Selective Accumulation of Aggregation-Prone Proteasome Substrates in Response to Proteotoxic Stress^{▿ †}

Cell culture and transfections.

The human melanoma Mel JuSo cell lines were cultured in Iscove's modified Dulbecco medium (Sigma-Aldrich) supplemented with 10% fetal calf serum (Sigma-Aldrich), 10 units/ml penicillin, and 10 μg/ml streptomycin (Sigma-Aldrich). Generation of stable cell lines expressing Ub-R-YFP (where YFP is yellow fluorescent protein), Ub^G76V-YFP, and YFP-CL1 has been described previously (23). The Mel JuSo ZsProSensor-1 cell line was created by transfection of Mel JuSo cells with the ZsProsensor-1 plasmid (BD Bioscience Clontech) using Lipofectamine (Invitrogen). Clones were selected in the presence of 0.5 mg/ml G418 (Gibco) and screened for ZsGreen fluorescence upon administration of the proteasome inhibitor MG132 (Biomol).

For electroporation, cells were harvested and resuspended in 2 mM HEPES, 15 mM K₂HPO₄-KH₂PO₄, pH 7.2, 250 mM mannitol, and 1 mM MgCl₂. The cell suspension was mixed with 10 μg of plasmid DNA, transferred to 4-mm electroporation cuvettes (Bio-Rad), and subsequently pulsed with the Bio-Rad GenePulser (15 pulses; 350 V; 25 μF). Directly after electroporation, cultivation medium was added, and cells were cultivated for 24 h.

Where indicated, proteasome activity was inhibited with 200 nM epoxomicin (Biomol) or 10 μM MG132 (Biomol). Proteotoxic stress was induced by a 1-h incubation at 42°C or a 1-h incubation with 5 μg/ml puromycin (BD Biosciences). Inhibition of protein translation was performed by the addition of 50 μg/ml cycloheximide for the indicated times.

Fluorescence microscopy and flow cytometry.

Mel JuSo cell lines were grown on coverslips, fixed with 4% paraformaldehyde (Sigma) or 3:1 methanol-acetone for γ-tubulin detection, permeabilized with 0.2% Triton X-100, and stained with Hoechst 33258 (Molecular Probes). Immunostainings were performed using rabbit polyclonal ubiquitin and mouse monoclonal γ-tubulin antibodies (Sigma-Aldrich). Fixed cells were examined with a Zeiss LSM 510 confocal laser scanning microscope (Plan-Neofluar 40× oil objective with a 1.3 numerical aperture). Quantitative analyses of fluorescence intensities were performed using ImageJ software. Flow cytometry was performed with a FACSort flow cytometer (Becton, Dickinson, and Co.) and analyzed with Cellquest software.

Time-lapse imaging.

Confocal laser scanning microscopy was performed using a Zeiss LSM 510 META equipped with a cell culture microscopy stage at a constant temperature of 37°C. Photoactivation of PAGFP-Ub in transfected cells was performed by applying a single pulse with 405-nm laser light at full intensity to a small region in the cell. Prior to and directly after the 405-nm laser pulse, images were taken. The fluorescence intensities in nonexposed regions of the cellular compartments were quantified using ImageJ software.

For fluorescent recovery after photobleaching (FRAP), the GFP fluorescence in a 0.8-μm strip of a selected cell was bleached using a single pulse of the 488-nm argon laser at 100% intensity. GFP fluorescence was monitored every 20 ms using the 488-nm argon laser excitation at 0.5% intensity. The obtained fluorescence recovery curves were normalized to the prebleach fluorescence.

Protein solubility assay.

For determining protein solubility, cells were trypsinized and lysed in 1% Triton X-100 in phosphate-buffered saline supplemented with complete protease inhibitor cocktail (Roche Diagnostics). Supernatant and pellet fractions were separated by centrifugation at 10,000 × g for 10 min and analyzed by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and Western blotting.

Immunoprecipitation.

Untreated and thermally stressed cells were harvested and lysed in 1% NP-40, 150 mM NaCl, 50 mM Tris-HCl, pH 8.0, containing 20 mM N-ethylmaleimide (Sigma-Aldrich), 10 μM MG132 (Biomol), and complete protease inhibitor cocktail (Roche). Lysates were precleared by centrifugation at 10,000 × g for 10 min. The YFP-tagged proteins were immunoprecipitated using rabbit polyclonal anti-GFP antibodies (Molecular Probes).

Thermally induced proteotoxic stress causes global changes in the ubiquitin equilibrium.

Our observation that degradation of ubiquitin-dependent but not ubiquitin-independent substrates is hampered during proteotoxic stress and that there is an inverse correlation between the levels of free ubiquitin and the reporter substrate suggested that limited ubiquitylation may be involved in the stress-induced UPS dysfunction. In order to study the dynamics of ubiquitin in living cells, we used a cell line stably expressing GFP-Ub. Several studies showed that GFP-Ub behaves in a way very similar to endogenous ubiquitin (8, 34, 35). We observed the formation of GFP-Ub-positive inclusions in the perinuclear region of cells exposed to heat shock (Fig. ​(Fig.2A).2A). These structures were not observed with the conjugation-deficient GFP-Ub^K0,G76V fusion (see below). The most predominant of the GFP-Ub inclusions were associated with γ-tubulin, a localization typical for aggresomes (Fig. ​(Fig.2B).2B). Notably, aggresomes are the primary disposal sites for aggregation-prone proteins (18, 42), underscoring the notion that a large fraction of stress-induced ubiquitylated substrates are misfolded proteins (40).

Open in a separate window

FIG. 2.

Thermally induced proteotoxic stress causes global changes in the ubiquitin equilibrium. (A) Fluorescent micrographs of GFP-Ub untreated cells, cells subjected to heat shock, and cells 2 h post-heat shock. The lower panels are higher-magnification micrographs of representative cells, depicted in the upper panel by squares. Aggresome-like structures are indicated by arrows. Scale bars, 20 μm (top) and 10 μm (bottom). (B) Fluorescent micrographs of the γ-tubulin immunostaining of untreated GFP-Ub cells and GFP-Ub cells 2 h after heat shock. The fluorescent signals of GFP-Ub (right), γ-tubulin (middle), and merged images (right) are shown. GFP-Ub inclusions colocalized with the centrosomal marker γ-tubulin, indicated by arrows. Scale bar, 10 μm. (C) Fluorescent micrographs of untreated (upper panels) and heat-shocked (lower panels) cells expressing PAGFP-Ub before photoactivation in a selected region in the cytosol (prepulse) and directly after photoactivation (postpulse). The contours of the cells and the photoactivated regions (indicated by red-lined squares) are depicted. The images are shown in false colors to illustrate differences in fluorescence intensities. The look-up table bar is provided on the right. (D) Quantification of the increase in fluorescence intensities in the nucleus after photoactivation of cytosolic PAGFP-Ub and PAGFP-Ub^k0,G76V in untreated and heat-shocked cells. The overall intensity is averaged from the intensities in three nuclear regions of interest (ROI), indicated by the filled gray squares in the diagrams of panel C. Error bars are standard deviations (n = 5), and the Student's t test values are shown.

Next, we photoactivated PAGFP-Ub in a small region in the cytosol and quantified the instant redistribution of PAGFP-Ub to the nucleus. Conjugation of PAGFP-Ub to a protein substrate is likely to result in a product that exceeds the size exclusion limit for passive diffusion through the nuclear pore, whereas free PAGFP-Ub based on its molecular weight can, like untagged ubiquitin, freely diffuse between the compartments (8). We therefore used the instant exchange between the cytosolic and nuclear compartments as a surrogate marker for free ubiquitin levels. Although due to its larger molecular weight the behavior of PAGFP-Ub may be slightly different from untagged ubiquitin, we anticipated that the comparison of the diffusion of this fusion protein under two different conditions might nevertheless give useful information on the pool of free ubiquitin in living cells. We found that heat shock significantly decreased the amount of PAGFP-Ub that reached the nucleus (Fig. 2C and D) but did not affect the exchange of the conjugation-deficient PAGFP-Ub^K0,G76V mutant (Fig. ​(Fig.2D).2D). Thus, analysis of PAGFP-Ub in living cells supports the observation that heat shock causes a reduction in the pool of free ubiquitin.

Increased conjugation of ubiquitin to stress-induced immobile proteins.

In order to distinguish mobile from immobile ubiquitin pools, we performed FRAP experiments (36) in control and stressed cells stably expressing GFP-Ub or the GFP-Ub^K0,G76V mutant. The heat shock-induced formation of γ-tubulin, GFP-Ub-positive foci was dependent on ubiquitin conjugation as this was not observed in GFP-Ub^K0,G76V-expressing cells (see Fig. S3 in the supplemental material). FRAP analysis revealed that heat shock caused a substantial increase in the pool of immobile ubiquitin in the nuclear and cytosolic compartments (Fig. ​(Fig.3A,3A, upper panels). Proteotoxic stress did not affect the mobility of GFP-Ub^K0,G76V, indicating that the stress-induced immobilization is due to conjugation to ubiquitin to immobile substrates (Fig. ​(Fig.3A,3A, lower panels). Immobilization of nuclear ubiquitin was not attributed to conjugation of ubiquitin to histones (a major fraction of immobile nuclear ubiquitin) since heat shock has been shown to cause a global deubiquitylation of histones (8). This suggests that the immobilization of ubiquitin is due to conjugation to ubiquitin of stress-induced immobile targets that are localized in the nucleus and cytosol.

Open in a separate window

FIG. 3.

Increased conjugation of ubiquitin to stress-induced immobile proteins. (A) FRAP curves of GFP-Ub and GFP-Ub^K0,G76V in the nucleus and cytoplasm in untreated and heat-shocked cells. (B) FRAP curves of Luc-GFP and GFP in the nucleus and cytoplasm in untreated and heat-shocked cells. The curves in panels A and B represent the average of 10 cells. (C) Fluorescent micrographs from untreated and heat-shocked cells coexpressing mRFP-Ub and Luc-GFP. The merged images are shown in the lower panels. Scale bar, 10 μm. (D) Western blot analysis of Triton X-100 solubility assays of untreated and heat-shocked parental Mel JuSo cells. Ubiquitin conjugates are detected in total cell lysate (T), detergent-insoluble pellet fraction (P), and soluble fraction (S). (E) Western blot analysis of Triton X-100 solubility assays of untreated and heat-shocked Mel JuSo cells expressing GFP-Ub. GFP-Ub conjugates are detected in total cell lysate (T), detergent-insoluble pellet fraction (P), and soluble fraction (S). α, anti; Mw, molecular weight in thousands.

The major sources of proteasome substrates that arise during heat shock are thermolabile proteins (31). Therefore, we tested whether the increased immobilization of ubiquitin could be explained by conjugation of ubiquitin to unfolded proteasome substrates. Heat shock also immobilized a large fraction of GFP-tagged firefly luciferase (Luc-GFP), a thermolabile protein (28), in the nucleus and cytosol (Fig. ​(Fig.3B,3B, upper panels), while GFP remained highly mobile (Fig. ​(Fig.3B,3B, lower panels). Furthermore, Luc-GFP and mRFP-Ub cosequestered in aggresome-like structures in the perinuclear region of cells exposed to heat shock (Fig. ​(Fig.3C).3C). Heat shock caused a clear increase in the amount of Triton X-100-insoluble ubiquitin conjugates (Fig. ​(Fig.3D)3D) and GFP-Ub conjugates (Fig. ​(Fig.3E).3E). Together, these data indicate that the stress-induced disturbance of the ubiquitin equilibrium is accompanied by increased conjugation of ubiquitin to immobile and detergent insoluble substrates.

Restoration of ubiquitin levels is followed by clearance of soluble but not aggregation-prone proteasome substrates.

To gain better insight into the behavior of different substrates after heat shock, we followed the levels of our reporter substrates for several hours after the insult. A striking difference was observed between the levels of the soluble (Ub-R-YFP and Ub^G76V-YFP) and the aggregation-prone (YFP-CL1) reporter substrates after heat shock. The initial accumulation of Ub-R-YFP and Ub^G76V-YFP during the first 2 to 4 h after heat shock was followed by clearance of the reporters to low steady-state levels, suggesting full restoration of the UPS (Fig. ​(Fig.4A).4A). In contrast, the increased levels of YFP-CL1 persisted (Fig. ​(Fig.4A)4A) and remained high up to 16 h after heat shock, the latest time point assessed (data not shown). Similar patterns were found when steady-state levels of all reporters were analyzed by Western blotting (Fig. ​(Fig.4B4B).

Open in a separate window

FIG. 4.

Restoration of ubiquitin levels is followed by clearance of soluble but not aggregation-prone proteasome substrates. (A) Quantification of the YFP fluorescence from flow cytometric analyses of Ub-R-YFP, Ub^G76V-YFP, and YFP-CL1 cell lines up to 8 h after heat shock. The fluorescence intensities of the YFP reporters are normalized to 0% in untreated cells and 100% for cells treated with proteasome inhibitor for 8 h. Error bars are standard errors of the means (n = 5). (B) Western blot analyses of the Ub-R-YFP, Ub^G76V-YFP, and YFP-CL1 cell lines before and after heat shock for 0, 2, 4, 6, and 8 h. The reporter levels are detected with GFP antibodies. GAPDH levels are shown as a loading control. (C) Western blot analysis of the endogenous short-lived proteins p53 and c-Myc. (D) Turnover of p53 and c-Myc in control cells (lanes 1 to 3) and heat-shocked cells (lanes 4 to 6). Cycloheximide (CHX) was added to block protein synthesis. In the absence of CHX, cells accumulated p53 after heat shock (lanes 7 to 9). The blots were probed with p53 and c-Myc antibodies. GAPDH loading controls are shown in panels C and D. Untr, untreated; HS, heat shock; arrow, c-Myc band; asterisk, nonspecific band.

The inverse correlation between free ubiquitin and proteasome substrates was also observed for the endogenous proteasome substrates p53 and c-Myc (Fig. ​(Fig.4C).4C). Short-lived proteins in general are involved in signaling pathways, and elevation in their steady-state levels can reflect transcription upregulation. To determine whether increased p53 and c-Myc levels are caused by either UPS impairment or stress-induced transcription, we inhibited protein translation following heat shock by treating with cycloheximide. Upon inhibition of protein synthesis and in the absence of proteotoxic stress, cellular levels of p53 and c-Myc drastically dropped, showing that these proteins are short-lived in Mel JuSo cells (Fig. ​(Fig.4D,4D, lanes 1 to 3). Exposure of the cells to heat shock caused a striking stabilization of p53 and c-Myc, suggesting that proteasomal degradation of these proteins was blocked (Fig. ​(Fig.4D,4D, lanes 4 to 6). In the absence of translation inhibition, p53 and c-Myc levels increased after proteotoxic stress in line with stress-induced stabilization (Fig. ​(Fig.4D,4D, lane 7 to 9). These data demonstrate that both soluble and aggregation-prone proteasome substrates are stabilized following heat shock but that only soluble proteasome substrates are cleared once UPS function is restored.

Overexpression of ubiquitin prevents accumulation of soluble but not aggregation-prone proteasome substrates.

The inverse correlation between the levels of free ubiquitin and soluble substrates provides indirect evidence suggesting that the reduction in free ubiquitin may be responsible for UPS dysfunction in cells exposed to heat shock. An important prediction based on this model is that increasing the ubiquitin pool prior to the insult should prevent UPS dysfunction. Since total cellular ubiquitin levels are already high, overexpression of ubiquitin is not trivial. Accordingly, we found previously that expression of a single ubiquitin-encoding open reading frame from a cytomegalovirus promoter only marginally increased ubiquitin levels (8). To allow robust overexpression of ubiquitin, we cloned the open reading frame of a ubiquitin precursor encoding nine ubiquitin monomers in frame with mRFP. Since ubiquitin precursors are rapidly processed by ubiquitin C-terminal hydrolases (45), this chimeric Ub^[9]-mRFP^[1] precursor is expected to be rapidly converted into nine wild-type ubiquitin monomers and one mRFP protein (Fig. ​(Fig.5A5A).

Open in a separate window

FIG. 5.

Overexpression of ubiquitin prevents accumulation of soluble proteasome substrates. (A) Schematic representation of the strategy for quantitative overexpression of ubiquitin with the Ub^[9]-mRFP^[1] construct. (B) Fluorescence micrograph of cells expressing Ub^[9]-mRFP^[1]. Fluorescence of mRFP (upper panel) and immunostaining for ubiquitin (lower panel) are shown. The contours of untransfected cells are outlined. Scale bar, 20 μm. (C) The mRFP fluorescence intensities of Ub^[9]-mRFP^[1]-transfected cells are plotted against the corresponding ubiquitin immunostaining intensities. Each spot represents one cell. The correlation coefficient (R²) is shown. (D) Quantitative analysis of ubiquitin immunostaining intensities in untreated parental cells after thermal stress induction (8 h after heat shock) and expressing Ub^[9]-mRFP^[1]. Error bars are standard deviations (n = 50). (E) Western blot analysis of ubiquitin and mRFP levels in parental Mel JuSo cells and in mRFP-N1- or Ub^[9]-mRFP^[1]-transfected cells. The band corresponding to free ubiquitin is indicated. GAPDH levels are shown as a loading control. (F) Analysis of the band densities corresponding to free ubiquitin on Western blots of untransfected (−) and mRFP-N1- or Ub^[9]-mRFP^[1]-transfected cell extracts. Error bars are standard error of the means (n = 10), and a Student's t test value between untransfected and Ub^[9]-mRFP^[1]-transfected cells is shown. (G to I) Quantitative analysis of the YFP reporter accumulation of mRFP- and Ub^[9]-mRFP^[1]-expressing Ub-R-YFP (G), Ub^G76V-YFP (H), and YFP-CL1 (I) cells; untreated (Untr) cells, cells 2 h after heat shock (post-HS), and cells incubated with proteasome inhibitor (Inh) were analyzed. YFP intensities from RFP-negative and -positive cell populations are shown. Error bars represent standard errors of the means (n = 50 to 100). There is a significant differences between Ub^[9]-mRFP^[1]-expressing cells, control, and mRFP-positive cells (unpaired Student's t test, P < 0.001). AU, arbitrary units; α, anti; CMV, cytomegalovirus; Mw, molecular weight in thousands.

Ubiquitin immunostaining of cells transiently transfected with the Ub^[9]-mRFP^[1] construct showed that mRFP-positive cells had increased levels of ubiquitin (Fig. ​(Fig.5B).5B). Quantitative analysis revealed a linear correlation between mRFP and ubiquitin levels in transfected cells (Fig. ​(Fig.5C)5C) and showed that ubiquitin levels from Ub^[9]-mRFP^[1]-expressing cells exceeded the levels that were obtained by stress-induced expression (Fig. ​(Fig.5D).5D). Western blot analysis on total lysates of transiently transfected Mel JuSo cells confirmed that expression of Ub^[9]-mRFP^[1] caused an increase in the density of the free ubiquitin band (Fig. 5E and F).

We next used expression of the Ub^[9]-mRFP^[1] precursor to increase the ubiquitin levels before the proteotoxic insult. Microscopic analysis showed that heat-shocked Ub-R-YFP (Fig. ​(Fig.5G;5G; see also Fig. S4A in the supplemental material) and Ub^G76V-YFP (Fig. ​(Fig.5H)5H) cells expressing Ub^[9]-mRFP^[1] did not accumulate the reporter substrate, unlike untransfected cells or cells expressing mRFP. Ubiquitin overexpression did not prevent accumulation of reporters in response to proteasome inhibitor, which was anticipated since the proteasome is acting downstream of ubiquitylation (Fig. 5G and H). Strikingly, we consistently found that overexpression of ubiquitin failed to prevent the accumulation of the aggregation-prone YFP-CL1 substrate (Fig. ​(Fig.5I;5I; see also Fig. S4B in the supplemental material). We conclude that changes in the ubiquitin equilibrium are responsible for UPS impairment during thermal proteotoxic stress. The failure to sustain degradation of aggregation-prone substrates in stressed cells even when the ubiquitin levels had been experimentally increased suggests that these substrates are intrinsically different from soluble proteasome substrates.

Stress-induced ubiquitin disequilibrium can cause UPS impairment.

The data presented in this study show that ubiquitin is implicated in heat shock-inflicted impairment of the UPS in mammalian cells and that overexpression of ubiquitin alone is sufficient to prevent UPS dysfunction, conclusively showing that ubiquitin is a limiting factor during thermal proteotoxic stress. Although this may seem surprising given the high intracellular levels of ubiquitin in metazoans (7), it is well established that two ubiquitin precursor encoding genes are stress inducible (12) and that the yeast Saccharomyces cerevisiae has specific mechanisms to increase disassembly of polyubiquitin chains during proteotoxic stress (15), suggesting an intimate link between cellular stress and the ubiquitin homeostasis.

Despite the fact that cells express large quantities of ubiquitin (7), it appears that the levels of ubiquitin available for ubiquitylation can nevertheless be limiting under certain conditions (Fig. ​(Fig.7A).7A). Ubiquitin modifications are reversible due to the counteracting activity of deubiquitylation enzymes, which serve to replenish the pool of free ubiquitin (45). This dynamic process of ubiquitylation and deubiquitylation appears to be essential for cellular homeostasis since a large number of ubiquitin-dependent processes compete for the limited amount of free ubiquitin (8). We have proposed that this delicate ubiquitin equilibrium may be of functional significance as it links various ubiquitin-dependent cellular processes in an intricate multilayered network with ubiquitin as a common regulator (14). The present study implies that the bulk of the proteasome substrates degraded under normal conditions face a shortage of ubiquitin during heat shock, resulting in stabilization of soluble and aggregation-prone short-lived proteins (Fig. ​(Fig.7B7B).

Open in a separate window

FIG. 7.

Schematic representation of the UPS impairment induced by thermal proteotoxic stress. Model for the functionality of the UPS in the absence of proteotoxic stress (A), in cells experiencing proteotoxic stress (B), and during the recovery phase after the proteotoxic insult (C).

Several observations in this study suggest that a large fraction of the stress-induced ubiquitylated proteins may not be targeted for proteasomal degradation. First, the increase in polyubiquitin adducts is primarily due to an increase in large, insoluble, and immobile polyubiquitylated proteins, characteristics that do not match the features of typical proteasome substrates, which are generally soluble and mobile entities (32). Second, whereas protein aggregates resist proteasomal degradation (43) and can obstruct proteasome activity (1), the proteasomes in thermally stressed cells are operative since increasing ubiquitin levels restore degradation, making it unlikely that these proteasomes are actively engaged in the attempted degradation of these insoluble and immobile polyubiquitylated proteins. Although we cannot exclude at this point that ubiquitylation of these proteins reflects a misplaced and failed attempt to target them for proteasomal degradation, it seems plausible that a fraction of the stress-induced ubiquitylation may serve another function in the stress response. In this respect, it is noteworthy that a number of recent studies suggest that formation of stress granules (21) and aggresomes (19, 29), as well as the autophagy-mediated clearance of aggregated proteins (2, 30), relies on polyubiquitylation. Thus, the enhanced ubiquitylation of insoluble, immobile substrates may reflect the cell's attempt to redirect these acute substrates from proteasomal degradation to storage in inclusion bodies or clearance by the autophagosomal/lysosomal pathway. Unfortunately, our data do not allow us to draw conclusions on when aggregation-prone substrates are targeted for ubiquitylation. In the light of a potential function in handling aggregation-prone proteins, determining whether polyubiquitylation takes place before or after their incorporation in inclusions would be informative.

Persistent accumulation of aggregation-prone substrates.

An intriguing observation is that whereas cells rapidly removed accumulated soluble proteasome substrates once the free ubiquitin levels were restored, they failed to do so with the accumulated aggregation-prone substrates (Fig. ​(Fig.7C).7C). This condition is not specific for proteotoxic stress that is accompanied by ubiquitin stress since a similar phenomenon was observed in puromycin-treated cells. This could reflect the intrinsic problems that the proteasome encounters in degradation of these aggregation-prone proteins in stressed cells (43). Thus, it is feasible that during proteotoxic stress, as a consequence of improper folding of proteins in combination with molecular crowding, these proteins form stable aggregates that resist proteasomal degradation. This finding may also provide an explanation for the selective accumulation of aggregated proteins in conformational diseases (6) since episodes of proteotoxic stress, which some cells may encounter as part of their normal physiology, may result in the selective precipitation of aggregation-prone proteins that have been redirected from proteasomal degradation to inclusion bodies.

We thank Harm Kampinga, Jacques Neefjes, and Lars-Gunnar Larsson for reagents and stimulating discussions and the members of the Dantuma lab for helpful suggestions.

The work in the Dantuma lab was supported by the Swedish Research Council, the Swedish Cancer Society, the Hereditary Disease Foundation, the Marie Curie Research Training Network (MRTN-CT-2004-512585), the Wenner-Gren Foundation, and the Karolinska Institute. N.P.D. is supported by the Swedish Research Council. J.P.T. is supported by NIH grant NS053825.