Interplay between pVHL and mTORC1 pathways in clear-cell renal cell carcinoma

Tissue processing

Tumor and normal renal cortex samples were collected from surgical specimens following informed consent of an IRB-approved tissue collection protocol and were flash frozen in liquid nitrogen and stored at −80°C. From each sample, immediately flanking sections from two sides perpendicular to each other were processed for H&E staining and analyzed by a pathologist (W.K.) for composition (tumor content) and quality (hemorrhage, necrosis). Tissues were selected and processed according to Peña-Llopis et al., in preparation.

VHL reconstitution

786-O, A498, and Caki-2 cells were transfected using TransIT®-LT1 Transfection Reagent (Mirus, Madison, VI) with pcDNA3.1/Hygro/HA-VHL (laboratory database ID #586) or empty vector (pcDNA3.1/Hygro; ID #338) and polyclonal populations were selected and maintained in Hygromycin (250 μg/ml).

siRNA transfections

siRNA oligonucleotides were from Dharmacon and Dicer-substrate siRNAs (DsiRNA) duplexes from IDT (Coralville, IA). Transfections were performed using Lipofectamine 2000 (Invitrogen) for A498 or DharmaFECT reagent 3 (Dharmacon, Lafayette, CO) for 786-O and Caki-2 cells according to manufacturer instructions using 220 pmol of siRNA and 20 pmol of DsiRNA per well of a 6-well plate. Sequences or catalog numbers are listed in Supplementary Table 2.

Sequencing of TSC1, TSC2, REDD1 and PTEN

Bidirectional DNA sequencing was performed following PCR amplification using primers designed to produce 400–600 bp amplicons and to include at least 50 bp from intron/exon boundaries. Unless otherwise indicated, only bidirectionally observed somatic mutations are reported.

Statistics

All data are presented as means with standard deviations unless otherwise specified. P-values are calculated by two tailed Student’s t test assuming equal variances unless otherwise indicated. Correlations were calculated using Spearman’s rho in SPSS Statistics 17.0.

For additional information, see Supplementary Material.

REDD1 regulation by pVHL

Recently, we reported that intravenous administration of adenovirus-Cre (Ad-Cre) to Vhl^F/F mice (also referred to as Vhl^floxP/floxP) led to Vhl inactivation and constitutive Hif activation in hepatocytes phenocopying the lipid accumulation observed in ccRCC (25). Using this system, we examined whether Vhl loss was sufficient to induce Redd1. Vhl disruption (see Fig. 1A) led to Hif activation (as determined by the upregulation of the Hif target gene Glut1; Fig. 1B) and Redd1 induction, which was observed at the mRNA (Fig. 1B) and protein levels (Fig. 1C). Thus, Vhl loss is sufficient to induce Redd1 expression.

To examine REDD1 regulation by pVHL in ccRCC, we selected a panel of ccRCC-derived cell lines (786-O, A498 and Caki-2). These cell lines have undergone loss of heterozygosity (LOH) at the VHL locus and contain a single mutant VHL allele (26). However, whereas pVHL function is completely disrupted in 786-O and A498 cells, which harbor truncating mutations upstream of the α-domain, which contains the elongin C binding motif, the α-domain is only partially truncated in Caki-2 cells (26). Nevertheless, the VHL mutation in Caki-2 cells (c.529A>T, Supplementary Fig. 1) is likely to be pathogenic as other somatic mutations in ccRCC have been identified downstream (both missense as well as truncating) (27). Another difference among the cell lines is that whereas in 786-O and A498 HIF-2α is upregulated and HIF-1α is undetectable, in Caki-2 cells, the reciprocal pattern is observed.

To determine whether REDD1 was upregulated in ccRCC as a consequence of VHL loss, we examined the effects of stable reconstitution with wild-type VHL. Ectopic VHL was expressed at different levels across the cell lines (Fig. 2A and data not shown), and as expected, the levels were lower than in previously selected monoclonal populations of reconstituted 786-O cells (28). Nevertheless, VHL reconstitution uniformly downregulated the levels of HIF-α and its target Glut-1 (Fig. 2A–D, Supplementary Fig. 2). In addition, VHL expression similarly downregulated REDD1 (Fig. 2A–D, Supplementary Fig. 2). Furthermore, VHL reconstitution restored the normal regulation of HIF-α and REDD1 by hypoxia (Fig. 2A–D, Supplementary Fig. 2). In keeping with the idea that endogenous mutant VHL in Caki-2 cells retains some functionality, baseline REDD1 levels were lower in these cells (Fig. 2A). Furthermore, the effects of VHL reconstitution on REDD1 induction by hypoxia in Caki-2 cells were incremental (Fig. 2D). Taken together these data show that REDD1 is induced in ccRCC cell lines as a consequence of VHL disruption and that REDD1 levels and its normal regulation can be restored by VHL reconstitution.

HIF-α-dependent regulation of REDD1 in ccRCC

To unequivocally determine whether REDD1 regulation by pVHL involved HIF and whether both HIF-1 and HIF-2 were implicated, we performed both chromatin immunoprecipitation (ChIP) as well as knockdown experiments. ChIP was conducted using antibodies against the aryl hydrocarbon receptor nuclear translocator (ARNT, also called HIF-1β) as well as the corresponding HIF-α subunit (HIF-2α [786-O and A498] and HIF-1α [Caki-2]). VHL-reconstituted cell lines were used as a control. As a reference, the amount of HIF bound to the HIF target genes phosphoglycerate kinase (PGK1) and lactate dehydrogenase A (LDHA) was measured.

As expected, ARNT and HIF-α were observed bound to PGK1 and LDHA sequences, and the amount decreased following VHL reconstitution (Fig. 2E, see also an independent experiment in Supplementary Fig. 3). Similarly, increased levels of ARNT and HIF-α were detected on REDD1 (Fig. 2E, Supplementary Fig. 3).

To determine whether HIF was required for REDD1 upregulation, the effects of depleting HIF-α subunits were evaluated using two independent siRNAs. Knockdown of either HIF-2α (in 786-O and A498) or HIF-1α (in Caki-2) downregulated REDD1 levels (Fig. 2F). Taken together these data show that VHL disruption results in a HIF-dependent direct upregulation of REDD1. Furthermore, unlike some genes that appear to be selectively regulated by HIF-1 or HIF-2 (29, 30), our results show that both HIF-1 and HIF-2 can mediate the upregulation of REDD1 in ccRCC.

REDD1 is highly expressed in VHL-deficient ccRCC tumors

Next we evaluated whether REDD1 expression was upregulated in ccRCC tumors. For these experiments we analyzed two previously reported, publicly available, microarray datasets (31, 47). We selected tumors with mutant VHL and evaluated REDD1 expression levels by comparison to normal kidney samples. The pattern of REDD1 expression was correlated to that of a cohort of hypoxia-inducible genes, largely HIF-1 and HIF-2 targets (30, 33–35), found to be upregulated in ccRCC. In addition as a reference, a second gene cohort was selected of genes implicated in ccRCC pathogenesis that for the most part are not regulated by HIF or upregulated in ccRCC (see Fig. 3A). This analysis showed that REDD1 was upregulated in VHL-mutant ccRCCs (q=6·10⁻¹²) and that REDD1 upregulation was undistinguishable from that of other HIF target genes (Fig. 3A).

In an independent tumor cohort, we compared tumors harboring VHL mutations or VHL gene methylation, to normal renal cortices (from where ccRCC is thought to arise) as well as to ccRCCs with presumed intact VHL function. This analysis showed that REDD1 was upregulated in VHL-deficient ccRCCs, by comparison to normal cortices, and that substantially more heterogeneity existed in REDD1 levels in ccRCCs with presumed wild-type VHL status (Fig. 3B).

To further establish that REDD1 was upregulated in VHL-deficient ccRCC, we evaluated the expression of REDD1 in a cohort of fresh-frozen ccRCC tumors containing VHL mutations and compared the levels to matched normal renal cortices from the same patients. As a reference, quantitative real-time PCR (qRT-PCR) was performed for several HIF targets. By comparison to the corresponding normal cortices, CAIX and VEGFA expression was upregulated in tumors (P = 5.5 × 10⁻⁵ and 1.2 × 10⁻³, respectively, paired t test, Fig. 3C). Similarly, REDD1 was upregulated in VHL-deficient ccRCCs with an overall increase in expression of approximately 3.3-fold (P = 3.4 × 10⁻³, paired t test, Fig. 3D).

We evaluated REDD1 protein levels in tumor lysates and tissue sections using a monoclonal anti-REDD1 antibody we generated. This antibody recognized REDD1 both by western blot and immunoprecipitation and appeared to be highly specific (Supplementary Fig. 4). REDD1 protein levels were consistently upregulated in VHL-deficient ccRCCs (Fig. 3E). In normal kidneys, REDD1 was found to be expressed in normal renal tubular cells (Fig. 3F), and these results were consistent with observations in hypoxic mice (36). Overall, the pattern of staining appeared to be diffusely cytoplasmic (Fig. 3F). In tumors, REDD1 levels were substantially elevated (Fig. 3F). Next, we examined REDD1 levels by IHC in an unselected cohort of 78 ccRCCs. REDD1 was highly expressed in the vast majority of tumors (Supplementary Table 1).

We previously reported that REDD1 is a critical negative regulator of mTORC1, and that REDD1 overexpression was sufficient to inhibit mTORC1 (8). However, mTORC1 has been reported to be active in 60–85% of ccRCCs (1, 2). We examined the same tumor cohort for mTORC1 activation. As a readout, we evaluated the phosphorylation of S6 at S235/236 and S240/244. A statistically significant correlation was found between phospho-S6^S240/244, which is thought to be more specific for mTORC1(37), and phospho-S6^S235/236 (p<0.001), and some level of S6^S240/244 phosphorylation was observed in 80% of tumors (Supplementary Table 1). Not unexpectedly, given that the majority of tumors stained positively for phospho-S6 and REDD1, there was significant overlap (Supplementary Table 1).

mTORC1 regulation by REDD1 in ccRCC

To begin to explore this paradox, we asked whether REDD1 was involved in regulating mTORC1 in ccRCC. Should REDD1 be regulating mTORC1 in ccRCC cell lines, REDD1 knockdown should increase mTORC1 activity. For these experiments, cell lines were grown in 1% serum, which more likely mimics the growth factor milieu of tumors. In both A498 and Caki-2 cells, REDD1 knockdown increased S6 phosphorylation (Fig. 4A) and reduced 4E-BP1 binding to eIF4E (Fig. 4B). Thus, REDD1 normally inhibits mTORC1 in these cells. By contrast, REDD1 knockdown had practically no effect on mTORC1 readouts in 786-O cells (Fig. 4A and data not shown).

Because the TSC1/TSC2 complex is required for mTORC1 regulation by REDD1 (8, 16, 17), we examined the effects of TSC2 depletion. Similar to the effects of REDD1 knockdown, S6 phosphorylation was upregulated in TSC2-depleted A498 and Caki-2 cells, but not in 786-O cells (Fig. 4C). These data show that in both A498 and Caki-2 cells mTORC1 activity is restrained by REDD1 and the TSC1/TSC2 complex. By contrast, mTORC1 activity appears to be uncoupled from both REDD1 and TSC1/TSC2 in 786-O cells, which are PTEN deficient (38). No mutations are observed in TSC1 or TSC2 in 786-O cells (38), and we speculate that TSC1/TSC2 may be inactivated post-translationally as a result of increased Akt activity due to PTEN loss. As the TSC1/TSC2 complex is required for REDD1-induced mTORC1 inhibition (8, 16, 17), these results would provide an explanation for mTORC1 activation in these cells despite high REDD1 levels. Furthermore, the finding that in 786-O cells mTORC1 was uncoupled from REDD1 provided an explanation for the coexistence of high REDD1 levels with mTORC1 activation in ccRCC. However, PTEN mutations are rare in ccRCC (39–41), and we sought to explore this further.

TSC1 is a tumor suppressor in sporadic ccRCC

We set out to examine other components of the mTORC1 pathway. A previous study failed to identify mutations in TSC1 or TSC2 in ccRCC (21). However, the study involved a small number of tumors, and we were encouraged by the finding during a whole-genome sequencing study (Peña-Llopis et al. in preparation) of a somatically acquired mutation in TSC1, which we sought to characterize. The mutation, which was accompanied by LOH, was a splice site mutation in the exon 5 splice acceptor (IVS211-2A > T; Fig. 5A). This mutation, as determined by RT-PCR and cDNA sequencing, led to exon 5 skipping (Fig. 5B and C). Exon 5 loss, however, would be predicted not alter the reading frame. To assess the results of exon 5 skipping, a TSC1^ΔEx5 cDNA was generated by site-directed mutagenesis and was introduced into immortalized Tsc1-deficient mouse embryo fibroblasts (MEFs). By contrast to Tsc1^−/− MEFs reconstituted with wild-type human TSC1, TSC1^ΔEx5-reconstituted MEFs expressed very low protein levels (Fig. 5D). This was not due to low TSC1^ΔEx5 mRNA expression (Figs 5E, F) and experiments with a proteasome inhibitor suggested that the TSC1^ΔEx5 protein was unstable (Fig. 5G). Since TSC1 functions to stabilize TSC2, not surprisingly, TSC2 protein levels were substantially reduced in TSC1^ΔEx5-reconstituted MEFs (Figs 5D and G).

Next we tested the functional consequences of TSC1^ΔEx5 expression on mTORC1. Serum starvation, which downregulated mTORC1 activity in wild-type TSC1 expressing cells, failed to downregulate mTORC1 in TSC1^ΔEx5-reconstituted cells (Fig. 5H). Finally, to determine whether TSC1^ΔEx5 would uncouple mTORC1 from REDD1, we evaluated the response of TSC1^ΔEx5-expressing cells to hypoxia, which results in a Redd1-dependent inhibition of mTORC1 (8). As shown in Fig. 5I, despite substantial upregulation of Redd1 in response to hypoxia, TSC1^ΔEx5-reconstituted cells, like empty vector reconstituted cells, failed to inhibit mTORC1. Taken together these data show that TSC1 (IVS211-2A > T) results in exon 5 loss and an unstable protein leading to constitutive mTORC1 activation and its uncoupling from REDD1.

Having established a precedent for an inactivating somatically acquired TSC1 mutation in ccRCC, and while suspecting based on the previous study (21) that the TSC1 mutation frequency in ccRCC may be low, we set out to determine whether additional TSC1 mutations could be found in ccRCC. For these experiments, we performed PCR amplification and bidirectional capillary sequencing analyses of TSC1 coding sequences (and splice sites) in the aforementioned cohort of 77 fresh-frozen ccRCCs specimens, including 71 tumors for which we had matched normal samples. Interestingly, three additional mutations in TSC1 were identified (Table 1). Matched normal samples were available for two tumors, which permitted confirmation that the mutations were somatically acquired. These mutations were both predicted to be truncating, and since truncating TSC1 mutations further downstream have been associated with TSC syndrome (42), they are most likely pathogenic.

We sought to compare the mutation frequency of TSC1 to that of PTEN, which has been previously implicated in ccRCC (39, 40). As shown, in Table 1, PTEN sequencing of the same tumor cohort revealed a single PTEN mutation (p.G165V). This substitution was previously reported in association with Cowden syndrome (43) and is therefore pathogenic. Thus, TSC1 mutations in ccRCC appear to occur at a frequency similar to, if not greater than, those in PTEN.

Because the TSC1/TSC2 complex may also be inactivated by mutations in TSC2, TSC2 coding and splice site sequences were similarly sequenced in the tumor cohort. However, no somatically acquired mutations were found in TSC2. Failure to identify TSC2 mutations was not due to suboptimal sequencing as, equivalent to TSC1 and PTEN, 100% of the TSC2 amplicons were successfully sequenced across every one of the tumor samples.

Finally, we examined REDD1. As for PTEN, a single mutation was found in REDD1 (Table 1). This mutation was predicted to be truncating and, since the C-terminus of REDD1 is essential for its function (9), the mutation would be predicted to be inactivating.

While no correlation was found between mutations and mTORC1 activation in tumors (see Supplementary Table 1), only a few tumors had mutations, and we hypothesize that other mechanisms, genetic or epigenetic, account for mTORC1 activation despite REDD1 induction in many other samples. Conspicuously, there were no concurrent mutations in TSC1, REDD1 and PTEN, a fact that could be explained simply by the low individual gene mutation frequency, but which may also indicate that the simultaneous occurrence of these mutations does not confer an additive advantage in ccRCC development and is therefore not selected for.