Yeast cells reveal the misfolding and the cellular mislocalization of the human BRCA1 protein

Breast and ovarian cancers are a major health concern, with respectively 1.67 and 0.24 million new cases diagnosed worldwide in 2012 (http://globocan.iarc.fr/). Around 10% of these cases are associated with familial transmission, which frequently involves a germline monoallelic mutation in the BRCA1 gene (MIM# 113705). This gene encodes an 1863-amino-acid protein that includes two small conserved and structured domains, the RING domain and the BRCT tandem repeats, forming respectively the N-terminal and C-terminal extremities of BRCA1, as well as a large central region considered as intrinsically disordered (Mark et al., 2005). Two functional nuclear localization signals (NLSs) and two functional nuclear export sequences (NESs) have also been identified (Chen et al., 1996; Fabbro et al., 2002; Thompson et al., 2005).

Although extensively investigated, the cellular localization and function of BRCA1, as well as its connection with cancer development remains incompletely elucidated. The broadly accepted function of BRCA1 is in the maintenance of the genome through regulation of the DNA double-strand break repair pathway. The nuclear BRCA1 protein participates in DNA repair channeling towards the homologous recombination pathway at the expense of the non-homologous end-joining pathway during the S and G2/M phases of the cell cycle. This involves the cell-cycle-dependent localization of BRCA1 at DNA damage sites (Zimmermann et al., 2013) followed by the recruitment of RAD51 through an interaction with PALB2–BRCA2 (Orthwein et al., 2015). Concerning the cellular localization, BRCA1 seems to shuttle between the cytoplasm and the nucleus (Henderson, 2005; Thompson, 2010). The nuclear transport occurs directly, because of the internal NLSs of BRCA1, or indirectly, through interaction with BARD1 (Fabbro et al., 2002).

Sequencing the BRCA1 gene uncovered numerous allelic variations. In the BRCA Share database (http://www.umd.be/BRCA1/), 2020 different DNA variants have been recorded to date. Among them, missense mutations that result in a single amino acid change are the most represented, with 616 different records (31%). Genetic or epidemiological methods allowed the classification of 20% of these records (124/616) either as pathogenic (39/124) or as neutral (85/124), meaning that the medical status of the remaining 498 missense variants is currently unknown. Such lack of information complicates the clinical decision-making after genetic testing of patients and their relatives.

Recently, we proposed evaluating the pathogenic relevance of BRCA1 missense variants by using the yeast Saccharomyces cerevisiae model organism, through four different assays: the colony size, liquid medium, spot formation and yeast localization assays (Millot et al., 2011; Thouvenot et al., 2016). The colony size and liquid medium assays monitor the growth inhibition of yeast cells expressing the full-length wild-type (WT) BRCA1 protein, either by plating the cells on agarose medium and measuring the number of cells in the formed colonies (colony size assay) or by monitoring the concentration of cells grown in liquid medium (liquid medium assay). In these assays, pathogenic mutations, formally classified by genetic or epidemiological methods, have a tendency to alleviate the growth inhibition observed with the WT BRCA1 protein. The spot formation and yeast localization assays rely on the observation that the BRCA1–mCherry fusion protein accumulates in a single aggregate inside the nucleus of yeast cells. This aggregate is referred to as a ‘spot’ due to its visual signature using fluorescence microscopy. Pathogenic missense mutations decrease the proportion of cells showing one fluorescent spot (spot formation assay) and increase the delocalization of brca1 nuclear aggregation into the cytoplasm (yeast localization assay). Assessment of BRCA1 missense variants in yeast offers three advantages: (1) a rapid, large-scale and cost-effective variant classification, (2) the possibility to evaluate variants located in both the N-terminal and C-terminal part of the protein, and (3) a variant classification accuracy of up to 92.5%, as estimated for the colony size assay (Thouvenot et al., 2016). The statistical validation of functional assays is a determinant step for clinical inclusion, which has been completed for these four yeast assays (Thouvenot et al., 2016). However, little is known about the biological relevance of the results of these assays. In other words, the reason why pathogenic mutations modify the formation of fluorescent spots and the growth inhibition remains to be elucidated. Here, we provide evidence that both formation of fluorescent spots and growth inhibition are sensitive to the misfolding effects of pathogenic mutations. In addition, we show that misfolding impairs the transport of BRCA1 into the nucleus of yeast and human cells. Finally, we decipher several aspects of the growth inhibition induced by the expression of BRCA1 in yeast.

In a previous study, we speculated that the full-length BRCA1–mCherry fluorescent spot is a protein aggregate, excluded from the nucleolus, and presumably amorphous rather than amyloid-like, as it is sensitive to sodium dodecyl sulfate (SDS) and to sarkosyl (Millot et al., 2011). To confirm that the fluorescent spot visualized in microscopy is an aggregate, we analyzed the organization of large nuclear structures. We found that the nuclear localization of proteasomal proteins was disturbed by the BRCA1–mCherry aggregate. As observed with fluorescence microscopy, the fluorescent signal of Pre6–GFP, a component of the 20S catalytic core of the 26S proteasome, had an irregular appearance in the presence of BRCA1–mCherry but was homogeneous in its absence (Fig. 1A–D). Pre6–GFP seemed to be excluded from the aggregate and tended to accumulate around the structure, which resulted in a ‘hole’ pattern in fluorescence microscopy. The same was observed with Pre9–GFP, another component of the 20S core of the proteasome (Fig. S1). The genomic DNA was also excluded from the aggregate (Fig. 1E,F). Interestingly, although the BRCA1–mCherry aggregation filled up to one-third of the volume of the nucleus, this did not affect the localization of 50–100-kDa nuclear proteins. Nab2–GFP and Rat1–GFP exhibited a diffusive signal in fluorescence microscopy, with a similar intensity inside and outside the aggregate (Fig. S1). Even the large Rpo21–GFP protein, weighing 191 kDa, was not affected by the BRCA1–mCherry accumulation, suggesting that only very large structures, like the nucleolus, the proteasome and the genomic DNA, are excluded from the amorphous nuclear aggregate.

Next, we benefited from the Pre6–GFP exclusion to determine whether the untagged BRCA1 protein also aggregates in yeast, with the advantage of monitoring living cells, which prevents any potential artifact stemming from cell fixation and anti-BRCA1 antibody labeling. Strikingly, the ‘hole’ pattern was clearly seen in Pre6–GFP cells expressing BRCA1 (Fig. 1A–D). Thus, aggregation of BRCA1–mCherry is not caused by the mCherry fusion protein.

We also recovered additional evidence that the BRCA1–mCherry fluorescent spot is an aggregate. Aggregation occurs when the quality-control network of the cell is overwhelmed. This quality-control system is involved in protein homeostasis. With the contribution of molecular chaperones, it facilitates the folding or refolding of misfolded protein species and regulates protein aggregation and degradation (Tyedmers et al., 2010). Here, we show that the double-deletion of the yeast chaperones ssa1 and ssa2 decreased the number of cells with fluorescent spots (Fig. S2A,B), as well as overexpression of Hsp26 (Fig. S2C,D). However, the Hsp104 protein chaperone had no clear influence on the aggregate formation (Fig. S2E,G). We also tested the potential involvement of the proteolytic degradation pathways. Deletions of either ATG8 or ATG1, which impair the autophagy pathway (Kraft et al., 2012), had no consequence on the formation of fluorescent spots (Fig. S3A–D). Interestingly, inactivation of the proteasome by the deletion of UMP1 (Ramos et al., 1998) significantly raised the number of cells with fluorescent spots (Fig. S3E,F). The proportion of cells showing two spots was increased four-fold in ump1Δ cells compared to WT cells. Taken together, these results confirm that the fluorescent spot visualized in microscopy is an aggregate.

We next investigated whether the two endogenous NLSs present in the BRCA1 sequence were necessary for the nuclear localization of the protein in yeast cells. The two NLSs were inactivated by introducing four missense mutations in each sequence (NLSm–mCherry), as previously described (Fabbro et al., 2002). Compared to BRCA1–mCherry, yeast cells expressing NLSm–mCherry exhibited a decrease in the proportion of cells harboring one spot and an increase in the proportion of cells with two spots as assessed by fluorescence microscopy (Fig. 2A,B). Strikingly, fluorescent spots were systematically detected in the cytoplasm and never in the nucleus (Fig. 2A,C). This indicates that at least one of the two endogenous NLSs of BRCA1 is functional in yeast. It also confirms our previous results using the deficient nucleoporin Nup49–GFP construct, which revealed that an active transport of BRCA1–mCherry from the cytoplasm towards the nucleus is necessary for nuclear aggregation (Millot et al., 2011). NLS inactivation completely impaired this mechanism, resulting in cytoplasmic aggregation.

The GFP-tagged Hsp104 chaperone is known to accumulate around amyloid-prone cytoplasmic aggregates (Kaganovich et al., 2008). Here, Hsp104–GFP colocalized with cytoplasmic but not nuclear aggregates of BRCA1 (Fig. S4A). The same was observed with the Ssa1–GFP chaperone protein (Fig. S4B), with a particularly strong accumulation around NLSm–mCherry cytoplasmic aggregates. In correlative light-electron microscopy (CLEM), the nuclear BRCA1–mCherry spot, visualized in fluorescence, was undetectable in electron microscopy (Fig. S4C). However, the cytoplasmic NLSm–mCherry spot appeared as a large structure that was devoid of membrane and that excluded the ribosomal granulations darkening the cytoplasm (Fig. S4D). Taken together, these results support the notion of an amorphous BRCA1 aggregation in yeast cells, with a nuclear localization of this event that depends on functional BRCA1 NLSs.

Next, we examined whether the conformation of the BRCA1 protein is important for the formation of fluorescent spots. According to the literature (Williams et al., 2003; Lee et al., 2010), we selected four missense mutations that strongly destabilize the structure of the BRCT domain (T1691K, G1706E, R1751P and W1837C) and four missense mutations that exhibit minor effects (R1699W, R1751Q, V1804D and P1859R). The three mutations T1691K, G1706E and W1837C induced a strong decrease in the proportion of cells harboring fluorescent spots compared to WT BRCA1 (Fig. 3A,B and F, left panel). The same result, but to a lesser extent, was observed with R1751P, whereas negative controls exhibited no effects. Thus, misfolding of BRCA1 impairs the formation of fluorescent spots. We confirmed this result using data from 18 additional BRCA1 missense mutations located in the BRCT domain (Fig. S5C; Table S2). From this, we conclude, first, that yeast cells are able to detect misfolding of the heterologous BRCA1 protein, and second, that the lack of formation of nuclear fluorescent spots is an effective readout of misfolding events. Strikingly, the three mutations T1691K, G1706E and W1837C induced a complete delocalization of the spots into the cytoplasm compared to WT BRCA1 (Fig. 3C; Fig. S6A). Again, the same result, but to a lesser extent, was observed with R1751P. Thus, the misfolding of the BRCT domain prevents the BRCA1 protein transport into the nucleus. Taken together, these results show that the spot formation and delocalization are efficient reporters of BRCA1 misfolding. In addition, pathogenic mutations R1699W, M1775K and M1775R do not strongly disturb the structure of BRCA1, and do not substantially alter the formation of fluorescent spots in the nucleus or the delocalization of the spots into the cytoplasm (Fig. S5C,D), which indicates that yeast cells do not reveal the pathogenicity of the mutations per se but the misfolding effect of missense mutations.

To determine whether the delocalizing effect of misfolding mutations is restricted to yeast cells, we introduced the eight mutations in a mCherry–BRCA1 expression vector specific to human cells and evaluated the effect of these mutations in human SV40-transformed fibroblast RG37 cells. The cytoplasmic localization of BRCA1 was clearly increased with the four misfolding mutations but not with the four controls (Fig. 3D; Figs S6B and S7), indicating that misfolding impairs the transport of BRCA1 towards the nucleus. Correlation between results obtained in human and yeast cells reached 0.97 (Fig. 3E). This shows that the the lack of spot formation, as well as the spot delocalization into the cytoplasm, in yeast cells is an effective readout of results obtained either in vitro, concerning BRCA1 misfolding, or in human cells concerning BRCA1 nuclear transport defects.

We next wondered whether the two large structured domains of BRCA1, namely the RING and BRCT domains, located at each extremity of the protein, were involved in the nuclear aggregation observed in yeast cells. Surprisingly, truncation of both extremities of BRCA1 had no impact on the formation of fluorescent spots when compared to the full-length protein (Fig. 4, central construction). This indicated that the intrinsically disordered central part alone is sufficient to promote BRCA1 aggregation in yeast. In contrast, truncation of the intrinsically disordered central part of BRCA1 (denoted the Nter-Cter construction) fully impaired the formation of fluorescent spots in the nucleus (Fig. 4). Instead of a signal restricted to a subdomain of the nucleus, a diffuse signal was visualized filling both the nuclear and cytoplasmic cell compartments, with occasionally an additional small fluorescent spot in the cytoplasm. Taken together, we conclude that the structured domains do not contribute to BRCA1 aggregation. Using the Nter-Cter construction, keeping the first endogenous NLS of BRCA1 (Nter-NLS-Cter construction), the signal visualized in fluorescence microscopy systematically filled the entire nucleus, which is the feature of diffusive nuclear proteins. This confirms that the first endogenous NLS of BRCA1, at least, is operational in yeast. This also shows that the RING or the BRCT domain does not promote aggregation in yeast, even when located in the nucleus.

How can we reconcile the fact that structured domains are not involved in the formation of fluorescent spots and that misfolding mutations in the structured domains impair this spot formation? To answer this question, we integrated misfolding mutations in the Nter-NLS-Cter construction (Fig. 5A). The Y1853X nonsense mutation results in the loss of the ten C-terminal amino acids of BRCA1, which destabilizes the BRCT structure (Williams et al., 2001) and delocalizes the full-length BRCA1 aggregation into the cytoplasm in yeast cells (Millot et al., 2011). Strikingly, both W1837C and Y1853X mutations strongly impaired the nuclear localization of the Nter-NLS-Cter peptide (Fig. 5B,C). Thus, mutations that delocalize the BRCA1 aggregation into the cytoplasm delocalize the Nter-NLS-Cter peptide in the same manner. This demonstrates that the nuclear aggregation of BRCA1 and the delocalization triggered by misfolding mutations are two distinct events. From this, we conclude that yeast cells are able to detect misfolding of synthetic peptides, independently of any aggregation mechanism.

It has been known for a long time that expression of the full-length human BRCA1 protein in yeast cells compromises cell growth (Humphrey et al., 1997; Caligo et al., 2009; Millot et al., 2011). However, the origin of this growth inhibition is poorly understood. Here, we show a connection between nuclear aggregation and growth inhibition. Compared to the WT BRCA1 protein, cell growth was recovered with NLSm (Fig. 6A) and with mutations that misfold the BRCT domain (Fig. 6B). Strikingly, expression of the intrinsically disordered central part of BRCA1 alone was sufficient to promote a growth inhibition comparable to that in cells expressing the full-length protein (Fig. 6C). Moreover, growth rates negatively correlate with the proportion of cells exhibiting one spot in fluorescence microscopy and correlate with spot delocalization (Fig. 6D,E; Fig. S5F,G). Finally, just as with the proportion of cells showing one fluorescent spot and spot delocalization, growth rates negatively correlated with the structural stability of the protein (Fig. S5B,E). All of these results point to the connection between nuclear aggregation and growth inhibition, which is also confirmed by the results previously obtained with the defective Nup49 nucleoporin (Millot et al., 2011). This suggests that, when taking place in the nucleus, aggregation of BRCA1 mediated by its central domain inhibits cell proliferation, and that events leading to decreases and/or delocalization of the aggregation, like misfolding, restore cell growth.

To confirm the connection between nuclear aggregation and growth inhibition at the cellular level, we scrutinized cells using fluorescence video microscopy. In a period of 6 h, 89% of individual cells exhibiting a fluorescent spot did not divide whereas the proportion was 8% when no spot was detected (Movie 1; Fig. 6F). Most of the time, the cell cycle arrest occurred in G1 (Fig. S8A). Thus, the nuclear aggregation of BRCA1 is frequently associated with a cell cycle arrest in G1. This arrest is independent of Mec1, the yeast homolog of ATR involved in G1/S and S/G2 checkpoints (Sidorova and Breeden, 1997), indicating that the DNA damage response is not activated (Fig. S9A,B).

However, a potential cellular toxicity due to visible BRCA1 aggregates, is not sufficient to fully explain the growth inhibition observed. As shown in Figs 2B and 3B, approximately one third of cells exhibit a nuclear spot after 4 h of BRCA1–mCherry expression. This means that in long-term cultures, cells exhibiting the fluorescent spot and blocked in G1 would be expected to be diluted among the cells that do not exhibit the spot and that proliferate. Fig. S10 confirms this dilution effect. However, video microscopy in transillumination (Movies 2 and 3) revealed that the growth inhibition is a combination of cells that grow slowly (compared to vector cells) and cells that are blocked in G1 (Fig. S8B–E). Long-term cultures on plates and in liquid medium confirmed the absence of cells showing rapid cell cycles when expressing BRCA1 (Fig. S8F,G). Finally, this suggests that the growth rate of cells expressing BRCA1, without visible nuclear aggregation, is also hampered, but to a lesser extent than cells exhibiting the nuclear aggregate. Thus, visible nuclear aggregation is not sufficient to fully explain the growth inhibition induced by WT BRCA1 expression.

The plasmid used to express BRCA1 is a 2 µ plasmid, widely employed in yeast. The main advantage of 2 µ plasmids is the copy number per cell, which can reach 100. However, it is also known to be unstable and copy numbers can strongly vary from cell to cell (Futcher and Cox, 1984). We suspected that cells exhibiting the nuclear fluorescent spot, and blocked in G1, are those that contain a high number of plasmids and thus that express a high level of BRCA1 protein. To verify this, we integrated the GAL1p-BRCA1 construction as a single copy in the yeast genome. In this configuration, cell proliferation was unaffected (Fig. 7A, right panel). Surprisingly, the BRCA1 protein could not be detected by western blot analysis (Fig. 7B, middle lane), even through the mRNA is transcribed (Fig. S11). Experiments using the translational inhibitor cycloheximide (CHX) confirmed the strong instability of the BRCA1 protein in yeast, as compared to the tubulin protein, with an approximate half-life of 2 h (Fig. S12A). Such a high instability is consistent with the pattern of degradation visible in western blots, below the major band (e.g. Fig. 7B, left lane). Proteasomal degradation is, at least in part, responsible for the BRCA1 instability, given that the proteasome inhibitor MG132 hampered the decrease of BRCA1 protein levels in cells treated with CHX for 4 h (Fig. S13A). Interestingly, the fluorescent spot formation was reduced concomitantly with protein levels in CHX experiments (Figs S12E,F; S13D,E). This result supports the requirement of strong BRCA1 expression for efficient aggregation.

We also replaced the GAL1 promoter (GAL1p) on the 2 µ plasmid by the constitutive ADH1 promoter (ADH1p). In yeast cells transformed with the ADH1p-BRCA1 construct, the BRCA1 protein was visible in western blot analysis (Fig. 7B, right lane), but the band intensity indicated that, on average, the protein amount per cell was lower than with the GAL1 promoter. In addition, the BRCA1–mCherry spot formation was less efficient with the ADH1 promoter than with the GAL1 promoter (Fig. 7C,D). Finally, a decrease of proliferation was observed with the ADH1 promoter, compared to empty vector cells (Fig. 7A, left panel), but the result is less dramatic than the one classically obtained with the GAL1 promoter (Fig. 7A, middle panel).

All these results show that strong expression is necessary for efficient BRCA1 aggregation and growth inhibition. The cellular heterogeneity regarding nuclear aggregation and cell cycle arrest (Movie 1) is probably the consequence of plasmid copy number variation: both events are obtained only within cells carrying a sufficient number of plasmids. From this, we propose that the increasing expression of BRCA1 leads to a graduated phenotype: from no consequence (single BRCA1 copy inserted), growth slowed down, growth slowed down and nuclear aggregation (rare event in Movie 1), up to cell cycle arrest and nuclear aggregation.

Do the average levels of BRCA1 protein per cell reflect the need of a strong expression system for efficient BRCA1 aggregation and growth inhibition? As shown by western blot analyses, variations in protein levels are observed with truncated forms of BRCA1 but the connection with cytoplasmic aggregation and growth inhibition is not evident (Fig. S14A–C; Millot et al., 2011). In contrast, the misfolding mutations T1691K, G1706E and W1837C, which induce strong delocalization of aggregation into the cytoplasm, as well as strong cell growth restoration, all exhibited particularly high levels of full-length proteins (Fig. 3G). Although the opposite results were expected, this confirms previous results obtained with other missense mutations (Millot et al., 2011; Thouvenot et al., 2016). Interestingly, such results indicate that high protein levels detected in western blots appear to be an additional signature of misfolding induced by missense mutations. As shown in Fig. 3H, right panel, and Fig. S5I, such a readout possesses low sensitivity (not all of the misfolding mutations exhibit a high protein level, as shown by the low protein levels of R1751P) but 100% specificity (a high protein level is support for a missense mutation that generates misfolding events). Intriguingly, the higher levels of mutated BRCA1 protein, when aggregation occurs in the cytoplasm, cannot be explained by better stability of the protein. Indeed, in CHX experiments, the half-life of the full-length BRCA1 protein is similar with or without the presence of the misfolding mutation T1691K (Fig. S12A,B). Moreover, in both cases, the formation of fluorescent spots is strongly affected (Fig. S12E–H) and sensitivity to proteasome inhibition is observed (Fig. S13). In fact, the higher levels of BRCA1 protein occurring with cytoplasmic aggregation could be explained by using the conclusion from the precedent section: strong expression of the WT BRCA1 protein induces a progressive decrease of the proportion of yeast cells carrying a high number of 2 µ expression plasmids during liquid culture due to cell cycle arrest in G1. This diluting effect of cells that contain high levels of BRCA1 proteins would not happen with T1691K, G1706E or W1837C, as cell growth is much less inhibited. This explanation is supported by the analysis of the complete lanes in western blots (blue numbers in Fig. S12A–C), given that the level of peptide is increased 2.94 times between 2 h and 6 h of galactose induction with the T1691K misfolding mutation, whereas this level is stable with the WT BRCA1 protein and the R1751Q control mutation.

From these results, we conclude that a high protein level, observed by western blotting, is a signature of misfolding for certain BRCA1 missense mutations, but that the level of full-length BRCA1 protein per se is not the sole determining factor of growth inhibition. In fact, a crucial point provided by this study seems to be the level of BRCA1 protein transported into the nucleus. Impairment of this transport through endogenous NLS mutations (Fig. 2), Nup49 defect (Millot et al., 2011) or misfolding (Fig. 3 and Fig. S5), as shown by the resulting cytoplasmic aggregation, alleviates the growth inhibition (Fig. 6).

Taken together, these results allow us to propose a model regarding the expression of the full-length BRCA1 protein in yeast cells (Fig. 8): after synthesis in the cytoplasm, the protein is transported into the nucleus, using at least one endogenous NLS, where a high protein concentration triggers amorphous aggregation through the intrinsically disordered central part of BRCA1, as well as a growth inhibition. Reducing the protein concentration in the nucleus diminishes both nuclear aggregation and growth inhibition. This reduction might have different origins, including misfoldings that occur in structured domains of BRCA1, which hinder nuclear transport.

In this work, we show that yeast cells detect misfolding of the full-length human BRCA1 protein. We illustrated this using a total of 14 missense mutations known to misfold the BRCT domain (four mutations in Fig. 3 and ten more in Fig. S5C–E), but the results are also corroborated by analysis of truncations. Indeed, the complete truncation of the RING domain (Millot et al., 2011), BRCT domain (Millot et al., 2011) or both (Figs 4 and 6) does not affect the nuclear spot formation and growth inhibition, contrary to internal truncations in these domains (Millot et al., 2011). This suggests that the conformational state of these domains is essential to the formation of the nuclear spot and the growth inhibition.

We also show that misfolding mutations hamper the nuclear transport of BRCA1 in yeast cells and in the human RG37 cell line. These results consolidate previous results about BRCA1 delocalization obtained in other human cell lines using missense or truncating mutations (Rodriguez et al., 2004), suggesting a connection between the protein structural stability and a nuclear transport clearance. In tumor and normal cells from patients with BRCA1 germline mutations, contradictory results have been reported, like cytoplasmic (Pérez-Vallés et al., 2001), diffusive (Tulchin et al., 2013) or absence of signal (De Brakeleer et al., 2007), probably because of variations in the sensitivity and specificity of the anti-BRCA1 antibodies used and lack of robust results when repeating BRCA1 immunohistological staining experiments (Wilson et al., 1999; Milner et al., 2013). Thus, no sound confirmation or refutation can be currently drawn from the delocalization effect of misfolding mutations using these kinds of experiments.

This study uncovers yeast as a simple and rapid model to assess the structural stability of BRCA1 mutations. The fact that yeast cells detect the misfolding of the Nter-NLS-Cter peptide (Fig. 5) also has interesting prospects in functional assessment of mutations. Indeed, this means that the delocalization of BRCA1 triggered by misfolding could also be observed for other human nuclear proteins that do not necessarily aggregate in yeast. Such a yeast assay, based on the nucleo-cytoplasmic delocalization readout, could be easily derived from the system presented here, the limitation being the length of the peptide analyzed (40–50 kDa minimum) to prevent a passive nucleo-cytoplasmic diffusion through the nuclear membrane (Shulga et al., 2000).

In a previous study, we estimated the clinical relevance of four different yeast assays (colony size, liquid medium, spot formation and yeast localization assays), based on the formation of fluorescent spots, spot localization or growth inhibition readouts, in the classification of BRCA1 missense variants (Thouvenot et al., 2016). Here, we show that misfolding detection is an important component of the readouts of these four assays, as shown by the correlation values obtained (Fig. 3F; Fig. S5B–E). This has several consequences regarding the use of yeast cells to classify BRCA1 variants. First, the fact that the four yeast functional assays are adapted to the assessment of missense mutations located in the RING domain of BRCA1 (Thouvenot et al., 2016), suggests that yeast cells are able to detect misfolding taking place in this domain. Second, we predict that the four functional assays are poorly adapted to missense mutations located in the large intrinsically disordered central part of BRCA1 (1500-amino-acid sequence between the RING and the BRCT domains). Third, the four BRCA1 functional assays could be considered as misfolding more than pathogenic indicators. This means that these yeast assays could also be used as complementary assays to decipher the deleterious effects of variants previously classified as pathogenic by genetic or epidemiological methods or by other functional assays. Finally, other unidentified phenomena, other than misfolding, could also be part of the readouts of the four assays, which would explain the few discrepancies observed between the assay results (Thouvenot et al., 2016).

The toxicity induced by expression of the full-length BRCA1 in yeast has long been a mystery (Humphrey et al., 1997). With the concordance between BRCA1 nuclear aggregation and growth inhibition, which are systematically and concomitantly hampered by mutations, truncations or nucleoporin defects (this study; Millot et al., 2011; Thouvenot et al., 2016), it is tempting to consider a direct contribution of the BRCA1 nuclear aggregate to this toxicity. However, we provide evidence, here, that the growth rate of cells expressing BRCA1 without visible nuclear aggregation is also hampered, but to a lesser extent than cells exhibiting the nuclear aggregate. Thus, it seems likely that nuclear aggregation merely reflects intranuclear BRCA1 concentrations, and that the toxicity is due to a soluble fraction of the complete form of the BRCA1 protein, or of a fragment from degradation. Importantly, we show that the level of BRCA1 protein is a determinant factor of toxicity, provided that these proteins are transported into the nucleus. The endogenous NLSs of BRCA1 are required for the transport. However, the fact that deficiency in Nup49 alleviates the growth inhibition (Millot et al., 2011) might be an indication that the toxicity is not the consequence of importin titration by these NLSs. Bennett et al. proposed that the toxicity results from a Spt4-dependent BRCA1 interaction with the yeast RNA polymerase II, inducing a protease activity that cleaves this polymerase (Bennett et al., 2008). We cannot exclude an effect of the nuclear Spt4 protein in the BRCA1 toxicity, which would corroborate the importance of the BRCA1 localization in the nucleus, but we were unable to confirm the RNA pol II cleavage in western blot experiments (Fig. S14D). However, contrary to previous statements (Humphrey et al., 1997; Bennett et al., 2008; Couch et al., 2008), we prove here that the toxicity induced by the full-length BRCA1 is not mediated by the BRCT domain. This indicates that the toxicity resulting from the full-length BRCA1 protein and from small synthetic NLS-BRCT constructs (Humphrey et al., 1997; Coyne et al., 2004) have distinct origins. Finally, we suspect that the concordance between BRCA1 nuclear aggregation and growth inhibition depends on a specific interaction between the GAL1 promoter and the 2 µ plasmid used. Indeed, in long-term cultures, the nuclear aggregate is hardly seen using the GAL1 promoter (Fig. S10), whereas 10% of the cells exhibit the nuclear aggregate with the ADH1 promoter (Fig. 7D). Further investigation will be required to completely decipher the mechanism of the toxicity induced by the full-length BRCA1 protein in yeast cells, but we show here that the GAL1 promoter-2 µ strategy used in the BRCA1 functional assays is the most efficient in terms of formation of fluorescent spots and growth inhibition.

Finally, this study highlights the BRCA1 expression in yeast as an interesting model of nuclear aggregation, a phenomenon less investigated than cytoplasmic aggregation. We previously reported that the BRCA1 aggregation was amorphous and microtubule dependent (Millot et al., 2011). Here, we show that the cytoplasmic chaperones Ssa1, Ssa2 and Hsp26 are involved in the regulation of this process, suggesting that the nuclear aggregation is, at least in part, controlled in the cytoplasm. Importantly, we believe that our model is adapted to high-throughput screening of chemical compounds that misfold BRCA1, or that would refold mutated forms of BRCA1, similar for instance to the effect of the NSC319726 molecule on the R175H mutated allele of p53 (Yu et al., 2012). In addition, recent studies have reported that chaperone proteins might provide a capacitor effect, which is defined by an ability to mask deleterious effects of mutations that are released when capacitors are absent (Rutherford and Lindquist, 1998; Jarosz and Lindquist, 2010). Given that most of the pathogenic missense mutations seems to induce BRCA1 misfolding (Lee et al., 2010), we suspect that the misfolding effect of missense mutations could be a key component of this capacitor effect. This could explain the low penetrance observed for some missense-mutation-dependent genetic diseases, which would be related to individual variations in chaperone levels. Again, the model of BRCA1 expression in yeast could be adapted to such investigations.

Plasmids are listed in Table S3. Except for the 12 final plasmids of this table, all of the plasmids are derived from pJL45 and pGM40, a modified version of pESC-URA containing the human full-length BRCA1 cDNA (Millot et al., 2011). In these plasmids, expression of the cDNA is controlled by the GAL1 promoter, which is induced by galactose and repressed by glucose, except for pGM35, pGM36 and pGM96, for which the GAL1 promoter has been replaced by the constitutive ADH1 promoter. Missense mutations were generated as in Thouvenot et al. (2016). The two BRCA1 NLSs were replaced by a cassette incorporating the mutated NLSs and coming from the pFlag-BRCA1-NLSm plasmid (gift from Berik R. Henderson, Westmead Institute for Medical Research, Australia). Deletions in the 5′ or 3′ BRCA1 cDNA sequence were produced by standard PCR. In the Nter-NLS-Cter Δ(101-494; 517-1639) construct, the first deletion was replaced by a Ser-Gly₉-Ser-Gly₂-Ser₂ linker, and the second deletion by a Gly-Ser₂-Gly₁₀ linker. In the Nter-Cter Δ(101-1639) construct, the deletion was replaced by a Ser-Gly₉-Ser-Gly₂-Ser₂-Gly₁₀ linker. In the yeast mCherry fusion constructs, the mCherry DNA sequence (GenBank AY678264.1) was separated from the 3′ end cDNA by a Gly₁₀ linker. The human expression plasmids (pGM137, pGM140 and pBB9 to pBB16) are derived from the pEGFP-C1 plasmid, in which the GFP sequence was replaced by BRCA1 sequences, using the AgeI/SmaI digestion. The mCherry sequence was separated from the 5′ BRCA1 cDNA by a Gly₁₀-Lys-Leu-Thr linker. The inserted sequences are under the control of the cytomegalovirus (CMV) promoter. The pGM35, pGM36 and pGM96 plasmids were obtained by replacing the GAL1 promoter of the pJL48, pJL45 and pGM40, respectively, by the ADH1 promoter. The pJL62 and pJL59 are derived from the pRS304 plasmid (Sikorski and Hieter, 1989).

Cells listed in Table S4 were treated as in Millot et al. (2011). To check that the coding sequence was correctly deleted in the genome (Winzeler et al., 1999) or that the coding sequence was correctly fused to the GFP sequence (Huh et al., 2003), a PCR using two primers flanking the integration area was performed on the genomic DNA, and the PCR product was sequenced. Promoters of the HSP26 and HSP104 genes were replaced by the GPD promoter using the pYM-N14 and pYM-N15 plasmids, according to Janke et al. (2004). The ssa1Δssa2Δ strain was built according to (Winzeler et al., 1999). BRCA1 constructs were integrated in the yeast genome at the TRP1 locus, after digestion of pJL59 and pJL62 by EcoRV. In experiments using galactose induction, cells were grown in glycerol lactacte medium (GL), which allows a rapid BRCA1 expression (30 min as shown in Fig. S10D).

Growth inhibition was monitored in liquid medium as described in Thouvenot et al. (2016). Briefly, cells were pre-stimulated in galactose or glucose for 7 h, were resuspended at 10⁶ or 0.5×10⁶ cells/ml, respectively, and were grown for another 15 h. Cell concentration was then measured by optical density at 600 nm (1 OD₆₀₀ corresponding to 1.15×10⁸ cells/ml). For experiments involving the ADH1 promoter, cultures were performed using DO-URA instead of GL-URA medium (Millot et al., 2011). Of note, contrary to in microscopy experiments, the proteins expressed in cell growth experiments are not fused to mCherry, for two reasons: first, these kinds of constructions are used in the colony size and liquid medium assays (Thouvenot et al., 2016) and second, cell growth results are globally similar with or without the mCherry fusion [correlation of 0.96, as shown in Millot et al. (2011)]. For fluorescence microscopy analyses, cells were induced for 4 h with galactose, unless otherwise specified. In Hsp104 inhibition experiments (Ferreira et al., 2001), cells were grown for 3 days in GL-URA medium (Millot et al., 2011) supplemented with 5 mM guanidine hydrochloride. Dilution of the culture was performed twice per day to prevent cell growth saturation. Then, galactose was added and fluorescence microscopy analyses were performed 4 h later.

RG37 cells are SV40-transformed human fibroblasts (Dumay et al., 2006). Cells were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum, 2 mM glutamine, and 200 IU/ml penicillin, and incubated at 37°C under 5% CO2. Thawed batches were tested for contamination prior use. Cells were plated 24 h prior to mCherry–BRCA1 transient transfection (JetPEI, Polyplus, Ozyme). For irradiation, cells were exposed to 10 Gy ionizing radiation 24 h after transfection using a ¹³⁷Cs source (2,8 Gy/min) and a IBL-637 (CIS-BioINternational) gamma irradiator (662 keV photons).

Cells were grown on glass coverslips. At 24 h after transfection, or 4 h or 7 h after irradiation, the cells were washed in PBS and fixed in 4% paraformaldehyde for 15 min at room temperature. The cells were then incubated with PBS containing 1% Triton X-100 for 5 min at room temperature. After blocking in PBS containing 2% BSA and 0.05% Tween-20 solution for 30 min at room temperature, immunostaining was performed using the rabbit anti-mCherry (Institut Curie, APR#13, 1:250 dilution) primary antibody for 1.5 h at room temperature with antibodies diluted in PBS containing 1% BSA and 0.05% Tween-20. Of note, we bypassed the BRCA1 antibody concerns regarding artefactual BRCA1 localizations (Wilson et al., 1999), using mCherry–BRCA1 constructs and the mCherry antibody. Next, the coverslips were incubated for 45 min with Alexa-Fluor-568-conjugated anti-rabbit-IgG secondary antibodies (Life Technologies) at room temperature and mounted in mounting medium (Dako) supplemented with 4′,6-diamidino-2-phenylindole (DAPI). Images were captured using a Zeiss Axio Imager Z1 microscope with a 30× objective equipped with a Hamamatsu camera. Acquisition was performed using AxioVision (4.7.2.). Images were analyzed using the ImageJ software.

Live fluorescence microscopy analyses were performed as previously described (Millot et al., 2011). Unless otherwise specified, cells were induced for 4 h with galactose before microscopy analysis. For each field (images or movies), the final GFP and mCherry pictures obtained were inverted maximum projections of 16 acquisitions along the z-axis, every 300 nm. The number of cells analyzed is shown in Table S1. Cells were confluent during the period of live picture acquisition to increase the number of cells per field and thus to increase the statistical power of picture analysis. The cytoplasmic delocalization of the fluorescent spot was evaluated as previously described (Thouvenot et al., 2016), except that images were not deconvoluted (segmentation of the spots improved). In protein stability experiments, the final concentration of cycloheximide, diluted in ethanol 100%, was 200 µg/ml (Figs S12 and S13). The final concentration of MG132, diluted in pure dimethyl sulfoxide (DMSO), was 100 µM (Fig. S13).

The nuclear DNA exclusion in living cells (Fig. 1E) was observed following the method of Koszul et al. (2009). Briefly, Nup133–GFP cells expressing BRCA1–mCherry were incubated overnight at 4°C in the medium culture, supplemented with 20 µg/ml DAPI, before direct analysis. The montage represented in Fig. 1E is a single slice from 22 deconvoluted acquisitions along the z-axis every 200 nm.

Analysis of fixed cells (Fig. 1F) was carried out as follows. Wild-type (WT) yeast cells expressing BRCA1–GFP were induced for 4 h with galactose. Next, cells were resuspended at 200,000 cells/ml in freshly made fixation buffer (20 mM KPO₄ pH 6.4, 4% formaldehyde) for 1 h at room temperature. Cells were washed twice with 0.1 M KPO₄ and incubated for 3 min in 100 mM Tris-HCl pH 9.4, 10 mM dithiothreitol (DTT). Next, cells were washed in sphero-buffer (100 mM KPO₄, 1.2 M sorbitol) and converted into spheroplasts using sphero-buffer+0.2 mg/ml Zymolyase 100T for 10 min at 30°C. Spheroplasts were fixed on poly-L-lysine-coated coverslips and incubated in 1× PBS, 0.1% Triton X-100 for 10 min, washed in 1× PBS, incubated in 1 µg/ml DAPI for 5 min, washed in 1× PBS and finally mounted for fluorescence microscopy. The montage represented in Fig. 1F is an inverted maximum projection of 16 deconvoluted acquisitions along the z-axis every 200 nm.

The method is described in Heiligenstein et al. (Heiligenstein et al., 2014). Briefly, cells expressing BRCA1–mCherry or NLS–mCherry were induced for 4 h with galactose. Then, cells were vitrified by high-pressure freezing and embedded in HM20 Lowicryl resin to preserve the mCherry fluorescence in the sample. Semi-thin sections (100 nm) were collected on an electron microscopy grid and fluorescence pictures were taken using a 60× oil immersion epifluorescence microscope. Fluorescence images were de-noised and flattened using the eC-CLEM software. The sections were then contrasted for electron microscopy (Reynolds lead citrate and Uranyl acetate) and electron micrographs of the target cells were acquired. Fluorescence pictures were overlaid onto the electron microscopy micrographs using eC-CLEM.

Western blots were performed as previously described (Millot et al., 2011). Unless otherwise specified, cells were induced for 4 h with galactose before cell lysis and total protein analysis. We showed that the mCherry fusion has little impact on western blot results (Thouvenot et al., 2016). Thus, lysates from cells expressing BRCA1 fused to mCherry were sometimes used in western blot analyses: (1) when the anti-BRCA1 antibody epitope was not present (i.e. constructions depicted in Fig. S14B,C) and (2) in Figs S12 and S13. The primary antibodies used were mouse anti-BRCA1 (MS110, Calbiochem, Billerica, MA, ref. OP92, 1:200 dilution), rat anti-Tubulin (YL1/2, AbD serotec, Oxford, UK, ref. MCA77G, 1:2000 dilution), rabbit anti-mCherry (Institut Curie, APR#13, 1:1200 dilution), goat anti-Rad53 (YC19, Santa Cruz Biotechnology, Dallas, TX, ref SC-6749, 1:600 dilution) and mouse anti-RNA pol II (8WG16, Covance, Emeryville, CA, ref MMS-126R, batch E10015AF, 1:500 dilution). The secondary peroxidase-conjugated antibodies were anti-mouse-IgG (Jackson Immunoresearch, West Grove, PA, USA, ref 115-035-146, 1:10,000 dilution), anti-rat-IgG and anti-rabbit-IgG (Jackson Immunoresearch, ref 112-035-062, ref 711-035-152, ref 1:5000 dilution), and anti-goat-IgG antibodies (Promega, Madison, WI, USA, ref V0851, 1:5000 dilution). The ImageJ software was used to quantify signal intensities.

Statistical analyses were performed as previously described (Millot et al., 2011). Results are summarized in Table S1. A total of three independent clones were analyzed if a significant difference was obtained using the first clone.

The authors would like to sincerely thank Beric R. Henderson for the pFlag-BRCA1-NLSm plasmid (Westmead Institute for Medical Research, Australia; Fabbro et al., 2002), Anne Peyroche and Benoît Le Tallec for the BLT67 strain (Service de Biologie Intégrative et Génétique Moléculaire, Commissariat à l'Energie Atomique, France; Le Tallec et al., 2007), Michel Toledano (Service de Biologie Intégrative et Génétique Moléculaire, Commissariat á l'Energie Atomique, France) for the Hsp104–GFP strain, Allyson Holmes and Benoît Le Tallec for critical reading of the manuscript, Angela Taddei's team, Frank Perez, Sophie Loeillet, Marc Lecuit, Thomas Boudier, Judith Lopes, Lélia Soter, Patricia Le Baccon, Jean-Baptiste Sibarita, Vincent Fraisier and Lucie Sengmanivong, as well as the Nikon Imaging Center and the PICT-IBiSA imaging facility, for technical assistance.