Aggregation dynamics and identification of aggregation-prone mutants of the von Hippel–Lindau tumor suppressor protein

The human VHL gene encodes three different isoforms from two transcripts: the mRNA variant 1 encodes the full-length pVHL213, which includes a hydrophobic β domain and a hydrophobic α domain, and a shorter pVHL160 version produced from an internal translation start site. The mRNA variant 2 encodes pVHL172, in which exon 2 is excluded by an alternative splicing event. We have recently shown that these isoforms are co-expressed in human renal cells (Chesnel et al., 2015). Mutations in the von Hippel–Lindau tumor suppressor gene (VHL) gene cause the von Hippel–Lindau (VHL) disease, an inherited cancer syndrome in which highly vascularized tumours, such as clear cell carcinoma, hemangioblastoma and pheochromocytoma (Latif et al., 1993; Richard et al., 2013), are frequently observed. The human VHL protein (VHL, hereafter referred to as pVHL) acts as a tumor suppressor and is mutated in 80% of sporadic clear cell renal carcinoma (ccRCC). pVHL is a component of the E3 ubiquitin ligase complex VBC that includes also elongin B and elongin C (ELOB and ELOC, also known as TCEB2 and TCEB1, respectively), cullin-2 and RBX1. The VBC complex regulates the cellular response to oxygen concentration by targeting the transcription factor HIFα for proteasomal degradation (Kamura et al., 2000).

In eukaryotes, 20–30% of proteins adopt folded conformations only after interaction with binding partners (Tyedmers et al., 2010). In the absence of partners, proteins may be misfolded and incorporated into aggregates. Full-length pVHL (pVHL213; 213 amino acids) is an aggregation-prone protein (Shmueli et al., 2013). When properly folded after sequential interaction with the HSP70 chaperone and the TCP-1 ring complex (TRiC) chaperonin, pVHL binds to ELOB or ELOC to form the VBC complex (Feldman et al., 2003, 1999; McClellan et al., 2005; Melville et al., 2003). Misfolded pVHL cannot bind to ELOC–ELOB and is degraded by a pathway involving the HSP90 chaperone (McClellan et al., 2005; Schoenfeld et al., 2000). pVHL mutations that have been detected in ccRCC might compromise its native folding and conformation, and alter its binding to TRiC (Feldman et al., 2003, 1999).

The budding and fission yeast expression systems allow study of the mutation-driven aggregation properties of a given protein and identification of molecular pathways involved in the cellular fate of proteins through genetic screens (Summers and Cyr, 2011). Human pVHL213 expressed in budding yeast does not fold properly because its cognate human partners are not endogenously expressed. Instead, it aggregates in two different subcellular compartments – in the insoluble protein deposit (IPOD), which is close to the vacuole, and the juxtanuclear quality control compartment (JUNQ). In either, aggregated proteins are either refolded by chaperones or degraded by the proteasome (Kaganovich et al., 2008; Specht et al., 2011). Moreover, a new intra-nuclear inclusion (INQ) has been described recently for the deposition of nuclear and cytosolic misfolded proteins (Miller et al., 2015). The yeast Hsp104 chaperone binds to stress-induced protein aggregates nonspecifically. In yeast, stress-induced misfolded proteins first accumulate as stress foci or Q bodies, before Hsp104-dependent fusion with IPOD or JUNQ (Escusa-Toret et al., 2013; Spokoini et al., 2012). In mammalian cells, the excess of misfolded protein is incorporated into JUNQ or IPOD, depending on the nature of the misfolded protein. Specifically, overexpressed pVHL is essentially found in JUNQ in different mammalian cell lines (Ogrodnik et al., 2014; Weisberg et al., 2012).

Ubiquitylation is a key regulatory mechanism to sort misfolded proteins. Ubiquitylated and soluble proteins are found in the JUNQ, where they can rapidly transit back to the cytosol. Conversely, terminally aggregated insoluble proteins are incorporated into the IPOD and targeted for degradation (Kaganovich et al., 2008). However, the exact role of ubiquitylation in targeting misfolded pVHL in specific inclusions has been questioned (Miller et al., 2015).

Reports have previously described pVHL213 folding and aggregation in vitro (Shmueli et al., 2013; Sutovsky and Gazit, 2004). However, the aggregation propensity of the pVHL160 and pVHL172 isoforms has not been studied yet. As protein aggregation depends on macromolecular cytoplasm crowding, we overexpressed human pVHL in fission yeast, which does not require proteasome inhibition or stress to promote protein aggregation, different from budding yeast. We found that pVHL proteins are incorporated into small dynamic aggregates and in large slow-moving inclusions. We then identified two pVHL mutations in ccRCC (P86L and P146A) that promote pVHL aggregation. P86L stimulated pVHL incorporation into small aggregates, whereas P146A exacerbated large inclusion formation. Conversely, the I151S mutation, which is located in the same aggregation-prone region as P146, reduced large inclusion formation and led to protein destabilization. Finally, we showed that the conserved gene pac10, encoding a chaperone prefoldin subunit, controlled pVHL stability.

In the absence of ELOB and ELOC, pVHL213 aggregates in proteasome-deficient budding yeast cells (Kaganovich et al., 2008). To investigate the mechanism of pVHL aggregation in the fission yeast Schizosaccharomyces pombe in the absence of cofactors, yeast cells that expressed the Hsp104–GFP chaperone, a marker of damaged proteins (hsp104-GFP cells), were transformed with pREP1-VHL213 plasmid (Table S1) controlled by the thiamine-repressed strong nmt1 promoter. Western blot analysis of cells cultured without thiamine for 24 h confirmed the induction of pVHL213 expression (VHL213 ON, Fig. 1A) and showed that it was partially insoluble (VHL213 ON, lane labelled P, Fig. 1A). Concomitant with increased expression of pVHL213, the Hsp104–GFP expression level also was upregulated, and a higher amount was recovered in the pellet fraction (VHL213 ON, lane labelled P, Fig. 1A), which was different from that observed with control cells (vector and VHL213 OFF, Fig. 1A). Moreover, immunofluorescence analysis showed that untagged pVHL213 colocalized with Hsp104–GFP in a large cytoplasmic structure (Fig. 1B). Colocalization with Hsp104 in inclusions was still observed in hsp104-GFP yeast cells that also expressed pVHL213 tagged with red fluorescent protein (RFP–VHL; Table S1, Fig. S1). This indicated that fusion of pVHL213 with a fluorescent tag did not alter its aggregation propensity. RFP–VHL-positive inclusions always colocalized with Hsp104–GFP. Some Hsp104–GFP inclusions did, however, not colocalize with RFP–VHL, suggesting that RFP–VHL aggregates do not saturate protein quality control sites.

In exponentially growing hsp104-GFP yeast cells (control), Hsp104–GFP was evenly expressed in the nucleus and was also observed as one or two dots per cell, close to the nucleus (Fig. 1C, arrowhead) or to the ends of the cell (Fig. 1C, arrow). These dots were detected in 55.6% of cells (Table S2; 46.1±8.0% of cells with a single dot and 9.5±7.6% of cells with two dots; mean±s.d.) and are likely to correspond to protein quality control (PQC) inclusions, where damaged proteins accumulate.

In hsp104-GFP­ yeast cells that overexpressed untagged pVHL213, the percentage of cells with Hsp104–GFP dots increased to 72.8% (Fig. 1C, pVHL213 ON; Table S2, pVHL213) owing to the strong increase (three times) in the proportion of cells with two dots (Table S2, n=76, 28.2±5.9% vs 9.5±7.6% in control cells). Upon repression of pVHL expression through addition of thiamine for 8 h (Fig. 1C, pVHL213 OFF; Table S2, pVHL213 OFF 8 h), the percentage of cells with inclusions decreased to 46.3%, and the fraction of cells with two dots decreased to 5.5±5.1%, similar to control cells (Table S2, pREP1 OFF 8 h, 9.2±6.5%). The mean size of Hsp104–GFP inclusions was 0.25±0.13 μm² in control cells (mean±s.d.) (Table S2, pREP1 ON) and was increased in pVHL213-overexpressing cells (0.58±0.37 μm², Table S2, pVHL213 ON, Mann-Witney test, P<0.01). After repression of pVHL expression with thiamine for 8 h, the mean size decreased to 0.30±0.24 μm² in pVHL213-overexpressing cells (Table S2, pVHL213 OFF 8 h), but was not different compared to that in control cells (0.28±0.20 μm², Table S2, pREP1 OFF 8 h).

Our results show that pVHL213 expression promotes the increase in the number and size of Hsp104–GFP-positive inclusions, and pVHL is likely to incorporate into large Hsp104-positive PQC inclusions.

To further characterize the inclusions, wild-type fission yeast cells were transformed with pREP41-GFP-VHL plasmids in which expression was controlled by the mild inducible nmt41 promoter (Table S1). GFP–VHL isoform expression was ascertained by western blot analysis (see Fig. 2A for a schematic representation of the three pVHL isoforms, Fig. 2B).

After overnight induction of expression, GFP–VHL213 was detected as small highly dynamic puncta, referred to as ‘small dynamic aggregates’ (SDA) that showed Brownian movements (Fig. 2C, arrows; Movie 1). GFP–VHL213-positive SDAs were detected in 20% of live cells.

GFP–VHL213 also formed large slow-moving cytoplasmic aggregates (LSAs) (Fig. 2C, arrowhead; Fig. 2D, arrow and arrowhead) that were not observed in cells that had been transfected with the empty vector (see below). Like Hsp104–GFP dots, GFP–VHL213 LSA appeared as a single large spherical aggregate at the end of cells (cell end LSA, ceLSA) (arrow, Fig. 2D) or close to the nucleus (juxtanuclear LSA, juLSA) (arrowhead, Fig. 2D). Overexpression of GFP–VHL213 in yeast cells that expressed Amo1–RFP (a nuclear envelope marker fused to RFP) indicated that juLSA were outside the nucleus (Fig. 2E). Analysis of 99 confocal microscopy sections showed that juLSA were close to but never within the nucleus outlined by Amo1–RFP. After induction of expression overnight, juLSA- and ceLSA-containing cells represented approximately 10% and 15% of all cells, respectively. Some cells (4%) contained both a ceLSA and a juLSA. This percentage was lower than the fraction of hsp104-GFP-expressing cells with dots, suggesting that Hsp104 inclusions incorporate GFP–VHL213 only when expression goes above a certain threshold. To quantify LSA subcellular localization, the cell surface was arbitrarily divided into five equal segments, including (a) the nucleus, (b) the intermediate regions between the nucleus and the cell periphery, and (c) the cell periphery (the most distal segments at the cell ends; Fig. 2F, schematic representation of a cell). This showed that pVH213 LSA were more frequently present in the intermediate areas than at the cell periphery (73% and 25%, respectively, n=347; Fig. 2F). To distinguish new and old cell poles (the new cell pole is produced during the last cell division), cells were stained with the fluorescent cell wall marker Aniline Blue, and LSA distribution was assessed relative to the two poles. GFP–VHL213-positive LSA were located to equal extent at the new and old cell ends [48±8.3% and 52±8.3%, respectively (mean±s.d.); n=126; see below].

It has been previously shown that large inclusions of stress-induced misfolded proteins are the result of the fusion of small puncta. This fusion has been shown to be reduced to 50% in the absence of the small heat shock protein coded for by the hsp16 gene (Coelho et al., 2014). In budding yeast, Q body formation has been shown to depend, at least in part, on the hsp16 homolog HSP42 (Escusa-Toret et al., 2013). However, in fission yeast, we observed that both GFP–VHL213 SDAs and LSAs were detected in hsp16Δ cells (Fig. 2C) in identical proportions to those seen in wild-type cells, strongly suggesting that SDAs are not counterparts of budding yeast Q bodies.

Disruption of the microtubule (MT) cytoskeleton with the MT-depolymerizing drug methyl benzimidazol-2-yl-carbamate (MBC) did not have any effect on LSA formation [38.4% of LSA-containing cells (n=429) vs 36.7%, (n=452) for DMSO-treated cells] or localization, indicating that MTs are not involved in LSA biogenesis and subcellular localization (Fig. S2A). Similarly, incubation with latrunculin B (LatB) to depolymerize the actin cytoskeleton did not significantly alter LSA formation [28.3% of LSA-containing cells (n=420) vs 33.9% (n=318) for DMSO-treated cells] or localization (Fig. S2B). Neither MBC nor LatB affected SDA dynamics, consistent with previous results (Coelho et al., 2014).

The intracellular movements of ceLSA were mainly slow (Movie 2) and were probably driven only by the cytoplasmic flow and cytoplasmic constraints. The movements of juLSA were very limited, possibly owing to a physical link with the nucleus. LSA slow movements suggest that their localization might reflect the site of formation.

Finally, the aggregation propensity of a protein is mainly dependent on the proportion of hydrophobic residues in the primary sequence. We thus tested whether overexpression of two unrelated GFP-tagged proteins with a similar proportion of hydrophobic amino acids to pVHL213 (24.8%) would promote their aggregation in fission yeast cells. Overexpression of Xenopus laevis Cl2 (Swissprot ID Q91715; 102 amino acids with 25.5% of hydrophobic residues) and of fission yeast Rad25 (270 amino acids with 23.4% of hydrophobic residues), driven by the nmt41 inducible promoter (Table S1), did not give rise to detectable fluorescent aggregates (Fig. S3), supporting the hypothesis that pVHL aggregation is an intrinsic property promoted by a specific primary sequence.

It has been shown that juxtanuclear inclusions contain only ubiquitylated proteins (Kaganovich et al., 2008), but this view has been recently challenged (Miller et al., 2015). In pVHL213, there are three lysine residues that could be subjected to ubiquitylation, sumoylation and/or neddylation (Cai and Robertson, 2010; Cai et al., 2010; Park et al., 2015; Stickle et al., 2004). To determine whether pVHL ubiquitylation regulates its spatial distribution in cellular inclusions, we expressed a GFP–VHL213 variant in which these three lysine residues were mutated into arginine (K159R K171R K196R, hereafter referred to as 3KR, Fig. 2G; Table S1). Western blot analysis showed that wild-type and 3KR GFP–VHL213 were expressed at similar levels, with comparable proportions of insoluble and soluble pools (Fig. 2H). Likewise, the overall percentage of cells with wild-type or 3KR GFP–VHL213 LSAs was similar (approximately 30%). However, in 3KR-expressing cells, the proportion of ceLSAs was higher than in cells expressing wild-type GFP–VHL213 (Fig. 2G, t0). This result suggests that pVHL ubiquitylation, neddylation and/or sumoylation potentially influences its spatial distribution in different inclusions, although incorporation exclusively in a single inclusion type was not observed, which is different to that observed in budding yeast (Kaganovich et al., 2008).

Monitoring GFP–VHL213 LSA formation dynamics over time indicated that LSA were assembled and disassembled in approximately 30 min in a hemicell (Fig. 3A and B; Movies 3 and 4). These observations suggest that LSA metabolism is limited in space. We then asked whether proteins that are incorporated into inclusions are terminally aggregated insoluble products. After induction of expression overnight, GFP–VHL213 expression was repressed with thiamine to prevent de novo LSA formation, and protein mobility in LSAs was monitored using fluorescence recovery after photobleaching (FRAP). After photobleaching of an area in the cytoplasm, cytoplasmic GFP–VHL213 diffused freely with fluorescence recovery in 30 s (Fig. 3C). Conversely, after photobleaching of an identical area corresponding to a whole GFP–VHL213 LSA (Fig. 3D), very limited fluorescence was recovered, indicating that bleached proteins within the LSA were immobile and could almost not be replaced by proteins from the cytoplasmic GFP–VHL213 pool. The observation that the fluorescence of a nearby unbleached LSA was unaffected by cytoplasm bleaching also strongly suggests that proteins could not diffuse from the cytoplasm into the inclusion (Fig. 3C). Finally, partial photobleaching of LSAs did not result in fluorescence redistribution within the aggregate (Fig. 3E), as described in budding yeast (Kaganovich et al., 2008). We conclude that GFP–VHL proteins that had been incorporated into large inclusions (LSA) were immobile and could be terminally aggregated products.

The nmt41 promoter is fully repressed 1 h after addition of thiamine in the medium. The expression of GFP–VHL213 was reduced in both the soluble and insoluble fractions, compared to that in induced cells, 3 and 6 h after thiamine addition (Fig. 4A, OFF). Analysis of cells containing GFP–VHL213-positive LSAs at 3 (t3) and 6 h (t6) after addition of thiamine showed that juLSA had already been lost at t3, whereas ceLSAs progressively decreased over time (Fig. 4B). This indicates that the degradation rates of these structures are different and strongly suggests that the formation of juLSAs and ceLSAs correspond to two different protein quality control processes. Similar results were obtained with 3KR GFP–VHL213, suggesting that ubiquitylation does not influence the rate of aggregate clearance (Fig. 2G, OFF).

Aggregated proteins are cleared through two different pathways – autophagy and proteasome degradation. To identify the clearance pathway of pVHL inclusions, pVHL213 was expressed in atg1Δ cells, an autophagy-deficient fission yeast mutant (Mukaiyama et al., 2009). The formation of pVHL-positive aggregates (SDAs and juLSAs or ceLSAs) and their clearance were unaffected in atg1Δ cells compared to wild-type cells (Fig. 4C). Conversely, inclusions persisted longer in pVHL213-expressing cells that had been incubated with the proteasome inhibitor bortezomib than with vehicle alone (DMSO, Fig. 4D and E), demonstrating that proteasome degradation is the main clearance pathway of pVHL inclusions. Strikingly, SDA formation was dramatically increased after 1 h of incubation with bortezomib (Fig. 4D and F), suggesting that these aggregates are extremely sensitive to proteasomal degradation. Consistently, the formation of SDAs and LSAs, including cells with two or more LSAs (2+ LSA, Fig. 4G), was strongly promoted in mts3-1 cells, a thermosensitive proteasome mutant.

In fission yeast cells that expressed GFP–VHL172 (Fig. 5A, white arrowhead), we detected only rare small dynamic puncta that were reminiscent of pVHL213 SDAs and inclusions of intermediate size compared to LSAs (Fig. 5B). Consistently, overexpressed GFP–VHL172 was almost completely absent in the insoluble fraction (Fig. 5A, VHL172, lane labelled P). This indicates that the main aggregation-prone region of pVHL lies within exon 2 (amino acids 114 to 154). GFP–VHL172-positive inclusions were still detected in the autophagy-deficient mutant. Consistently, untagged pVHL172 (pREP1-VHL172 plasmid, Table S1) colocalized with Hsp104–GFP, like pVHL213 (Fig. 5C). GFP–VHL172-positive inclusions were mainly located in the central region (Fig. 5D). We observed that soluble GFP–VHL172 occasionally accumulated in the nucleus, suggesting that the pVHL172 isoform has specific nuclear import signals or lacks nuclear export signals. Following protein expression repression, GFP–VHL172 aggregates were rapidly degraded. An aggregation-prone region is still present in the pVHL172 isoform because, as for GFP–VHL213, GFP–VHL172 formed SDA-like structures and intermediate inclusions, and the number of SDA-like puncta was also strikingly increased after incubation with bortezomib (Fig. 5E).

GFP–VHL160 that had been overexpressed in fission yeast cells (Fig. 5A, black arrowhead) was incorporated into highly dynamic very dense small puncta (Fig. 5F, arrow) throughout the cytoplasm. These puncta were of more heterogeneous sizes and intensities compared to GFP–VHL213 SDAs. GFP–VHL160 was also found in very large LSA-type inclusions (Fig. 5F, arrowhead, LSA; Movie 5). The shape of GFP–VHL160-positive LSAs was irregular, different from the round GFP-VHL213- and GFP-VHL172-positive inclusions. Their formation started as early as 9 h after promoter de-repression. pVHL160 puncta formation was followed by the appearance of large and irregular LSAs, although puncta-containing cells were still observed. These inclusions were never located close to the nucleus. Most pVHL160-positive large aggregates (Fig. 5D, 70%, n=183) were in the cell periphery. Strikingly, almost all the GFP signals were present in aggregates, suggesting that most of the GFP–VHL160 pool was insoluble. This was confirmed by the accumulation of GFP–VHL160 in the insoluble fraction (Fig. 5A,P). Interestingly, there was a strong bias towards accumulation of GFP–VHL160 LSAs in the old cell pole in contrast to the situation with GFP–VHL213 LSAs (Fig. 5G). These results suggest that the subcellular localization of inclusions might be regulated through different assembly site specifications and/or tethering mechanisms, the nature of which remains to be identified. The percentage of cells containing GFP–VHL160-positive LSA-type inclusions decreased after 18 h, as observed for the two other pVHL isoforms, whereas small puncta persisted longer. However, the overall percentage of GFP–VHL160-expressing cells with aggregates –either small puncta or inclusions – did not decline (present in approximately 90% of yeast cells), contrary to that observed for other pVHL isoforms. This suggests that the threshold level for GFP–VHL160 aggregate formation is low and that their degradation might be slower compared to that for other pVHL aggregates. These observations suggest that GFP–VHL160 aggregates are strongly resistant to degradation and persist in cells, probably reflecting a higher propensity of GFP–VHL160 to aggregate compared to the other isoforms.

In contrast to pVHL213 and pVHL172, the pVHL160 isoform showed a distinct behaviour if it was not GFP tagged. When expressing pVHL160 in the absence of the tag, we did not observe bigger Hsp104–GFP inclusions in cells nor Hsp104–GFP in the insoluble protein fraction. The untagged pVHL160 exogenously expressed protein was present in the soluble fraction (our unpublished results). As the behaviour of the other GFP fusion proteins was the same for the corresponding untagged pVHL isoforms, the different aggregation behaviour of pVHL160 with or without a tag could be explained by a specific effect of the tag on the structural conformation.

In conclusion, pVHL172 has a low propensity to form aggregates, whereas pVHL160 aggregates in dense dynamic SDA-type and large insoluble LSA-type structures.

To identify potential aggregation-prone regions (APR), the TANGO algorithm was used (Fernandez-Escamilla et al., 2004), and the highest pVHL aggregation propensity score corresponding to the hydrophobic stretch IFANIT within exon 2 (amino acids 147–152, APR3; Fig. 6A) was revealed. Two other motifs with lower scores in TANGO analysis were located within exon 1 (APR1 and APR2) and corresponded to amino acids 72–78 (SQVIFCN) and 87–91 (VWLNF), respectively (Fig. 6A). Then, the TANGO overall aggregation scores of VHL missense mutants that have been identified in individuals with ccRCC were calculated (Table S3). The scores were compared with two alternative algorithms (Conchillo-Solé et al., 2007; Tartaglia et al., 2008), and three mutations were selected: P86L, P146A and I151S (Fig. S4). The position of P86 and P146 in the pVHL structure based on the 1LM8 dataset in Protein Data Bank corresponding to pVHL160 (Fig. 6A, upper panel,) suggested that these residues were involved in the formation of β-sheets that are essential for the structural organization of the exon1-to-exon2 linker region. Therefore, they could contribute to the tertiary structure of pVHL. In addition, the P86L mutation creates a continuous hydrophobic stretch between APR1 and APR2, whereas the P146A mutation increased the TANGO score of APR3 by two-fold (Fig. S4). Finally, I151S was identified as an aggregation-suppressive mutation that completely abolished the aggregation propensity predicted in APR3 (Fig. S4). However, the I151 mutation did not seem to be involved in the pVHL secondary structure, based on its position (Fig. 6A).

We tested the identified APRs and the potential impact of the selected ccRCC mutations by overexpressing pVHL213 mutants in our fission yeast inducible system. Western blot analysis showed equivalent expression levels for GFP–VHL213 P86L and GFP–VHL213 P146A compared to wild-type GFP–VHL213, whereas the expression of the I151S mutant was moderately reduced (Fig. 6B).

The number of cells containing SDAs was increased in a population expressing GFP–VHL213 P86L compared to a wild-type GFP–VHL213-expressing population (Fig. 6C, 41.9% vs 16.9%). Conversely, the number of LSAs was not different, suggesting that the P86L mutation affected only the formation of SDAs. After fixation, no SDA-containing cell was detected in a wild-type GFP-VHL213-expressing population, whereas SDAs were still observed in some GFP–VHL213-P86L-expressing cells (Fig. 6C, lower panels). Consistent with the higher aggregation propensity score of APR3 harbouring the P146A mutation (Fig. S4), larger LSAs were observed in cells expressing GFP–VHL213-P146A than in cells expressing wild-type GFP–VHL213 [Fig. 6D; mean sizes: 0.595±0.33 μm² (n=254) vs 0.543±0.38 μm² (n=294) respectively; Fig. 6E, Mann–Whitney test, P=0.005]. The fraction of cells with LSAs was not significantly changed, suggesting that the maximal number of cells containing pVHL inclusions had been reached under our experimental conditions. Interestingly, the proportion of cells with LSA was lower in a GFP–VHL213-I151S-expressing population than in a population expressing wild-type GFP-VHL213, and LSA appeared smaller [Fig. 6D and E, mean size: 0.441±0.24 μm² (n=182), although not significantly different from the wild type, Mann–Whitney test, P=0.09)].

The conserved multi-protein chaperone complex prefoldin comprises six different subunits (two α and four β subunits; PFDN1 to PFDN6 in humans) that are organized in a double-barrel structure, and the complex is involved in protein folding and aggregation of misfolded proteins. The prefoldin subunit PFDN3 (or VBP-1 for VHL-binding protein 1) is a direct pVHL binding partner in human cells (Tsuchiya et al., 1996), and Pac10 is the VBP-1 homolog in fission yeast. To investigate the potential role of the prefoldin subunit in pVHL folding, GFP–VHL isoforms were overexpressed in a pac10Δ mutant. Compared to wild-type cells, pac10Δ cells showed a strong reduction in GFP–VHL expression (Fig. 7A). Moreover, formation of GFP–VHL213-positive inclusions was dramatically reduced in pac10Δ cells [Fig. 7B, left panel, 6.5±4.5% (n=757) and 32.4±12.3% (n=610) of LSA-containing cells in pac10Δ and wild-type populations, respectively; mean±s.d.]. This indicated that Pac10 contributed to their formation. However, GFP–VHL213- and GFP–VHL160-positive small aggregates were still detected in pac10Δ cells, strongly suggesting that small puncta and/or SDA were Pac10-independent structures. In pac10Δ cells, the GFP–VHL213 P146A mutant could form significantly more LSAs than wild-type GFP–VHL213 (Fig. 7B right panel vs Fig. 7C, 19.5±1.6% vs 6.5±4.5%, respectively), even though expression levels were similar (Fig. 7A), suggesting that the aggregation-prone P146A mutation bypassed the requirement of Pac10 for pVHL aggregation. Specifically, Pac10 regulated pVHL stability and pac10Δ cells could maintain wild-type pVHL expression below the threshold required for large aggregate formation. However, the aggregation-prone P146A mutation might lower this threshold, also allowing the formation of LSAs in pac10Δ cells.

Misfolded pVHL forms aggregate in vitro (Shmueli et al., 2013; Sutovsky and Gazit, 2004), in yeast cells (Escusa-Toret et al., 2013; Kaganovich et al., 2008; Miller et al., 2015; Specht et al., 2011; Spokoini et al., 2012) and also in mammalian cells (Ogrodnik et al., 2014; Weisberg et al., 2012). In cancer cells, protein quality control mechanisms are increasingly altered with the pathology severity. Therefore, proteins with a high susceptibility to unfold, such as pVHL, might be more destabilized in tumor cells. Furthermore, misfolded proteins can also co-aggregate with normal native partner proteins that consequently cannot carry out their functions (Olzscha et al., 2011). Moreover, some VHL missense mutations do not alter pVHL function directly but affect protein stability, leading to loss of function. Targeting pVHL stabilization is of potential therapeutic value (Ding et al., 2014; Yang et al., 2013). Thus, we investigated potential aggregation-prone pVHL mutations and pVHL aggregation dynamics in fission yeast. We show that pVHL can aggregate in two forms: small dynamic aggregates (SDAs) and large inclusions (LSAs), and these are, at least, due to two distinct parts of the protein (see model Fig. 8). Furthermore pVHL mutations found in individuals with ccRCC promote distinct pVHL aggregation patterns.

Expression of heterologous aggregation-prone proteins has been widely studied in yeast models. We set up a new easy assay to monitor pVHL aggregation in which MG132 and heat stress are not required to promote aggregation as in budding yeast. In contrast to a recent report in budding yeast (Brock et al., 2015), we did not observe any impact of the putative fission yeast Elongin C homolog on pVHL213 stability (data not shown).

Our data are consistent with the results of size exclusion chromatography using recombinant pVHL213 protein that showed the existence of aggregates of different sizes (Shmueli et al., 2013). SDAs may correspond to a local aggregation at a short distance from the site of co-translational production of the polypeptides. We hypothesize that they form immediately after ribosome exit in cells with overwhelmed protein quality control mechanisms due to excessive protein expression. SDAs are very proteasome-sensitive and thus they are only very transiently detected. These structures are formed by an aggregation mechanism involving the N-terminal aggregation-prone regions (APR1 and APR2) in exon 1 that are first exposed to the cytoplasm during protein synthesis. Furthermore, pVHL-positive SDA formation does not require stress conditions nor Hsp104 and might thus only depend on the intrinsic properties of the pVHL primary sequence. We propose that these structures self-assemble because we observed them in all situations, including in hsp16Δ and pac10Δ cells, where the formation of large aggregates is inhibited. SDAs resembled stress-induced Q bodies that are described in budding yeast (Escusa-Toret et al., 2013). In fission yeast, SDAs were not observed in hsp104-GFP- or hsp104-mCherry-expressing (our unpublished results) cells that also overexpressed pVHL213, indicating that they correspond to Hsp104-free aggregates. Fusion of stress-induced aggregates is reduced to 50% in hsp16Δ cells (Coelho et al., 2014). We found that the proportion of GFP–VHL213 SDA- and LSA-containing cells was indistinguishable between wild-type and hsp16Δ cell populations, strongly suggesting that LSAs are not the result of an Hsp16-dependent fusion of SDAs. Thus, we considered that, in fission yeast, SDAs and LSAs correspond to two structurally unrelated pVHL-containing inclusions that form independently. We observed that wild-type GFP–VHL213 and GFP–VHL172 were lost after formaldehyde fixation but that GFP–VHL213 P86L or GFP–VHL160 were not, suggesting that there may be several populations of small aggregates with distinct structural properties.

Even under normal growth conditions, fission yeast cells exhibited Hsp104-positive dots, as seen in budding yeast (Saarikangas and Barral, 2015). Upon pVHL overexpression, large Hsp104-positive pVHL inclusions were located either at the cell ends or close to, but clearly outside of, the nucleus, in contrast with the recent observations in budding yeast (Miller et al., 2015). Spokoini et al. report a link between the IPOD and the large vacuole of budding yeast (Spokoini et al., 2012). However, fission yeast contain approximately eighty small vacuoles that are homogeneously distributed in the cytoplasm. Therefore, in fission yeast, the mechanisms determining the site of ceLSA formation remain to be elucidated.

The disruption of MTs or F-actin did not impact the LSA or SDA localization and dynamics, suggesting cytoskeletal-independent behaviours. Similarly, both the MT and F-actin networks are not involved in stress-induced aggregate formation in fission yeast (Coelho et al., 2014). Moreover, there was no biased localization of pVHL213 LSAs towards one cell pole, suggesting that LSA localization is independent of new cell pole establishment.

Using in silico methods, we identified three potential APRs in pVHL. In yeast, both pVHL160 and pVHL213 form large inclusions, whereas pVHL172, which lacks APR3, only rarely formed inclusions and of small size. This suggests that APR3 is crucial for the formation of large pVHL inclusions. Indeed, pVHL mutations located within or close to APR3 affected its aggregation. The P146A mutation increases the aggregation score of APR3, and this mutant formed larger inclusions than the wild-type pVHL. Conversely, the I151S mutation lowered the aggregation score of APR3 and seemed to decrease the inclusion size, suggesting a correlation between the APR3 aggregation score and the size of inclusions in cells.

The APR1 and APR2 regions are present in all three pVHL isoforms. Because all pVHL isoforms formed SDAs, we hypothesize that APR1 and APR2 are involved in SDA formation. In agreement, the P86L mutation strongly enhanced the APR1–APR2 aggregation score by bridging these two regions in a single long hydrophobic stretch, and we consistently observed a higher proportion of SDAs in the pVHL-P86L-expressing cells.

APR3 lies in the β-strand region of exon 2, whereas APR1 and APR2 are in the N-terminal part of pVHL. The folding of the APR3 region might be slower than that of the N-terminal part and might require distinct chaperones and mechanisms. Our structural analysis of pVHL (Fig. 6A) suggests that the P86L mutation would not disrupt but only disturb the interaction between β-sheets 3 and 7. In contrast, the P146A mutation would disrupt the interaction between β-sheets 8 and 9 and β-sheet 2, leading to an incomplete terminal folding of the pVHL protein. These distinct structural effects of the mutations might explain the formation of different types of aggregates.

The chaperonin TRiC binds to the exon 2 region of pVHL and is involved in pVHL213 folding (Feldman et al., 1999). This essential protein complex is conserved in fission yeast and thus might influence pVHL213 aggregation. However, Feldman et al. (2003) report that alanine substitutions in amino acids 144–147 of pVHL do not alter TRiC binding, whereas the I151S mutation reduces the binding efficiency of pVHL to TRiC (Feldman et al., 2003). Thus, the effect of the P146A mutation on aggregation might not involve an interaction with TRiC. Conversely, the reduced binding of TRiC to pVHL I151S might contribute to the lower stability of this mutant in cells. Defining the exact contribution of TRiC in this process would require conditional mutant alleles.

Ubiquitylation on Lys171 and Lys196 residues has been reported in mammalian cells (Cai and Robertson, 2010; Park et al., 2015). Park et al. recently showed that the 3KR mutant was not ubiquitylated, in contrast to wild-type pVHL213 (Park et al., 2015). These residues are highly conserved in different species, suggesting that ubiquitylation might also be conserved. After verifying that the aggregation score of the 3KR mutant was indistinguishable from that of the wild-type protein, we investigated the aggregation of a 3KR mutant. The 3KR mutation had no effect on pVHL stability, and the percentage of LSA-containing cells was unchanged. Ubiquitylation did not control a specific subcellular localization of pVHL aggregates, but we detected an increase in ceLSA over juLSA formation in 3KR-expressing populations, suggesting that fine-tuned targeting mechanisms might operate to sort aggregate proteins into ceLSA and juLSA.

The prefoldin complex acts in mediating protein folding by targeting substrate proteins to the chaperonin TRiC. Here, we found that in the absence of Pac10 (the VBP-1 fission yeast homolog), pVHL can still form SDAs but not LSAs. This suggests that Pac10 plays a role in the incorporation of misfolded pVHL into large inclusions, possibly through direct interaction with pVHL, as in human cells (Tsuchiya et al., 1996). Moreover, the pVHL protein level was reduced in pac10Δ cells, suggesting that the interaction with Pac10 contributes to pVHL stability. In its absence, misfolded pVHL might be rapidly degraded because SDAs are proteasome-sensitive. We hypothesize that pVHL binds transiently to Pac10 before folding and that this step could be crucial to prevent its rapid degradation. The observation that pVHL P146A, but not P86L or I151S, has a higher propensity to form aggregates than the wild-type protein suggests that this mutation bypasses the protective effect of Pac10. It will be of interest to study the impact of modulating the expression of the prefoldin subunit VBP-1 on the half-life of endogenous pVHL in mammalian cells. Characterization of the prefoldin-mediated folding pathway for pVHL should prove important for a better understanding the VHL disease and for the identification of new pharmacological targets.

It has been shown that VHL mRNA levels are unaffected in some VHL-related pathologies despite loss of pVHL function (Yang et al., 2013). These observations were sometimes correlated to an increased rate of degradation of mutated pVHL. This rate could vary depending on the mutation and/or the protein quality control efficiency of cells. Thus, the exact impact of the loss of pVHL function on the disease remains difficult to predict. This situation warrants a better understanding of the mechanistic principles of pVHL folding and an in-depth characterization of missense pVHL mutations affecting them. Here, we show that some pVHL mutations can result in the incorporation of pVHL213 into insoluble aggregates. Aggregation of pVHL213 into many small aggregates or a large aggregate has the same effect as degradation, which is to lower the cytoplasmic fraction available to perform its tumor suppressor function.

S. pombe strains used in this study are listed in Table S4. Media and genetic methods were as described previously (Moreno et al., 1991). Cells were grown at 30°C. Thiamine (Sigma-Aldrich) was added to a final concentration of 2 μM. For non-standard growth conditions, cells were grown in 1 mM bortezomib (ApexBio), 20 μM LatB (Calbiochem), 25 mg/ml MBC (Sigma-Aldrich) or DMSO (Sigma-Aldrich).

Plasmids used in this study are listed in Table S1. The pCMV-hVHL213 (a kind gift from Dr Alexander Buchberger, Würzburg, Germany), which contains human VHL open reading frame (ORF) cDNA (encoding amino acids 1–213), was used as template for PCR amplification with the VHL1S and VHL3AS primers (pVHL213), or with the VHL2S and VHL3AS primers (pVHL160) (Table S5). The amplicons were subcloned in the pGEM-T vector (Promega) or directly digested with the appropriate restriction enzymes (NdeI and BamHI; New England Biolabs) for ligation into pREP1 or pREP41GFP. The pREP41-VHL172 and pREP41GFP-VHL172 plasmids were obtained after PCR amplification of the pVHL172 ORF (encoding amino acids 1–113+155–213 of pVHL213) from the IMAGE cDNA clone (Genbank BC058831) with VHL1S and VHL3AS. To generate RFP–VHL fusion constructs, a 650-bp fragment was amplified (VHL11SalIS and VHL13SmaIAS primers). This SalI-SmaI-digested PCR fragment was cloned into pSLF279 (generously provided by Kyung-Sook Chung, Korea Research Institute of Bioscience & Biotechnology, Korea). Point mutations (P86L, P146A, I151S or K159R K171R K196R) were introduced into the pVHL213 cDNA that had been inserted into vector pREP41GFP using a sequential two-step PCR approach (Table S5). All constructs were confirmed by Sanger sequencing and transformed into the appropriate yeast strains using the lithium–acetate method.

Cell cultures (50 ml) were harvested by centrifugation at 800 g at 20°C for 2 min, washed in 5 ml ice-cold STOP buffer (10 mM EDTA, 150 mM NaCl, 50 mM NaF, 0.05% NaN₃). Dried cell pellets were stored at −70°C. Cell pellets were lysed in lysis buffer (30 mM HEPES, pH 8.0, 150 mM NaCl, 1% glycerol, 1 mM DTT, 0.5% Triton, ‘Cocktail Set IV’ Calbiochem protease inhibitor) with 425–600-μm glass beads (Sigma-Aldrich) in a precellys24 homogenizer (Bertin) at speed 6100 for 15 s three times and clarified by centrifugation at 6000 g at 4°C for 5 min. Supernatants were divided into 40-μl aliquots. One aliquot was used as total protein; another aliquot was spun at 16,000 g at 4°C for 30 min and designated as the soluble fraction (according to Kaganovich et al., 2008). Pellets were re-solubilized by heating in 50 μl 1× SDS sample buffer. 10 μl of 5× SDS sample buffer was added to the total protein and soluble fraction samples. Equal amounts of each fraction were resolved with SDS-PAGE and transferred onto PVDF Immobilon-P membranes. Membranes were blotted in TBS with 5% milk and 0.05% Tween-20. Anti-VHL (Chesnel et al., 2015) and anti-GFP (11 814 460 001, Roche) antibodies were used at 1:1000 and 1:4000 dilutions, respectively. As loading control, the mouse monoclonal anti-PSTAIR antibody (anti-Cdc2, P7962, Sigma-Aldrich) was used at 1:5000. Secondary antibodies were conjugated to alkaline phosphatase or horseradish peroxidase and revealed with Enhanced ChemiFluorescence (GE Healthcare) or West Dura (Pierce), respectively.

For time-lapse video microscopy, 2 μl of exponentially growing cells were mounted on 2% Edinburgh minimal medium agarose pads, and images were captured at 30°C. GFP, mCherry and RFP signals were captured using spinning disk Nikon TE2000, Leica DMRXA, Leica DMIRB or confocal LEICA DMI 6000 CS microscopes. Nuclei were stained with 200 ng/ml DAPI (Sigma-Aldrich). For detection of the cell wall, cells were fixed with 4% formaldehyde (Sigma-Aldrich) for 30 min, washed in PBS followed by staining with 2.5 μg/ml Alexa-Fluor-488-conjugated isolectin (I21411, Molecular Probes). Images were acquired with the Metamorph software. Aggregate sizes were estimated from the area measured using the ImageJ software. FRAP data were acquired using a spinning disk Nikon TE2000 microscope and the Metamorph software – GFP signals were recorded every second for 3 min, the bleach was applied for 0.5 s after 10 s. Normalized FRAP data were calculated with the ImageJ software. For indirect immunofluorescence experiments, cells were fixed with 4% formaldehyde and processed as described previously (Balasubramanian et al., 1997). Anti-VHL 6030 antibody (Chesnel et al., 2015) and Texas-Red conjugated anti-mouse IgG antibodies (Molecular Probes) were used at 1:50 and 1:200 dilutions, respectively.

We thank Fred Chang (UCSF, San Francisco, CA), Damien Coudreuse (Institut de Génétique & Développement de Rennes, Rennes, France), Colin Gordon (Western General Hospital, Edinburgh, UK), Paul Nurse (The Francis Crick Institute, London, UK), Kenneth Sawin (Institute of Cell Biology, Edinburgh, UK), Per Sunnerhagen (University of Gothenburg, Göteborg, Sweden), Iva Tolic (Ruder Bošković Institute, Zagreb, Croatia), and Kaoru Takegawa (Kagawa University, Kagawa, Japan) for the kind gift of yeast strains, and A. Buchberger, K. S. Chung and Makoto Kawamukai (Shimane University, Matsue, Japan) for plasmids. We thank members of the Molecular Bases of Tumorigenesis: VHL Disease team at Institut de Génétique & Développement de Rennes for helpful discussions. Fluorescence microscopy imaging was performed on the SFR Biosit CNRS UMS3480 microscopy platform. We thank Renata Hancsovszki (University of Debrecen, Hungary) for her technical help, and Sylvain Prigent (Biogenouest) and Stéphanie Dutertre (Biosit) for help with image processing.