Vulnerability of newly synthesized proteins to proteostasis stress

A key pathologic feature of neurodegenerative diseases including Alzheimer's disease, fronto-temporal dementia (FTD), Parkinson's disease, amyotrophic lateral sclerosis and Huntington's disease is the presence of intracellular and/or extracellular accumulations of insoluble protein aggregates (for a review, see Muchowski and Wacker, 2005). Because of the similarities in the pathologies of these diseases, it has been proposed that they share common pathogenic mechanisms. However, no single common mechanism has yet emerged; the literature contains reports that describe inhibition of proteasome function (Bence et al., 2001), damage to mitochondria (reviewed by Cho et al., 2010), damage to membrane organelles (reviewed by Butterfield and Lashuel, 2010), compromised chaperone function (reviewed by Kikis et al., 2010; Voisine et al., 2010) and autophagic failure (reviewed by Wong and Cuervo, 2010). Thus, an emerging concept is one of multisystem failure as a component of neurodegeneration.

One hypothesis that could underpin the multi-system failure aspects of neurodegenerative disease is that the production and/or accumulation of these misfolded proteins might lead to an imbalance in the cellular protein quality control network or proteostasis network. This network functions in the folding, post-translational modification and degradation of proteins involved in diverse functional pathways (reviewed by Balch et al., 2008; Kikis et al., 2010). A common pathologic feature of neurodegenerative disease is the accumulation of misfolded proteins within vulnerable neuronal or glial cell populations. In each disease, the pathologic protein aggregates are composed of a specific protein, or in some cases multiple proteins, that pathologically define the disease. The accumulation of such proteins is in-and-of-itself an indication that protein homeostasis is compromised, but recent studies have introduced the concept of collateral misfolding in which the accumulation of a specific misfolded mutant protein disrupts protein homeostasis in a manner that compromises the ability of other bystander proteins to correctly fold (Gidalevitz et al., 2006, 2009).

The concept of bystander misfolding was first uncovered in studies involving C. elegans that express the first exon of the human huntingtin (HTT) gene, which encodes a CAG repeat sequence that codes for glutamine (Gidalevitz et al., 2006). Expansion of the CAG repeat is causative for Huntington's disease (The Huntington's Disease Collaborative Research Group, 1993) and produces a protein that is prone to aggregate (Scherzinger et al., 1999). The expression of mutant Htt in the muscle wall of C. elegans produces inclusion pathology with muscle dysfunction (Satyal et al., 2000). Gidalevitz and colleagues have demonstrated that expression of mutant Htt exon-1 in C. elegans produces a burden on the proteostatic network such that co-expressed proteins harboring temperature-sensitive folding mutations are unable to achieve functional structure at temperatures that such mutants are normally functional (Gidalevitz et al., 2006). The authors hypothesized that, in settings of proteostatic stress, proteins harboring rare sequence variations might be vulnerable to misfolding due to an inherently greater dependence on proteostatic machinery to achieve functional conformations (Gidalevitz et al., 2006).

In a previously published proteomic analysis of two human neural cell lines, we embarked on an effort to identify human proteins that might be vulnerable to proteostatic stress (Xu et al., 2012). Using a thermal stress paradigm in two independently derived human cell lines, we identified ∼30 cytosolic and nuclear proteins that lose solubility upon heat shock (Xu et al., 2012), indicating that vulnerability to proteostatic stress is not restricted to rare-variant proteins but applicable to a larger population of normal proteins. Among the proteins identified as sensitive to thermal proteostatic stress were several that emerged as possible biomarkers that consistently lose solubility, including ubiquitin, CDK1, FEN1 and TDP-43 (also known as TARDBP). In cells held at 37°C, these proteins are predominantly, sometimes exclusively, found in PBS soluble fractions (Xu et al., 2012), whereas in the heat-stressed cells these proteins are detected in detergent-insoluble fractions. In the present study, we used human HEK293 cell models to compare the expression of mutant huntingtin (Q103) as a proteostasis stressor to thermal stress. The HEK293 cell model has been routinely used as an expression system to study protein aggregation related to neurodegenerative disease, including mutant Htt (Bence et al., 2001; Schilling et al., 2007; Waelter et al., 2001). We demonstrate that ∼5–10% of the soluble cytosolic and nuclear proteome in HEK293 cells is vulnerable to misfolding when proteostatic function is overwhelmed by thermal stress. Using this paradigm further in HEK293 cells, we demonstrate that inhibiting new protein synthesis 30 min prior to heat-shock prevented the appearance of insoluble cytosolic proteins. Focusing on a subset of thermal-sensitive biomarker proteins, we show that the overexpression of mutant Htt (Q103) induces the misfolding of the same proteins that are sensitive to thermal stress. As was the case in the thermal stress paradigm, inhibiting new protein synthesis reduced the impact of mutant Htt on proteostatic function. These findings suggest that stressors that burden the proteostatic network produce the greatest impact on newly made proteins, with a substantial portion of the proteome being vulnerable to such stress.

HEK293 cells and their derivatives, such as 293T and 293FT cells, have been extensively used as model cells for the expression of mutant proteins implicated in neurodegenerative disease, including expression of mutant Htt (Bence et al., 2001; Schilling et al., 2007; Waelter et al., 2001), mutant superoxide dismutase 1, for modeling familial amyotrophic lateral sclerosis (Karch and Borchelt, 2008; Prudencio et al., 2009; Wang et al., 2003), α-synuclein, for modeling Parkinson's disease (Paxinou et al., 2001), and tau to model various tauopathies (Mirbaha et al., 2015). In each of these cases, expression of a mutant protein, or in some cases a wild-type (WT) protein, produces aggregate inclusions that resemble structures seen in tissues of humans with the disease and relevant animal models. Given that the HEK293 cell is a workhorse cell line in these types of studies, we sought to identify proteins in HEK293 cells that might be vulnerable to proteostatic stress. Following protocols established in our previous study of human SH-SY5Y and CCF-STTG1 cells (Xu et al., 2012), we conducted a liquid chromatography tandem mass spectrometry (LC-MS/MS) analysis of proteins in HEK293 cells that lose solubility after thermal stress. For comparison, as an alternative means to produce proteostatic stress, cells were also treated with MG132 to inhibit the proteasome, causing an accumulation of misfolded proteins that would normally have been degraded (for a review, see Balch et al., 2008). Sequential detergent extraction was used to separate proteins by solubility in phosphate-buffered saline (PBS), then Nonidet P-40 (NP40), then sodium deoxycholate (DOC), with a final extraction of the DOC-insoluble fraction in SDS (Xu et al., 2012). This sequential approach of fractionation was used to identify the detergent that provided the best separation of soluble and insoluble proteins. The goal was to identify an insoluble fraction that had relatively low complexity, enabling greater efficacy in protein identification (Xu et al., 2012). The SDS-insoluble fractions (SDS-P) contained very little protein and were not analyzed further. Portions of each fraction were separated by SDS-PAGE and immunoblotted for ubiquitin to identify which fractions contained insoluble polyubiquitin (Fig. 1A). Interestingly, most of the ubiquitylated proteins in cells treated with MG132 were soluble in PBS (Fig. 1A, PBS-S, condition 2). Notably, although ubiquitin immunoblotting revealed a smear of immunoreactivity in the insoluble fractions, Coomassie-Blue-stained gels demonstrated well-defined protein bands present in all fractions (Fig. 1B). In comparing the effect of MG132 to thermal stress, in Coomassie-Blue-stained gels, we clearly observed a larger number of proteins in each of the insoluble fractions from heat-shocked cells (Fig. 1B; compare condition 2 to 3). Thus, by comparison, thermal stress seemed to be a more severe insult than proteasome inhibition.

To identify the normally soluble proteins that lose solubility upon heat-shock, we separated PBS-soluble (PBS-S) fractions from cells held at 37°C and DOC pellet (DOC-P) fractions of heat-shocked and control cells by SDS-PAGE before cutting each lane into six to eight pieces for trypsin digestion and LC-MS/MS. Compilation of the proteins identified from these fractions resulted in the identification of 970 proteins from the HEK293 cells with a relatively low false-positive rate (FPR) of less than 0.1% for proteins and less than 0.4% for peptides from 18,559 spectra. Among these 970 proteins (Table S1; a Scaffold file of the complete list of 970 proteins can be provided upon request), 315 proteins were found in both the PBS-S and DOC-P fractions, and 321 proteins were found only in insoluble DOC-P fractions. Regardless of cell treatment, there were 334 proteins found only in the soluble PBS-S fractions. The most abundant peptides for this latter class of proteins were from DYNC1H1 (dynein), TUBB3B (tubulin), HUWE1 (E3 ubiquitin ligase), PKM2 (pyruvate kinase), FLNA (filamin), PLS3 (plastin) and TPR (nucleoprotein). Ultimately, there were 44 proteins identified in HEK293 cells as proteins that seemed to specifically lose solubility in thermal stress (Table S2); the most abundant of these was ubiquitin. Given the much larger number of proteins for which the only peptides identified were in soluble fractions, we conclude that ∼10–20% of the soluble proteome might be vulnerable to thermal stress.

To determine which types of proteins lose solubility in these HEK293 cells, we analyzed the LC-MS/MS data using a statistical method as previously described (Xu et al., 2012, 2013). Apart from ubiquitin, we identified 13 additional proteins in which the frequency of peptide identification in the DOC-insoluble fraction of heat-shocked cells was significantly (P<0.05) more abundant than predicted in two experiments (Table 1). There were 33 additional proteins in which we identified no peptides in the DOC-insoluble fractions from 37°C cells in both LC-MS/MS experiments, but the frequency of peptide identification in the DOC-insoluble fraction of one experiment was less than five and thus did not achieve statistical significance in both experiments (Table S2). If we include the larger dataset in analyses of protein classification, most of the proteins identified as showing shifts in solubility were nuclear proteins (Fig. S1), which is consistent with our previous findings in thermal stressed SH-SY5Y and CCF-STTG1 cells (Xu et al., 2012). In comparing the list of proteins identified by LC-MS/MS as sensitive to thermal stress in HEK293 cells to those identified in the two neural cell lines, we note that six proteins that lost solubility were common to all three cell lines (Fig. 2). An additional seven proteins were identified by LC-MS/MS as losing solubility in both the HEK293 and SH-SY5Y cells. Based, in part, on the commercial availability of specific antibodies, we identified five proteins whose losses of solubility could be used as biomarkers of proteostatic stress, these being ubiquitin, MATR3, FEN1, TDP-43 and CDK1.

To determine whether proteostatic stress caused by misfolded protein pathology produces bystander misfolding in mammalian cells, we examined the effect of expressing fragments of mutant Htt (Htt-103Q) on the solubility of the biomarkers proteins identified above (ubiquitin, MATR3, FEN1, TDP-43 and CDK1). HEK293FT cells were transfected with N-terminal fragments of Htt fused to cyan fluorescent protein (CFP) to allow us to flow sort the transfected cells and isolate them from untransfected cells (Fig. 3A). We compared the effects of expressing Htt–CFP encoding a 25Q repeat, to the effects of expressing Htt–CFP encoding a 103Q repeat. Immunoblots of sorted cells demonstrated the expression of both the Htt25Q–CFP and Htt103Q–CFP protein fragments (Fig. 3B). In cells expressing the Htt103Q–CFP fragment, the expressed protein formed inclusions, whereas the Htt25Q–CFP fragment showed a diffuse distribution in the cytoplasm (Fig. 3C). Cells transfected with either Htt25Q–CFP or Htt103Q–CFP showed increased levels of polyubiquitin, but the intensity of immunoreactivity was consistently higher in the cells expressing the Htt103Q–CFP protein (Fig. 4A). These cell lysates were then subjected to detergent extraction and centrifugation to separate soluble and detergent-insoluble proteins, and then these fractions were examined by SDS-PAGE and immunoblotting. In cells expressing the Htt103Q–CFP protein, we observed increased levels of insoluble MATR3, FEN1, TDP-43 and CDK1 (Fig. 4B–D). In unstressed cells, these proteins were either exclusively or largely, found in PBS-soluble fractions. These findings indicate that the aggregation of mutant Htt fragments produces a sufficient level of proteostasis stress to cause other cellular proteins to lose solubility.

Previous studies in yeast have demonstrated that inhibiting protein synthesis for 30 min prior to thermal stress prevents the accumulation of polyubiquitin after heat-shock (Medicherla and Goldberg, 2008), and thus we turned back to the thermal stress paradigm to determine whether the proteins identified in HEK293 cells as vulnerable to thermal stress were newly-made versions of these proteins. To accomplish this goal, we treated cells with cycloheximide (CHX), a protein synthesis inhibitor, 30 min prior to heat-shock. As predicted from the yeast work (Medicherla and Goldberg, 2008), treatment of cells with cycloheximide before heat-shock prevented the accumulation of insoluble polyubiquitylated proteins (Fig. 5A, last 2 lanes; Fig. 5B,C). Similarly, treatment of cells with CHX prevented the appearance of CDK1, FEN1, TDP-43 and MATR3 in insoluble fractions (Fig. 5B, lanes 4–6; Fig. 5C). Adding CHX to the cells just prior to harvest did not prevent the loss of solubility (Fig. S2, lanes 7–9), indicating the effects of CHX on protein solubility required at least 30 min of prior incubation. Immunoblots of the soluble fractions from these cells demonstrated that levels of CDK1, FEN1, TDP-43 and MATR3 in the soluble fractions of cells treated with CHX were similar to that of the untreated cells (Fig. S3). Thus, the absence of these proteins in insoluble fractions of cells treated with CHX was not due to the absence of the protein. Rather, the data show that pre-existing forms of these proteins (older than 30 min) retain solubility after heat-shock.

To determine whether proteostatic stress caused by mutant huntingtin expression shows a similar pattern of vulnerability, we transfected HEK293FT cells with the huntingtin vectors encoding the N171 fragment of Htt (with 18 and 82Q) (Schilling et al., 2007), waited 24 h, and then treated some of the cells with cycloheximide for 6 h before detergent extraction and immunoblotting for ubiquitin (Fig. 6). Similar to the heat-shock paradigm, cells expressing mutant huntingtin that were treated with cycloheximide showed lower levels of insoluble polyubiquitin immunoreactivity (Fig. 6A, compare lanes 7 and 8). Quantification and statistical analysis of replicate experiments demonstrated that cycloheximide treatment significantly reduced the levels of insoluble ubiquitin (Fig. 6A, graph). Immunoblots of the insoluble fractions from these cells demonstrated that these fractions contained mutant Htt fragments as expected for this aggregation prone protein (Fig. 6B). Immunoblots of the PBS soluble fractions demonstrated high levels of Htt-N171-18Q fragments along with soluble forms of Htt-N171-82Q fragments (Fig. 6B). Additionally, these blots demonstrated that the cycloheximide treatment did not dramatically reduce the overall accumulation of Htt in these transfected cells. Immunofluorescence stains of transfected cells demonstrated that in both cultures most of the expressed Htt was diffusely distributed in the cells (Fig. 6C), a pattern consistent with soluble protein as indicated by the immunoblots. However, in the cells expressing Htt-N171-82Q we observed that a subset of cells harbored inclusions (Fig. 6C, arrows). Overall, these findings are consistent with our data from the heat-stress paradigms and suggest that newly synthesized proteins represent the major component of the proteome that is vulnerable to misfolding during proteostatic stress caused by mutant huntingtin expression.

In the present study, using a heat-shock paradigm in HEK293 cells, we identified multiple proteins that rapidly lose solubility upon thermal stress. Many of the same proteins were previously identified as losing solubility upon thermal stress in human SH-SY5Y and CCF-STTG1 cells (Xu et al., 2012). On the basis of consistency among the cell lines examined and on the availability of commercial antibodies, we identified a set of 5 proteins whose losses of solubility could be used as biomarkers of proteostatic stress, these being ubiquitin, MATR3, FEN1, TDP-43 and CDK1. Here we demonstrate that these proteins lose solubility when N-terminal fragments of mutant huntingtin are expressed at a level that induces intracellular inclusion formation. Thus, these biomarker proteins for thermal stress are also vulnerable to bystander misfolding caused by pathologic protein aggregation. Using protein synthesis inhibitors, we found that it is the newly synthesized versions of these biomarker proteins that are selectively sensitive to proteostatic stress. Apart from CDK1, FEN1, TDP-43 and MATR3, our LC-MS/MS data from thermal stress experiments suggests that 10–20% of the soluble proteome (defined as proteins that are normally soluble in PBS) might be transiently vulnerable to proteostatic stress. Although it comes as no surprise that newly made proteins are heavily dependent upon the proteostatic network, our data indicate that it is these proteins that represent the population that is most vulnerable to collateral misfolding when an aggregating proteotoxin such as mutant Htt accumulates.

The proteins we identify as being sensitive to thermal stress are involved in multiple cellular functions and although the modest loss of function of any one of these proteins might be inconsequential, it is difficult to know what the consequences would be of a concerted partial loss of function of multiple proteins in a particular functional pathway. For example, CDK1 is a master regulator of mitosis and is responsible for the phosphorylation of a large number of substrates (for a review, see Domingo-Sananes et al., 2011), FEN1 is essential for excision DNA repair (for a review, see Tsutakawa et al., 2011), and MATR3 and TDP-43 are RNA-binding proteins that regulate mRNA splicing and possibly other RNA metabolic processes (Coelho et al., 2015; Polymenidou et al., 2011). Partial loss of function of these activities and potentially many others could compromise cellular function or viability. In acute proteostatic stress settings, if only newly made proteins are vulnerable to the stressor, then one would not expect sustained loss of function to occur as the system would reset once the source of stress was removed. However, in settings of chronic proteostasis stress, one could envision that the proteins we identified as vulnerable could have a substantial loss of functional molecules that could produce biological effects. Cultured cells are not well suited for modeling chronic insults and future effort to understand the impact of chronic insults, such as Htt aggregation, on proteostatic function and bystander misfolding will require applying these techniques to in vivo models.

We recognize that our LC-MS/MS sampling of the proteome is not all inclusive. There is sampling error (Xu et al., 2012), and proteins of low abundance are inherently difficult to quantify by the spectral counting approach we have used here because spectral numbers below five generally do not achieve statistical significance. Additionally, our method focuses on identifying cytosolic and nuclear proteins that lose solubility and thus we have not sampled the membrane-bound component of the proteome and can make no estimate on their vulnerability to proteostatic stress. Overall, our data provide the first estimation of the number and types of cytosolic and nuclear proteins that might be vulnerable to bystander misfolding.

Our principal means of identifying misfolded proteins relies on the loss of protein solubility in detergent. In studies of proteins implicated in neurodegenerative disease, there are many examples in which loss of solubility in detergent distinguishes a misfolded form of a protein from a more natively folded form (Ishihara et al., 1999; Johnston et al., 2000; McKinley et al., 1991; Scherzinger et al., 1999; Wang et al., 2003). Unlike these disease-specific examples, we believe that the change in solubility observed in our experimental paradigm occurs when exposed hydrophobic domains of non-natively folded proteins drive non-specific interactions to produce large heterogenous complexes. Because we analyzed the cells during continuous thermal stress, we did not expect to see inclusions containing these proteins to form. Among the four proteins that we identified as vulnerable to proteostasis stress, one of the most obvious candidate proteins to produce inclusion bodies is TDP-43. This protein has been shown to localize to cytosolic stress granules upon heat-shock (McDonald et al., 2011), which could change its solubility (Liu-Yesucevitz et al., 2010). TDP-43 is also a component of inclusion pathology found in individuals with fronto-temporal dementia (FTD-U) that present with ubiquitin immunoreactive inclusions that are tau negative (Neumann et al., 2006). Subsequently, this protein has been found in inclusions in sporadic ALS (Neumann et al., 2006) and mutations were then discovered in TDP-43 in patients with familial ALS and FTD (Kabashi et al., 2008; Rutherford et al., 2008; Sreedharan et al., 2008). More recently, TDP-43 immunoreactive inclusion pathology has been reported in patients with Alzheimer's disease, Pick's disease and Huntington's disease (Freeman et al., 2008; Hasegawa et al., 2007; Uryu et al., 2008). A notable feature of TDP-43 is that it encodes a domain with high sequence similarity to yeast prion domains of Sup35 and Psi (King et al., 2012). In yeast, overexpression of wild-type TDP-43 produces inclusion structures (Couthouis et al., 2011), and TDP-43 is a major component of inclusion body pathology of ALS and FTD. Thus, TDP-43 would seem to be a prime candidate to form inclusions upon proteostasis stress. However, when we immunostained our heat-shocked cells with antibodies to TDP-43 to determine whether cells form TDP-43 immunoreactive stress granules, or inclusions, within the time-frame of our experiments none were observed (Fig. S4). Thus, we conclude that the insoluble TDP-43 observed in our stressed-cell models is likely to have been dispersed in small structures that coalesce into large heterogeneous aggregates when cells are lysed.

The forms of CDK1, FEN1, TDP-43 and MATR3 that were detected in the insoluble fractions migrated at the expected size and were thus not ubiquitylated. Additionally, as previously reported for the SH-SY5Y and STTG-1 cells, in heat-stressed HEK293 cells almost all of the ubiquitin peptides possessing the K-GlyGly motif, which identifies a site of ubiquitin conjugation, were linked to ubiquitin itself at K48 or K63 to form polyubiquitin chains. The K-GlyGly modification of K48 was identified from 16 mass spectrometry samples (corresponding to 16 gel pieces) with 20 unweighted spectra. Ubiquitylation at K63 was also found but to a much lesser extent. As expected, most of the ubiquitin-conjugated peptides (both K48 and K63) identified by LC-MS/MS were found in insoluble fractions after 42°C treatment. At present, we do not know the identity of the ubiquitylated proteins that lose solubility. Apart from ubiquitin itself, only histone H3.1 was identified definitively to contain K-GlyGly peptide conjugates that are signatures of ubiquitinylation. Given that inhibiting new protein synthesis also inhibited the accumulation of polyubiquitin, we assume that ubiquitin must be conjugated to misfolded protein substrates, but if 10–20% of the soluble proteome is misfolded by thermal stress to become a potential substrate for ubiquitination, then it might be that no particular protein is ubiquitylated a sufficient number of times to allow detection. It is also important to note that it is clear that most ubiquitin is conjugated to itself. It is possible that sequestering the ubiquitin from possible conjugation to misfolded protein substrates is a means of limiting the volume of proteins to be degraded, or is a means to delay degradation, allowing induced chaperones an opportunity to repair some of the proteins that lost solubility as result of proteostatic stress.

In conclusion, we demonstrate here that the overexpression of mutant Htt in human cells can induce what appears to be the equivalent of the bystander misfolding that was first described in C. elegans models (Gidalevitz et al., 2006). In cells in which proteostatic stress has been induced by thermal stress or mutant Htt expression, newly made components of the proteome appear to be the most vulnerable to ‘secondary’ misfolding. These findings imply that reducing the burden of newly made proteins might be a way to allow the proteostatic network to regain balance. Inhibitors of protein synthesis would surely have short-term toxicity issues, but a treatment that substantially reduced protein synthesis for some period might be one means to push the system back in balance.

Human embryonic kidney 293 (HEK293) cells [CRL-1573™, American Type Culture Collection (ATCC), Rockville, MD, USA] (3–4 passages from frozen seed stock) were cultured in Dulbecco's modified Eagle's medium (DMEM) with 10% fetal bovine serum (FBS) and 4 mM L-glutamine at 37°C in 5% CO₂. Cells used to generate samples for LC-MS/MS were cultured in 100-mm dishes. Cells were heat-shocked at 42°C for 1 h before harvesting, or held in 37°C incubators as controls. To inhibit the proteasome, cells at 70–80% confluency were treated with culture medium containing 10 nM MG-132 [N-(benzyloxycarbonyl)-leucinylleucinylleucinal, Z-Leu-Leu-Leu-al] (EMD Millipore, Billerica, MA) for 12 h before harvesting.

For experiments in which cells were treated with cycloheximide {CHX; 3-[2-(3,5-Dimethyl-2-oxocyclohexyl)-2-hydroxyethyl]glutarimide; Sigma-Aldrich, St Louis, MO} before heat shock, HEK293 cells were cultured in 60-mm dishes at 37°C. After the cells had reached 90% confluency, the medium was replaced with fresh medium and then 5 µl DMSO containing 2.5 mg CHX was added into each dish and mixed well. The cells were kept in a 37°C incubator for 30 min, and then moved to a 42°C incubator for 3 h. Control experiments included cells in which 5 µl of DMSO was added before the 3 h of heat shock at 42°C, and cells in which 5 µl of DMSO+CHX was added right after heat shock and just before harvesting (see Table S3).

For Htt transfection experiments, 293FT cells (Invitrogen, Carlsbad, CA; 3-4 passages from seed stock) were used because they transfect more efficiency than the original HEK293 cells. The 293FT cells were plated in poly-D-lysine-coated 60-mm dishes. When cells were 80% confluent, they were transfected with expression vectors for Htt103Q–CFP and Htt25Q–CFP (kind gift of Ron Kopito, Stanford University, CA), using Lipofectamine 2000 (Invitrogen) following the manufacturer's protocol; 100 units/ml penicillin and 100 μg/ml streptomycin were added to avoid contamination. At 48 h post-transfection, the cells were harvested by trypsin treatment, then diluted into complete growth medium to stop the reaction, and then washed in PBS. The PBS-suspended cells were subjected to fluorescence-activated cell sorting (FACS) with a BD FACSARIA II sorter (BD Biosciences, San Jose, CA) to isolate transfected cells. The same number of fluorescent and non-fluorescent cells for each transfection was collected for analysis.

The sequential method of detergent extraction and ultracentrifugation followed previously published protocols (Xu et al., 2012). Briefly, cells cultured in 100-mm dishes were harvested by scraping in PBS, then sequentially extracted in buffers containing three different detergents to fractionate the proteins according to their solubility (Xu et al., 2012). The sequential extraction procedure produced eight samples: total crude cell lysate, PBS-S (PBS soluble), NP40-S (NP-40 soluble), DOC-S (deoxycholate soluble), SDS-S (SDS soluble); PBS-P (PBS insoluble fraction), NP40-P (PBS soluble but NP-40 insoluble fraction), and DOC-P (NP-40 soluble but deoxycholate insoluble fraction). The pellet after SDS extraction contained little protein, and this fraction was not analyzed. All the samples used for LC-MS/MS were generated by above protocols.

In experiments in which cells were transfected, or treated with CHX or other drugs, smaller culture plates were used (60 mm), and thus fewer cells were extracted. The initial homogenization in PBS for these cells was conducted in 200 µl, then 400 µl TEN with 1% NP-40 was used to extract the PBS insoluble pellet, followed by centrifugation at top speed in a benchtop microcentrifuge for 10 min. The NP40 insoluble pellet was then extracted in TEN with 2% DOC with the resuspended protein homogenate transferred to a new 1.5 ml centrifuge tube before centrifugation at 16,000 g for 10 min. Finally, each of the DOC insoluble pellets were resuspended in 65 µl 1× Laemmli buffer with 15 µl loaded on SDS-PAGE for immunoblot assay.

The methods used for mass spectrometry have been described by Xu et al. (2012). Briefly, 45 µl of each fraction to be analyzed (e.g. PBS-S or DOC-P) was loaded per lane for SDS-PAGE (4–20% Tris-Glycine gel) followed by Coomassie Blue staining. The lanes for each sample were cut longitudinally, and then each gel strip was cut into five to eight pieces, from low to high molecular mass. Standard in-gel trypsin digestion was used prior to LC-MS/MS following a protocol used by the Protein Chemistry Core (ICBR, University of Florida, Gainesville, FL). The digested peptides from all the samples were analyzed with a QSTAR^® XL instrument (Applied Biosystems, Foster City, CA), a hybrid quadrupole time-of-flight mass spectrometer. A 90-min gradient was used for liquid chromatography separation. Tandem mass spectra were extracted by Analyst QS (version 1.1). Charge state deconvolution and deisotoping were not performed. All MS/MS samples were analyzed using Mascot (Matrix Science, London, UK; version Mascot) and X! Tandem (The GPM, thegpm.org; version 2007.01.01.1). X! Tandem was set up to search a subset of the ipi.HUMAN.v3.80 database assuming trypsin. Mascot was set up to search the IPI_human_20070808 database (3.32, 67,524 entries) (Kersey et al., 2004) assuming the digestion enzyme trypsin. Mascot and X! Tandem were searched with a fragment ion mass tolerance of 0.30 Da and a parent ion tolerance of 0.30 Da. Iodoacetamide derivative of cysteine was specified in Mascot and X! Tandem as a fixed modification. S-carbamoylmethylcysteine cyclization (N-terminus) of the N-terminus, deamidation of asparagine and glutamine, oxidation of methionine and ubiquitination residue of lysine were specified in Mascot and X! Tandem as variable modifications.

Scaffold (version 3_0_07, Proteome Software Inc., Portland, OR) was used to validate MS/MS-based peptide and protein identifications. Peptide identifications were accepted if they could be established at greater than 95.0% probability as specified by the Peptide Prophet algorithm (Keller et al., 2002). Protein identifications were accepted if they could be established at greater than 99.0% probability and contained at least two identified peptides. Protein probabilities were assigned by the Protein Prophet algorithm (Nesvizhskii et al., 2003). Proteins that contained similar peptides and could not be differentiated based on MS/MS analysis alone were grouped to satisfy the principles of parsimony.

The numbers of unweighted spectrum counts per protein were tabulated from the unfiltered Scaffold data for comparison. The method of spectral counting was used for relative protein quantitation to test whether the abundance of a particular protein was significantly higher in one sample than another sample (Prokai et al., 2009). The method for relative quantitation followed published protocols (Higgs et al., 2005; Old et al., 2005), in which the change in abundance was determined by the ratio of the total number of identified MS/MS spectra (unweighted spectral count from Scaffold) for a particular protein in the heat-shock treated and control groups, respectively. A statistical G-test (likelihood ratio test for independence) was then utilized to determine the statistical probability that the abundance of a particular protein in a particular fraction was higher or lower than expected (Old et al., 2005; Sokal and Rohlf, 1995). To increase statistical power for G-test analysis of the initial high-confidence protein identifications (99% protein confidence, 95% peptide confidence and containing two unique peptides), we included in the data sets peptides with lower Mascot scores that represent true positive identifications at 50% probability to match. Differences in protein composition between fractions were considered highly significant if the G-test significance was P<0.05.

The gene ontology analysis of both the heat-induced insoluble proteins used the PANTHER (Protein ANalysis THrough Evolutionary Relationships) system (http://www.pantherdb.org) (Thomas et al., 2006).

For the samples used for LC-MS/MS, equal volumes of each fraction (10 µl) were loaded per well, and the proteins were separated by 10–20% Tris-Glycine SDS-PAGE. One gel was stained with Coomassie Blue and a parallel gel was used for immunoblotting using monoclonal antibodies against ubiquitin [5-25 (1:1000), Covance, Princeton, NJ]. For SDS-PAGE of samples for immunoblotting we used either 10–20% or 4-20% Tris-Glycine gels. Antibodies used in immunoblotting were against ubiquitin [rabbit anti-ubiquitin (1:1000), cat. no. Z045801-5, DAKO, Carpinteria, CA; and mouse monoclonal anti-ubiquitin antibody, cat. no. MCA-UBi-1(1:1000), Encor Biotech Inc. Gainesville, FL], CDK1 (rabbit polyclonal, 1:500, cat. no. HPA003387, Sigma-Aldrich), FEN1 antibody (rabbit polyclonal, 1:500, cat. no. SAB4500881, Sigma-Aldrich), TDP-43 antibody (mouse monoclonal, 1:5000, cat. no. MCA-3H8, Encor Biotech. Inc. Gainesville, FL), MATR3 (rabbit monoclonal, 1:10,000, cat. no. ab151714, Abcam, Cambridge, UK) and Htt (monoclonal, 1:1000, clone 2B4, cat. no. MAB5492, Millipore, Boston, MA) were used according to standard protocols.

The culture medium was removed from the cells in 60-mm dishes, then 2 ml 4% paraformaldehyde was added to the cells and held at 25°C for 20 min, followed by washing three times in PBS (5–10 min each). After blocking in 3% normal goat serum in PBS with 0.05% Triton X-100, anti-TDP-43 antibody (1:1000 dilution, MCA-3H8, Encor Biotech) or anti-Htt antibody (2B4, 1:3000 dilution, Millipore) was added and then incubated at 4°C overnight. After washing in PBS with 0.05% Triton X-100, Alexa-Fluor-586-conjugated goat anti-mouse-IgG was added and incubated at 25°C in darkness for 1 h. After three washes in PBS, all liquid was drained and one drop of mounting medium with DAPI was added to the dish before coverslips were placed onto samples. An Olympus BX60 epi-fluorescence microscope was used to capture the fluorescence images.

The authors are grateful to Interdisciplinary Center for Biotechnology Research (ICBR) in University of Florida for assistance with LC-MS/MS analysis. We thank Stanley Stevens (University of South Florida) for helpful suggestions regarding data analysis. We are grateful to Dr Gerry Shaw for providing antibodies to ubiquitin and TDP-43.