Long-term knockdown of FLCN increases tumorigenesis
To mimic long-term haploinsufficiency of FLCN in kidney cells within BHD patients, we continuously grew HK2 cells, a human proximal tubule cell line, with and without FLCN knockdown for one year in tissue culture to allow for long-term effects of FLCN knockdown to become apparent. To compare short-term effects, cells were grown for < 2 months after FLCN knockdown. To investigate whether long-term FLCN-deficiency increased tumorigenicity, we compared the growth of these FLCN knockdown HK2 cells in soft agar (Fig. 1A). Short-term FLCN knockdown did not alter colony formation, while long-term FLCN knockdown led to much larger colonies, with a mean diameter > 100 µm. Colony diameters within the long-term FLCN knockdown population were more divergent, showing heterogeneity in colony formation.
To further assess anchorage-independent growth, these HK2 cells were grown as self-aggregated spheroids (Fig. 1B). We observed enhanced tumorigenity after long-term FLCN knockdown, when compared to short-term knockdown. We next analyzed spheroid growth of additional non-target and FLCN shRNA knockdown clones (Supplementary Fig. 1) that supports our finding that long-term FLCN knockdown cells showed the greatest increase in spheroid diameter. This data reveals that long-term FLCN knockdown has higher capacity to promote tumorigenicity, when compared to short-term knockdown.
FLCN knockdown dramatically altered the transcriptome profile over time, showing gene enrichment of cell cycle genes.
Transcriptional changes between non-target and FLCN knockdown in HK2 cells were investigated. A PCA and Euclidean cluster dendrogram plot of DESeq2 normalized samples showed that experimental replicates cluster tightly in distinct groups. The most significant factor of interest (PC1) separates the long-term FLCN knockdown group from all other conditions (Supplementary Fig. 2A-B). While short-term FLCN knockdown altered expression of 216 genes, this strikingly rose to 1063 genes after long-term FLCN knockdown (Fig. 1C, ≥ 2.5 fold change up or down, padj < 0.05 with false rate discovery (FDR) correction applied). A WNT signaling signature was evident after both short- and long-term FLCN knockdown. Differentially expressed genes from long-term FLCN knockdown showed greatest enrichment in ‘Aberrant regulation of mitotic G1/S transition in cancer due to RB1 defects’, Reactome: R-HSA-9659787 (> 100 fold enrichment and 7.65E-04 FDR correction applied). Therefore, we investigated RB1/E2F-regulated genes involved in S-phase entry. When comparing short-term FLCN knockdown to its wild-type control (Fig. 2A), TGFA, CCND1, PPARGC1A (also known as PGC1a) and CDKN1C were upregulated with FLCN knockdown. After long-term FLCN knockdown, many more genes were differentially expressed. Three HOX genes showed the greatest fold increase, while TP53, BMP2, TGFA and CCND1 were downregulated (Fig. 2B). Changes of E2F-regulated genes: CCND1, TP53, PPARGC1A, TGFA, c-Jun, RPA1 and RBL1 [26] are shown in Fig. 2C, along with p21 (CDKN1A) and FOXN3 that can arrest cells at the G1 checkpoint [27, 28]. mRNA expression of CCND1, TP53, FOXN3, p21 (CDKN1A), TGFA and CCNE1 were downregulated after long-term FLCN knockdown. Supporting these observed alterations in cell cycle regulatory components, FLCN knockdown cells showed small but significant changes in 2N (G0/G1) and 4N (G2/M) by flow cytometry, when compared to wild-type (Supplementary Fig. 2C).
To obtain a sense of this dysregulation at a pathway level, we assessed the signaling flow from FLCN knockdown to E2F changes (Fig. 3A). Upon long-term FLCN knockdown we observe marked gene expression alterations in multiple pathways that have been previously linked to BHD: (i) upregulation of HIF1A [12] towards gene targets, SLC2A2 (11.2 fold) and DDIT4 (1.6 fold). (ii) Dramatic reduction of DEPTOR (by 93%), the negative regulator of mechanistic target of rapamycin complex 1 (mTORC1). (iii) Enhanced expression of TGFβ (2.1 fold increase) and SMAD3 (1.8 fold increase), supporting elevated TGFβ-SMAD signaling [30]. (iv) 20% reduction in SIRT1 expression (negative regulator of PPARGC1A) and a marked 2.7 fold increase in PPARGC1A expression, and subsequent increase of PPARGC1A-regulated genes: NR1H3, SOX9, and HMOX1 (> 2.5 fold change and < 0.0001 adjusted pvalue). This observation supports previously published work where PPARGC1A was observed to drive mitochondrial biogenesis and metabolic transformation in BHD cell models [11]. More than half of the G1/S regulatory genes show dysregulation following FLCN knockdown, with notable changes to (v) INK4 family members that negatively regulate CCND1 (graphed in Fig. 3B, CDKN2A-D) and (vi) cell cycle regulators that control RB1 phosphorylation and E2F activation. We also see evidence of FLCN knockdown cells employing feedback mechanisms to reduce G1/S regulatory component expression. For example, although initially upregulated upon FLCN knockdown (Fig. 2C and Supplementary Fig. 4), CCND1 mRNA was strikingly downregulated by 97% after long-term FLCN knockdown (Fig. 2C and 3A). As there is a bidirectional relationship between metabolism and cell cycle progression [29], we speculate that the altered metabolism observed in FLCN-deficient cells [11, 12] could underlie the defects in E2F-regulated cell cycle gene expression that becomes more marked and pro-tumorigenic over time.
To assess whether alterations in spheroid growth correlated with the observed alterations in cell cycle pathway dysregulation, we generated lysates from the spheroids grown in Fig. 1B and analyzed a panel of cell cycle proteins. We found that spheroids from long-term FLCN knockdown had elevated expression of CCND1 protein and RB1 phosphorylation (Fig. 3C). These alterations after long-term FLCN knockdown could correlate with their increased growth properties. As negative regulators of cyclin:cyclin dependent kinase (CDK) complexes, we analyzed TP53 and cyclin-dependent kinase inhibitor 1A (CDKN1A, known as p21/WAF1). We found them highly expressed following FLCN knockdown, and in both cases the levels were higher after long-term knockdown. This contrasts with the RNA data, where CCND1, TP53 and CDKN1A mRNA expression were all downregulated with long-term FLCN knockdown (Fig. 2C). Long-term FLCN knockdown also elevated the expression of β-catenin showing increased WNT signaling, in line with our RNA sequencing findings (Fig. 1C). Sequestosome 1 (SQSTM1, known as p62) mRNA was also increased by 1.5 fold (Fig. 2C) and protein expression was markedly enhanced (Fig. 3C). AMPK and Acetyl-CoA carboxylase (ACC) phosphorylation was enhanced following FLCN knockdown that was more evident after long-term knockdown (Fig. 3C). High levels of SQSTM1 and AMPK activation is consistent with metabolic alterations and energy stress, as previously reported in BHD models [11].
FLCN interacts with components of the cell cycle and DDR
To further define FLCN as a tumor suppressor, mass spectrometry of GST-tagged FLCN-interacting partners was performed (Fig. 4A). This uncovered 603 potential FLCN-binding proteins, and successful enrichment for FLCN binding proteins was confirmed via the identification of previously characterized FLCN interactors, such as FLCN interacting protein 1 (FNIP1) [31] and FNIP2 [32], as well as PKP4 [13, 14] (Fig. 4B) and the BRCA1 A complex component, BRE [33]. When considering identified interactors where more than one unique peptide was identified, 541 potential FLCN binding proteins were mapped by DAVID analysis into functional clusters (Fig. 4C). This revealed that the strongest enrichment was in proteins involved in translation, in keeping with the known function of FLCN in the mTORC1 pathway. Of interest, we identified protein folding chaperones as highly enriched in our dataset, including all 8 components of the TRiC/CCT chaperonin complex and two isoforms of HSP90, plus the co-chaperone CDC37. The chaperonin CCT helps fold both mLST8 and Raptor, part of mTORC1 [34], while the R2TP complex (which includes RUVBL1 and RUVBL2), together with heat shock protein 90 (HSP90), is a chaperone for the assembly of protein complexes including phosphatidylinositol 3-kinase (PI3K)-like kinases (PIKKs) such as TOR and PRKDC, more commonly called DNA-dependent protein kinase (DNA-PK) (highlighted in Fig. 4B and 4D) [35]. Additionally, FLCN was previously described as a HSP90 client protein, where the HSP90-FLCN interaction enhances the stability of FLCN [36]. Several CCT substrates are cell cycle proteins, particularly at the G1/S phase (discussed in [37]), so it was interesting to observe that cell division was also a top 10 enriched process (Fig. 4C). Comparing different stages of the cell cycle, there was a higher proportion of FLCN-binding proteins linked to the G1/S phase (Supplementary Fig. 3A), including CDK1, CDC20, TP53 and the RB binding protein 7 (RBBP-7). This analysis supports our RNA sequencing findings, indicating that FLCN is linked to the G1/S cell cycle. One cell cycle protein identified, DNA-PK, was our top interactor, with the most peptides identified (Supplementary Fig. 3B). This protein also has a role in the DNA repair, another enriched process amongst our interactome (Fig. 4C). Indeed, a number of interactors had involvement in metabolism as well as cell cycle and DNA damage (Fig. 4D).
As our top inteactor, with 66 unique peptides and 17.9% coverage (Supplementary Figs. 3B and C), we focused our attention on DNA-PKcs. It plays an important role in cell cycle check point control during DNA damage, which correlates with our transcriptomic work that links FLCN knockdown to cell cycle dysregulation (Fig. 2B) and interaction network with FLCN (Fig. 4B-D). As there are limitations to over-expressed protein purification methods, i.e., potential non-specific/artefact protein binding to either beads or mislocalized over-expressed proteins, we carried out further FLCN-binding validation experiments on DNA-PKcs. We initially examined whether endogenous DNA-PKcs co-purified with GST-tagged FLCN (Fig. 5A). We observed robust interaction of DNA-PKcs with GST-FLCN and saw no DNA-PKcs in the empty vector control, showing no non-specific/aretfact binding of DNA-PKcs to beads. Other PIKK family members, ATM or ATR, did not immunoprecipitate with FLCN. Further validating the DNA-PKcs interaction, we observed interaction of endogenous DNA-PKcs with immunoprecipitated endogenous FLCN (Fig. 5B). Next, we compared association of DNA-PKcs with wild-type FLCN and BHD patient-derived mutants, Y463X and H429X (Fig. 5C). Both C-terminal truncation FLCN mutants showed interaction with endogenous DNA-PKcs. We then considered that FLCN might be a direct substrate of DNA-PKcs. Therefore, in vitro DNA-PK kinase assays were performed, using TP53 as a DNA-PK substrate control (Fig. 5D). Supplementation of dsDNA was used to further enhance DNA-PK’s kinase activity. Unlike TP53, where DNA-PK-mediated phosphorylation was enhanced with supplementation of dsDNA, we observed a much weaker level of [32P]-incorporation into FLCN. Inclusion of FNIP1 or FNIP2 did not further enhance this low level of phosphorylation (data not shown). Given the low levels of [32P]-incorporation into FLCN that was not further enhanced with supplementation of dsDNA, FLCN is unlikely a direct substrate of DNA-PK in these assays.
Cell cycle progression is dysregulated following long-term loss of FLCN
As DNA-PKcs regulates the phosphorylation of H2AX in response to cell cycle progression and DNA damage [38], we next analysed γH2AX (H2AX phosphorylated at Ser139). Following short-term FLCN knockdown γH2AX was enhanced and long-term FLCN knockdown elevated γH2AX further (Fig. 6A). γH2AX is classically regarded as a marker of DNA damage involved in the surveillance and repair of double strand breaks, such as those induced by ionising radiation (IR). However, γH2AX has also been reported to occur independently of double strand DNA breaks [39], in mitotic cells [38] and in response to serum starvation [40]. Therefore, we next assessed the interaction of DNA-PKcs and GST-FLCN following IR and found that endogenous DNA-PKcs dissociated from GST-FLCN after IR treatment at 5 and 10 Gy (Fig. 6B). This reveals that FLCN/DNA-PK association is regulated by DNA damage. 5 Gy IR treated cells were analyzed for markers of DNA damage to determine whether FLCN knockdown altered DDR signaling. IR enhanced γH2AX as expected, with a higher basal and IR-induced level of γH2AX in the FLCN-deficient HK2 cells (Fig. 6C). DNA-PKcs autophosphorylation is essential for the appropriate regulation of DNA strand end processing, enzyme inactivation, and complex dissociation from DNA (see review [41]). However, no change in DNA-PK autophosphorylation was observed upon FLCN knockdown. While phosphorylation of TP53 at Ser15 was induced under IR, there was no difference when comparing FLCN knockdown to wildtype (Fig. 6C). This data shows that although γH2AX is elevated following FLCN loss, FLCN loss has no direct impact on IR induced DNA-PK signaling, i.e., elevated γH2AX is unlikely to be reflective of double strand DNA breaks. When the RNA sequencing data with and without FLCN knockdown was run through Mutect2, there was no evidence of enhanced DNA mutations (data not shown). This indicates that FLCN loss is unlikely to be enhancing DNA damage, but might be more related to cell cycle control linked to DNA-PK.
To examine FLCN’s role in an alternative system, we analysed C. elegans with and without Flcn RNAi knockdown. C. elegans has previously been used as a model organism to characterise the effects of DNA damage and cell cycle check point control in germ cells (reviewed in [42]). While we observed no change in the relative number of mitotic germ cells upon Flcn knockdown, Flcn knockdown worms were defective in cell cycle arrest by UV-induced DNA damage. We observed by a small increase in the relative number of mitotic germ cells after UV (Fig. 6D). This supports a hypothesis that Flcn knockdown results in dysregulation of cell cycle control following DNA damage. To determine whether cell cycle arrest defects following FLCN loss is also linked to the G1/S phase transition in mammalian cells, we examined HK2 cells for CCND1 expression and RB1 phosphorylation following IR treatment (Fig. 6E). CCND1 expression was basally higher after FLCN knockdown, in line with previous work [16]. Upon long-term FLCN knockdown, CCND1 expression was further increased (Fig. 6E). In healthy cells, CCND1 protein levels are typically reduced after IR, as part of a normal DNA damage cell cycle checkpoint control mechanism [43]. However, in FLCN knockdown cells following IR treatment, CCND1 protein levels remained elevated, indicating a defect in the normal control of CCND1 after DNA damage. Increased RB1 phosphoryation was also observed after FLCN knockdown, implying an elevated level of active G1/S cyclin-CDK complexes. RB1 phosphorylation leads to E2F activation and entry into S phase, suggesting that knockdown of FLCN favours G1/S checkpoint slippage. Supporting this observation, transcriptomic analysis showed enrichment of E2F regulated genes in FLCN knockdown HK2 cells (Fig. 2B). Overall, FLCN knockdown leads to cell cycle dysregulation with and without DNA damage, where we observe heightened levels of CCND1 and RB1 phosphorylation as well as enhanced E2F-mediated transcription of S-phase genes.