An unpredicted aggregation-critical region of the actin-polymerizing protein TRIOBP-1/Tara, determined by elucidation of its domain structure

Abstract

Aggregation of specific proteins in the brains of patients with chronic mental illness as a result of disruptions in proteostasis is an emerging theme in the study of schizophrenia in particular. Proteins including DISC1 (disrupted in schizophrenia 1) and dysbindin-1B are found in insoluble forms within brain homogenates from such patients. We recently identified TRIOBP-1 (Trio-binding protein 1, also known as Tara) to be another such protein through an epitope discovery and proteomics approach by comparing post-mortem brain material from schizophrenia patients and control individuals. We hypothesized that this was likely to occur as a result of a specific subcellular process and that it, therefore, should be possible to identify a region of the TRIOBP-1 protein that is essential for its aggregation to occur. Here, we probe the domain organization of TRIOBP-1, finding it to possess two distinct coiled-coil domains: the central and C-terminal domains. The central domain inhibits the depolymerization of F-actin and is also responsible for oligomerization of TRIOBP-1. Along with an N-terminal pleckstrin homology domain, the central domain affects neurite outgrowth. In neuroblastoma cells it was found that the aggregation propensity of TRIOBP-1 arises from its central domain, with a short “linker” region narrowed to within amino acids 324–348, between its first two coiled coils, as essential for the formation of TRIOBP-1 aggregates. TRIOBP-1 aggregation, therefore, appears to occur through one or more specific cellular mechanisms, which therefore have the potential to be of physiological relevance for the biological process underlying the development of chronic mental illness.

Introduction

The TRIOBP⁴ (Trio-binding protein) gene encodes for two distinct proteins (1, 2), both of which are implicated in the modulation of actin, although the exact mechanisms by which they do this remain unclear (3, 4). Of these TRIOBP-4, encoded by the 5′ end of the gene, is a largely disordered protein (5) expressed principally in the inner ear, with mutant forms implicated in deafness (1, 2, 4), whereas the 3′-encoded TRIOBP-1 protein is predicted to be folded (3) and is more ubiquitously expressed (1, 2). The TRIOBP-1 protein is also known as Tara (Trio-associated repeat on actin). Longer proteins containing the reading frames of both TRIOBP-1 and TRIOBP-4 also exist (1, 2) but are less well characterized.

Recently, we performed an antibody-based screen for proteins that specifically form insoluble aggregates in the brains of patients with schizophrenia (6). This was based on the hypothesis that disrupted protein homeostasis in post-mitotic neurons is likely to characterize subtypes of chronic mental illness (7). In this manner, an antibody against TRIOBP-1 was found to show specificity for the pooled purified insoluble protein fractions of post-mortem brain samples from schizophrenia patients compared with an equivalent preparation prepared from brain samples of control individuals (8). Further analysis determined the TRIOBP-1 protein, but not TRIOBP-4, to readily form insoluble aggregates when expressed in cell culture systems or primary neurons, implicating aggregation of TRIOBP-1 in mental health (8).

The specific roles of TRIOBP-1 in the brain are not well studied; however, more generally it is known to be a critical promoter of actin polymerization, binding directly to polymerized fibers of F-actin (3, 9). It is also involved in chromosome segregation during mitosis (10) as well as in cell migration. The latter occurs through its interaction with nuclear distribution element-like 1 (NDEL1) (11), a protein of significant importance in neurodevelopment that has been implicated in schizophrenia (12). Small, but statistically significant, increases in TRIOBP transcript expression have also been detected in post-mortem brain tissue of schizophrenia patients from two independent samples (13).

The misassembly of TRIOBP-1 protein into insoluble aggregates in mental illness could arise either because the protein is naturally “sticky,” with a tendency to self-interact in a nonspecific manner or as the result of a specific physiological mechanism. Such mechanisms could include, for example, the addition or removal of a protein-binding partner, a post-translational modification, interaction with a small molecule, or an error during protein folding. We hypothesized that, were a specific mechanism involved, it should be possible to isolate individual regions of the protein that were involved in this process (either small motifs or entire folded domains) and that are, therefore, required for aggregation of TRIOBP-1.

In this manuscript we, therefore, investigate the domain structure of TRIOBP-1 and establish that such an aggregation-critical region exists that can be narrowed down to a stretch of just 25 amino acids (aa), strongly supporting the idea that a specific mechanism exists by which TRIOBP-1 forms insoluble aggregates.

Results

Predicted coiled-coil structure of TRIOBP-1 and inclusion of its N terminus

Previous predictions have suggested that TRIOBP-1 (RefSeq accession number NP_008963.3) contains a pleckstrin homology (PH) domain near its N terminus, whereas the majority of its C-terminal half forms α-helices, most likely in the form of coiled-coil domains (3). The potential coiled-coil composition of the C-terminal half of TRIOBP-1 was investigated theoretically using PSIPRED (14, 15) and COILS (16) (Fig. 1, A–C), with reference also to Paircoil2 (17). In this manner we predicted the existence of six distinct putative coiled coils (Fig. 1D) in human TRIOBP-1, which will be referred to as CC1-CC6. Of these, CC3, CC4, and CC5 are comparatively long stretches (running from approximately aa 395–465, aa 473–550, and aa 566–617, respectively), whereas CC1, CC2, and CC6 are shorter (aa 303–322, aa 352–372, and aa 623–642 respectively).

Figure 1.

Predicting the coil structure of TRIOBP-1. A, prediction of coiled-coil-forming propensity of the 652-amino acid TRIOBP-1 reading frame using the COILS server with a 21-amino acid sliding window. B, prediction of α-helix-forming propensity using the PSIPRED server. C, prediction of β-sheet-forming propensity using the PSIPRED server. D, proposed coiled-coil regions of TRIOBP-1 based on this data, shown in line with the graphs. The predicted PH domain is also displayed.

Note that in this manuscript the amino acids of human TRIOBP-1 are numbered based on a 652-aa long reading frame. This is 59-aa longer than the 593-aa reading frame considered in many other publications (3, 10, 11, 18–20) as a result of the existence of two putative Kozak sequences and methionine residues at the 5′ end of the TRIOBP-1 transcript. Amino acid numbers quoted here for the 652-aa protein can, therefore, be converted directly to positions in the 593-aa protein considered elsewhere by subtracting 59. We chose to include this additional 59-aa N-terminal region in the amino acid numbering system as it is relatively well conserved and maintaining its reading frame throughout mammalian species (Fig. 2A). Nevertheless, TRIOBP-1 species containing this region appear to represent a minority in cells (Fig. 2B) and, therefore, arise either from a less common translation event or else they represent the approximate position of a “pro” domain, which is normally cleaved off in the cell. This 59-aa N-terminal region is rich in both prolines and positively charged residues and localizes to both the cytoplasm and nucleus when expressed in neuroblastoma cells (Fig. 2C).

Figure 2.

The extreme N terminus of TRIOBP-1. A, the putative initial 59 aa of TRIOBP-1, as present in multiple mammalian lineages. Notably, despite variation in amino acid sequence and particularly in a poly-glycine stretch (green), these proteins all maintain the same reading frame. Positively charged residues are shown in red, negatively charged residues are in blue, and prolines are in purple. Aligned using Clustal Omega (48) and then manually curated. RefSeq accession number for the sequences shown are, in order displayed: NP_008963.3; XP_008065523.1; XP_009215517.1, XP_003126108.2; XP_006914811.1; XP_004410876.1; XP_007454108.1; XP_004279508.1; XP_004312195.1; XP_008849513.1; XP_005067025.1; XP_005067025.1; NP_001019887.1; XP_006242072.1; XP_005354277.1; XP_006979187.1; XP_005322324.1. B, SH-SY5Y cells inducibly expressing the 652 aa species of TRIOBP-1 with and without induction of expression. The endogenous TRIOBP-1 species is seen to be of a slightly lower molecular weight than the 652-aa overexpressed protein, although a small amount of endogenous protein of this size is present, consistent with only a small portion of endogenous TRIOBP-1 protein expressing or maintaining the extreme N terminus. C, expression of the N-terminal 59 aa of TRIOBP-1 alone in SH-SY5Y cells with a FLAG tag.

The domain structure of TRIOBP-1 as determined by solubility and stability of recombinant protein fragments

To determine which regions have the potential to exist as distinct folded domains in vivo, different fragments of TRIOBP-1 were expressed as recombinant proteins. This was based on the premise that compactly folded protein domains are more likely to form soluble recombinant proteins, with distinct oligomeric states, both in bacteria and in vitro. In contrast, constructs encoding only part of such a region would instead be more likely to either degrade or form insoluble aggregates. In support of this idea, we began by investigating the strongly predicted PH domain at the N terminus of TRIOBP-1. Expression of the whole PH domain (aa 60–189) led to a soluble recombinant protein that stably formed a dimer (apparent molecular mass of ≈30 kDa compared with a predicted molecular mass of 17.3 kDa) as analyzed by size exclusion chromatography (SEC, Fig. 3A). Circular dichroism (CD) analysis confirmed this protein to be structured, with a high β-sheet content, ∼50% as predicted in DichroWeb using the CDSSTR method and SMP180 dataset (21–25), which is consistent with the typical structure of PH domains (Fig. 3B). In contrast, expression of a shorter construct lacking the N-terminal-most predicted β-strand of the PH domain (aa 93–189) led to an unstable recombinant protein. Specifically, the protein formed a high-oligomeric-state void-volume species (Fig. 3A), excepting for a portion of the protein that instead broke down to yield a smaller soluble product (Fig. 3C).

Figure 3.

Completeness of the PH domain is required for a stable recombinant protein fragment. A, SEC of recombinant proteins equating to the complete (aa 60–189) or incomplete (aa 93–189) PH domain of TRIOBP-1. The complete domain exists as a single clear species, whereas the incomplete domain exists as multiple species, labeled A–C, including one in the void volume. B, CD measurement of the aa 60–189 TRIOBP-1 fragment shows it to be folded and principally helical. C, Western blot of aa 93–189 species from part A reveals only a breakdown product to exist in the soluble fractions as opposed to the void volume. mAU, milliabsorbance units.

Based on the principle that solubility and stability of a recombinant protein fragment in vitro can indicate the presence of distinct and complete folded domains in vivo, as exemplified here by the PH domain, the structure of the C-terminal half of TRIOBP-1 was investigated. When this whole coiled-coil region (aa 281–652) was expressed, it broke down within the bacteria into a number of discrete species (Fig. 4, A and B), suggesting that the coiled-coil regions consists of at least two distinctly folded domains. Based on the sizes of the fragments detected with antibodies raised against different regions of the coiled-coil region, a break between CC4 and CC5 was predicted. In agreement with this, expression of the extreme two C-terminal coiled coils of TRIOBP-1 (CC5 and CC6, aa 556–652) led to a single stable monomeric protein (Fig. 4C, observed molecular mass ≈20 kDa, predicted molecular mass 14 kDa; the discrepancy likely arising from coiled-coil protein forming an elongated structure), which has a predominantly α-helical structure as determined by CD (Fig. 4D). In contrast, inclusion of CC4 in the construct (CC4-CC6, aa 467–652) led to an unstable protein that broke down before purification (Fig. 4E). Together this implies that CC5–6 constitutes a distinctly folded structure that we refer to as the “C-terminal domain.”

Figure 4.

TRIOBP-1 contains two distinct coiled-coil domains, as determined by expression of recombinant protein fragments. A, SEC of recombinant TRIOBP-1 coiled coils (CC), 1–6, aa 281–652. mAU, milliabsorbance units. B, SDS-PAGE and Western blot of the whole lysate from bacteria expressing this fragment, nickel affinity-purified protein (Input), and the chromatography fractions indicated in A after being concentrated. The protein broke down into distinct fragments. C, SEC of CC5–6, aa 556–652, revealing a single stable protein species. D, circular dichroism profile of the major CC5–6 species. E, recombinant TRIOBP-1 CC4–6, aa 467–652) broke down before nickel affinity purification.

Expression of CC1-CC4 instead led to a predominant multimeric species with an apparent molecular mass of ∼200 kDa (Fig. 5A, aa 281–555, predicted molecular mass for a monomer: 34.2 kDa), suggestive of a hexamer. CD demonstrated it to be principally α-helical (Fig. 5B). This multimer was prone to precipitation at high concentrations, and there was some evidence of degradation of the protein (80–100 ml in Fig. 5A, also visible in Fig. 6E), suggesting that this protein fragment was largely, but not completely, stable in vitro. A further truncated recombinant protein containing CC1-CC3 was highly stable even at concentrations of >10 mg/ml and existed as seemingly the same single oligomeric species as the CC1-CC4 protein (Fig. 5C, aa 281–466, apparent molecular mass ≈160 kDa; predicted molecular mass of a monomer 24 kDa, again suggestive of a hexamer) and also had a high helical content (Fig. 5D). Although these in vitro analyses may not necessarily completely reflect the situation in vivo, they strongly suggest that CC1-CC3 of TRIOBP-1 forms a folded “central domain,” potentially also including CC4 but distinct from the C-terminal domain of CC5 and CC6.

Figure 5.

The central domain of TRIOBP-1 is responsible for its oligomerization. SEC and CD measurements of different combinations of the coiled coils that make up the central domain of TRIOBP-1. A, SEC of aa 281–555. B, CD of the principle oligomeric state of aa 281–555. C, SEC of aa 281–466. D, CD of aa 281–466. E, SEC of aa 281–382. F, CD of aa 281–382. G, SEC of aa 349–466. H, CD of aa 349–466. mAU, milliabsorbance units.

Figure 6.

The oligomeric state of the central domain, analyzed by analytical ultracentrifugation. A, SEC sedimentation velocity data performed on three fragments of the TRIOBP-1 central domain, with c(s) fit results shown. AU, absorbance units. B, sedimentation profiles (dots) over time for the aa 281–382 fragment, 65 μm, experiment in A, from purple (initial scan) to red (final, scans taken at 3-min intervals) together with the c(s) fit results (lines). Below the graphs the corresponding residuals are shown. C, equivalent sedimentation profile for the aa 349–466 fragment, 55 μm. D, equivalent sedimentation profile for the aa 281–466 fragment, 50 μm. E, Western blot of the soluble TRIOBP-1 fragments used for these assays, with polyclonal antibodies raised against different parts of the protein used to distinguish them. Note that some degradation of the aa 281–555 construct can be seen, yielding a fragment corresponding approximately to the aa 281–382 or 324–382 constructs. This is not seen for the aa 281–466 construct. The left two panels were stained from one membrane, and the right two panels were from another displaying identical samples.

When only the short CC1 and CC2 coiled regions were expressed, the ensuing protein appeared to be either a trimer or an extended dimer (Fig. 5E, aa 281–382, apparent molecular mass ≈40 kDa, predicted molecular mass for a monomer 14.1 kDa) and to be reasonably stable, although less so than the CC1-CC3 protein and with a lower α-helix content (Fig. 5F). Expression of CC2 and CC3 led to a similarly stable trimeric protein (Fig. 5G, aa 349–466, apparent molecular mass ≈55 kDa, predicted molecular mass for a monomer 16.2 kDa), which consisted almost entirely of α-helical structure (Fig. 5H).

To further investigate the oligomeric state of the central domain, these constructs were investigated using analytical ultracentrifugation (AUC). AUC sedimentation velocity experiments showed the CC1-CC2 fragment to exist predominantly as a species with an estimated molecular mass of 26.0 kDa (Perrin factor, s_20,w = 1.15 S; Fig. 6, A and B), consistent with a dimer. It had a frictional ratio of 1.61, indicative of an extended conformation. Under the same experimental circumstances, most of the CC2-CC3 construct adopted an oligomeric state consistent with a monomer, and only 1.8% of the protein remained as a species matching the one seen by SEC, with a predicted molecular mass of 49.4 kDa (s_20,w = 3.46 S; Fig. 6, A and C). This region was even more elongated that CC1-CC2, with a frictional ratio of 1.82, consistent with an extended coiled coil structure. When CC1-CC3 was similarly examined it did not maintain the same high-order oligomeric state seen by SEC, with an apparent trimer being the largest species (observed molecular mass of 76.3 kDa, compared with a theoretical molecular mass of 23.9 kDa; Fig. 6, A and D). This was also highly elongated (frictional ratio: 1.63). It, therefore, appears that the central domain trimerizes through interactions of CC2-CC3 but is also capable of forming higher order oligomers, most likely hexamers, through the additional dimer-forming capacity contained within CC1-CC2. The nature of all soluble TRIOBP-1 constructs used here could be confirmed by Western blot using anti-TRIOBP antibodies raised against different sections of the protein (Fig 6E).

Functional analysis of the individual domains of TRIOBP-1

After identification of two distinct coiled-coil domains of TRIOBP-1, in addition to the previously predicted PH domain, initial experiments were carried out to investigate their functions in the cell, the results of which are summarized in Table 1.

Table 1.

Summary of functional assays performed on isolated domains of TRIOBP-1

Results are based on data found in Figs. 7 and 8.

	PH domain	Central domain	C-terminal domain
Location on TRIOBP-1
Amino acids	60–189	281–555	556–652
Coiled coils	N/A	1–4	5–6
Actin assay
Segregates with F-actin	No	Yes	Yes
Segregates with G-actin	Yes	Yes	Yes
Effect on F-actin depolymerization	No effect	Inhibition	Trend inhibition
Effect on G-actin polymerization	No effect	No effect	Trend promotion
Neurite assay
Effect on neurite number	Increase	Decrease	No effect
Effect on longest neurite length	Trend increase	No effect	No effect

The canonical role of TRIOBP-1 is as a modulator of actin polymerization, leading to increased fibular F-actin compared with globular G-actin (3), although the mechanism by which it does this has so far been unknown. The effect of the individual domains of TRIOBP-1 on actin polymerization was, therefore, investigated. NLF neuroblastoma cells were transfected with constructs encoding the PH domain, central domain (CC1–4), and C-terminal domain (CC5–6) of TRIOBP-1 and lysed in a buffer that stabilizes polymerized F-actin (26). The ensuing lysates were then separated by ultracentrifugation into an F-actin containing pellet and a G-actin containing supernatant (Fig. 7A). The supernatant was incubated under conditions that promote polymerization before being separated once again by ultracentrifugation into F- and G-actin-containing fractions and determining how much of the G-actin had re-polymerized in the presence of each TRIOBP-1 construct. Similarly, the original F-actin-containing pellet was resuspended and equilibrated in conditions that favor de-polymerization (27) and then separated by ultracentrifugation to determine what proportion of the polymeric F-actin had broken down into monomeric G-actin in the presence of each TRIOBP-1 construct (Fig. 7A).

Figure 7.

The effect of TRIOBP-1 subdomains on actin dynamics. A, scheme of the actin assay performed. Cell lysate from NLF neuroblastoma cells was separated by ultracentrifugation into F-actin and G-actin-containing fractions. The stability of the separated F-actin was assayed by incubating it in a depolymerization buffer and then separating by ultracentrifugation into a (stable) F-actin-containing fraction and a (de-polymerized) G-actin fraction. Similarly, the stability of G-actin was assayed by incubating it in polymerization buffer and then separating it into a (stable) G-actin-containing fraction and a (re-polymerized) G-actin fraction. B, Western blots of the ensuing fractions showing the abundance of actin and of FLAG-tagged TRIOBP-1 fragments which were expressed by transfection. All actin blots are shown under equivalent conditions as are the top three FLAG-blots. The lowest FLAG blot is shown under a higher gain due to the much lower level of the aa 556–652 protein after transfection (∼5-fold lower than aa 281–555). C, quantification of the proportion of actin in the original F-actin fraction, which transferred to the G-actin fraction after incubation under de-polymerization conditions. D, quantification of the proportion of actin in the original G-actin fraction, which transferred to the F-actin fraction after incubation under polymerization conditions. In both graphs n = 3. *, p < 0.05, according to a paired t test after correction for multiple testing.

The central (CC1-CC4) and C-terminal (CC5-CC6) domains of TRIOBP-1 were found to segregate with both F- and G-actin in the assay, whereas the PH domain was found exclusively in soluble fractions, implying that both coiled-coil domains, but not the PH domain, can interact with F-actin (Fig. 7B). As would therefore be expected, the presence of the PH domain had no effect on the dynamics of actin in this assay when compared with mock-transfected control cells (Fig. 7, B–D). The central domain had no effect on the polymerization of G-actin (Fig. 7, B and D) but did cause an almost 3-fold reduction in the amount of F-actin that depolymerized during the experimental time period (Fig. 7, B and C). Expression of the C-terminal region was at a lower level than the other two constructs but nevertheless showed putative effects in limiting actin de-polymerization to the central domain (Fig. 7, B and C) and in causing an increase in polymerization of G-actin (Fig. 7, B and D), although neither of these remained significant after correction for multiple testing. We, therefore, demonstrate for the first time that the central domain and potentially the C-terminal domain, but not the PH domain, is individually capable of interacting with F-actin and protecting it against de-polymerization, at least under these experimental conditions.

The exact role of TRIOBP-1 in neurons remains obscure; however, overexpression of full-length TRIOBP isoforms has been shown to influence neurite outgrowth (8). To determine which domain(s) could be involved in this process, Neuroscreen-1 cells were transfected with constructs encoding these domains or an empty vector control and then induced to differentiate for 72 h with nerve growth factor. The morphology of the ensuing transfected cells was then examined by immunofluorescent microscopy and quantified in a blinded manner (Fig. 8A). Notably, expression of the PH domain led to a significant increase of ∼30% in the number of neurites per cell (Fig. 8B, only neurites >20 μm in length were considered), suggesting this domain to play an active role in neurite initiation. There was also an increase in the length of the longest neurite per cell following expression of the PH domain, although this did not survive correction for multiple testing (Fig. 8C). In contrast, expression of the central domain led to an ∼30% decrease in neurite number relative to control cells (Fig. 8B). The most likely explanation is that this is a dominant negative effect of expressing the actin-binding domain in isolation and, thus, in competing with endogenous TRIOBP-1. The C-terminal domain did not affect neurite outgrowth in this experiment.

Figure 8.

The effect of TRIOBP-1 subdomains on neurite outgrowth. A, sample images of Neurosreen-1 cells transfected with either an empty vector (Mock, n = 171) or one of the following TRIOBP-1 fragments: aa 60–189 (PH, n = 174), aa 281–555 (CC1–4, n = 161), and aa 556–652 (CC5–6, n = 152). Color has been deliberately overexposed to ensure all neurites are visible. Scale bars represent 50 μm. B, the mean number of neurites per cell (defined as any outgrowth >20 μm in length) shows the PH domain alone to increase neurite number and the central domain to decrease it. *, p < 0.05; **, p < 0.01 after correction for multiple testing. C, mean length of the longest neurite in each cell. The PH domain has a trend effect, significant before but not after correction for multiple testing. All values were compared by one-way ANOVA with Bonferroni correction to the Mock value.

The aggregation propensity of TRIOBP-1 is dependent on a 25-amino stretch between coiled coils 1 and 2

In our previous paper we found evidence that insoluble TRIOBP-1 may accumulate specifically in the brains of at least a subset of patients with schizophrenia (8). Furthermore, it was demonstrated that full-length TRIOBP-1 or TRIOBP-1 lacking its PH domain can be made to aggregate when overexpressed in neuroblastoma cells, thus providing a potential model for insoluble TRIOBP-1 in the brain (8). Similar aggregated accumulations of exogenously expressed TRIOBP-1 have also recently been reported by others (28). If the aggregation propensity of TRIOBP-1 arises as a result of a specific cellular mechanism, such as being modulated by a protein-protein interaction or post-translational modification, then it should be possible to identify a specific region of TRIOBP-1 that is critical for this aggregation event to occur. To investigate this, further constructs encoding fragments of TRIOBP-1 were generated, this time fused to N-terminal FLAG tags, and were expressed in human SH-SY5Y neuroblastoma cells.

Notably, although full-length TRIOBP-1 forms aggregates when overexpressed (Fig. 9A), the N-terminal region of TRIOBP-1 showed no such tendency, instead being found in the cytoplasm and at the periphery of the cell where it co-localized with actin (aa 1–280; Fig. 9B). When the PH domain was expressed in isolation, it showed a similar cell periphery/actin localization with no sign of aggregation but was also found prominently in the nucleus (aa 60–189; Fig. 9C). This was in agreement with our previous finding that deleting the PH domain of TRIOBP-1 did not ablate its aggregation propensity (8).

Figure 9.

Subcellular localization and aggregation propensity of TRIOBP-1 fragments in neuroblastoma cells. FLAG-tagged TRIOBP-1 protein fragments were transfected into SH-SY5Y cells and then visualized using an anti-FLAG antibody and immunofluorescence microscopy. F-actin was visualized using florescent-labeled phalloidin. For each construct, the aa position in full-length TRIOBP-1 is displayed. All scale bars represent 20 μm. A, the full-length protein, aa 1–652, which aggregates. B, the N-terminal region, aa 1–280. C, the PH domain, aa 60–189. D, CC1–6, aa 281–652, which forms aggregates. E, CC1–4, aa 281–555, which forms aggregates. F, CC5–6, aa 556–652. G, CC5, aa 556–619.

In contrast, a construct encoding CC1-CC6 of TRIOBP-1 was seen to form large accumulations in the cell (aa 281–652, Fig. 9D), as did a construct encoding just CC1-CC4, the central domain (281–555 Fig. 9E). Reinforcing the idea that aggregation propensity of TRIOBP-1 arises from the central domain; the C-terminal domain alone formed a diffuse cytoplasmic pattern when the two coiled-coil regions of this domain were expressed (CC5 and CC6, aa 556–652; Fig. 9F) or when CC5 was expressed alone (aa 556–619; Fig. 9G).

To determine which section of the TRIOBP-1 central coiled domain was involved in aggregation, further truncation constructs were expressed in neuroblastoma cells. Notably, neither constructs containing only CC3–6 or CC3–4 formed aggregates (aa 383–652 and aa 383–555; Fig. 10, A and B), implying that an aggregation-critical region or motif lies within the first two short coiled coils. CC1 and CC2 alone were unstable and rapidly degraded when expressed in the cell but could be stabilized by the addition of a CFP fusion protein (Fig. 10C). Alone, this region did not show clear aggregation, which was not a quenching effect of the CFP, as full-length TRIOBP-1 fused to CFP formed clear aggregates (Fig. 10D). Instead, fine mapping of the aggregation critical region was performed by systematic removal of sections of the CC1-CC6 construct from the N-terminal end. Two constructs expressing CC2-CC6 were, therefore, generated: one incorporating a 25-aa stretch that lies between CC1 and CC2 and one lacking this short section. Dramatically, although the construct featuring this region formed aggregates (aa 324–652; Fig. 10E), the protein lacking it instead adopted a more natural cytoplasmic expression pattern (aa 349–652; Fig. 10F).

Figure 10.

Further subcellular localization and aggregation propensity of TRIOBP-1 fragments in neuroblastoma cells. FLAG-tagged TRIOBP-1 protein fragments were transfected into SH-SY5Y cells and then visualized using an anti-FLAG antibody and immunofluorescence microscopy. F-actin was visualized using florescent-labeled phalloidin. For each construct, the aa position in full-length TRIOBP-1 is displayed. All scale bars represent 20 μm. A, CC3–6 of TRIOBP-1, encoded by aa 383–652. B, CC3–4, aa 383–555. C, CC1–2, aa 281–382; however for this image an enhanced cyan fluorescent protein (ECFP)-fused construct was used, and so anti-TRIOBP-1 antibody was used alongside the ECFP florescence to visualize the protein. D, the full-length protein, aa 1–652, with an ECFP fusion protein. Note, that although ECFP stains the whole aggregate, an anti-TRIOBP-1 antibody stains only the periphery. This is consistent with previous reports of TRIOBP-1 aggregates (28) and likely represents the antibody as unable to penetrate the core of the densely packed protein aggregate. E, CC2–4, including the 25-aa linker between CC1 and CC2, aa 324–652, which forms aggregates. F, CC2–4, excluding the 25 aa linker between CC1 and CC2, aa 349–652.

To determine whether these punctate structures really represented insoluble TRIOBP-1 species, both constructs were expressed in cells that after lysis were subjected to a series of solubilization buffers and centrifugation steps to remove all but the most insoluble proteins. Upon examination by Western blot, both the constructs encoding aa 324–652 and aa 349–652 were seen to be present in cell lysates, but only 324–652 was retained in the insoluble protein fraction (Fig. 11A), consistent with its representing an insoluble, aggregated TRIOBP-1 species. This 25-aa linker region between CC1 and CC2, therefore, appears to be critical for its aggregation propensity. To further verify this, SH-SY5Y cells were transfected with five previously employed TRIOBP-1 truncation constructs in a blinded manner, and the number of TRIOBP aggregates per cell was quantified. The presence of this 25-aa region was shown to have a highly significant effect on the number of such aggregates in the cytoplasm (Fig. 11B), with three constructs containing the region having a mean of at least 1.3 aggregates per cell compared with 0.1 or less for two constructs lacking this region. This, therefore, reinforces the idea that this subsection of the coiled central region is essential for aggregation. Among the cells displaying TRIOBP-1 aggregates (which for the purposes of this experiment were defined as any intense area of anti-FLAG signal >0.5 μm in diameter in the cell body) there was variation in the size of TRIOBP-1 aggregate seen but with no significant differences between the different TRIOBP-1 fragments used (Fig. 11C).

Figure 11.

Confirmation of an aggregation-critical region between coiled coils 1 and 2 of TRIOBP-1. A, cell lysates and corresponding purified insoluble protein pellets of SH-SY5Y cells expressing FLAG-tagged TRIOBP-1 aa 324–652 or 349–652. B, in a blinded experiment, cells were transfected with TRIOBP-1 fragments, and the mean number of aggregates per cell was quantified. S.E are shown along with p values determined by one-way ANOVA, with Bonferroni correction for multiple testing. Constructs used encoded TRIOBP-1 aa 190–652 (n = 65), aa 281–652 (n = 71), aa 324–652 (n = 72), aa 349–652 (n = 71), or aa 383–652 (n = 65). Over the whole dataset, p < 0.001. n.s., not significant. *, p < 0.05; ***, p < 0.001. C, in the same experiment, the mean diameter of FLAG-positive aggregates did not vary significantly between constructs. D, Western blot of lysates from SH-SY5Y cells transfected with the TRIOBP-1 constructs used in the blinded aggregation assay. E, FLAG-tagged TRIOBP-1 protein fragment aa 190–652 transfected into SH-SY5Y cells. F, transfection of an identical construct lacking the aggregation critical region aa 324–348. G, transfection of a construct encoding TRIOBP-1 aa 1–382. All scale bars represent 20 μm.

When tested by Western blot, there was some variation in the level of expression of proteins from these (Fig. 11D), likely representing variable levels of nonsense-mediated decay; however, there was no obvious correlation (positive or negative) between expression level and aggregation propensity.

To further confirm the requirement of this region for aggregation, long TRIOBP-1 constructs containing the entire coiled-coil region (aa 190–652; Fig. 11E) were confirmed to aggregate when expressed in SH-SY5Y but not when this 25-aa region was deleted (aa 190–652 Δ324–348; Fig. 11F). Note that these fragments could not be readily studied by SEC as, like the aa 281–652 construct (Fig. 4, A and B), they degraded into smaller stable fragments when purified from bacteria. The N-terminal region plus CC1–2 did not aggregate however (aa 1–382; Fig. 11G), demonstrating that although this 25-aa linker region is essential for aggregation to occur, it is not in itself sufficient to induce aggregation, requiring the presence of at least some other part of the central domain. The reason for this lack of aggregation is likely due to loss of a more C-terminal section of TRIOBP-1 rather than a stabilizing effect of the PH domain, given that full-length TRIOBP-1 aggregates (Figs. 9A and 10D) and previous experiments have shown that deletion of the PH domain has no effect on aggregation propensity (8).

Discussion

Many chronic neurodegenerative conditions can be characterized by the presence of insoluble aggregates of specific proteins as evidence for aberrant proteostasis. The concept that aberrant proteostasis may also exist in psychiatric illness is an emerging one, with TRIOBP-1 being a potential example of such a protein, having been identified via a hypothesis-free affinity proteomics approach using brain samples of patients with schizophrenia (8).

Here, we have demonstrated that a specific region of TRIOBP-1, bounded by aa 324–348, is an absolute requirement for its aggregation, indicating that it is likely to occur through a specific physiological process as opposed to the random aggregation of naturally sticky proteins. This was revealed through the expression of truncated TRIOBP-1 constructs in neuroblastoma cell lines. Although these truncated forms of TRIOBP-1 are not believed to be expressed in vivo, their similarity in expression pattern to full-length TRIOBP-1 suggests that they may nevertheless be a useful tool to study its protein aggregation. This 25-aa region lies between two short predicted coiled coils, CC1 and CC2 (Fig. 12A), which form a dimeric structure when expressed in isolation but appear to be part of the larger hexameric central coiled-coil domain. In our previous analyses (8), this portion of TRIOBP-1 was one of the less strongly predicted to contain an “aggregation motif,” with only the sequence around aa 230–236 predicted to have some aggregate-forming propensity by the AGGRESCAN (29), FoldAmyloid (30), and ProA (31) servers. It was not predicted as such by TANGO (32, 33). In all instances, however, at least four sites elsewhere in the protein were more strongly predicted to be capable of inducing aggregation; thus, this location for an aggregation-critical region was an unexpected result. The 25-aa region contains multiple proline residues (Fig. 12B), suggesting that it likely represents a turn between CC1 and CC2. It is also rich in charged amino acids, allowing for the possibility that TRIOBP-1 aggregation is dependent on protein-protein interactions, of which several have previously been reported for TRIOBP-1 (3, 18, 19). That this region was not strongly predicted to have an innate aggregation propensity might further implicate the involvement of additional proteins or other cellular factors in the aggregation process, with such factors instead interacting with TRIOBP-1 in a manner dependent on this 25-aa region.

Figure 12.

Summary of the domain structure of TRIOBP-1. A, the PH (purple) domain and coiled-coil regions of TRIOBP-1 are shown. Coiled coils are divided between those, which constitute the central domain (CC1-CC3 and possibly CC4, red) and the C-terminal domain (CC5-CC6). Regions of interest are indicated in green, including the aggregation critical region and the oligomeric states of selected fragments of TRIOBP-1 when expressed as recombinant proteins. B, sequence of the 25-amino acid region found to be essential for aggregation. It is rich in prolines (shown in purple) and charged residues (red for positive charge, blue for negative).

The large size of the protein aggregates generated when TRIOBP-1 constructs incorporating this region were expressed combined with the high variability in the sizes of these protein accumulations is consistent with these objects representing unfolded or misfolded TRIOBP-1 species. The exact mechanism by which TRIOBP-1 aggregation occurs remains to be determined, but it is likely to be tied into its oligomeric state and/or its protein interaction partners. An E3 ubiquitin ligase, HECTD3 (homologous to E6-AP carboxyl terminus protein D3), has also been reported to be responsible for degradation of TRIOBP-1 (19), and it is, therefore, likely that disruption of this process could lead to aberrant TRIOBP-1 protein homeostasis, potentially including aggregation.

We have also more generally explored the domain organization of TRIOBP-1, which was strongly predicted to consist of an N-terminal PH domain and a large C-terminal coiled-coil region. The predicted PH domain is now confirmed to be a distinctly folded region when expressed in vitro and appears to be of functional relevance for the generation of neurites based on neuron-like cells investigated in culture. Furthermore, we have now found the coiled-coil region to consist of two distinct domains, the central domain consisting of CC1-CC3 and potentially also of CC4, whereas the C-terminal domain consists of CC5-CC6 (Fig. 12A). The central region is responsible for the assembly of TRIOBP-1 oligomers, forming an oligomeric state comparable with that previously described for the full-length protein (18) and containing the aggregation-critical region. CC4 corresponds to the region of TRIOBP-1 recently identified to be required for interaction with NDEL1 and subsequent effects on cell migration (11) and also contains a known phosphorylation site of importance for mitotic progression (10). The central domain is also involved in neurite outgrowth, with its expression in isolation leading to a reduction in neurite number, possibly as a result of the truncation mutant having a dominant negative effect.

That expression of TRIOBP-1 leads to an increase in cellular polymerized F-actin is well established (3), and knockdown of TRIOBP-1 can be used as a tool to prevent F-actin formation in the cell (9, 34); however, the mechanism by which it does this is unclear. Here we demonstrate that at least when expressed in isolation, both the central and C-terminal domains can bind to F-actin with the central domain limiting its depolymerization. The C-terminal domain also showed a trend toward the same effect. These data suggest that the TRIOBP-1 protein promotes F-actin formation indirectly by inhibiting its degradation into monomeric G-actin. This is broadly in agreement with previous results showing that amino acids 72–444 of TRIOBP-1 (adapted to the numbering system used here and incorporating the PH domain plus approximately CC1–3) are sufficient to bind actin (3). That the central and C-terminal domains are each capable of modulating actin dynamics in isolation is analogous to the fact that both the TRIOBP-1 and TRIOBP-4 proteins modulate actin despite sharing no common amino acid sequence. It thus appears that longer isoforms of TRIOBP, such as TRIOBP-6, could contain at least three distinct actin-binding and -modulating domains: the central and C-terminal regions of TRIOBP-1 plus the R1 motif of TRIOBP-4 (5).

To the best of our knowledge, four other proteins have so far been identified that may also form protein aggregates in mental illness. Of these, both DISC1 and dysbindin-1B are proteins encoded for by genetic risk factors for major mental illness, which were later found in insoluble protein fractions of subsets of patients with mental illness (35, 36). CRMP1 (collapsin response mediator protein 1) was identified through a combination of genetics and our antibody-based proteomics approach (6), whereas NPAS3 (neuronal PAS domain protein 1) aggregation appears to occur in the presence of a point mutation that is found in a family with schizophrenia (37, 38). In all these cases, like TRIOBP-1, overexpression of the protein in mammalian cells leads to the formation of protein aggregates (6, 36, 38), and in the case of DISC1 and dysbindin-1B this has also been seen when transgenic human proteins are expressed in rodents (39–41). NPAS3 (neuronal PAS domain protein 1) aggregation, like that of TRIOBP-1, appears to be dependent on a particular region of the protein, in this case narrowed to an ∼100-aa-long region that includes the aggregation-related point mutation (38). Although such experiments have not been directly undertaken for the other proteins, early work on DISC1 did describe a central region of the protein as required for its subcellular localization to “puncta” (42) reminiscent of protein aggregates. Furthermore, a seemingly distinct region of the C terminus has been shown to be responsible for the aggregation propensity of recombinant DISC1 proteins (43).

TRIOBP-1 is, therefore, one of several proteins that appears to form aggregates in cell culture systems through specific, although not yet elucidated, cellular processes and that are also implicated in forming insoluble complexes in the brain. If, as we would predict, these insoluble accumulations in the brain correspond in form and effect to the aggregated protein structures seen in cell culture systems, then further study of such aggregates would yield significant insight into the pathophysiological events underlying chronic mental illness.

Experimental procedures

Plasmid constructs

The human TRIOBP-1 652 aa open reading frame (RefSeq accession number NP_008963.3) was cloned from the cDNA of human SH-SY5Y neuroblastoma cells, and then regions of the gene were subcloned. These reading frames were transferred into the pDONR/Zeo vector using BP clonase (plasmid and enzyme from Thermo Fischer Scientific, Darmstadt, Germany) except for the full-length protein, which was cloned and ligated into the pENTR1A no ccDB vector (44) (E. Campeau, #17298, Addgene, Cambridge, MA). These reading frames were then transferred using LR clonase II (Thermo Fischer Scientific) into one or more of the following destination vectors: pETG10A (A. Geerlof, EMBL, Heidelberg, Germany), pdcDNA-FLAGMyc (B. Janssens, #LMBP 4705, BCCM/LMBP Plasmid Collection, Zwijnaarde, Belgium), or pdECFP (45) (S. Wiemann, #LMBP 4548, BCCM/LMBP Plasmid Collection). All constructs were confirmed by sequencing, and the expressed proteins were confirmed by Western blot.

Recombinant protein expression and purification

Recombinant TRIOBP-1 fragments were expressed from pETG10A vectors in BL21 Star (DE3) or Rosetta (DE3) pLysS Escherichia coli (respectively, from Thermo Fisher Scientific and Merck Millipore, Darmstadt, Germany) and induced with 0.2–1 mm isopropyl β-d-1-thiogalctopyranoside under conditions optimized for each construct. Bacterial pellets were resuspended in 20 mm Tris, pH 7.4, 500 mm NaCl, 25 mm imidazole, and 20 mm MgCl₂ containing protease inhibitors and DNase I and lysed by the addition of 1 mg/ml lysozyme and 1% Triton X-100. Insoluble material was removed by centrifugation, and the soluble fraction was incubated for 16 h at 4 ºC with nitrilotriacetic acid-agarose resin (Qiagen, Hilden, Germany). Resin was then transferred to a gravity flow column and washed with 20 mm Tris, pH 7.4, 500 mm NaCl, 25 mm imidazole. Protein was eluted by the addition of the same buffer containing 250 mm imidazole. Further purification was performed by SEC on a HiLoad 16/600 Superdex 200pg column (GE Healthcare) pre-equilibrated with 20 mm Tris, pH 7.4, 150 mm NaCl. This was performed using an ÄKTA Pure system (GE Healthcare) that was kept cooled to 4–8 ºC at all times. Inferred molecular weights of protein peaks were calculated based on calibration with protein standards and compared with theoretical molecular weights calculated using the ExPASy ProtParam tool (46). For CD and AUC analyses, proteins were transferred to 20 mm sodium phosphate, pH 7.4, 150 mm sodium fluoride using a HiPrep 26/10 Desalting column (GE Healthcare).

Circular dichroism

Protein samples were analyzed on a J-815 spectrometer (Jasco, Groß-Umstadt, Germany). Measurements were recorded at λ = 260 to 185 nm using a 1-nm resolution, 20-nm/min scan speed, 1-mm optical path length, an integration time of 0.5 s, and a controlled temperature of 20 °C. An average reading was taken from 10 spectra per experiment. A buffer-only spectra was subtracted from these results, and the data were then transformed to units of mean residue ellipticity. Deconvolution of the data was performed in DichroWeb (21, 22) using the CDSSTR method (23, 24) and reference dataset SMP180 (25).

Analytical ultracentrifugation

Measurements were carried out using a Beckman Optima XL-A ultracentrifuge (Beckman-Coulter, Brea, CA) equipped with an absorbance detector and an eight-hole rotor. For sedimentation velocity experiments, 400-μl samples were loaded into a 12-mm double-sector aluminum cell. All samples were equilibrated to 20 °C, and the centrifugation was performed at 20 °C at 50,000 rpm. The absorbance readings of all samples were recorded at 285 nm with a radial resolution of 30 μm. Data were analyzed by Sedfit (version 15.01b) using the continuous c(s) distribution model. The partial specific volumes of the target proteins, buffer density, and viscosity were determined by Sednterp (version 20130813BETA). All graphs were generated by GUSSI (version 1.2.1), and the reported s-values were normalized to s_20,w values.

Western blot, gels, and quantification

Acrylamide gels were run, transferred to nitrocellulose membranes, and blocked according to standard protocols. Commercial antibodies were used against α-actin (#A2066, Sigma, raised in rabbit), the FLAG tag (#F1804, Sigma, raised in mouse), the His₆ tag (#27E8, Cell Signaling Technology, Danvers, MA, raised in mouse), TRIOBP-1 (#SAB2102579, #HPA019769, and #HPA003747, Sigma, raised in rabbit against human TRIOBP-1 recombinant protein fragments containing amino acids 144–193, 279–379, and 504–635, respectively), and α-tubulin (#T9026, Sigma, raised in rabbit). Antibody signal was visualized and quantified where necessary using IRDye secondary antibodies, an Odyssey Clx instrared imaging system, and associated analysis software (LI-COR Biosciences, Bad Homburg, Germany). Total protein in acrylamide gels was visualized using InstantBlue (Expedeon, Swavesey, UK).

Cell culture

The human neuroblastoma cell line SH-SY5Y (Leibniz Institute DMSZ, Braunshweig, Germany) was cultured in DMEM/F-12 containing 10% fetal calf serum, 1× MEM non-essential amino acids, penicillin, and streptomycin. The human neuroblastoma cell line NLF (Children's Hospital of Philadelphia, PA) were grown in RPM1 1640 medium supplemented with 10% fetal calf serum, 2 mm l-glutamine, penicillin, and streptomycin. The rat pheochromocytoma cell line Neuroscreen-1 (Thermo Fischer Scientific) was grown in RPM1 1640 medium supplemented with 10% horse serum, 5% fetal calf serum, 2 mm l-glutamine, penicillin, and streptomycin. Transfections were performed for 24 h using Lipofectamine 2000 according to manufacturers' instructions. For neurite outgrowth experiments, Neuroscreen-1 cells were given 50 ng/ml nerve growth factor for 72 h after transfection. All cell culture chemicals were from Thermo Fischer Scientific. Cells expressing full-length TRIOBP-1 in a doxycycline-inducible manner were established using the Retro-X Tet-On Advanced Inducible Expression System (Clontech Laboratories, Mountain View, CA).

Immunocytochemistry

Cells were seeded on glass coverslips, fixed with PBS and 4% paraformaldehyde, permeabilized with PBS and 0.5% Triton X-100, and blocked with PBS and 10% goat serum. Primary antibodies were applied and detected using Alexa Fluor secondary antibodies (488 or 596 nm, Thermo Fischer Scientific). Actin was visualized using acti-stain 488 phalloidin (#PHDG1, Cytoskeleton, Inc., Denver, CO) according to manufacturer's instructions. This was also used to visualize total cell body and neurites in quantification experiments. Coverslips were mounted using ProLong Gold with DAPI (Thermo Fischer Scientific) and viewed on an LSM-510 confocal microscope (Carl Zeiss Microscopy, Göttingen, Germany) at room temperature. Images were taken using a Plan-Apochromat ×100 objective, Immersol 518 immersion oil, and ZEN 2 pro software (all from Carl Zeiss Microscopy). All images displayed are representative of at least three independent experiments.

For the quantification experiments, plasmid DNA preparations were prepared and then coded so that the researcher performing transfections and immunocytochemistry was blinded as to which TRIOBP-1 construct was used for each set of cells. Photographs were then taken of the first transfected cells identified by microscopy, and quantified measurements were taken from these images (using ImageJ, version 1.45s; Ref. 47) before unblinding of the samples. Data were analyzed by one-way ANOVA, with Bonferroni correction for multiple testing.

Insoluble protein fraction purification

Purification of the insoluble fractions of SHSY-5Y cells was performed as described previously (8). Briefly, cells were lysed and washed with a series of buffers variously containing 1% Nonidet P-40, 0.2% Sarcosyl, DNase I, and/or 1.5 m NaCl, with centrifugation steps to remove all proteins solubilized by each buffer. The final pellet was then resuspended in loading buffer and probed by Western blot alongside samples of the original, unfractionated cell lysate.

Actin polymerization assays

NLF neuroblastoma cells were transfected with constructs encoding regions of TRIOBP-1 or an empty control vector and grown for 48 h. Cells were then lysed at 37 °C with 50 mm PIPES, pH 6.9, 50 mm NaCl, 5 mm MgCl₂, 5 mm EGTA, 5% glycerol, 0.1% Nonidet P-40, 0.1% Triton X-100, and 1% Tween 20 plus protease inhibitor mixture for 10 min with gentle shaking. Cell debris was removed by centrifuging at 350 × g for 10 min. Cell lysate was then ultracentrifuged at 100,000 × g for 1 h at 37 °C, and the G-actin-containing supernatant was removed. The supernatant was incubated for a further 60 min at 37 °C and then ultracentrifuged again to test for additional polymerization of actin. Meanwhile, the F-actin-containing pellet was resuspended in 5 mm Tris, pH 8.0, 0.2 mm CaCl₂, 0.2 mm ADP plus protease inhibitor mixture. This was incubated for 4 h at 4 °C to facilitate depolymerization of actin and then ultracentrifuged at 100,000 × g for 1 h at 4 °C. Pellets from this second round of ultracentrifugation were resuspended in their respective buffers. Actin and TRIOBP-1 content was assayed by Western blot. The experiment was repeated three times.

Author contributions

N. J. B. performed the experiments except for the circular dichroism and analytical ultracentrifugation experiments, which were designed by L. N-S. and performed by A. S. K. Y. and T. Z. N. J. B. and R. M. generated novel reagents. N. J. B. and C. K. designed the experiments and wrote the paper with input from the other authors. All authors analyzed the results and approved the final version of the manuscript.