1640 media (Genesee Scientific; 25–506) supplemented with 10% Fetal Bovine Serum (FBS) (Genesee Scientific; 25–514H) and 1% penicillin streptomycin (Genesee Scientific; 25–512). Cell line authentication was performed by short tandem repeat (STR) profiling (ATCC), and routine mycoplasma testing was carried out using the ATCC Universal Mycoplasma Detection Kit (30–1012K).

METHOD DETAILS

TEtranscripts

Command line parameters

Transposable element expression at the family and subfamily level was quantified using TEtranscripts.⁵⁰ Briefly, paired-end RNA-seq fastqs were aligned to the hg19 reference genome using STAR 2.7.0f, with parameters –outFilterMultimapNmax 20 –outFilterMismatchNmax 999 –outFilterMismatchNoverReadLmax 0.04 –alignIntronMin 20 –alignIntronMax 1250000 –alignMatesGapMax 1250000 –twopassMode Basic.⁴⁸ For A704, single-end alignment was performed. Resulting genome-aligned BAM files were used as input for TEtranscripts. TEtranscripts was run in mode multi, and the Ensembl GRCh37 Repeat Masker TE gtf file was used for TE annotation. The docker image used to run TEtranscripts was pulled from quay.io/biocontainers/tetranscripts: 2.2.1–pyh864c0ab_1.

Quality control

RNA-seq samples with less than 1M input reads, less than 70% uniquely mapped reads, or more than 40% multi mapped reads based on STAR alignment were excluded. Only samples with mutation annotations, RECIST (in case of IO cohort) and survival annotations were kept. Counts matrices were filtered to include only TEs that are expressed in at least half the corresponding cohort and also have a mean expression count of ≥10 counts.

For the TCGA cohort, mutation annotations and purity estimates were downloaded from cBioPortal^61–63 and from Smith et al.²¹ Samples previously reclassified as chRCC were excluded.⁶² For CCLE, metadata was downloaded from cBioPortal,⁶¹ Broad DepMap³³ and Cellosaurus,⁶⁴ whereas for A704, metadata was obtained from Gao et al.³⁴ TEs belonging to the LTR, LINE, SINE and DNA transposons families were retained for analysis, while TEs annotated as “DNA?”, “RNA”, “SINE?”, “Unknown”, “Other”, “RC”, “Satellite” were excluded due to very low member numbers or undefined annotations.

hervQuant

Command line parameters

ERV expression was quantified in RNA-seq samples with hervQuant.²¹ Briefly, paired-end fastqs were aligned to the hg19 transcriptome and a reference of 3,173 ERVs using STAR 2.6.0c, with custom parameters –outFilterMultimapNmax 10 and –outFilterMismatchNmax 7 as optimized in hervQuant.¹⁹ The output BAM files were filtered retaining reads mapping to ERVs, and the filtered BAMs were quantified against the ERV reference using Salmon 0.9.1 with parameters –incompatPrior 0.0 –seqBias –gcBias –reduceGCMemory.⁴⁹

Quality control

Samples with less than 1M input reads based on STAR alignment, less than 70% uniquely mapping reads based on Salmon quantification, and more than 0.5% of multimapping reads mapping to ERV loci were excluded. ERV multimapping was determined as the proportion of multimapping reads mapping to ERV loci. The counts matrices were filtered such that ERVs had non-zero expression in at least a quarter of the corresponding cohort and had mean counts ≥1.

For the TCGA cohort, mutation data was downloaded from cBioPortal.⁶¹ Raw expression counts and STAR/Salmon alignment and quantification data was obtained from Smith et al.²¹ Samples with less than 1M reads based on STAR alignment and less than 70% uniquely mapping rate based on Salmon quantification were excluded. In addition, samples reclassified as chRCC were excluded.⁶² For CCLE and A704, samples with less than 1M input reads based on STAR alignment, less than 70% unique mapping rate based on Salmon quantification and greater than 0.5% ERV multimapping rate were excluded. Mutation data for CCLE was downloaded from cBioPortal.⁶¹ ERVs expressed in less than a quarter of the cohort or with mean count <1 across the cohort were excluded.

SQuIRE

We quantified subfamily-level and locus-level TE expression in the CCLE ccRCC dataset utilizing SQuIRE as previously described.⁵¹ For sub-family level analyses, we filtered out any TE subfamilies with expression <10 fpkm. For locus-level analyses, we filtered out any TEs with expression <5 fpkm.

Differential expression analyses

Differential expression was conducted with DESeq2⁵² (v 1.36.0). Covariates included in the differential expression model, where applicable, were: sex, age (≥65 or <65), cohort of origin (IO cohort only; CM009, CM010, CM025 or Miao et al.), tumor purity (binned in deciles starting from <0.2; IO and TCGA cohorts only), clinical stage (TCGA cohort only; T1/T2 or T3/T4). Loss of function mutations were defined as frameshift indels, nonsense and splice site. In addition, missense mutations predicted by at least two of three tools Mutation Assessor,⁵³ Polyphen⁵⁴ and SIFT⁵⁵ to be deleterious were considered as loss of function. Log2-fold change estimates were shrunken using the apeglm package⁵⁶ (v 1.18.0). Genes and TEs were considered significantly differentially expressed if the adjusted p-values using the Benjamini-Hochberg method with an FDR of 0.05 were ≤0.05 and had a |log2-fold change| > 0.

Survival analyses

CoxPH models

CoxPH models were fit using the survival R package (v 3.3–1). For each ERV using DESeq-normalized expression as a continuous variable, a CoxPH model was fit predicting OS or PFS including sex, age (≥65 or <65), and purity bins as covariates. To determine the association of ERV expression and PBRM1 mutations with survival, expression of ERV 544 and 2014 was categorized as high or low for each patient if its expression was > or ≤ the mean expression of that ERV within the specific cohort. For the CoxPH model fitting, the erv&pbrm1 variable was defined as high&mut vs. other, where ‘other’ includes samples defined as low&mut, high&wt, low&wt. In addition, sex, age and purity bins were included as covariates.

Kaplan-meier analyses

The survminer R package (v 0.4.9) was used to plot the survival curves after Kaplan-Meier survival analysis using the survival R package (v 3.3–1). Using the same erv&pbrm1 classifications as in the CoxPH models, a survival model was fit using this variable as a predictor of PFS. For IMMotion 150, this analysis was conducted in the atezolizumab (IO) arm.

Single cell accessibility analyses

Quality control and cell type annotation

We performed read filtering, alignment, and cell calling with Cell Ranger ATAC v2.0.0. We used MACS2 v2.2.9.1 within Signac v1.10.0 using default parameters to construct a per-sample accessible peak set.^57,58 We identified putative doublets using scDblFinder v1.12.0.⁵⁹ After removing cells identified as doublets, we set objective quality control (QC) thresholds on additional QC metrics computed with Signac, using a statistical approach based on the median and median absolute deviation (MAD). On a per-sample basis, cells were excluded if they had less than 3,000 or more than the median plus three times the MAD of fragments overlapping peaks. This threshold was established to ensure sufficient genomic coverage while avoiding excessive signal that might indicate an artifact. Similar criteria were applied to other metrics, with cells excluded for having a fraction of reads in peaks below the median minus 1.25 MAD, blacklist fractions or nucleosome signals above the median plus 3 MAD, and transcription start site enrichment below the median minus 3 MAD. While annotating broad and granular cell types and states, we further defined a set of cells not to be included in further analysis. These cells included those that appeared to be contaminants, ambiguous identities, or clustered solely by a QC metric. To operate in a shared feature space, we quantified each sample on a peak set derived from an on-going multi-cohort RCC scATAC analysis. We merged each sample’s count matrix and performed normalization and dimension reduction using Signac’s RunTFIDF and RunSVD commands to perform latent semantic indexing (LSI). Then, we batch corrected the LSI components using Harmony with sample ID as the batch covariate.⁴⁷

We annotated broad cell lineages (e.g., lymphoid, myeloid, stromal, tumor) by first performing unsupervised clustering with the smart local moving algorithm on the batch corrected embeddings. To identify non-malignant cell lineages, we evaluated, for each cluster, differentially accessible genes using gene activity scores (Wilcoxon test) and accessibility over known marker genes. We used Copy-scAT v0.30 -derived copy number estimates to identify clusters of putative tumor cells.⁶⁰ CD45⁺ cells were defined as cells from the lymphoid or myeloid lineages. Downstream analysis, statistical testing, and visualization of scATAC data were primarily performed using Signac v1.10.0.

ERV coverage

We defined samples as having a putative loss-of-function PBRM1 mutation if the mutation types reported were nonsense, frameshift and splice site.⁴³ Missense and in-frame indels were not included in the analyses due to the undefined consequences on protein function of these distinct types of mutations. Remaining samples were defined as having wildtype PBRM1.

scTE-based TE quantification

For each ccRCC scATAC-seq sample from the CPTAC cohort,⁴³ we sorted the CellRanger ATAC pipeline output BAM by read name using Picard’s SortSam tool. We then utilized an adapted scTE method (available at https://github.com/sabrinacamp2/scTE) to assess TE accessibility. This adaptation involved defining a maximum fragment length for paired-end alignments used in the TE quantification. With this adapted scTE method, we used the following command to produce a TE counts matrix: ‘scTEATAC -CB CB -UMI False -p 6 -i ${bam} –expect-cells 20000 -x ./hg38.te.atac.idx -o ${sample_id}’. We used Seurat v4.3.0 for all processing and analysis of the scTE-produced TE counts matrix.

scTE data filtering and normalization

Cells were retained if they had TEs quantified from scTE, passed our peak-based QC, and were annotated as either tumor or CD45⁺ cells from our peak-based annotation, as detailed in the manuscript Methods section. We log-normalized the TE counts with the scale factor set to the median number of TEs detected per cell.

scTE clustering, visualization, differential accessibility analysis

For panel A analysis, we scaled the data and performed PCA, selecting PCs 1 through 15 to use in downstream analysis based on variance explained in an elbow plot. We used these PCs to compute a nearest neighbors graph and UMAP visualization. Clusters were identified using Seurat’s FindClusters function with the Louvain algorithm at a resolution of 0.1. Differential accessibility of TEs between tumor and CD45⁺ cells was assessed using the logistic regression framework in Seurat’s FindAllMarkers function, adjusted for sequencing depth. We focused on TEs with a minimum log-fold change of 0.1 and an adjusted p-value less than 0.05.

For panel B analysis, we subset the data to tumor cells with either a pLOF PBRM1 mutation or no PBRM1 mutation. We then scaled the data, ran PCA, and used Harmony to adjust the PCs for sample-specific variation.⁵⁸ Adjusted PCs 1 through 10 were selected for use in downstream analysis based on variance explained. We used these adjusted PCs to compute a nearest neighbors graph and UMAP visualization. Clusters were identified using Seurat’s FindClusters function with the Louvain algorithm at a resolution of 0.2. Differential accessibility of TEs between PBRM1-mutant and wild type cells was assessed using the logistic regression framework, adjusted for sequencing depth. We focused on TEs with a minimum log-fold change of 0.1 and an adjusted p-value less than 0.05.

Experimental validations

CRISPR-Cas9 mediated knockout of PBRM1 via RNP electroporation

Guide RNAs targeting human PBRM1 were designed using the Benchling CRISPR gRNA Design Tool. The crRNA sequences used are listed below.

Guide RNA	crRNA sequence (5′ to 3′)
PBRM1 sg1	GACTATTATGAAGTGGTTTC
PBRM1 sg2	TCAACCAGACTATTATGAAG
PBRM1 sg3	TTGGATTTTCATCAAGTCAA
PBRM1 sg4	TTGCATTGTTAAAAAGAAGC

Synthetic crRNAs (Dharmacon) were annealed to tracrRNA (Dharmacon, U-002005–05) at a 1:1 molar ratio by incubation at 37°C for 30 min to generate sgRNAs. Cas9 nuclease protein NLS (Dharmacon, CAS12205) was added to the sgRNA mixture at a 1:2 molar ratio (Cas9: sgRNA) to generate ribonucleoprotein (RNP) complexes at a final concentration of 50 pmol. RNPs were incubated for 15 min at 37°C and pooled (four guides per gene) prior to electroporation.

786-O cells were resuspended at 0.5 × 10⁶ cells/20 μL in SF buffer (Lonza V4XC-2012). Pre-assembled RNP complexes were added to the cell suspension and transferred to a single well of a 16-well Nucleocuvette Strip. Electroporation was performed using the Lonza 4D Nucleofector with program EN-138. Immediately after electroporation, 80 μL of pre-warmed complete RPMI medium was added to each well, and cells were allowed to recover at 37°C for 15 min before transfer to 6-well plates for culture. Knockout efficiency was verified by Western blot and RT-qPCR.

RT-qPCR

Total RNA was extracted from cell lines using the RNeasy Mini Kit (Qiagen, 74104) following the manufacturer’s instructions. cDNA was synthesized from 2 μg of RNA using SuperScript IV VILO Master Mix (Thermo Fisher Scientific, 11756050). RT-qPCR was performed in technical triplicates using Power SYBR Green PCR Master Mix (Thermo Fisher Scientific, 43676459) in a 384-well plate on a ViiA 7 Real-Time PCR System (Thermo Fisher Scientific). PCR primers are listed in the table below. Relative transcript levels were determined using the 2-ΔΔCT method with ACTB (B-actin) as the endogenous control.

Using Integrative Genomics Viewer, we aggregated RNA-seq reads for ccRCC cell lines mapped to ERV 544 and 2014 and determined areas of high transcription to target for primer design. Next, we generated multiple primers in the identified areas of interest which met the following criteria: GC: 40–60% (optimal 50%), Tm 58–65%, Size: 20, three nucleotide 3′ GC clamp, amplicon size: 75–160 (optimal 120). To determine the specificity of these primers, Basic Local Alignment Search Tool (BLAST) was employed and primers with matches elsewhere in the genome were discarded. From here, four specific primer pairs were selected for ERV 544 and ERV 2014 respectively to proceed with experimentation. We tested these four primers sets in multiple experiments including comparing a PBRM1 wild type cell line (786-O) to a PBRM1 mutant cell line (CAKI-2). The relationships demonstrated were consistent across all four primer sets for a given ERV. Ultimately, we choose the primer pair with the best melt curves for ERV 544 and 2014 respectively to optimize specific adherence. These primers amplified the following regions: ERV 544: chr2: 57,423,6555–57,423,737 (83 base pairs). ERV 2014: chr5:147,252,057–147,251,970 (88 base pairs).

Gene	Primer	Sequence
B-actin65	Forward	5′-ACCAACTGGGACGACATGGAGAAA-3′
B-actin65	Reverse	5′-TAGCACAGCCTGGATAGCAACGTA-3′
ERV 544	Forward	5′-TTACACAAGCGCCAGGACCG-3′
ERV 544	Reverse	5′-ACGCAGACATGAGTGCTAGGC-3′
ERV 2014	Forward	5′-AAACACCTGAACCGCAGCCG-3′
ERV 2014	Reverse	5′-GCCTGGTGGTCAGATTTCTGGC-3′
ERVE-411	Forward	5′-GCAGATCCTGGGAGCACTCT-3′
ERVE-411	Reverse	5′-TGTTCAACCGCTGTGTTAATTCTC-3′
ERV3.211	Forward	5′-TACAAGGAGGCAGGTGGAAG-3′
ERV3.211	Reverse	5′-TTCCCATGTGTGTTCCCTGA-3′
hERV470011	Forward	5′-GACGCTCCCAGCAGAATAAA-3′
hERV470011	Reverse	5′-CCGGTCAGGAAACCAAGAAA-3′
PBRM166	Forward	5′-ACGGAAAATCAACATGAGTG-3′
PBRM166	Reverse	5′-TGCCTTCATATTCTGCTTTC-3′

QUANTIFICATION AND STATISTICAL ANALYSIS

All analysis considerations can be found in the corresponding figure legends and method details sections.

Supplementary Material

Supplemental information can be found online at https://doi.org/10.1016/j.celrep.2026.117808.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Biological samples
786-O	ATCC	CRL-1932
Deposited data
CheckMate 025	Motzer et al.31	Motzer et al.31
CheckMate 010	Motzer et al.30	Motzer et al.30
CheckMate 009	Choueiri et al.32	Choueiri et al.32
TCGA-KIRC	Creighton et al.5	https://portal.gdc.cancer.gov/
CCLE	Ghandi et al.33	Ghandi et al.33
A704	Gao et al.34	Gao et al.34
IMmotion 150	McDermott et al.46	McDermott et al.46
CPTAC scATAC	Wu et al.43	Wu et al.43
Oligonucleotides
See STAR methods
Software and algorithms
Harmony	Korsunsky et al.47	Korsunsky et al.47
STAR	Dobin et al.48	Dobin et al.48
Salmon	Patro et al.49	Patro et al.49
TEtranscripts	Jin et al.50	Jin et al.50
hervQuant	Smith et al.21	Smith et al.21
SQuIRE	Yang et al.51	Yang et al.51
DESeq2	Love et al.52	Love et al.52
Mutation Assessor	Reva et al.53	Reva et al.53
PolyPhen	Adzhubei et al.54	Adzhubei et al.54
SIFT	Ng et al.55	Ng et al.55
apeglm package	Zhu et al.56	Zhu et al.56
Signac	Stuart et al.57	Stuart et al.57
MACS	Zhang et al.58	Zhang et al.58
scDblFinder	Germain et al.59	Germain et al.59
Copy-scAT	Nikolic et al.60	Nikolic et al.60

Highlights.

• Global transposable element expression is consistent across ccRCC datasets

• Distinct endogenous retroviruses are expressed in ccRCC mutational subtypes

• Expression of specific ERVs associates with clinical responses to immunotherapy

• Aberrant ERV expression may stem from increased chromatin accessibility in ccRCC