Abstract

   For clear cell renal cell carcinoma (ccRCC), lipid deposition plays
   important roles in the development, metastasis, and drug resistance.
   However, the molecular mechanisms underlying lipid deposition in ccRCC
   remain largely unknown. By conducting an unbiased CRISPR-Cas9
   screening, we identified the epigenetic regulator plant homeodomain
   finger protein 8 (PHF8) as an important regulator in ccRCC lipid
   deposition. Moreover, PHF8 is regulated by von Hippel–Lindau
   (VHL)/hypoxia-inducible factor (HIF) axis and essential for VHL
   deficiency–induced lipid deposition. PHF8 transcriptionally
   up-regulates glutamate-ammonia ligase (GLUL), which promotes the lipid
   deposition and ccRCC progression. Mechanistically, by forming a complex
   with c-MYC, PHF8 up-regulates TEA domain transcription factor 1 (TEAD1)
   in a histone demethylation–dependent manner. Subsequently, TEAD1
   up-regulates GLUL transcriptionally. Pharmacological inhibition of GLUL
   by l-methionine sulfoximine not only repressed ccRCC lipid deposition
   and tumor growth but also enhanced the anticancer effects of
   everolimus. Thus, the PHF8-GLUL axis represents a potential therapeutic
   target for ccRCC treatment.
     __________________________________________________________________

   Targeting PHF8/TEAD1/GLUL axis represents a therapeutic target for
   ccRCC by inhibiting lipid deposition and tumor progression.

INTRODUCTION

   About 430,000 people suffer from and more than 179,000 die of kidney
   cancer each year worldwide, and these numbers have been increasing in
   the past decades ([60]1). Clear cell renal cell carcinoma (ccRCC)
   characterized by increased lipid deposition in the cytoplasm accounts
   for more than 75% of all subtypes of kidney cancer ([61]2). It has been
   well established that the intracellular lipid droplets mainly composed
   of triglycerides and cholesterol esters play important roles in energy
   homeostasis, membrane synthesis, and cellular signaling during
   proliferation ([62]3). More recent evidence indicates that extra lipid
   deposition in ccRCC is involved not only in tumor initiation and
   progression but also in metastasis and resistance to the antiangiogenic
   drug ([63]4, [64]5).

   Inactivation of von Hippel–Lindau (VHL) caused by either somatic
   mutation or promoter methylation occurs in ~90% of ccRCC ([65]6). Loss
   of function of VHL leads to accumulation of hypoxia-inducible factors
   (HIFs), the major driving force in ccRCC development ([66]7). In
   addition, metabolic reprogramming driven by VHL/HIF-mediated
   pseudo-hypoxia makes ccRCC a metabolism-related disease ([67]8, [68]9)
   evidenced by increased glucose usage (Warburg effect) and extra lipid
   deposition ([69]3). Multiple lines of evidence showed that aberrant
   activation of VHL/HIF axis is the main cause of lipid deposition
   ([70]3, [71]10). However, the underlying mechanism in VHL/HIF-mediated
   lipid deposition in ccRCC remains to be revealed.

   Glutamine is the most abundant amino acid in the blood stream and
   involves in multiple metabolic pathways ([72]11). Comparing to
   noncancer cells, increased accumulation and metabolism of glutamine in
   cancer cells ([73]2) can serve as a precursor for tricarboxylic acid
   (TCA) cycle and the syntheses of lipids, glutathione, and nonessential
   amino acids ([74]12), providing a metabolic fuel to sustain rapid
   cancer cell proliferation, progression, and metastasis ([75]11,
   [76]13). In VHL-deficient cells including ccRCC, glutamine is the major
   source for lipid synthesis via reductive carboxylation ([77]8, [78]14).
   Although exogenously supplied glutamine favors lipid deposition and
   tumor cell growth, the endogenously produced glutamine plays
   indispensable roles in multiple metabolic pathways ([79]15, [80]16).
   Glutamate-ammonia ligase (GLUL), also known as glutamine synthase, is
   the only identified enzyme responsible for de novo glutamine synthesis
   by ligating a glutamate with an ammonium ion (NH[4]^+) ([81]12).
   Previous studies have showed that GLUL-mediated glutamine synthesis
   plays a crucial role in tumor growth via regulating amino acid
   transportation and nitrogen metabolism such as nucleotide synthesis and
   recycling excessive ammonia ([82]17, [83]18), whereas the roles of GLUL
   in ccRCC are unclear.

   It has been reported that dysfunction of epigenetic regulators
   including PBRM1, SETD2, BAP1, and KDM5C are associated with ccRCC
   ([84]19). Alterations of the VHL/HIF axis is known to affect several
   Jumonji C (JmjC)-containing epigenetic regulators ([85]20, [86]21),
   including plant homeodomain finger protein 8 (PHF8), also known as
   KDM7B. PHF8 transcriptionally regulates target gene expression by
   demethylating H3K9me2/1, H3K27me2, or H4K20me1 ([87]22, [88]23) and
   involves in neuronal differentiation, cognitive ability development,
   cell adhesion, and cytoskeleton organization ([89]24–[90]26). More
   recent data suggest that up-regulated PHF8 is correlated with a variety
   of malignancies including breast cancer ([91]27), gastric cancer
   ([92]28), hepatocellular carcinoma ([93]29), acute lymphoblastic
   leukemia ([94]30), and prostate cancer ([95]31). In this study, we
   demonstrated that PHF8 promotes lipid deposition and tumor growth in
   ccRCC. Mechanistically, PHF8 is recruited by c-MYC to the promoter
   regions of TEA domain transcription factor 1 (TEAD1) to
   transcriptionally up-regulate TEAD1. Subsequently, TEAD1 up-regulates
   GLUL transcriptionally. In addition, we have demonstrated that
   targeting the PHF8-GLUL axis not only inhibited lipid deposition and
   tumor growth of ccRCC but also synergized with the therapeutic reagent
   everolimus (EVE).

RESULTS

A CRIPSR-Cas9 screen reveals a key role of PHF8 in the lipid deposition of
ccRCC

   Given the important roles of lipid deposition and epigenetic
   reprogramming in ccRCC, we decided to investigate whether epigenetic
   regulators are involved in ccRCC lipid deposition. To this end, we
   first carried out an unbiased CRISPR-Cas9 screening of 1136 epigenetic
   regulators with four small guide RNAs (sgRNAs) per gene (table S1). The
   786-O cells (VHL-null cells) were transfected with the sgRNA pool,
   followed by BODIPY 493/503 staining and flow cytometry to sort the
   cells into two subpopulations based on the levels of cellular fatty
   acid ([96]Fig. 1A). The primary screening identified 39 genes (table
   S2) exhibiting altered sgRNA frequencies (P < 0.05; [97]Fig. 1B).
   Furthermore, results from real-time quantitative polymerase chain
   reaction (RT-qPCR; fig. S1A) confirmed that the top six genes
   {−log[10][P value] > 2 and log[2][fold change (FC)] < −1} including
   TES, PHF8, FRG1, GRIPAP1, LIN28A, and N6AMT1 were significantly
   down-regulated in the cells with reduced fatty acid. Knockdown either
   one of these six genes using small interfering RNAs (siRNAs) could
   reduce the lipid deposition in 786-O cells revealed by BIODIPY 493/503
   staining ([98]Fig. 1C and fig. S1, B and C). Of note, silencing PHF8
   resulted in the biggest decrease of lipid deposition. This finding
   suggested that PHF8 might be the most important determinant for lipid
   deposition and prompted us to further investigate the role of PHF8 in
   lipid deposition in ccRCC. The 786-O and 769-P cells, two VHL-null cell
   lines, were chosen, because both had higher levels of both PHF8 (fig.
   S1D) and lipid deposition (fig. S1, E and F) compared with other cell
   lines. We then depleted PHF8 in 786-O and 769-P cells using sgRNA
   ([99]Fig. 1D) and found that the lipid droplet formation was
   significantly inhibited revealed by both oil red O (ORO) staining
   ([100]Fig. 1, E and F) and Triglyceride-Glo assay ([101]Fig. 1G). We
   have also generated 11 ccRCC organoids derived from different patients
   (table S3) and found that organoid ccRCC02 carries a VHL mutation
   (c.232_239delAATCGCAG) with concurrent high PHF8 expression (fig. S1, G
   and H). The role of PHF8 in lipid deposition was further confirmed by
   manipulating the levels of PHF8 in this organoid ([102]Fig. 1H and fig.
   S1I). Last, we conducted lipidomic analysis with the 786-O cells
   transfected with either PHF8-sgRNA or Ctrl-sgRNA. The levels of 149
   lipid compounds were significantly altered after PHF8 knockout, and 75
   (48.7%) of them were in the family of triglycerides ([103]Fig. 1, I and
   J, and table S4). These results demonstrated that PHF8 is an important
   epigenetic regulator involving in lipid deposition in ccRCC.

Fig. 1. A CRIPSR-Cas9 screen reveals a key role of PHF8 in the lipid
deposition of ccRCC.

   [104]Fig. 1.
   [105]Open in a new tab

   (A) Scheme of CRISPR-Cas9–based epigenetic library screening to
   identify the determinant of lipid deposition of 786-O cells. (B)
   Volcano plot showing the genes with significantly increased small guide
   RNA (sgRNA) frequencies in low fatty acid group with criteria of
   log[2][fold change (FC)] < −1 and –log[10](P value) > 2. (C)
   Quantification data of lipid using BODIPY 493/503 staining in 786-O
   cells transfected with small interfering RNAs (siRNAs) targeting
   indicated genes, with scrambled siRNA as control. (D) Western blotting
   (WB) assay showing the efficiency of plant homeodomain finger protein 8
   (PHF8) depletion using CRISPR-Cas9 method in 786-O and 769-P cell
   lines. GAPDH, glyceraldehyde-3-phosphate dehydrogenase. (E and F)
   Representative images (E) and quantification data (F) of lipid droplets
   stained with oil red O (ORO; red) in clear cell renal cell carcinoma
   (ccRCC) cells (786-O and 769-P) transfected with control (Ctrl)–sgRNA
   and PHF8-sgRNA. Nuclei were stained with Mayer’s hematoxylin (scale
   bar, 100 μm in 400×). (G) Relative intracellular triglyceride (TG)
   levels revealed by using Triglyceride-Glo assay in 786-O and 769-P cell
   lines transfected with Ctrl-sgRNA and PHF8-sgRNA. (H) Representative
   fluorescence images of lipid droplets stained with BODIPY 493/503
   (green) and corresponding quantification in ccRCC02 organoids
   transfected with PHF8-sgRNA or Ctrl-sgRNA. Nuclei were stained with
   DAPI (4′,6-diamidino-2-phenylindole; blue; scale bar, 50 μm in 400×).
   (I) Pie chart illustrating the lipid subclasses of the significantly
   altered lipid compounds after depleting PHF8 in 786-O cells revealed by
   lipomics. PC, phosphatidylcholine. (J) Heatmap showing that 75
   identified lipid compounds belonging to TAG (triacylglycerol) subclass
   were decreased after PHF8 depletion in 786-O cells (n = 6). In (C) and
   (F) to (H), data are presented as means ± SD from at least three
   independent experiments or biological replicates, and two-tailed
   Student’s t test was used. *P < 0.05; **P < 0.01; ***P < 0.001; ****P <
   0.0001.

PHF8 is essential for lipid deposition in VHL-deficient ccRCC

   Multiple lines of evidence indicate that VHL/HIF axis plays a crucial
   role in the initiation and development of ccRCC. Consistent with more
   recent findings, results from our preliminary study revealed that VHL
   reexpression (ReVHL) in 786-O cells led to decreased lipid droplet
   formation ([106]Fig. 2A and fig. S2A), substantiating the hypothesis
   that VHL deficiency is associated with lipid deposition in ccRCC
   ([107]10). Because the VHL/HIF axis is involved in the regulation of
   JmjC-containing epigenetic regulators ([108]20, [109]21), we detected
   21 JmjC-containing epigenetic regulators and found that seven of them
   were down-regulated significantly (mRNA FC [ReVHL/EV] < 0.5) after
   ReVHL in 786-O cells ([110]Fig. 2B and fig. S2B). When we knocked down
   these seven genes including JHDM1A, JHNDM2A, JARID1A, JARD1C, UTX,
   PHF8, and PHF2 (fig. S2C), PHF8 is found to be the key JmjC-containing
   epigenetic regulator involved in lipid deposition of ccRCC, revealed by
   flow cytometry using the BODIPY 493/503 probe ([111]Fig. 2C).
   Overexpression of PHF8 in 786-O ReVHL cells reversed ReVHL-induced
   reduction of lipid deposition ([112]Fig. 2D and fig. S2D), suggesting
   that PHF8 is essential for VHL deficiency–induced lipid deposition.
   Furthermore, we found that the levels of PHF8, HIF1α, and HIF2α were
   all down-regulated by reexpressed VHL in VHL-null 786-O cells
   ([113]Fig. 2E). Silencing either HIF1A or HIF2A resulted in
   down-regulation of the mRNA levels of PHF8, as well as their
   corresponding target genes, while overexpression either HIF1A or HIF2A
   led to the opposite effects (fig. S2, E and F), indicating that HIF1α
   and HIF2α play important role in PHF8 up-regulation in ccRCC.
   Furthermore, chromatin immunoprecipitation (ChIP)–qPCR results
   demonstrated that besides their well-known downstream genes, both HIF1α
   and HIF2α were recruited to the promoter of PHF8 ([114]Fig. 2, F and
   G), and luciferase reporter assay showed that overexpressed HIF1α and
   HIF2α can enhance PHF8 transcription ([115]Fig. 2H). In addition, the
   mRNA and protein levels of PHF8 ([116]Fig. 2, I and J, and fig. S2G),
   as well as lipid deposition ([117]Fig. 2K), can be rescued by
   overexpression of the undegradable HIF1α (HIF1A^P402A/P564A, mutHIF1A)
   and HIF2α (HIF2A^P405A/P531A, mutHIF2A) in 786-O ReVHL cells. These
   data suggest that VHL-regulated PHF8 is HIF1α and HIF2α dependent and
   PHF8 is essential for lipid deposition in VHL-deficient ccRCC.

Fig. 2. PHF8 is essential for VHL deficiency–induced lipid deposition of
ccRCC.

   [118]Fig. 2.
   [119]Open in a new tab

   (A) Representative images and quantification data of lipid droplets
   stained with BODIPY 493/503 (green) in 786-O with or without VHL
   reexpression (ReVHL). Nuclei were stained with DAPI (blue; scale bar,
   50 μm). EV, empty vector. (B) Real-time quantitative polymerase chain
   reaction (RT-qPCR) analysis of relative mRNA expressions of seven
   JmjC-containing epigenetic regulators in 786-O with or without ReVHL.
   (C) Flow cytometric analysis of lipid droplets in 786-O cells
   transfected with siRNAs targeting indicated genes using BODIPY 493/503.
   (D) Representative images and quantification data of lipid droplets
   stained with ORO (red) in indicated 786-O cells. Nuclei stained with
   Mayer’s hematoxylin (scale bar, 100 μm in 400×). (E) WB analysis of the
   indicated proteins in 786-O cells with or without ReVHL. (F) Chromatin
   immunoprecipitation (ChIP)–qPCR analysis of hypoxia-inducible factor 1α
   (HIF1α) binding on the promoter regions of PHF8 and GLUT1. IgG,
   immunoglobulin G. (G) ChIP-qPCR analysis of HIF2α binding on the
   promoter regions of PHF8 and EPO. (H) PHF8 promoter–driven luciferase
   reporters and luciferase activity (LUC) in 786-O cells with
   overexpressed HIF1α and HIF2α. (I and J) RT-qPCR analysis of relative
   mRNA expressions of indicated genes in 786-O ReVHL cells with or
   without overexpression of nondegradable mutHIF1A (P402A/P564A) (I) or
   mutHIF2A (P405A/P531A) (J). (K) Representative images and
   quantification data of lipid droplets stained with ORO (red) in
   indicated 786-O cells. Nuclei stained with Mayer’s hematoxylin (scale
   bar, 100 μm in 400×). In (A), (B), (D), and (F) to (K), data are
   presented as means ± SD from at least three independent experiments or
   biological replicates. Two-tailed Student’s t test was used for (A),
   (B), and (F) to (H), and ordinary one-way analysis of variance (ANOVA)
   was used for (D) and (I) to (K). *P < 0.05; **P < 0.01; ***P < 0.001;
   ****P < 0.0001.

PHF8 promotes ccRCC tumor growth and lipid deposition in a
demethylation-dependent manner

   Previous studies have shown that lipid deposition can promote tumor
   growth by maintaining endoplasmic reticulum integrity ([120]32) and
   cell membrane fluidity ([121]33) as well as activating fatty acid
   receptors and subsequent downstream signaling cascade ([122]34,
   [123]35). By interfering lipid synthesis using siRNA targeting fatty
   acid synthase (FASN; fig. S3A), we found that reduction in lipid
   deposition led to compromised ccRCC cell proliferation (fig. S3, B and
   C). Moreover, knockdown of FASN in PHF8-overexpressed cells repressed
   both PHF8-induced lipid deposition and cell proliferation (fig. S3, D
   to F). Therefore, lipid deposition is closely associated with
   PHF8-induced ccRCC tumor growth. Given that PHF8 is a well-established
   histone demethylase, we then investigated its effect on lipid
   deposition and tumor growth through epigenetic regulation. We depleted
   PHF8 in 786-O and 769-P cells using sgRNA and found that both cell
   viability and proliferation reduced greatly ([124]Fig. 3, A to D). PHF8
   silencing in the cells derived from organoid ccRCC02 also impaired the
   formation of organoid ([125]Fig. 3E). We then reintroduced either the
   wild-type (WT) or catalytically inactive mutant (H283A for
   [126]NP_001171825.1) PHF8 to the 786-O cells transfected PHF8-sgRNA and
   found that only the WT PHF8 but not mutant PHF8 can demethylate
   H3K9me2/1 ([127]Fig. 3F) and able to rescue cell viability and
   proliferation ([128]Fig. 3, G and H) as well as the in vivo phenotype
   ([129]Fig. 3, I and J) evidenced by the intensity of Ki67 staining
   ([130]Fig. 3, K and L). In addition, the WT PHF8 but not mutant PHF8
   can rescue the lipid droplet formation induced by PHF8 knockout in
   786-O cells in vitro (fig. S3G) and 786-O cell–derived xenografts in
   vivo (fig. S3H). Together, these results demonstrated the essentiality
   of PHF8’s demethylase activity in promoting ccRCC tumor growth and
   lipid deposition.

Fig. 3. PHF8 promotes ccRCC tumor growth in a demethylation-dependent manner.

   [131]Fig. 3.
   [132]Open in a new tab

   (A and B) Representative images (A) and quantification data (B) of
   colony formation assay of ccRCC cell lines (786-O and 769-P)
   transfected with PHF8-sgRNA or Ctrl-sgRNA. (C and D) Representative
   images (C) and quantification data (D) of 5-ethynyl-2′-deoxyuridine
   (EdU; green) assay of ccRCC cell lines (786-O and 769-P) transfected
   with PHF8-sgRNA or Ctrl-sgRNA. Nuclei were stained by DAPI (blue; scale
   bar, 100 μm in 400×). (E) Representative images and quantification data
   of organoids transfected with PHF8-sgRNA or Ctrl-sgRNA (scale bar, 100
   μm in 200×). (F) WB analysis of indicated proteins in 786-O cells
   transfected with Ctrl-sgRNA, PHF8-sgRNA with or without wild-type (WT)
   PHF8, or PHF8^H283A overexpression. (G) Relative optical density at 450
   nm (OD[450]) values of indicated 786-O cells detected by cell counting
   kit-8 (CCK-8) assay. n.s., not significant. (H) Representative images
   and quantification data of EdU (green) assay of indicated 786-O cells.
   Nuclei were stained by DAPI (blue; scale bar, 100 μm in 400×). (I and
   J) Images (I) and tumor volumes (J) of indicated 786-O cell–derived
   xenograft tumors (n = 7 per group). (K and L) Representative images (K)
   and quantification data (L) for immunohistochemical (IHC) staining of
   Ki67 in indicated 786-O cell–derived xenograft (scale bar, 100 μm in
   400×). In (B), (D), (E), (G), (H), (J), and (L), data are presented as
   means ± SD from at least three independent experiments or biological
   replicates. Two-tailed Student’s t test was used for (B), (D), and (E),
   and ordinary one-way ANOVA test was used for (G), (H), (J), and (L). *P
   < 0.05; **P < 0.01; ***P < 0.001.

PHF8-GLUL axis plays an essential role in ccRCC tumor growth and lipid
deposition

   To further investigate the underlying molecular mechanism in
   PHF8-mediated ccRCC tumor growth and lipid accumulation, RNA sequencing
   (RNA-seq) was conducted, and 3575 differentially expressed genes were
   identified when PHF8 is knocked out in 786-O cells (triplicates per
   group; data are deposited in [133]GSE214183). Results from Kyoto
   Encyclopedia of Genes and Genomes (KEGG) pathway analysis showed that
   26.3% of differentially expressed genes were metabolism related
   ([134]Fig. 4A) in accordance with the important role of PHF8 in lipid
   metabolism. Of note, metabolic pathways involving alanine, aspartate,
   and glutamate were the most markedly affected ([135]Fig. 4B), and this
   notion is further confirmed by RT-qPCR (fig. S4A). Among the
   differentially expressed genes, GLUL, encoding an enzyme responsible
   for glutamine de novo synthesis, is affected most severely (log[2]FC =
   −10.98) at both mRNA ([136]Fig. 4C) and protein ([137]Fig. 4D). In line
   with that GLUL-mediated glutamine synthesis is important for tumor cell
   proliferation ([138]16, [139]18), we found that silencing GLUL in 786-O
   and 769-P cells not only reduced cell proliferation but also the levels
   of glutamine (fig. S4, B to D). Reintroduced GLUL in PHF8-depleted
   786-O and 769-P cells ([140]Fig. 4D) is capable of rescuing glutamine
   and lipid deposition (fig. S4, E to H), cell proliferation in vitro
   ([141]Fig. 4E), and tumor growth in vivo ([142]Fig. 4, F and G). The
   restored GLUL is also capable of rescuing organoid formation and lipid
   deposition ([143]Fig. 4H and fig. S4, I and J). In addition, acetyl-CoA
   acetyltransferase 2 and enoyl-CoA delta isomerase 1, two enzymes
   responsible for lipolysis, were up-regulated when PHF8 is knocked out
   (fig. S4, K and L). Because PHF8 is also recruited to their promoter
   (fig. S4M), PHF8 likely plays some role in lipolysis by acting as a
   co-repressor as reported previously ([144]36). However, results of
   carbon flux by labeling ^13C[5] glutamate substantiated that reductive
   carboxylation signaling is decreased in GLUL knockdown cells, evidenced
   by that m+5 α-ketoglutarate and m+5 citrate, two important intermediate
   metabolites in reductive carboxylation signaling, were lower when GLUL
   was silenced (fig. S4N and table S5). Given the essential roles of
   reductive carboxylation signaling in fatty acid synthesis in ccRCC
   ([145]8, [146]14), these results together indicated that GLUL plays an
   indispensable role in PHF8-regulated ccRCC tumor growth and lipid
   deposition.

Fig. 4. GLUL expression is controlled by PHF8 and is essential for ccRCC cell
proliferation.

   [147]Fig. 4.
   [148]Open in a new tab

   (A) Pie chart showing the metabolism-related gene clusters of the
   differential genes using Kyoto Encyclopedia of Genes and Genomes (KEGG)
   method based on RNA sequencing (RNA-seq) after depleting PHF8 in 786-O
   cells. (B) Histogram showing the top 15 significantly enriched pathways
   using KEGG method. TNF, tumor necrosis factor. (C) Heatmap showing the
   top 15 differential genes. (D) WB analysis of indicated proteins in
   786-O and 769-P cells transfected with Ctrl-sgRNA and PHF8-sgRNA with
   or without glutamate-ammonia ligase (GLUL) overexpression. (E)
   Representative images and quantification data of EdU (green) assay of
   ccRCC cell lines (786-O and 769-P) transfected with Ctrl-sgRNA and
   PHF8-sgRNA with or without GLUL overexpression. Nuclei were stained by
   DAPI (blue; scale bar, 100 μm in 400×). (F) Images and tumor volumes of
   indicated 786-O cell–derived xenograft tumors (n = 5 in Ctrl-sgRNA
   group and n = 6 in both of PHF8-sgRNA and PHF8-sgRNA + GLUL groups).
   (G) Representative images and quantification data for IHC staining of
   Ki67 in indicated 786-O cell–derived xenografts (scale bar, 100 μm in
   400×). (H) Representative images and quantification data of organoids
   transfected with Ctrl-sgRNA and PHF8-sgRNA with or without GLUL
   overexpression (scale bar, 100 μm in 200×). In (E) to (H), data are
   presented as means ± SD from at least three independent experiments or
   biological replicates, and ordinary one-way ANOVA test was used. *P <
   0.05; **P < 0.01.

PHF8 up-regulates GLUL by transcriptional up-regulating TEAD1

   Consistent with the RNA-seq data, PHF8 knockout led to reduced mRNA
   levels of GLUL in both 786-O and 769-P cells (figs. S4A and S5A),
   suggesting that PHF8 may up-regulate GLUL transcriptionally. However,
   ChIP-qPCR and ChIP sequencing (ChIP-seq; data are deposited in
   [149]GSE217161) failed to show the binding of PHF8 on the promoter
   region of GLUL, suggesting that PHF8 likely regulates GLUL indirectly
   (fig. S5, B and C). By conducting KEGG pathway enrichment analysis
   using gene set enrichment analysis (GSEA), we found that the hippo
   signaling pathway is most severely affected by PHF8 knockout
   ([150]Fig. 5A), which is further validated by GSEA analysis (enrichment
   score = −0.63, adjust P = 0.0155; [151]Fig. 5B). It has been reported
   that Yes1 associated transcriptional regulator (YAP1)/transcriptional
   coactivator with PDZ-binding motif/TEAD complex could regulate GLUL
   transcription by binding to its promoter ([152]37). We performed
   ChIP-qPCR and dual luciferase reporter assays (fig. S5, D and E) and
   confirmed that YAP1 occupies GLUL promoter and regulates its
   expression. Results from the RNA-seq data indicate that most of the
   genes in hippo signaling pathway including YAP1, TEAD1, TEAD2, TEAD3,
   CYR61, and CTGF were down-regulated in the absence of PHF8 (fig. S5F).
   We then focused on TEAD1, because (i) reduced mRNA and protein levels
   of TEAD1 were consistently observed when PHF8 is knocked out
   ([153]Fig. 5C and fig. S5, F and G) and (ii) high copy number of TEAD1
   is associated with worse ccRCC prognoses ([154]Fig. 5C). Results from
   ChIP-seq targeting TEAD1 (ENCODE database, [155]www.encodeproject.org)
   suggested that TEAD1 is enriched at the genomic loci of GLUL promoter
   (fig. S5H). We conducted ChIP-qPCR and dual luciferase reporter assays
   to verify the recruitment of TEAD1 to the GLUL promoter (fig. S5, D and
   E). Moreover, reexpressed PHF8 in PHF8-sgRNA cells enhanced the
   recruitment of TEAD1 to the GLUL promoter (fig. S5I). In addition,
   overexpressed TEAD1 can up-regulate the mRNA ([156]Fig. 5D) and protein
   ([157]Fig. 5E) levels of GLUL in 786-O PHF8-sgRNA cells. Similar
   results were obtained when TEAD1 is overexpressed in the ccRCC02
   PHF8-sgRNA organoid ([158]Fig. 5F). Overexpressed TEAD1 is also capable
   of enhancing cell proliferation, tumor growth ([159]Fig. 5, G to I; and
   fig. S5, J and K), and lipid deposition ([160]Fig. 5J). To ascertain
   the necessity of TEAD1 in PHF8/GLUL-mediated cell growth and lipid
   deposition, we reexpressed PHF8 and knocked down TEAD1 in 786-O
   PHF8-sgRNA cells and showed that down-regulating TEAD1 can repress
   PHF8-induced GLUL expression ([161]Fig. 5K), cell growth, and lipid
   deposition evidenced by results from 5-ethynyl-2′-deoxyuridine (EdU;
   [162]Fig. 5L) and ORO staining ([163]Fig. 5M). Moreover, overexpressed
   GLUL in TEAD1 knockdown 786-O cells is capable of restoring lipid
   deposition and cell proliferation, which further demonstrated the
   essentiality of GLUL in TEAD1-mediated phenotypes (fig. S5, L to N).
   Consistent with above-mentioned essentiality of demethylase activity,
   the mRNA levels of TEAD1 and GLUL can only be rescued by WT PHF8 but
   not the mutant PHF8 ([164]Fig. 5N). These results suggested that PHF8
   up-regulates GLUL by transcriptionally up-regulating TEAD1 in a
   demethylase-dependent manner.

Fig. 5. The regulatory effect of PHF8 on GLUL depends on TEAD1.

   [165]Fig. 5.
   [166]Open in a new tab

   (A) Histogram showing the top 10 enriched pathways based on RNA-seq.
   (B) Gene set enrichment analysis (GSEA) enrichment plot of the hippo
   signaling pathway. (C) Integration of mRNA (RNA-seq) and protein
   changes after PHF8 knockout and the prognosis in ccRCC. The prognosis
   was evaluated by analyzing the association of copy number segments of
   indicated genes with survival data from The Cancer Genome Atlas. (D)
   RT-qPCR analysis of indicated genes in indicated 786-O cells. (E) WB
   analysis of indicated proteins in indicated 786-O cells. (F) RT-qPCR
   analysis of indicated genes in organoids transfected with indicated
   lentivirus. (G) EdU (green) assay of indicated 786-O cells. Nuclei were
   stained by DAPI (blue; scale bar, 100 μm in 400×). (H) Images and tumor
   volumes of indicated 786-O cell–derived xenograft tumors (n = 5 in both
   of Ctrl-sgRNA and PHF8-sgRNA + TEAD1 groups and n = 6 in PHF8-sgRNA
   group). (I) Representative images and quantification data of the
   organoids transfected with indicated lentivirus (scale bar, 100 μm in
   200×). (J) Representative images and quantification data of lipid
   droplets stained with ORO (red) in indicated cells. Nuclei stained with
   Mayer’s hematoxylin (scale bar, 100 μm in 400×). (K) WB analysis of
   indicated proteins in indicated 786-O cells. (L) EdU (green) assay of
   indicated cells. Nuclei were stained by DAPI (blue; scale bar, 100 μm
   in 400×). (M) ORO (red) staining showing lipid droplets in indicated
   cells. Nuclei stained with Mayer’s hematoxylin (scale bar, 100 μm in
   400×). (N) RT-qPCR analysis of indicated genes in indicated 786-O
   cells. In (D), (F) to (J), and (L) to (N), data are presented as means
   ± SD from at least three independent experiments or biological
   replicates. Ordinary one-way ANOVA test was used. *P < 0.05; **P <
   0.01; ***P < 0.001; ****P < 0.0001.

PHF8 is recruited by c-MYC to up-regulate TEAD1

   On the basis that (i) both c-MYC and PHF8 play important roles in
   metabolism ([167]38), (ii) PHF8 can form a complex with c-MYC
   ([168]26), and (iii) altered signature of c-MYC (P = 0.0149) was found
   after PHF8 knockout in 786-O cells when GSEA analysis was conducted
   with “HALLMARK_MYC_TARGETS_V2” gene set ([169]Fig. 6A), we examined
   whether PHF8 regulates TEAD1 through the PHF8/c-MYC complex.
   Endogenously expressed PHF8 and c-MYC form a complex in both 786-O and
   769-P cells ([170]Fig. 6B and fig. S6A). However, the interaction
   between PHF8 and c-MYC did not require its demethylase activity
   evidenced by that both WT and mutant PHF8 complexed with c-MYC in human
   embryonic kidney 293T (HEK293T) cells ([171]Fig. 6C and fig. S6B).
   Moreover, down-regulated mRNA and protein levels of TEAD1 were observed
   when c-MYC was silenced (fig. S6, C and D). Consistently, c-MYC
   overexpression can regulate the mRNA levels of TEAD1 (fig. S6E).
   Results from ChIP-seq in 786-O cells (data are deposited in
   [172]GSE217161) showed that ~71% of c-MYC binding sites overlap with
   that of PHF8 (P = 0.0012; [173]Fig. 6D and fig. S6F) and metagene plot
   analyses showed highly overlapped chromatin-binding profiles around the
   transcriptional start sites (TSSs) between PHF8 and c-MYC
   ([174]Fig. 6E). In addition, [175]Fig. 6F shows that PHF8 and c-MYC
   co-occupy the TEAD1 promoter region. Results from ChIP-qPCR indicate
   that both PHF8 and c-MYC were enriched on the TEAD1 promoter
   ([176]Fig. 6G). Because knockdown of c-MYC reduced PHF8 enrichment on
   the TEAD1 promoter ([177]Fig. 6H) and PHF8 knockout did not affect
   c-MYC enrichment ([178]Fig. 6I), although silencing either of them
   reduced RNA–polymerase II (pol II) enrichment on TEAD1 promoter, we
   conclude that PHF8 is recruited to the TEAD1 promoter by c-MYC. This
   conclusion is further substantiated by analyzing the PHF8 ChIP-seq in
   c-MYC knockdown cells and c-MYC ChIP-seq in PHF8-sgRNA cells. The
   enrichment of PHF8 at genomic loci including TEAD1 promoter declined
   markedly when c-MYC is knocked down ([179]Fig. 6, F and J), while the
   enrichment of c-MYC remained unchanged when PHF8 is knocked out
   ([180]Fig. 6, F and K). Of note, the levels of H3K9me1 and H3K9me2 on
   TEAD1 promoter increased significantly when PHF8 is knocked down
   ([181]Fig. 6I). The results from TEAD1 promoter–driven luciferase
   assays demonstrated that TEAD1 is under the control of both PHF8 and
   c-MYC ([182]Fig. 6L) and that the demethylase activity of PHF8 is
   indispensable for its regulation ([183]Fig. 6M). These results indicate
   that PHF8, recruited to the TEAD1 promoter by c-MYC, up-regulates TEAD1
   through epigenetic modification.

Fig. 6. PHF8 interacts with c-MYC to up-regulate TEAD1 in a
demethylation-dependent manner.

   [184]Fig. 6.
   [185]Open in a new tab

   (A) GSEA enrichment plot of the HALLMMARK_MYC_TARGETS_V2 gene set based
   on the RNA-seq data. (B) WB analysis of whole-cell lysates (WCLs) and
   immunoprecipitates (IPs) with PHF8 and c-MYC antibodies in 786-O cells.
   (C) WB analysis of WCLs and immunoprecipitates with Flag and HA
   antibodies in 293T cells transfected with the indicated plasmids. (D)
   Venn diagram showing the common peaks of PHF8 and c-MYC in the ChIP-seq
   results. (E) Metagene plot of ChIP-seq profiles of PHF8 and c-MYC peaks
   on the transcriptional start site (TSS) region. RPKM, reads per
   kilobase per million mapped reads. (F) ChIP-seq tracks from indicated
   cells at the genomic loci of TEAD1. (G) ChIP-qPCR assays of PHF8 and
   c-MYC binding on the promoter region of TEAD1. (H) ChIP-qPCR analysis
   of PHF8, c-MYC, and RNA–polymerase II (Pol II) binding on the promoter
   region of TEAD1 in 786-O cells transfected with siRNA targeting c-MYC
   and with scrambled siRNA as control. (I) ChIP-qPCR analysis of PHF8,
   c-MYC, RNA-Pol II, H3K9me1, and H3K9me2 binding on the promoter region
   of TEAD1 in 786-O cells transfected with Ctrl-sgRNA and PHF8-sgRNA. (J)
   Metagene profile and bar plot of PHF8 binding in 786-O cells
   transfected with siRNA targeting c-MYC and with scrambled siRNA as
   control. PC, peak center. (K) Metagene profile and bar plot of c-MYC
   binding in 786-O cells transfected with or without PHF8-sgRNA. (L)
   TEAD1 promoter–driven luciferase reporter assay in 786-O cells
   transfected with EV, WT PHF8 or c-MYC. OE, overexpressed. (M) TEAD1
   promoter–driven luciferase reporter assay in 786-O cells transfected
   with EV, WT PHF8, or PHF8^H283A. In (G) to (M), data are presented as
   means ± SD from at least three independent experiments or biological
   replicates. Two-tailed Student’s t test was used. *P < 0.05; **P <
   0.01; ***P < 0.001; ****P < 0.0001.

The PHF8/TEAD1/GLUL axis is clinically relevant in ccRCC

   To evaluate the clinical significance of PHF8/TEAD1/GLUL axis, we
   compared the mRNA level of PHF8 and protein levels of PHF8, TEAD1, and
   GLUL in tumor and their adjacent normal tissues by RT-qPCR and Western
   blotting (WB), respectively ([186]Fig. 7, A to C), and found that all
   of them are higher in tumor tissues. In addition, immunohistochemical
   (IHC) staining ([187]Fig. 7D) revealed that the levels of PHF8, TEAD1,
   and GLUL are not only positively correlated with each other
   ([188]Fig. 7E) but also highly associated with both higher Fuhrman
   grade ([189]Fig. 7F) and poorer prognosis ([190]Fig. 7G). These
   findings were further substantiated by the results from analyzing the
   public datasets. The mRNA levels of GLUL and TEAD1 are positively
   correlated with that of PHF8 (fig. S7A). The levels of PHF8, GLUL, and
   TEAD1 are higher in tumor tissues when proteomic data from Clinical
   Proteomic Tumor Analysis Consortium
   ([191]http://ualcan.path.uab.edu/index.html) ([192]39) were analyzed
   (fig. S7, B to D). Higher copy number segments of GLUL and TEAD1
   correlates with shorter survival (fig. S7, E and F;
   [193]https://xena.ucsc.edu/kaplan-survival-analysis/). These data imply
   the clinical importance of PHF8/TEAD1/GLUL axis in ccRCC prognosis.

Fig. 7. PHF8/TEAD1/GLUL axis is clinically important for ccRCC progression.

   [194]Fig. 7.
   [195]Open in a new tab

   (A) RT-qPCR analysis of relative mRNA levels of PHF8 in paired adjacent
   and tumor tissues (n = 20). (B and C) WB analysis (B) and mean gray
   scale values (C) of indicated proteins in paired adjacent and ccRCC
   tissues (n = 20). (D) Representative images for IHC staining of PHF8,
   TEAD1 and GLUL in our ccRCC tissue microarray (TMA; n = 117; scale bar,
   100 μm in 400×). (E) Heatmap showing the H score of each ccRCC sample
   (n = 117) and illustrating the Spearman’s correlation coefficient (R).
   (F) Violin plot showing that H score distribution between different
   Fuhrman grades. (G) Kaplan-Meier plot of survival based on the H score
   of PHF8, TEAD1, and GLUL (high group: H score > 4, low group: H score ≤
   4; n = 117). HR, hazard ratio; CI, confidence interval. In (A) and (C),
   data are presented as scatter point. In (F), data are presented using
   violin plots. In (G), data are presented using Kaplan-Meier plots of
   survival. Paired two-tailed Student’s t test was used for (A) and (C),
   Kruskal-Wallis test was used for (F), log-rank test was used for (G),
   and Spearman’s correlation analysis was used for (E). **P < 0.01; ***P
   < 0.001; ****P < 0.0001.

Targeting GLUL is an efficacious therapeutic strategy against ccRCC

   To test whether GLUL can serve as a potential therapeutic target for
   ccRCC, we examined the effect of an irreversible selective GLUL
   inhibitor l-methionine sulfoximine (MSO) on cell proliferation. The
   786-O and 769-P cells transfected with either control (Ctrl)–sgRNA or
   PHF8-sgRNA were treated with MSO in glutamine-free medium as reported
   previously ([196]16). The cell proliferation was inhibited
   significantly by MSO only when PHF8 is intact but not silenced
   ([197]Fig. 8, A and B, and fig. S8A). The size of the tumors
   ([198]Fig. 8C) and the number of Ki67-positive cells in the tumor
   ([199]Fig. 8D) derived from the 786-O xenograft reduced significantly
   by MSO treatment. However, these effects disappeared when PHF8 is
   silenced. Similar findings were obtained when the organoids were
   treated with MSO ([200]Fig. 8E). Consistent with previous studies
   ([201]40, [202]41), we found that GLUL inhibition also activated the
   mechanistic target of rapamycin kinase (mTOR) signaling pathway in
   786-O cells ([203]Fig. 8F). We hypothesized that MSO may function
   synergistically with the mTOR signaling pathway inhibitor EVE, a
   second-line drug for ccRCC. To test our hypothesis, we conducted a
   series of in vitro and in vivo experiments with either MSO or EVE alone
   or in combination. Our results showed that the number of proliferating
   cells ([204]Fig. 8G and fig. S8, B and C), the size of the organoids
   ([205]Fig. 8H) and 786-O–derived xenograft ([206]Fig. 8I) or
   patient-derived xenografts (PDXs; [207]Fig. 8J), as well as the number
   of Ki67-positive cells ([208]Fig. 8, K and L) can be repressed by
   either of them alone and combination of these two drugs that functioned
   synergistically. In addition, MSO can inhibit lipid deposition
   significantly when the cells expressing intact PHF8 but not when PHF8
   is silenced (fig. S8, D to G), whereas EVE cannot repress the lipid
   deposition (fig. S8, F and G). These data together demonstrated that
   MSO alone, but more efficiently in combination with EVE, can repress
   not only lipid deposition but also ccRCC progression.

Fig. 8. GLUL inhibition suppresses ccRCC growth and enhances the anti-cancer
effects of EVE.

   [209]Fig. 8.
   [210]Open in a new tab

   (A and B) Representative images (A) and quantification data (B) of EdU
   (green) assay in 786-O and 769-P cells treated with indicated regimens.
   Nuclei were stained by DAPI (blue; scale bar, 100 μm in 400×). (C)
   Images and tumor volumes of xenograft tumors formed by indicated 786-O
   cells with or without l-methionine sulfoximine (MSO) treatment (n = 5
   in Ctrl-sgRNA group and n = 6 in PHF8-sgRNA group as well as Ctrl-sgRNA
   and PHF8-sgRNA treated with MSO groups). ‰, per mil. (D) Ki67 staining
   in indicated 786-O cell–derived xenografts with indicated treatments
   (scale bar, 100 μm in 400×). (E) Organoids transfected with Ctrl-sgRNA
   and PHF8-sgRNA treated with indicated regimens (scale bar, 100 μm in
   200×). (F) WB analysis of the indicated proteins in 786-O cells treated
   with or without MSO (30 μM) for indicated time points. (G)
   Quantification data of EdU (green) assay of 786-O and 769-P cells
   treated with indicated regimens. EVE, everolimus. (H) Organoids treated
   with indicated regimens (scale bar, 100 μm in 200×). (I) Images and
   tumor volumes of xenograft tumors formed by 786-O cells treated with
   indicated regimens (n = 6 per group). (J) Images and tumor volumes of
   patient-derived xenografts (PDXs) treated with indicated regimens (n =
   5 per group). (K) Ki67 staining in 786-O cell–derived xenografts
   treated with indicated regimens (scale bar, 100 μm in 400×). (L) Ki67
   staining in PDXs treated with indicated regimens (scale bar, 100 μm in
   400×). Glutamine-free medium used in (A), (B), and (E) to (H). In (B)
   to (E) and (G) to (L), data are presented as means ± SD from at least
   three independent experiments or biological replicates. Ordinary
   one-way ANOVA test was used. *P < 0.05; **P < 0.01; ***P < 0.001; ****P
   < 0.0001.

DISCUSSION

   In this study, we reported a previously unidentified regulatory axis
   involved in tumor growth and lipid deposition in ccRCC. We identified
   PHF8 as an important epigenetic regulator involved in VHL/HIF-mediated
   tumor growth and lipid deposition. Mechanistically, PHF8 interacts with
   c-MYC and transcriptionally activates TEAD1 in a
   demethylation-dependent manner. Up-regulated TEAD1 subsequently
   enhances GLUL transcription by binding to its promoter directly and
   ultimately promotes tumor growth and lipid deposition ([211]Fig. 9).
   Activated PHF8/TEAD1/GLUL axis is also associated with higher Fuhrman
   grade and poorer prognosis of patients with ccRCC. Therapeutically,
   GLUL inhibitor MSO can efficiently suppress the tumor growth and lipid
   deposition as well as enhance the anticancer effect of EVE
   ([212]Fig. 9), which provides a previously unknown avenue for ccRCC
   treatment.

Fig. 9. Schematic illustration of the proposed model in which PHF8/TEAD1/GLUL
axis regulates VHL-mediated lipid deposition and tumor growth of ccRCC.

   [213]Fig. 9.
   [214]Open in a new tab

   Lipid deposition plays critical roles in development, progression, and
   drug resistance of multiple cancers including colorectal cancer, lung
   cancer, breast cancer, glioblastoma, and hepatocellular carcinoma
   ([215]4, [216]5, [217]42). Lipid deposition is recognized as a
   prominent feature and associated with malignancy, prognosis, and
   antiangiogenesis drug resistance in ccRCC ([218]3, [219]9). Qu et al.
   ([220]43) found that lipid deposition could promote ccRCC cells
   proliferation and xenograft growth. Li et al. ([221]44) reported that
   high lipid deposition contributes to metastasis of ccRCC. Lucarelli
   et al. ([222]45) showed that lipid accumulation is involved in ccRCC
   chemotherapy resistance. However, the underlying mechanism in lipid
   deposition is largely unknown. Recently, aberration of VHL/HIF axis has
   been linked to lipid deposition by reducing adipose triglyceride
   lipase–mediated lipolysis ([223]46) and carnitine palmitoyl transferase
   1A–regulated fatty acid β-oxidation ([224]10). Given that the important
   role of epigenetic regulation and VHL/HIF axis in ccRCC ([225]19), we
   explored whether epigenetic regulator is involved in VHL/HIF-mediated
   lipid deposition. By using unbiased CRISPR-Cas9 screening targeting
   1136 epigenetic regulators, we identified the histone demethylase PHF8
   as an important factor driving lipid deposition.

   Glutamine is crucial for cell proliferation, pH homeostasis, and redox
   balance by serving as an intermediate metabolite for protein, lipid,
   nucleotide, and glutathione ([226]47). Recently, the role of glutamine
   in tumorigenesis ([227]15), progression ([228]48), metastasis
   ([229]49), and drug resistance ([230]50) has attracted increasing
   attention. For the cells in solid cancers, glutamine demand could not
   be satisfied by the restricted availability of circulating glutamine
   ([231]17, [232]51). As a result, many tumors up-regulate GLUL to
   increase glutamine synthesis, especially at times of enhanced metabolic
   needs for high proliferation rate ([233]47). Tardito et al. ([234]15)
   found that tumor cells seem to prefer the endogenously synthesized
   glutamine over exogenous glutamine. Consistently, elevated levels of
   GLUL have been found in multiple cancer types with poor prognosis
   ([235]47). In this study, we first revealed that GLUL has important
   roles in ccRCC growth, and its expression is correlated with the
   severity and poor prognosis in patients with ccRCC. Given the fact that
   glutamine is accumulated while glutamine utilization is increased in
   ccRCC ([236]2, [237]8), GLUL-mediated endogenously glutamine synthesis
   might be the reason for glutamine accumulation. Therefore, GLUL
   inhibitor MSO is capable of repressing both tumor growth and lipid
   deposition. Notably, the inhibitory effect of MSO might depend on the
   expression of PHF8, because MSO treatment did not further reduce
   proliferation when GLUL is undetectable in PHF8 knockout cells
   ([238]Fig. 4D).

   Hippo and MYC signaling pathways are tightly related to multiple
   cancers. Hippo signaling pathway plays major roles in organ size
   control, cell proliferation, stem cell self-renewal, tumorigenesis, and
   drug resistance ([239]52, [240]53). Accumulative evidences suggested
   that hippo signaling pathway promotes ccRCC development, progression,
   invasion, and metastasis ([241]54). Carter et al. ([242]55) found that
   deletion of large tumor suppressor kinase 1/2, the important regulators
   of the hippo signaling pathway, leads to spontaneous RCC formation in
   mouse. On the other hand, MYC signaling involves in a wide range of
   oncogenic processes including proliferation, differentiation,
   metastasis, and apoptosis ([243]56). MYC signaling is also a
   well-documented pathway closely related to metabolic pathways including
   glutamine metabolism ([244]38). In our study, we found that there is a
   cross-talk between the hippo and MYC oncogenic signaling. We
   demonstrated that PHF8, as a “bridge” of the two oncogenic pathways, is
   required for the regulatory function of c-MYC on TEAD1.

   Although we have demonstrated that PHF8-GLUL axis plays important role
   in both tumor growth and lipid deposition and lipid deposition is
   closely related to tumor growth in ccRCC, the precise mechanism
   underlying how PHF8-induced lipid deposition drives ccRCC progression
   needs to be further investigated. Similarly, the importance of
   GLUL-mediated glutamine synthesis in ccRCC lipid deposition remains to
   be deciphered. Since GLUL can catalyze the synthesis of glutamine from
   TCA cycle–derived carbons ([245]15), it appears paradoxical that GLUL
   is crucial for lipid deposition, because glutaminase (GLS)-mediated
   glutamine catabolism via reductive carboxylation is the key process for
   lipid deposition in VHL-null cells. However, because GLUL is mainly
   located in the cytoplasm while GLS is mainly in the mitochondria
   ([246]www.proteinatlas.org) ([247]57), we speculate that GLUL might be
   involved in transforming excess cytoplasmic glutamate into glutamine,
   which is then transported to the mitochondria and subsequently becomes
   lipid synthesis precursors via GLS-mediated reductive carboxylation.
   Furthermore, GLUL might affect lipid synthesis indirectly, because GLUL
   is also involved in the transcriptional regulation of sterol regulatory
   element–binding protein 1 (SREBP1), the master regulator of cholesterol
   and lipid syntheses ([248]58). Although there could be other
   explanations for the important roles of GLUL in lipid deposition, the
   detailed mechanism underlying GLUL-involved lipid production needs to
   be further revealed, especially by labeling the endogenously
   synthesized glutamine and analyzing its subcellular localization.
   Further research in this field will not only better our understanding
   of the mechanism in GLUL-mediated lipid deposition in ccRCC but also
   pave the way for GLUL-related therapeutics.

MATERIALS AND METHODS

Experimental design

   These experiments were designed to reveal the roles of PHF8/TEAD1/GLUL
   axis in the lipid deposition and tumor progression of ccRCC and
   investigate the therapeutic effects of MSO targeting this axis on
   ccRCC. First, we used an unbiased CRISPR-Cas9 screening targeting 1136
   epigenetic regulators to identify that PHF8 is the determinant of lipid
   deposition of ccRCC. The subsequent experiments were focused on
   examining how PHF8 regulates the lipid deposition and tumor progression
   of ccRCC. Cell lines, organoids, PDX, and subcutaneous xenografts were
   used in this study. To investigate the underlying mechanism, we used
   co-IP (CoIP), ChIP-seq, ChIP-qPCR, dual luciferase reporter assay, and
   carbon metabolite analysis. Furthermore, we validated the important
   roles of this axis in ccRCC by using the data of patients with ccRCC
   from either our own cohort or public database. Last, we conducted a
   series of in vitro and in vivo experiments to demonstrate the
   therapeutic effect of MSO, targeting GLUL, alone, or in combination
   with EVE on ccRCC. The investigators were not blinded to the allocation
   of groups during experiments or subsequently during the analysis.
   Although statistical methods were not used to predetermine sample size,
   sample sizes were chosen on the basis of estimates from pilot
   experiments and previously published results. The n values and
   particular statistical methods are indicated in the figure legends and
   the “Statistical analysis” section.

Reagents and antibodies

   MSO (M5379) was purchased from Sigma-Aldrich (St. Louis, MO, USA), EVE
   (S1120) was purchased from Selleck Chemicals (Houston, TX, USA), BODIPY
   493/503 (D3922) was purchased from Thermo Fisher Scientific (Waltham,
   MA, USA). Primary antibodies used in this study are listed in table S6.

Experimental models and subject details

Cell culture

   The ccRCC cell lines 786-O [the American Type Culture Collection (ATCC)
   number: CRL-1932], 769-P (ATCC number: CRL-1933), ACHN (ATCC number:
   CRL-1611), HK-2 (ATCC number: CRL-2190), and HEK293T (ATCC number:
   CRL-11268) were purchased from Cell Bank of Shanghai Institutes for
   Biological Sciences (Shanghai, China). A498 (ATCC number: HTB-44), RCC4
   (European Collection of Authenticated Cell Cultures, cat. no.
   03112702), and OS-RC-2 (cell number: RCB0735, RIKEN Cell Bank, Tsukuba,
   Japan) were provided by G. Sun (Department of Urology, West China
   Hospital, Sichuan University). 786-O, 769-P, and OS-RC-2 cells were
   maintained in RPMI 1640 (C11875500BT, Gibco, Waltham, MA, USA)
   containing 10% fetal bovine serum (FBS; 10099141, Gibco). RCC4, ACHN,
   and HEK293T cells were cultured in Dulbecco’s modified Eagle’s medium
   (DMEM; 11965118, Gibco) supplemented with 10% FBS. A498, and HK-2 cells
   were cultured in MEM (SH30024.01, HyClone) supplemented with 10% FBS.
   The cells were incubated in 5% CO[2] at 37°C. According to the
   manufacturer’s instruction, ViaFect (E4981, Promega Corporation,
   Madison, WI, USA) was used for plasmids transfection, Lipofectamine
   2000 (11668019, Invitrogen, Carlsbad, CA, USA) was used for siRNA
   transfection, and polybrene (H9268, hexadimethrine bromide,
   Sigma-Aldrich) was used for lentivirus transfection.

Human tissue samples

   All procedures involving human participants were obeyed with the
   ethical standards of the institutional research committee, the 1964
   Helsinki Declaration and its later amendments, or comparable ethical
   standards. Written informed consent was obtained from each patient. All
   samples were collected by Department of Urology with approval from the
   Research Ethics Committee of Daping Hospital, Army Military Medical
   University (no. 2020-67, Chongqing, China). Briefly, fresh tumor
   tissues and matched normal tissues from 20 patients with ccRCC were
   subjected to immunoblotting and RT-qPCR. Human ccRCC tissue microarray
   (TMA) containing a total of 117 archived paraffin-embedded specimens
   were subjected to IHC staining. The information of the patients was
   listed in table S7.

Organoids

   Tumor tissues from 11 patients with ccRCC were collected to establish
   ccRCC organoids as described previously with some modifications
   ([249]59, [250]60). In brief, these tissues were cut into small pieces
   (1 to 2 mm in size), digested with type II collagenase (17101015,
   Gibco) for 120 min, and centrifuged at 350 g/s for 5 min at 4°C. The
   resuspended cells were embedded in 75% (v/v) Matrigel (Corning, NY,
   USA) solution and cultured with 500 μl of Advanced DMEM/F-12 (Gibco)
   supplemented with 5% FBS (Gibco), 1× B27 (17504044, Gibco), 10 mM
   Y-27632 dihydrochloride (M1817, AbMole BioScience, Houston, TX, USA),
   noggin (100 ng/ml), fibroblast growth factor 2 (20 ng/ml; 100-18B,
   PeproTech, Rocky Hill, NJ, USA), 1× ITS [insulin (5 μg/ml), transferrin
   (5 μg/ml), and selenium (5 ng/ml); I1884, Sigma-Aldrich], and 1×
   penicillin/streptomycin (15140122, Gibco) per well after Matrigel
   solidifying. The detailed clinical information of the patients was
   summarized in table S3.

Animals

   All animal experiments were approved by the Institutional Animal Care
   and Use Committee of Army Medical University (Chongqing, China) and in
   accordance with international laws (European Economic Community Council
   Directive 86/609, O.J. L 358.1, 12 December 1987; Guide for the Care
   and Use of Laboratory Animals, U.S. National Research Council, 1996).
   All mice were housed in the Animal Experimental Center of Daping
   Hospital with a 12/12-hour light/dark cycle at 25°C and were fed with
   standard chow and filtered sterile water. Athymic nude mice were
   obtained from Vital River company (Beijing, China).
   NOD-Prkdc(scid)IL2rgamma(null) (NPG) mice were purchased from Vitalstar
   (Beijing, China). 786-O cells (5 × 10^6) were injected subcutaneously
   into the flank of 4- to 6-week-old mice. Cell-derived xenografts were
   measured twice per week. Tumor volumes were calculated using the
   formula length × width × width/2.

Patient-derived xenograft

   Patients’ tumor samples were cut into 1 to 2 mm pieces, and the tissue
   fragments were seeded under the renal capsule for tissue amplification
   within 4- to 6-week-old NPG mice. For passaging or drug experiments,
   PDXs were collect, cut into 2-mm pieces, and implanted subcutaneously
   into the flank region. The drug treatment and measurement were
   conducted in accordance with the protocol of nude mice.

Method details

Gene silencing

   siRNAs targeting PHF8, JARID1A, JARID1C, JHDM1A, JHDMA2A, UTX, PHF2,
   TEAD1, GLUL, FASN, and c-MYC were purchased from RiboBio (RiboBio Co.
   Ltd., Guangzhou, Guangdong, China). siRNAs targeting TES, FRG1,
   GRIPAP1, LIN28A, and N6AMT1 were synthesized by Sangon (Sangon Biotech,
   Shanghai, China). All siRNAs were used to transiently knock down the
   indicated genes, with scrambled siRNA as the control. In brief, cells
   were cultured until 70 to 80% confluence in six-well were infected with
   50 nM siRNAs in presence of 5 μl of Lipofectamine 2000 (11668019,
   Thermo Fisher Scientific) according to the manufacturer’s protocol.
   After 24 hours of transfection, the cells were harvested for next
   manipulation. To stably knock out PHF8, sgRNA (four sgRNAs targeting
   human PHF8) was cloned into lentiCRISPR v2 vector and then expressed in
   lentivirus. Cells with 70 to 80% confluence were infected by lentivirus
   containing sgRNAs targeting PHF8 in presence of polybrene (8 ng/ml),
   and scrambled sgRNA was used as control. After 48 hours of
   transfection, puromycin was added to select infected cells until 5
   days. Then, a monoclonal cell population was isolated by limiting
   dilution method. The siRNA and sgRNA sequences are listed in table S8.

Gene overexpression

   The plasmids containing full-length cDNAs of VHL, GLUL, PHF8, and TEAD1
   were purchased from GeneCopoeia (Rockville, MD, USA). MYC-HIF2A was
   obtained from OriGene (RC208604). Hemagglutinin–c-MYC, Flag-HIF1A,
   Flag-PHF8, HIF1A^P402A/P564A, HIF2A^P405A/P531A, and PHF8^H283A were
   synthesized by Sangon. For transiently overexpression, the cells with
   70 to 80% confluence in six-well plate were infected with 3 μg of
   plasmids and 9 μl of ViaFect (Promega) for 24 hours. For stable
   overexpression, the plasmids were introduced into cells by lentiviral
   delivery in presence of polybrene (8 ng/ml). All the plasmids used were
   listed in table S6.

Epigenetic library screening in 786-O cells

   The sgRNAs of epigenetic library that targets 1136 genes (four sgRNAs
   per gene; table S1) was provided by B. Wang (Department of Gastrology,
   Daping Hospital) and cloned into lentiCRISPR v2 vector. 786-O cells
   were infected with lentivirus containing pooled epigenetic sgRNA for 12
   hours in the medium supplemented with polybrene (8 ng/ml). The cells
   were selected by puromycin (1 μg/ml) for 5 days and then harvested,
   fixed, and stained using BODIPY 493/503 (Thermo Fisher Scientific). The
   stained cells were classified into two subgroups with low or high fatty
   acid by flow cytometry. The sgRNA cassettes were amplified from genomic
   DNA using two-step PCR and deep sequencing adapters, and sample
   barcodes were added during the PCR. Last, sgRNA contents were measured
   with sequencing by GENEWIZ (Genewiz, Suzhou, China). Enrichment of
   sgRNAs in low fatty acid group was compared with that of high fatty
   acid group. Briefly, the libraries with different indices were
   multiplexed and loaded on an Illumina HiSeq/NovaSeq instrument
   according to the manufacturer’s instructions (Illumina, San Diego, CA,
   USA). Sequencing was carried out using a 2 × 150 paired-end
   configuration; image analysis and base calling were conducted by the
   HiSeq Control Software + Off-Line Basecaller + GAPipeline-1.6
   (Illumina) on the HiSeq instrument. The significantly changed genes
   were chosen by setting threshold at level of P < 0.05. The summary
   result was shown in table S2.

Mice treatment

   MSO was dissolved in phosphate-buffered saline (PBS) and injected
   intraperitoneally once every 2 days (10 mg kg^−1). EVE was dissolved in
   30% (v/v) propylene glycol, 5% (v/v) Tween 80, and 65% (v/v) ddH[2]O
   and given by gavage daily (2.5 mg kg^−1). For control group, mice were
   treated with 1 per mil (‰) dimethyl sulfoxide by gavage daily. All
   xenografts were measured twice per week, and their width and length
   were recorded. All xenografts were harvested and further subjected to
   IHC staining, ORO staining, or BODIPY 493/503 staining.

RNA extraction and RT-qPCR

   Total RNA was extracted using TRIzol Reagent (Jiangsu Cowin Biotech,
   Jiangsu, China) and purified using Ultrapure RNA Kit (CW0581M, CWBIO).
   First-strand cDNA was synthesized using the HiScript II Q RT SuperMix
   (R222-01, Vazyme, Nanjing, China) and 1 μg of purified RNA for each
   reaction. AceQ qPCR SYBR Green Master Mix (Q111-02, Vazyme) and primers
   were used for RT-qPCR. Each reaction was performed in at least
   triplicate. The relative mRNA expression was analyzed by using FC
   (2^−ΔΔCt). The primers used in our study were showed in table S8.

RNA sequencing

   Total RNA was extracted from 786-O cells transfected with Ctrl-sgRNA
   and PHF8-sgRNA. After quality controlling and RNA library preparing,
   triplicate samples were subjected to sequence by Shanghai NovelBio
   Bio-Pharm Technology Co. Ltd. (Shanghai, China). RNA-seq reads were
   filtered and mapped to human genome (GRCh38, National Center for
   Biotechnology Information) using Hierarchical Indexing for Spliced
   Alignment of Transcripts 2 ([251]61). The significance was listed by
   setting FCs threshold at a level of absolute value of log[2](FC) > 1
   and an FDR of <0.05 based on DEseq2 ([252]62). The differentially
   expressed gene lists were subsequently analyzed using gene ontology,
   KEGG, or GSEA method.

ChIP PCR

   The procedure of producing DNA fragments was according to the
   manufacturer’s introduction (53009, ChIP-IT Express Enzymatic kit,
   Active Motif). Briefly, about 1 × 10^7 cells were cross-linked in 1%
   formaldehyde (9027, 16% formaldehyde solution, Thermo Fisher
   Scientific) at room temperature for 10 min, following by 5 min of
   blocking with 1× glycine buffer. The cells were collected and lysed
   with lysis buffer on ice. The nuclei were collected through
   centrifuging at 5000 revolutions per minute (rpm) for 10 min at 4°C.
   After enzymatic digestion, shearing, and 0.5 M EDTA blocking, the
   sheared chromatin was incubated with proper antibody and magnetic beads
   on rotator at 4°C overnight. After wash and reverse cross-link, DNA was
   purified by using DNA Fragment purification kit (9761, Takara, Shiga,
   Japan) and quantified by RT-qPCR with specific primers (the primers
   used were listed in table S8). Results are presented as the percentage
   of the total input DNA [% input = 100 × 2^ (adjusted C[t] of input −
   C[t] of IP samples), adjusted C[t] of input = C[t] of input – log[2]
   dilution factor].

ChIP-seq

   ChIP-seq library preparation and data analysis were performed by Active
   Motif (Shanghai, China). In brief, cells were cross-linked with 1%
   formaldehyde and quenched with 125 mM glycine. The chromatin fragments
   were precleared and then immunoprecipitated with Protein G Magnetic
   Beads coupled with anti-PHF8 (A301-772A, Bethyl Laboratories) and anti
   c-MYC (9402, Cell Signaling Technology) antibodies. After reverse
   cross-linking, ChIP and input DNA fragments were end-repaired and
   A-tailed using the NEBNext End Repair/dA-Tailing Module [E7442, New
   England Biolabs (NEB)], followed by adaptor ligation with the NEBNext
   Ultra Ligation Module (E7445, NEB). The DNA libraries were amplified
   for 15 cycles and sequenced using Illumina NextSeq 500 with single-end
   1 × 75 as the sequencing mode. Raw reads were filtered to obtain
   high-quality clean reads by removing sequencing adapters, short reads
   (length: <35 base pairs), and low-quality reads using Cutadapt (v1.9.1)
   and Trimmomatic (v0.35). Then, FastQC is used to ensure high-quality
   reads. The clean reads were mapped to the human genome (assembly human
   genome GRCh38) using the Burrows-Wheeler Aligner software ([253]63).
   Peak detection was performed using the MACS (v2.1.1) peak finding
   algorithm ([254]64) with 0.05 set as the q value cutoff. Annotation of
   peak sites to gene features was performed using the ChIPseeker R
   package ([255]65).

CoIP and immunoblotting

   For CoIP, protein was extracted by Pierce IP Lysis Buffer (87787,
   Thermo Fisher Scientific) with cOmplete Tablets, Mini EDTA-free,
   EASYpack (04693159001, Roche, Basel, Switzerland), and PhosSTOP
   EASYpack (04906837001, Roche). Briefly, the lysates were incubated with
   certain antibodies at 4°C overnight. Then, lysates were incubated with
   Pierce Protein A Magnetic Beads (88846, Thermo Fisher Scientific) for
   an hour at room temperature. After being washed by lysis buffer, beads
   were heated with SDS loading and analyzed by WB. Antibodies used in
   this study were listed in table S6.

Western blotting

   Cultured cells or tumor tissues and matched normal tissues from
   patients with ccRCC were lysed with T-PER Tissue Protein Extraction
   Reagent (78510, Thermo Fisher Scientific) and a mixture of cOmplete
   Tablets, Mini EDTA-free, EASYpack, and PhosSTOP EASYpack. After
   centrifugation at 12,000 rpm and 4°C for 15 min, the supernatant of
   lysates was collected for WB according to standard procedures.
   Antibodies used in this study listed in table S6. Detection was
   performed by measuring chemiluminescence on ChemiDoc XRS+ System
   (Bio-Rad, Hercules, CA, USA).

IHC staining

   All fresh specimens were fixed in 4% paraformaldehyde (PFA) for 24
   hours, embedded by paraffin, and sectioned (3 μm) for staining
   according to the procedure of previous study ([256]31, [257]66). In
   brief, antigen retrieval was performed by boiling slides in citrate
   (MVS-0101, MaxVision, Fuzhou, China) or EDTA solution (MVS-0099,
   MaxVision) according to instruction of antibodies (table S6). Slides
   were blocked with 5% goat serum and incubated with primary antibody at
   4°C overnight and secondary antibodies anti-mouse/rabbit (Kit-5030,
   MaxVision) for 30 min. Then, the antigens were stained by
   3,30-diaminobenzidine. The stained slices were photographed under a
   light microscope (Olympus, BX53). Then, staining intensities were
   scored by certified urological pathologists (Q. Ma and P. Zhong) as
   described previously ([258]67).

ORO staining

   For ORO staining (S19039, Shanghai Yuanye Bio-Technology, Shanghai,
   China), fresh frozen tissues were sectioned at a thickness of 10 μm or
   cells were seeded on a six-well plate containing glass coverslips. In
   brief, frozen tissues or cells were fixed in 4% PFA for 15 min and
   washed three times with PBS. Then, lipid droplets in cells were stained
   with ORO working solution (ORO dissolved in 60% isopropanol and 40%
   water, v/v) for 30 min in room temperature. Specimens were rinsed with
   PBS for three times and stained with Mayer’s hematoxylin for 5 min. The
   stained slices were photographed by using microscope (Olympus, BX53),
   and lipid droplets were quantified by using the ImageJ software (Fiji).

BODIPY 493/503 staining

   Briefly, after culturing on glass coverslips in a six-well plate, cells
   were fixed in 4% PFA for 10 min, washed three times with PBS, and then
   stained with BODIPY 493/503 (2 μM) for 30 min at room temperature in
   dark room. Cells were then again washed three times with PBS, followed
   by DAPI (4i,6-diamidino-2-phenylindole) incubation for 5 min. Lipid
   droplets were visualized using a microscope (Olympus, BX53). For
   organoids staining, lipid droplets were visualized using Zeiss laser
   scanning confocal microscope (LSM 880, Carl Zeiss, Jena, Germany).
   Lipid droplets were quantified by using the ImageJ software (Fiji).

Glutamine and triglyceride quantitation assay

   In brief, 1 × 10^4 cells were seeded in 96-well plate and incubated
   overnight. Then, culture medium was removed, and cells were washed with
   PBS three times. Then, cells were treated with
   Glutamine/Glutamate-Glo-Assay (J8021) or Triglyceride-Glo Assay (J3160)
   according to the manufacturer’s instruction. Detection was performed by
   measuring chemiluminescence on SpectraMax iD3 spectrophotometer
   (Molecular Devices, San Jose, CA, USA).

Lipidomic analysis

   The 0.75-ml iced methanol and 2.5 ml of methyl tert-butyl ether was
   added into 100 μl of sample. The mixture was incubated for an hour at
   room temperature in a shaker. The 0.625-ml mass spectrometry (MS)–grade
   water was added to induce phase separation, and then the mixture was
   centrifuged at 1000g for 10 min. The upper (organic) phase was
   collected. After quality control, ultrahigh-performance liquid
   chromatography (UHPLC)–tandem MS (MS/MS) analyses were performed using
   a Vanquish UHPLC system (Thermo Fisher Scientific, Germany) coupled
   with an Orbitrap Q Exactive HF mass spectrometer (Thermo Fisher
   Scientific, Germany) in Novogene Co. Ltd. (Beijing, China). The raw
   data files generated by UHPLC-MS/MS were processed using the Compound
   Discoverer 3.01 (CD3.1, Thermo Fisher Scientific) to perform peak
   alignment, peak picking, and quantitation for each metabolite. Peaks
   were matched with the LIPID MAPS, LipidBlast, and Human Metabolome
   Database. After statistical analyses using statistical software R
   (version R-3.4.3), Python (Python 2.7.6), and CentOS (CentOS 6.6
   release), the metabolites with P < 0.05 and FC of ≥2 or ≤0.5 were
   considered to be differential metabolites.

Carbon metabolite analysis

   Using glutamine-free DMEM (11960044, Gibco) supplemented with 500 μM
   [^13C[5]]-l-glutamic acid (IR-30570, IsoReag, Shanghai Zzbio Co. Ltd.)
   and 10% FBS, we labeled 786-O cells (transfected with siRNA targeting
   GLUL or scrambled siRNA) for 24 hours. After washing and metabolites
   extraction, clear supernatant was collected and analyzed by Dionex
   ICS-6000 high-performance ion exchange chromatography–MS/MS
   (HPIC-MS/MS) with the method developed by Shanghai Biotree Biomedical
   Technology Co. Ltd. (Shanghai, China) ([259]68). Briefly, cells were
   washed with ice-cold normal saline and extracted with 500 μl of extract
   solvent (MeOH/H2O, 3:1, precooled at −40°C). After vortexing for 30 s,
   the samples were homogenized for 4 min at 35 Hz with a shaking table.
   The homogenized tissue extract was sonicated in ice water bath for 5
   min, followed by incubation at −40°C for 1 hour and centrifugation at
   12000 rpm at 4°C for 15 min. The supernatant was collected and analyzed
   by HPIC-MS/MS. Standard solutions of metabolites were used for
   calibration of PIC-MS/MS.

Colony formation assay

   Cells in logarithmic growth phase were digested with 0.25% trypsin
   (03-050-1ACS, Biological Industries), centrifuged, and counted after
   resuspending. About 1000 cells were seeded in each well of six-well
   plate. After 2 weeks, six-well plate was harvested by rinsing with PBS
   for three times, fixing with 4% PFA for 10 min. Colonies were stained
   with crystal violet (0.5%, C0121, Beyotime) for 2 min, and the numbers
   of colonies were counted by using ImageJ software.

CCK-8 assay

   Cells were seeded into 96-well plate at a density of 1 × 10^3 cells per
   well with 100 μl of complete medium. To evaluate cell viability, cell
   counting kit-8 (CCK-8) was added into each well and incubated for 2
   hours at 37°C under 5% CO[2]. The absorbance was read at 450 nm by
   using SpectraMax iD3 spectrophotometer (Molecular Devices).

EdU assay

   Cells were seeded into a six-well plate with glass coverslips and
   cultured overnight in complete medium. EdU cell proliferation kit
   (C0071S, Beyotime, Shanghai, China) was used according to the
   manufacturer’s instruction. Briefly, cells were treated with 1× EdU for
   2 hours. Then, DAPI was used to dye the cell nucleus. The stained cells
   were photographed by using fluorescence microscope (Olympus, BX53) and
   analyzed by using ImageJ software.

Flow cytometry

   Cell lipid content was measured by flow cytometry using the BODIPY
   493/503 probe. Briefly, cells were transfected with siRNA targeting
   PHF8, JARID1A, JARID1C, JHDM1A, JHDM2A, and UTX, PHF2, and scrambled
   siRNA were stained by BODIPY 493/503 for 30 min at room temperature.
   Cells were collected and washed with PBS for three times and
   subsequently resuspended with PBS. All cells were analyzed on a
   CytoFLEX (Beckman Coulter, Indianapolis, IN, USA) flow cytometer to
   measure the fluorescence intensity.

Dual-luciferase reporter assay

   HEK293T cells were seeded in 24-well plates at a density of 1 × 10^5
   cells per well. When confluence reached 70 to 80%, cells were
   cotransfected with 500 ng of plasmid (Flag-HIF1A, MYC-HIF2A, YAP1,
   c-MYC, PHF8, or PHF8^H283A) and promoter plasmid (PHF8, TEAD1, and
   GLUL) containing reporters (firefly luciferase and Renilla luciferase)
   in presence of ViaFect (Promega). Luciferase activities were measured
   at 48 hours after transfection using the Luc-Pair^TM Duo-Luciferase HS
   Assay Kit (GeneCopoeia, cat. no. LF004) according to the manufacturer’s
   instruction and were normalized to Renilla luciferase activities.

Gene set enrichment analysis

   GSEA ([260]www.gsea-msigdb.org/gsea/index.jsp) was performed to analyze
   the datasets involved in HALLMARK_MYC_TARGETS_V2 and Hippo signaling
   pathway defined by KEGG. Genes were preranked according to their
   expression of RNA-seq, and R package clusterProfiler 4.4.4 was used for
   GSEA analysis and visualization ([261]69).

Clinical outcome

   For TMA analysis, the immunohistochemically stained tissue sections
   were scored separately by two urological pathologists (Q. Ma and
   P. Zhong) blinded to the clinicopathological parameters. Briefly, the
   quantified method is based on a multiplicative index of the average
   staining intensity (0 to 3) and extent of staining (0 to 3) in the
   cores. The staining score of ≤4 was defined as low expression, whereas
   the score of >4 was referred as high expression. TMA data and clinical
   information were used for analyses, and survival analysis was performed
   using Kaplan-Meier method. The hazard ratio and confidence intervals
   (CIs; 95% CI) were computed.

Statistical analysis

   Unless otherwise specified in the figure legends, all experiments
   reported in this study were performed using at least three independent
   experiments or biological replicates. Analysis for the statistical
   significance between two groups were determined by two-tailed Student’s
   t test, analysis of variance (ANOVA), Pearson correlation, log-rank
   test, Fisher’s exact test, or Wilcoxon rank sum test. For all
   statistical tests, the 0.05 level of confidence was accepted for
   statistical significance.

Acknowledgments