Physicochemical properties that control protein aggregation also determine whether a protein is retained or released from necrotic cells

Amyloidogenic protein aggregation impairs cell function and is a hallmark of many chronic degenerative disorders. Protein aggregation is also a major event during acute injury; however, unlike amyloidogenesis, the process of injury-induced protein aggregation remains largely undefined. To provide this insight, we profiled the insoluble proteome of several cell types after acute injury. These experiments show that the disulfide-driven process of nucleocytoplasmic coagulation (NCC) is the main form of injury-induced protein aggregation. NCC is mechanistically distinct from amyloidogenesis, but still broadly impairs cell function by promoting the aggregation of hundreds of abundant and essential intracellular proteins. A small proportion of the intracellular proteome resists NCC and is instead released from necrotic cells. Notably, the physicochemical properties of NCC-resistant proteins are contrary to those of NCC-sensitive proteins. These observations challenge the dogma that liberation of constituents during necrosis is anarchic. Rather, inherent physicochemical features including cysteine content, hydrophobicity and intrinsic disorder determine whether a protein is released from necrotic cells. Furthermore, as half of the identified NCC-resistant proteins are known autoantigens, we propose that physicochemical properties that control NCC also affect immune tolerance and other host responses important for the restoration of homeostasis after necrotic injury.


Introduction
Proteins must fold into their native conformation to be fully functional. However, as the native conformation is labile, all proteins are also capable of misfolding. Protein misfolding exposes hydrophobic regions that form intra-and inter-molecular associations, which if left unchecked can multimerize into higher-order toxic aggregates [1]. Protein quality control mechanisms adequately deal with most forms and degrees of misfolding. Nonetheless, destabilizing mutations, extreme environments or ageing can overwhelm proteostasis leading to the unwanted accumulation of aggregated proteins. Protein aggregation, and in particular amyloidogenic aggregation, underlies many chronic age-related diseases including Alzheimer's disease and Parkinson's disease. In these chronic settings, the onset of protein aggregation is widely thought to elicit injury.
Recently, we [10,12,20,21] and others [22] have characterized the process of injury-induced protein aggregation. In particular, we found that injury can trigger a specific form of protein aggregation called nucleocytoplasmic coagulation (NCC) [10]. By definition, NCC is a temporally coordinated disulfidecrosslinking reaction that occurs during late-stage cell death. NCC causes many intracellular proteins, including actin, tubulin, GAPDH and HSP90b, to convert from a soluble form into high-molecular-weight insoluble species [10]. In vitro and in vivo data show that NCC-aggregated proteins promote activation of the extracellular protease plasmin on the surface of dead cells [10,20,21]. Surface-bound plasmin then initiates two forms of clearance: direct proteolytic degradation of dead cells [10,20] and signalling nearby dendritic cells to increase their phagocytic capacity [21]. Thus, NCC-aggregated proteins can be considered a novel 'damage-associated molecular pattern' that initiates humoral and cellular responses to sterile injury [23].
Despite these recent advancements the mechanisms of injury-induced protein aggregation are poorly understood. Accordingly, to ascertain the basis of injury-induced protein aggregation, we have used quantitative mass spectrometry to profile the insoluble proteome of three cell types across different injury conditions. We find that NCC is a specific consequence of necrosis. Moreover, our data suggest that NCC is the predominant form of injury-induced protein aggregation, which affects a much wider array of proteins than first thought. Surprisingly, a small subset of proteins avoids NCC-mediated aggregation. Comparison of NCC-resistant and NCC-sensitive proteins uncovers two fundamental principles: (i) that injury-induced aggregation is distinguishable from amyloidogenic aggregation, and (ii) that a protein's inherent physicochemical properties determines whether it is retained or released from necrotic cells.
Altogether, NCC represents a 'fail-safe switch' that broadly incapacitates cellular functions during necrosis and selectively retains debris at the site of injury. As a result, we propose that NCC is a beneficial event that pacifies the reactivity of necrotic cells. In support of this notion, and consistent with the fact that defective removal of dead cells is a strong risk factor for autoimmune disease [24][25][26][27][28][29][30], we note that many proteins which evade NCC are also recognized autoantigens. Future studies should now assess whether mutations that alter a protein's propensity to undergo NCC can in turn influence its ability to act as an autoantigen.

Profiling the insoluble proteome of injured cells
We first characterized injury and protein aggregation in human Jurkat lymphocyte cells treated with etoposide (a topoisomerase-II inhibitor), staurosporine (a pan-kinase inhibitor) or Fas ligand (FasL; a TNF death receptor agonist). All agonists induced apoptosis in Jurkat cells (indicated by an increase in phosphatidylserine exposure and caspase-mediated RIPK1 cleavage; figure 1a; electronic supplementary material, figure S1a,b). More specifically, FasL and staurosporine triggered a similar degree/rate of injury, with early stage apoptosis evident within 3 h of treatment (indicated by phosphatidylserine-positive, TO-PRO-3-negative staining; electronic supplementary material, figure S1a) and transition to a secondary necrotic state within 24 h of treatment (indicated by complete loss of metabolism, plasma membrane disruption, pronounced LDH release and the acquirement of TO-PRO-3positive staining; figure 1a; electronic supplementary material, figure S1a). Note that neither necroptosis (mediated by fulllength RIPK1) nor autophagic death (mediated by LC3 cleavage; electronic supplementary material, figure S1b) was evident after FasL or staurosporine treatment. By contrast, etoposide caused early stage apoptosis after 24 h of treatment with no signs of secondary necrosis (as indicated by LDH retention, plasma membrane impermeability, partial metabolic suppression, and modest elevations in phosphatidylserine exposure and RIPK1 cleavage; figure 1a; electronic supplementary material, figure S1a). Lastly, we confirmed protein aggregation in Jurkat cells during the later stages of staurosporine-treatment with synchronous conversion of HSP90b, a/b-tubulin and b-actin from a soluble state into insoluble high molecular weight disulfidecrosslinked species (electronic supplementary material, figure S1c). Disulfide-crosslinked insoluble forms of these same proteins were also observed after FasL treatment, but not after etoposide treatment (not shown). Note that the coordinated loss of protein solubility upon secondary necrosis represents a genuine aggregation event (rather than merely being a consequence of disulfide crosslinking), as these same insoluble proteins were observed in the presence of a reducing agent (electronic supplementary material, figure S1d). Thus, injuryinduced protein aggregation in Jurkat cells occurred specifically during necrosis, but not during early stage apoptosis. Moreover, as the vast majority of insoluble material was disulfidecrosslinked (electronic supplementary material, figure S1c), NCC appears to be the predominant form of injury-induced protein aggregation.
Next, we profiled the insoluble proteome of Jurkat cells after 24 h of treatment with vehicle, etoposide, staurosporine or FasL using the quantitative mass spectrometry-based technique of SILAC (stable isotope labelling by amino acids in cell culture [32]; see electronic supplementary material, figure S2 and Material and methods for experimental details). This approach allowed the identification of hundreds to thousands of insoluble proteins per experiment with measurement of their relative abundance between uninjured and injured cells. Figure 1b and electronic supplementary material, table S1 summarize the SILAC results. Proteins with an increased right-shifted SILAC ratio in figure 1b were assumed to have undergone injury-induced aggregation, as this shift was only observed in staurosporine-or FasL-treated Jurkat cells. To verify this assumption, 12 proteins identified as putative aggregators in the SILAC experiments were subjected to immunoblot analysis. As shown in figure 2a, all 12 proteins were confirmed to undergo injury-induced protein aggregation in necrotic Jurkat cells. Consistent with the initial description of NCC in neuronal cells [10], necrosis in Jurkat cells caused all tested proteins to convert into insoluble high-molecular-weight rsob.royalsocietypublishing.org Open Biol. 6: 160098 dithiothreitol-sensitive species (as assessed by immunoblot; figure 2b) and to relocate into distinct aberrant subcellular structures (as assessed by super-resolution microscopy; figure 2c) which stained strongly for the presence of oxidized thiols (as assessed by confocal microscopy; figure 2d). Thus, we concluded that proteins that exhibited log 2 SILAC ratios greater than þ0.6 were indeed 'NCC-participating' proteins (green colour in figure 1b). Moreover, as 30-46% of the identified proteins underwent NCC in necrotic Jurkat cells (figure 1b), injury-induced aggregation appears to affect a much larger portion of the proteome than prior studies suggest.
Closer inspection of the SILAC data revealed another population of proteins with log 2 SILAC ratios less than 20.6 (11-13% of identified targets; red colour in figure 1b). These 'NCC-resistant' proteins were decreased in abundance in the insoluble fraction of necrotic Jurkat cells relative to uninjured cells, suggesting that they were either degraded or avoided aggregation during injury. To assess which scenario was applicable, six putative NCC-resistant proteins were selected for immunoblot analysis. Consistent with the SILAC data, all six tested proteins were decreased in abundance in the insoluble fraction of necrotic Jurkat cells relative to uninjured cells ( figure 3a). Notably, the disappearance of these NCC-resistant proteins in necrotic cells did not result in the commensurate appearance of lower molecular weight immunoreactive species, suggesting that they were not being degraded (figure 3a). Next, we reasoned that if a protein avoided both NCC and degradation, then it should instead be released into the extracellular milieu upon necrosis. Consequently, we assayed for the presence of FUS (also called 'RNA-binding protein FUS' or By definition, apoptosis is an energetic process, whereas necrosis coincides with metabolic incompetency. Therefore, because 24 h incubation with staurosporine or Fas ligand abolished metabolism (in conjunction with high levels of LDH release, trypan blue uptake and TO-PRO-3-staining; electronic supplementary material, figure S1a), we have concluded that these agonists produced a secondary necrotic state by this time point. (b) Jurkat cultures were labelled with either heavy or light lysine isotopes and treated with vehicle (uninjured), etoposide, staurosporine or Fas ligand for 24 h. Triton-insoluble fractions were extracted and proteins identified using quantitative mass spectrometry. The abundance of identified proteins relative to that in isotopically different uninjured cultures was then plotted as the log 2 ratio of the stable isotope label (SILAC); where an increased ratio indicates more of a specific insoluble protein in injured cells relative to uninjured cells and a decreased ratio indicates less of a specific insoluble protein in injured cells relative to uninjured cells. Proteins with a log 2 ratio . þ0.6 were gated into the NCC-participating subproteome (green colour) and proteins with a log 2 ratio , 20.6 were gated into the NCC-resistant subproteome ( pink colour). This gating was also chosen based on the recommendations of Zhang & Neubert [31]. rsob.royalsocietypublishing.org Open Biol. 6: 160098 'Translocated in liposarcoma', TLS; which was the most easily detected NCC-resistant protein) in the conditioned media from injured Jurkat cells. As shown in figure 3b, full-length FUS, but not MCM7 (an NCC-participating protein), was readily detected in the conditioned media of necrotic Jurkat cells. To the best of our knowledge, this is the first report that the ubiquitous protein, FUS, is released from necrotic cells. It remains to be determined whether other NCC-resistant proteins are released from necrotic cells, but this seems highly likely (for reasons provided below with figures 4 and 5). The capacity of FUS and other NCC-resistant proteins to act as paracrine communicators of injury should now be assessed, especially given the influence of NCC on the proteolytic [10,20] and phagocytic [21] removal of dead cells. It will also be interesting to see if FUS released from necrotic cells has a role during FUS-mediated amyotrophic lateral sclerosis [37]. We next examined injury-induced protein aggregation in human U937 monocyte cells, where both secondary necrosis (electronic supplementary material, figure S3a) and NCC (electronic supplementary material, figure S3b) occurred after 24 h of treatment with either etoposide or staurosporine. SILAC experiments showed that similar, albeit more pronounced changes to the insoluble proteome occurred upon secondary necrosis in U937 cells (figure 1d; electronic supplementary material, figure S3c), with 53-61% of proteins participating in NCC and 16-30% of proteins avoiding NCC (electronic supplementary material, figure S3c). The reason for this increased aggregation of the U937 proteome is unknown, but may reflect the heightened sensitivity of U937 cells to injury (cf. figure 1a and electronic supplementary material, figure S3a), or could relate to the extreme differences in the metabolism of monocytes and lymphocytes [38]. It is telling that etoposide elicited neither necrosis nor NCC in Jurkat cells, but strongly induced both secondary necrosis and NCC in U937 cells. This observation emphasizes that necrosis per se is the major cellular correlate of NCC. It is also salient that, despite marked differences in the means by which etoposide, staurosporine and FasL injure cells, many of the same proteins underwent NCC when these stauro.  Staurosporine-induced necrosis causes NCC-proteins (a-tubulin, EF2 and MCM7) to relocate into aberrant and distinct structures. Arrows indicate EF2 deposits that exhibit little-to-no staining for tubulin or MCM7. Arrowheads indicate MCM7 deposits that exhibit little-to-no staining for tubulin or EF2. Note, the procedure used to stain oxidized thiols was not compatible with the sample preparation for super-resolution microscopy. (d) Cells were fixed and subjected to immunofluorescence with maleimide co-staining for oxidized thiols. Shown are representative x,y confocal micrographs. Staurosporine-induced necrosis leads to nuclear breakdown and a marked increase in oxidized thiols that colocalize with aberrantly deposited NCC-proteins (3PDGH, GAPDH, MCM7, EF1, EF2 and a-tubulin).
rsob.royalsocietypublishing.org Open Biol. 6: 160098 treatments caused necrosis (figure 1c; electronic supplementary material, figure S3d). Such significant overlap in the insoluble proteome across different cell types and necrotic conditions denotes that a common underlying mechanism governs the participation/resistance of proteins to NCC.

Elucidating the physicochemical basis of nucleocytoplasmic coagulation
All proteins are capable of aggregating; however, the propensity, kinetics and conformations of aggregation are largely determined by the inherent physicochemical properties of the protein in question. Therefore, a high-confidence list of NCCparticipating, NCC-resistant and intermediary 'unchanged' proteins was collated from the Jurkat and U937 SILAC datasets (electronic supplementary material, table S1), and their primary sequences were compared to ascertain the physicochemical basis of NCC. The most prominent feature of NCC-participating proteins was their relatively high cysteine content (figure 4a), with a significant correlation between cysteine content and the degree of injury-induced aggregation (figure 4b). This finding, in conjunction with our earlier publications showing that NCC-aggregates are disulfide-crosslinked [10,12], highlights the importance of cysteine residues to NCC. Conversely, NCC-resistant proteins were found to be devoid of native disulfide bonds (figure 4c), implying that unlike NCC-participating proteins, the functions/folding of NCC-resistant proteins may be independent of cysteine reactivity. With the exception of C, D, N and H residues, the amino acid content of NCC-participating proteins aligned well with that of amyloidogenic peptides, as defined in [33,39,40] (figure 4a and not shown). This observation suggests that amyloidogenic aggregation and NCC are similar, but nevertheless distinct processes. We propose that NCC is an 'off-shoot' of amyloidogenic aggregation (electronic supplementary material, figure S4). NCC and amyloidogenesis may even be at cross-purposes, given that amyloidogenesis involves the slow organized association of discrete protein segments in the absence of disulfide crosslinking, whereas NCC involves rapid multivalent associations in the presence of disulfide crosslinking (electronic supplementary material, figure S4). Further evidence for this hypothesis stems from our prior work showing that NCC-aggregates exhibit a prefibrillar oligomeric conformation with no subsequent appearance of fibrillar species [10].
Other characteristics of NCC-participating proteins include a higher representation of aromatic residues (F, Y, W; p , 0.001 via unpaired two-tailed t-test; figure 4a), aliphatic residues (I, V, L; p , 0.0001 via unpaired 2-tailed t-test; figure 4a) and increased hydrophobicity (electronic supplementary material, figure S5a)-attributes which explain the predicted high aggregation propensity of NCC-participating proteins relative to NCC-resistant proteins (figure 4d). Curiously, the intrinsic disorder content (figure 4e) and the molecular weight (electronic supplementary material, figure S5b) of NCC-participating proteins were lower than those of NCC-resistant proteins. These collective attributes further distinguish NCC from amyloidogenesis, which instead favours co-aggregation of large hydrophobic proteins with high intrinsic disorder (figure 4f) [34]. That diametrically opposed traits exist between NCCparticipating and NCC-resistant proteins shows that the same physicochemical forces control these protein fates during necrosis. Indeed, our data show that whether a protein is retained or released from necrotic cells is in part determined by its propensity to undergo injury-induced aggregation. This conclusion was independently supported by analysing the insoluble   figure S6 and table S1).

Identifying higher-order features of nucleocytoplasmic coagulation
Many factors besides primary sequence affect protein aggregation. Foremost among these factors is the concentration dependency of protein aggregation. In line with this tenet, NCC-participating proteins were found to be more abundant than NCC-resistant proteins (figure 5a). Next, we performed ontology enrichment analysis to gain a broader sense of what molecular functions (figure 5b; electronic supplementary material,  Only proteins quantified in at least two necrotic instances were considered. A fourth subset of 388 proteins randomly selected from the reviewed human nuclear/ cytoplasmic/mitochondrial proteome was also generated for control purposes (electronic supplementary material, table S1). The four subsets were used for physicochemical trait analysis. (a) Shown in green is the amino acid composition of NCC-participating proteins expressed relative to that of NCC-resistant proteins (arbitrarily normalized to 1). Shown in grey is the predicted relative contribution of individual amino acids to amyloidogenic aggregation as defined in AGGRESCAN [33] (arbitrarily normalized to 0). (b) Graph depicting a significant ( p , 0.0001) positive correlation between the relative cysteine content and the tendency for proteins to undergo NCC (i.e. SILAC ratio) as determined by Pearson's correlation analysis (where the black solid line represents the line of best fit and black dashed lines indicate 95% confidence intervals). (c-e) Graphs comparing the degree of native disulfide bonding (c), the predicted aggregation propensity as determined by AGGRESCAN [33] (d) and the intrinsic disorder content (e) across the four protein subsets. For the box plots in (d,e), the centre line indicates the median, box edges indicate the 25th/75th percentile and whiskers indicate the 10th/90th percentile. ****p , 0.0001 as determined by one-way ANOVA with Tukey's correction. (f ) Graph comparing the average molecular weight (kDa), hydrophobicity and % disorder for the NCC-participating, NCC-resistant and unchanged protein subsets with that of amyloid interacting proteins identified in [34] (another protein subset from electronic supplementary material, table S1).
rsob.royalsocietypublishing.org Open Biol. 6: 160098 Interrogation of the human interactome [41] suggests that NCC causes aggregation of seemingly unrelated proteins (i.e. the NCC-interactome has low connectivity; electronic supplementary material, figure S7b) that participate in a high number of protein-protein interactions (especially, homotypic interactions; electronic supplementary material, figure S7b). That NCC preferentially affects fundamental enzyme complexes agrees well with the observation that NCC-participating proteins are evolutionarily more ancient than NCC-resistant proteins (figure 5c). Altogether, our data suggest that NCC preferentially shuts down a wide array of ancient proteins with essential interactions and functions. NCC is therefore likely to be a rudimentary response that 'inactivates' necrotic cells.

Putative role for nucleocytoplasmic coagulation in attenuating autoimmunity
Maintaining tolerance after injury is a necessary and problematic process, especially given that cell death can allow the unwanted exposure/release of epitopes that would otherwise be sequestered from the immune system. Indeed, cell death is a potent trigger of inflammation, with autoimmune disease often preceded by injury [24][25][26][27]. Moreover, defective or inadequate removal of dead cells underlies the onset of many autoimmune disorders including system lupus erythematosus [28]. Accordingly, we hypothesized that NCC, by restricting the escape of constituents from necrotic cells and by promoting tolerogenic clearance [21], is a favourable response that mitigates autoimmunity. To support this hypothesis, we assessed the representation of known autoantigens (using a set of 2079 autoantigens from Backes et al. [36]) and found that autoantigens were far more prevalent among the NCC-resistant subpopulation (figure 5d). Furthermore, the primary amino acid composition of autoantigens closely mirrored that of NCC-resistant proteins, with cysteines ( p , 0.0001), aliphatic residues (I, V, L; p , 0.0001) and aromatic residues (F, Y, W; p , 0.0001) being under-represented, and charged residues (D, E, K, R; p , 0.0001) being over-represented in autoantigens (figure 5e relative amino acid content (autoantigen/control proteins) 0 N C C -p a r t i c i p a t i n g N C C p a r t i c i p a t i n g N C C -p a r t i c i p a t i n g u n c h a n g e d u n c h a n g e d u n c h a n g e d N C C -r e s i s t a n t N C C r e s i s t a n t N C C -r e s i s t a n t rsob.royalsocietypublishing.org Open Biol. 6: 160098 extracellular space upon necrosis. Collectively, these findings add support to the hypothesis that NCC is a favourable response that may mitigate autoimmunity. Future studies should now test whether mutations that increase a protein's tendency to undergo NCC also reduce its autoreactivity following necrosis-an approach that holds great appeal for investigating the basis of lupus and other injury-related autoimmune disorders.

Concluding remarks
A cascade of events that accounts for both our previous [10,12] and current findings is where NCC relies upon membrane disruption during necrosis (either primary or secondary necrosis). Membrane failure would cause a major loss of redox potential and widespread oxidation, which in turn would preferentially disulfide-crosslink abundant protein complexes within otherwise redox-reduced compartments. Our prior studies also suggest that initial misfolding events (e.g. linchpin residue oxidation) that precede necrosis may be a prerequisite for NCC [10,12]. We postulate that NCC represents a fail-safe switch that ensures irreversible metabolic and proteostatic incompetency during intense injury. The proteomic coverage of NCC is much wider than first thought, with a large number of essential proteins and macromolecular structures entrapped by NCC, thereby preventing their release into the extracellular space. Entrapment by NCC, however, is a selective process because a small proportion of the proteome (including FUS) avoids NCC and is therefore selectively released into the extracellular milieu upon necrosis. As the release of intracellular molecules upon necrosis is deleterious, NCC is likely to be a beneficial response that helps multicellular organisms restore tissue homeostasis after injury. This supposition is somewhat counterintuitive given that amyloidogenic protein aggregation promotes neurodegenerative diseases. However, based on the similar albeit distinct physicochemical signatures of NCC and amyloidogenesis, we hypothesize that NCC may be an 'offshoot pathway' that prevents deleterious amyloid formation in necrotic cells (electronic supplementary material, figure S4). Such interrelations between amyloidogenic aggregation and NCC should now be assessed. Beyond this, it is intriguing that features such as cysteine content, hydrophobicity, intrinsic disorder and abundance dictate whether a protein is retained or released from necrotic cells. This understanding broadens the biology of protein aggregation and indicates that reductionist biochemical approaches used in the field of protein misfolding may also be used to investigate the immunogenic and inflammatory nature of necrosis.

Cell culture
Jurkat and U937 cells were maintained as non-adherent cultures in RPMI supplemented with 10%(v/v) heat-inactivated FCS, 2 mM L-glutamine, 50 U ml 21

Cell injury assays (LDH, MTS and trypan blue assays)
LDH and MTS assay kits were from Promega and were performed according to the manufacturer's instructions. Trypan blue positivity was determined using the TC-10 Automated Cell counter (Bio-Rad) according to the manufacturer's instructions.

Flow cytometry
Jurkat and U937 cells (seeded and treated as in §3.2) were incubated for the indicated time period. Cells were then rsob.royalsocietypublishing.org Open Biol. 6: 160098 incubated for a further 5 min with 1 mg l 21 AlexaFluor488 conjugated-Annexin V and 250 mM TO-PRO-3 or 7-AAD and subject to flow cytometry (BD Accuri C6 Plus flow cytometer). Data were analysed off-line using FLOWJO v. 10.1r5.
Chloroacetamide was added to the lysis buffer to prevent inadvertent disulfide bond formation during/after cell homogenization and subsequent fractionation. Note, chloroacetamide was chosen over iodoacetamide as it reacts more specifically with cysteine thiols [42]. Homogenates were triturated and then stored at 2808C until further fractionation was necessary; 1 ml of lysis buffer was used per 0.6 Â

SDS-PAGE for tryptic digestion
Samples were boiled in 2% SDS-loading buffer with dithiothreitol and subjected to SDS-PAGE using a NuPAGE Novex Bis-Tris Mini gel (4-12%) with NuPAGE MOPS SDS-Running buffer supplemented with 1Â NuPAGE antioxidant (Life Technologies). After electrophoresis, the gel was stained with InstantBlue protein stain and cut into approximately 2 Â 6 mm slices then each slice further cut into 1 mm 2 pieces. Gel pieces were subjected to in-gel trypsin digestion as in [43].

Mass spectrometry
Liquid chromatography tandem mass spectrometry was performed as in [44] with minor modifications. Extracted peptides were analysed by Q-Exactive Orbitrap mass spectrometer (Thermo Scientific) coupled online with an RSLC nano-HPLC (Ultimate 3000, Thermo Scientific). Samples were injected onto a Thermo RSLC pepmap100, 75um id, 100 angstrom pore size, 50 cm reversed phase nano column with buffer A (2.5% acetonitrile, 0.1% formic acid) at a flow rate of 300 nl min 21 . The peptides were eluted over a 40-min gradient to 40% buffer B (80% acetonitrile 0.1%, formic acid). The eluate was nebulized and ionized using the Thermo nano electrospray source coated silica emitter with a capillary voltage of 1700 V. Peptides were selected for MS/MS analysis using XCALIBUR software (Thermo Fisher) in Full MS/dd-MS2 (TopN) mode with the following parameter settings: TopN 10, MSMS AGC target 1e5, 60 ms Max IT, NCE 27 and 2 m/z isolation window. Dynamic exclusion was set to 20 s. MS data were analysed with MAXQUANT [45] software v. 1.3.0.5. Search parameters included: specific digestion with trypsin with up to two missed cleavages, þ6 Lysine was set as label, protein N-terminal acetylation and methionine oxidation were set as variable modifications while cysteine alkylation was set as fixed modification. The searches were performed against UniProt human protein sequences (March 2013 version). The remaining search parameters were default settings. The search was then imported into PERSEUS software [45] for collation (electronic supplementary material, table S1) and for further annotation. Note that the gating of SILAC ratios for analysis is explained in the respective figure legends. These gating strategies were based on the recommendations of Zhang & Neubert [31], with more than 95% of the identified proteins having a ratio that fell within the 'unchanged' intermediary gate in the corresponding uninjured cell samples.

Immunoprecipitation
Jurkat cells were treated for 24 h and the conditioned media collected. Fifteen microlitres of Protein-G/-A beads (as a 50% slurry) was added to 1.3 ml of conditioned media and the mixture gently rotated (30 min, 48C). The beads were centrifuged (2 min, 3000g) and discarded; 20 ml of Protein-G/-A beads þ 1 mg of anti-MCM7 antibody was added to the supernatant and the mixture gently rotated (3 h, 48C). The beads were centrifuged (2 min, 3000g) and kept as the 'MCM7 pulldown' material; 20 ml of Protein-G/-A beads þ 1 mg of anti-FUS antibody was added to the supernatant and the mixture gently rotated (3 h, 48C). The beads were centrifuged (2 min, 3000g) and kept as the 'FUS pulldown' material. The beads were boiled for 10 min in 35 ml of SDS-loading buffer with dithiothreitol and the supernatants subjected to SDS-PAGE/ immunoblot analysis. Quantitation of the immunoprecipitation signal was performed using IMAGEQUANT v. 5.2 software (Molecular Dynamics) and expressed relative to the total amount of FUS present in the Triton-soluble lysate from uninjured Jurkat cells.

Protein property analysis
3.10.1. Protein sequence analysis Canonical FASTA sequences for the different protein subsets (electronic supplementary material, table S1) were retrieved from UniProt and entered into the Sequence Manipulation Suite [46] to determine amino acid content, hydrophobicity and molecular weight.

Protein aggregation prediction
Canonical FASTA sequences for the different protein subsets (electronic supplementary material, table S1) were retrieved from UniProt and uploaded to the AGGRESCAN [33] site to ascertain the predicted aggregation propensity. Note, only proteins with less than 1000 residues were analysed. AGGRESCAN provides the relative aggregation propensity of amino acids within both the central hydrophobic cluster of b-amyloid and within the context of a larger full-length soluble polypeptide [33]. AGGRESCAN's aggregation propensities are derived from in vivo experimental data [33]. Similar analyses (with similar results) were performed using the FoldAmyloid [39] and Zyggregator [40] databases (not shown).

Intrinsic disorder content analysis
UniProt IDs (electronic supplementary material, table S1) were used to query the MobiDB 2.0 database [47]. The percentage of each protein in disorder (either from curated data or from consensus predicted data when no curated data existed) irrespective of segment was determined.

Autoantigen analysis
UniProt IDs (electronic supplementary material, table S1) were cross-referenced against the full set of autoantigens from Backes et al. [36].

Protein abundance analysis
The different protein subsets (electronic supplementary material, table S1) were uploaded to the PaxDB [48] site and their organismal abundance was retrieved from the Human PaxDB integrated dataset.
3.10.6. Protein age analysis Protein age was determined using the ProteinHistorian [49] website, whereby protein subsets were individually analysed via the swisspfam-PTHR7_youngest database and the Dollo parsimony reconstruction algorithm.

Native disulfide bond analysis
Annotations from the UniProt [50] database were used to determine the number and location of native disulfide bonds for each protein of interest.

Ontology enrichment analysis
CYTOSCAPE v. 3.2.0 and the ClueGO v. 2.2.5 and CluePedia v. 1.2.5 plug-ins was used to perform ontology enrichment analysis as in [35]. Molecular function enrichment analysis was performed within GO term levels 3-5. Cellular compartment enrichment analysis was performed within GO term levels 4-10.

Protein-protein interaction analysis
The high-confidence NCC-participating protein subset was entered into the Mentha [51] interactome database. All human interacting proteins were extracted and visualized as a network using Cytoscape. Duplicate interactions were removed and the network rendered directionless. The topology and characteristics of the resultant network was analysed using the Network Analyser plug-in [52].

Statistical analyses
All results are based on three independent experiments unless stipulated otherwise. Analyses were performed with GraphPad PRISM 6 with p , 0.05 considered statistically significant. The analysis employed for each cohort is stated in the respective figure legend.
Data accessibility. All mass spectrometry files used in the manuscript will be deposited in mzML format on the ProteomeXchange website (http://www.proteomexchange.org/).