Conversion of Green Fluorescent Protein into a Toxic, Aggregation-prone Protein by C-terminal Addition of a Short Peptide*

A non-natural 16-residue “degron” peptide has been reported to convey proteasome-dependent degradation when fused to proteins expressed in yeast (Gilon, T., Chomsky, O., and Kulka, R. (2000) Mol. Cell. Biol. 20, 7214-7219) or when fused to green fluorescent protein (GFP) and expressed in mammalian cells (Bence, N. F., Sampat, R. M., and Kopito, R. R. (2001) Science 292, 1552-1555). We find that expression of the GFP::degron in Caenorhabditis elegans muscle or neurons results in the formation of stable perinuclear deposits. Similar perinuclear deposition of GFP::degron was also observed upon transfection of primary rat hippocampal neurons or mouse Neuro2A cells. The generality of this observation was supported by transfection of HEK 293 cells with both GFP::degron and DsRed(monomer)::degron constructs. GFP::degron expressed in C. elegans is less soluble than unmodified GFP and induces the small chaperone protein HSP-16, which co-localizes and co-immunoprecipitates with GFP::degron deposits. Induction of GFP::degron in C. elegans muscle leads to rapid paralysis, demonstrating the in vivo toxicity of this aggregating variant. This paralysis is suppressed by co-expression of HSP-16, which dramatically alters the subcellular distribution of GFP::degron. Our results suggest that in C. elegans, and perhaps in mammalian cells, the degron peptide is not a specific proteasome-targeting signal but acts instead by altering GFP secondary or tertiary structure, resulting in an aggregation-prone form recognized by the chaperone system. This altered form of GFP can form toxic aggregates if its expression level exceeds the capacity of chaperone-based degradation pathways. GFP::degron may serve as an instructive “generic” aggregating control protein for studies of disease-associated aggregating proteins, such as huntingtin, α-synuclein, and the β-amyloid peptide.

has been demonstrated in cell culture and in many transgenic mouse and invertebrate models (reviewed in Refs. [2][3][4]. We have shown previously that transgenic expression of the human ␤-amyloid peptide (A␤) 2 in Caenorhabditis elegans leads to the formation of intracellular aggregates and associated toxicity (5). Similar observations have been made for transgenic C. elegans animals expressing polyglutamine repeat proteins (6 -8), ␣-synuclein (9), or tau (10). One unanswered question for the C. elegans (as well as Drosophila and mammalian) disease models is the specificity of the observed toxicity, i.e. would any aggregating protein have the same effect? In theory, this question could be addressed by control experiments in which transgenic animals are constructed in parallel that express a nondisease-associated, "generic" aggregating protein. We show here that GFP can be converted into such a control aggregating protein and that expression of this aggregating GFP variant results in in vivo toxicity grossly similar to that observed for diseaseassociated aggregating proteins.
In a search for random peptides that would confer instability on proteins expressed in yeast, Gilon et al. (11) identified a non-natural 16-residue peptide (CL1) that conferred apparent ubiquitin-dependent degradation on the Ura3 protein. This short C-terminal "degron" peptide was subsequently used by Bence et al. (12) to engineer an unstable GFP variant (GFP-u) that underwent proteasome-dependent degradation in HEK 293 cells. When we attempted to engineer a similar construct to monitor proteasome function in C. elegans, we unexpectedly found that GFP::degron fusion protein aggressively formed persistent, perinuclear aggregates. We have therefore taken advantage of this unexpected result, and we used this novel, aggregating derivative of GFP to examine the generic response to protein aggregation in a C. elegans model.

EXPERIMENTAL PROCEDURES
DNA Constructions-The initial GFP::degron fusion was generated using C. elegans expression vector pPD118.20 (gift of A. Fire), which expresses a Ser-65 3 Cys GFP variant under the control of the C. elegans myo-3 promoter. A synthetic double-stranded oligonucleotide encoding the 16-residue CL1 degron peptide (cGCATGCAAGAACTGG-TTCTCTAGCCTGTCCCACTTCGTCATCCATCTGTAGg), bracketed by EcoRI-and NheI-compatible single-stranded extensions, was ligated to EcoRI/NheI-cleaved pPD118.20, resulting in the myo-3/GFP::degron vector pAT2. Blunt-end ligation was used to generate a control plasmid (pAT1) that contained a 7-nucleotide insert upstream of the degron sequence, resulting in an out-of-frame degron peptide. To generate a myo-3/GFP::PEST expression construct, the mouse ornithine decarboxylase 1 PEST region was recovered by PCR and inserted into the EcoRI/NheI sites of pPD118.20 as described above. To generate a GFP::degron construct with pan-neuronal expression, the myo-3 promoter region of pAT2 was replaced with a PstI/XbaI fragment from pCL35 (containing the snb-1 promoter and a synthetic 3Ј-untranslated intron), resulting in construct pCL54. Temperature-inducible expression of the GFP::degron was engineered by replacing a KpnI/NheI fragment of smg-1dependent myo-3 vector pPD118.60 (encoding unmodified GFP) with a KpnI/NheI fragment of pAT2 containing the GFP::degron fusion, resulting in pCL60.
To express the GFP::degron fusion protein in mammalian cells, an EcoRI/BstEI restriction fragment encoding the degron peptide was recovered from pCL54 and inserted into the downstream EcoRI/XmaI sites of vector pEGFP-C2 (Clontech). The resulting construct, pCL85, expresses an EGFP::degron protein under control of the CMV promoter. To engineer mammalian DsRed::degron expression, an AgeI/ EcoRI fragment from the pDsRed monomer (Clontech) encoding the DsRed monomer protein was swapped for the EGFP in pCL85. The DsRed stop codon was then removed by in vitro mutagenesis (Quick-Change kit, Stratagene), fusing the degron peptide in-frame with the DsRed monomer protein. The terminal peptide sequences of all expression constructs used in this study are shown in Table 1. For clarity, we will refer to the CMV-driven EGFP constructs as "GFP" in the text of this work, with the understanding that these constructs express the sequence-modified EGFP.
To engineer constitutive expression of HSP-16 in C. elegans muscle, the coding region of the hsp-16.2 gene (including its single intron) was amplified from genomic DNA and inserted between the KpnI and EcoRI sites of pPD118.20, thus replacing GFP with the hsp-16.2 coding sequence, generating pCL137. All constructions were checked by DNA sequencing.
Construction of Transgenic C. elegans Strains-Transgenic lines were generated by microinjection of the constructs described above. Plasmid pRF4, encoding dominant rol-6(su250) mutant collagen, was co-injected as a selectable morphological marker for all microinjections, with one exception; pCL137 (myo-3/hsp-16.2) was co-injected with a GFP marker plasmid (pCL26 (mtl-2/GFP)) that produces strong intestinal GFP expression. Lines with chromosomally integrated transgenes were generated by ␥-irradiation as described previously (5).
Primary cultures of dissociated neurons from E18 embryonic rat hippocampi were prepared essentially as described (13). Cells were maintained in Neurobasal medium (Invitrogen) supplemented with B-27 and Glutamax. After 7 days in vitro, cells were transfected with 2.0 g of plasmid DNA using Lipofectamine 2000, as per the manufacturer's instruction. Cells were maintained in a 37°C, 5% CO 2 incubator for 12 or 24 h to allow for protein expression.
The coverslips were examined by epifluorescence, and images of the EGFP, DsRed(monomer) and DAPI fluorescence were obtained as described below. These images were scored experiment-blind for the number of cells containing aggregated EGFP and the number containing aggregated DsRed. These values were normalized to the relative number of transfected cells by using the masking function of SlideBook software to determine the area in m 2 occupied by the fluor serving as the control for transfection efficiency.
As a further check on the results obtained by scoring the images blind, we devised a simple algorithm to detect aggregates objectively. We determined the mean intensity of both GFP and DsRed(monomer) in the area masked by the control fluor, we then set the threshold for visualization of each fluor at four times its mean intensity. The small points of fluorescence that exceeded that "4ϫ threshold" generally corresponded to aggregates identified by eye, but the following two issues made it less accurate than the human eye. First, cells with intense fluorescence exceeded the 4ϫ threshold regardless of whether they contained any aggregates (false positives). Some of these false positives were eliminated by specifying that true aggregates do not lie over the nuclei nor extend over more than half the cell. Second, some cells with weak overall fluorescence contained clear aggregates that did not exceed the 4ϫ threshold, which was based on the mean intensity of the population as a whole (false negatives). Despite the limitations of the algorithmbased approach, these objective measurements of aggregates paralleled the experiment-blind human scoring, validating our scoring methods.
For comparison of GFP fluorescence distribution in transfected primary neurons, three cultures were analyzed for both GFP and GFP::degron transfections. Exposure times were adjusted so that cell body fluorescence was within the linear range of the camera, and that cell body fluorescence was comparable between GFP and GFP::degron-transfected neurons. Fluorescence in all control GFPtransfected neurons was readily detectable in the distal axons and dendrites (12 of 12 cells, fluorescence at least 4ϫ higher than background), whereas fluorescence in GFP::degron-transfected cells was not detectable in these neurites above background (14 cells analyzed).
Hippocampal cells were fixed in a solution of 4% paraformaldehyde, 4% sucrose in phosphate-buffered saline and then blocked in 0.5% fish skin gelatin in phosphate-buffered saline for 1 h at 37°C. After blocking, coverslips were incubated with a polyclonal MAP2 (1:500, Chemicon AB5622) antibody overnight at 4°C (to identify neurons), which was followed by incubation in Cy3-conjugated goat anti-rabbit (1:500, The Jackson Laboratory) for 1 h at 37°C. Coverslips were mounted on glass slides in Elvanol containing 0.5 mg/ml DAPI. Images of GFP-expressing cells were acquired using a SPOT RT-SE (Diagnostics Instruments) cooled CCD camera controlled by Metamorph software (Universal Imaging) with a Ϫ60, 1.4 N.A. Plan Apo objective (Nikon).
Electron Microscopy-Synchronously hatched cultures of transgenic nematodes were propagated at 16°C for 36 h and then up-shifted as third larval stage animals to 25°C for 24 h to induce transgene expression, as described previously (15). Aliquots of induced fourth larval stage animals were frozen in a Balzers HPM 010 high pressure freezing apparatus as described previously (16). Samples were freeze-substituted in 2% osmium tetroxide and 0.05% uranyl acetate in acetone at Ϫ80°C for  Fig. 1A for corresponding Ponceau S total protein stain of this immunoblot. F, high magnification image of perinuclear fluorescent aggregates (arrows) in GFP::degron strain CL2292. G, electron micrograph of muscle cells expressing GFP::degron. Note perinuclear electron-dense deposits (arrows). A-D, living animals, digitally fused differential interference contrast (DIC) and epifluorescence (projected digitally deconvolved) images, size bar ϭ 50 m. F, live animal, digitally fused DIC and epifluorescence images, size bar ϭ 10 m. G, size bar ϭ 1 m. 5 days, gradually warmed to room temperature, infiltrated with Araldite/Embed-812 (Electron Microscopy Sciences), and polymerized. Thin sections were stained with 2% uranyl acetate and Reynold's lead citrate and imaged at 80 kV in a JEOL 100C or Phillips CM10 electron microscope.
Sequential Extraction of Nematode Homogenates-Homogenates were prepared as described previously for immunoprecipitation experiments (17) from equivalent size populations of transgenic worms and then extracted in 100 mM Tris, pH 8.6, 10% glycerol for 1 h at 4°C in the presence of protease inhibitors. Insoluble material was pelleted (25,000 ϫ g for 15 min) and then re-extracted in 1% SDS for 1 h at 4°C. The remaining insoluble material was pelleted (25,000 ϫ g for 15 min) and solubilized in 70% formic acid. Extracts were concentrated and desalted using Microcon centrifugal filter tubes (Millipore), fractionated on a NOVEX 4 -12% BisTris gel using MES SDS running buffer (Invitrogen), transferred to nitrocellulose, stained with Ponceau S to confirm that loading and transfer had been equivalent for all of the extracts, and immunoblotted with anti-GFP antibody (Qbiogene) (see supplemental Fig. S1).
Filter Trap Assays-Worm lysates were prepared using a Tris/Triton X-100 immunoprecipitation buffer as described previously (17). Lysates were centrifuged at 2,000 rpm in a microcentrifuge (325 relative centrifugal force) for 5 min to remove large insoluble material (cuticle fragments, etc.), and the supernatant was saved as the lysate supernatant. Gel fractionation of the lysate supernatants and subsequent Ponceau S staining were used to equalize levels of total protein between lysates. 30-l aliquots of lysate supernatant were run through a cellulose acetate membrane (0.2 m) pre-soaked in Tris/Triton X-100 immunoprecipitation buffer using vacuum pressure. The membranes were allowed to dry, and then the individual sample dots were placed into separate wells of a 96-well microtiter plate (clear, flat bottom wells). Total GFP fluorescence trapped on the membrane was quantified using a Tecan GENios microplate reader (485 nm excitation and 535 nm emission filters).
Immunoprecipitation Studies-Magnetic beads (Miltenyi Biotec) were cross-linked to anti-GFP 3E6 antibody (Qbiogene) and purified using a magnetized column (Miltenyi Biotec). Purified beads were mixed with 0.5 ml of worm lysate prepared as described previously (17). After incubation, the lysate/bead mixture was run through a magnetized column to purify the immunoprecipitate. This protocol involves no centrifugation or sedimentation, thereby avoiding nonspecific recovery of particulate material (e.g. protein aggregates). The MAC immunoprecipitate was washed four times with lysis buffer (1% Triton), once with 50 mM Tris, pH 7.5, and then eluted using a small volume of 95°C Laemmli Sample Buffer (4% SDS). The MAC immunoprecipitation eluate was run on a 12% BisTris gel using MES running buffer (Invitrogen) and immunoblotted with anti-GFP and anti-HSP-16 antibodies.
Scoring of Transgenic Worm Paralysis-Populations of transgenic worms were staged by a 2-h synchronous egg lay on 5-cm nematode growth media plates spotted with RW2 strain Escherichia coli. Worms were propagated at 16°C for 48 h after the end of the egg lay and then shifted to 25°C to induce transgene expression. Scoring for paralysis was initiated 20 h after temperature upshift. Animals were scored as paralyzed if they were associated with "halos" of cleared E. coli lawn around their heads (indicative of an insufficient body movement to access food) or if the animals did not move in response to prodding with a tapered platinum wire. Paralyzed animals were removed from the E. coli lawn at every time point to facilitate accurate counts of paralyzed animals.

RESULTS
GFP::Degron Aggregates in C. elegans Muscle-We fused the 16-residue CL1 peptide identified by Gilon et al. (11) to the C terminus of GFP and independently expressed either this fusion protein or control GFP in C. elegans body wall muscle using the promoter of the myo-3 gene. Unlike unmodified GFP, which is observed throughout the muscle cytoplasm (Fig. 1A), expression of GFP::degron results in perinuclear aggregates, with little detectable diffuse cytoplasmic staining (Fig. 1, B and F).
To determine the specificity of this effect, transgenic lines were generated expressing GFP fused to an out-of-frame degron peptide (Fig. 1C) or to the well characterized PEST domain of mouse ornithine decarboxylase (Fig. 1D). These control fusions displayed a cytoplasmic distribution like unmodified GFP. To investigate whether the aggregation of GFP::degron might be an artifact of exceptionally high level expression, total (formic acid-extractable) GFP levels were measured by quantitative immunoblot. As shown in Fig. 1E, the transgenic strain expressing unmodified GFP (lanes 1 and 3) has at least three times as much transgenic protein as does the strain expressing GFP::degron (lanes 2 and 4). Thus, the aggregates observed in the GFP::degron strain are because of some qualitative difference between the GFP::degron fusion protein and GFP. To examine these GFP::degron aggregates at the ultrastructural level, the GFP::degron strain was analyzed by high pressure freeze electron microscopy. As shown in Fig. 1G, we observed electron-dense, perinuclear deposits specifically in the muscle cells expressing GFP::degron. These cytoplasmic deposits are not membrane-bound and resemble aggresomes (18).
To determine whether the formation of visible GFP::degron aggregates coincided with the production of less-soluble forms of GFP, homogenates of transgenic worms expressing GFP or GFP::degron were subjected to sequential extraction using Tris/glycerol, 1% SDS, and formic acid buffers. As shown in Fig. 2A, only a small fraction of GFP::degron is extractable in the Tris/glycerol buffer (compare lanes 1 and 2), and only GFP::degron expression results in the formation of a pool of SDS-insoluble, formic acid-extractable material (compare lanes  5 and 6). Detergent-resistant GFP::degron aggregates were also observed in filter-trap assays using 0.2-m pore cellulose acetate filters (Fig. 2B). Quantification of retained fluorescence in replicate samples demonstrated that ϳ8-fold more GFP::degron was retained on the fil-  1, 3, and 5) and CL2292 (GFP::degron, lanes 2, 4 and 6) subjected to sequential extraction with Tris/glycerol (lanes 1 and 2), 1% SDS buffer ( lanes  3 and 4), and formic acid (lanes 5 and 6) and then fractionated by SDS-PAGE and immunoblotted with anti-GFP antibody. Note the increased fraction of GFP requiring detergent and formic acid solubilization in extracts from GFP::degron-expressing animals (lanes 4 and 6). See supplemental Fig. 1B for corresponding Ponceau S total protein stain of this immunoblot. B, Tris/Triton X-100 lysates from CL2179 (GFP, spot 1) and CL2292 (GFP::degron, spot 2) run through 0.2-m cellulose acetate filters. Note the ϳ8-fold increase in retained fluorescence for GFP::degron-expressing strain.
GFP::Degron Aggregation Occurs in Neurons and Mammalian Cells-Given the unexpected aggregation of GFP::degron in C. elegans muscle, we first investigated whether this fusion protein would also aggregate in C. elegans neurons. As shown in Fig. 3 (A and B), GFP that is expressed in ventral cord neurons is observed throughout the cell body and axons, whereas GFP::degron forms intense perinuclear deposits in the same neurons. We then asked whether GFP::degron would also aggregate in mammalian cells. The degron peptide was fused to EGFP in a CMV promoter-driven expression vector, and the resulting construct was used to transiently transfect undifferentiated mouse N2A cells and primary fetal rat hippocampal neurons. In both instances, transient GFP::degron expression resulted in perinuclear fluorescence different from the diffuse cytoplasmic fluorescence typically observed in cells transformed with control GFP (compare Fig. 3, C and E with D and F).
Transfection of HEK 293 Cells with Degron Fusion Constructs-The results described above differ from the reported fate of GFP::degron fusion protein GFP-u in HEK 293 cells (12). To determine whether this might be because of either cell type differences or subtle differences in the fusion protein sequences (see Table 1 for a comparison of GFP::degron and GFP-u terminal sequences), we transfected HEK 293 cells in parallel with GFP, GFP::degron, and GFP-u. The fraction of transfected cells containing visible aggregates was then scored 17 h after transfection. Cells transfected with either our GFP::degron construct or the previously described GFP-u construct had significantly more transfected cells with visible aggregates than cells transfected with GFP ( Fig.  4, A, C, and E, quantified in I). However, GFP itself was seen to occasionally form aggregates in transiently transfected cells (Fig. 4E), suggesting that GFP has a native capacity to aggregate when expressed at high levels in HEK 293 cells. This background aggregation of GFP complicates interpretation of the increased rates of GFP::degron and GFP-u aggregation. We therefore fused the degron peptide to DsRed(monomer) (19), which in our hands is much less aggregation-prone than GFP in HEK 293 cells. As shown in Fig. 4 (G and H, quantified in I), addition of the degron peptide to the DsRed(monomer) results in a dramatic increase (ϳ10-fold) in the fraction of transfected cells with aggregates, strongly supporting the view that fusion of proteins to the degron peptide predisposes them to aggregate.
Interaction of GFP::Degron with Chaperone Protein HSP-16-We have shown previously (20) that expression of the aggregation-prone human A␤ peptide in C. elegans muscle results in the formation of immunoreactive deposits, often with a perinuclear distribution similar to that of GFP::degron. A␤ expression also results in the induction and co-localization of the small heat shock protein HSP-16, a chaperone protein homologous to ␣B-crystallin. To determine whether GFP::degron expression has a similar effect on HSP-16 expression and

TABLE 1 End sequences of control and fusion proteins used in this study
Shown are the predicted end amino acid sequences of wild type GFP, expression vector GFP derivatives (pPD118.20 and EGFP-C2), GFP::degron fusions (pAT2, pCL85, and EGFP-u), DsRed(monomer), and DsRed(monomer)::degron fusion. Degron sequences are underlined. Note that unmodified expression vector GFPs contain peptide extensions as a result of the multiple cloning sites introduced into these vectors. These peptide additions (unlike the degron peptide) do not alter GFP fluorescence or solubility.

GFP
. . . HGMDELYK GFP (pPD118. 20) . . . HGMDELYKSPQALEFASRPYK GFP::degron (pAT2  localization, a population of CL2292 (myo-3/GFP::degron) animals was probed with anti-HSP-16 antibody. As shown in Fig. 5, GFP::degron expression likewise results in the induction of HSP-16 and its transient co-localization with the protein aggregates. The association of HSP-16 with GFP::degron aggregates was seen predominantly in young larval animals with weakly fluorescent GFP aggregates. As evident in the fused image (Fig. 5B), although a clear co-localization of GFP fluorescence and HSP-16 immunoreactivity was observed, there was not a one-to-one correspondence between the intensity of the GFP fluorescence and anti-HSP-16 signal, with some weakly fluorescent GFP deposits showing strong HSP-16 immunoreactivity and vice versa. To investigate further the nature of the association of GFP::degron and HSP-16, GFP::degron and control GFP protein were immunoprecipitated from populations of transgenic worms, fractionated, and probed with anti-HSP-16 antibody.
HSP-16 was co-immunoprecipitated much more strongly with GFP::degron than with GFP (Fig. 5D). (Consistent with this, we have shown previously that endogenous HSP-16 induced by heat shock is not efficiently co-immunoprecipitated with GFP (17).) As the immunoprecipitation experiments were performed under conditions that preserved the GFP::degron aggregates in the worm lysates, these studies demonstrated the association of HSP-16 with the aggregates but did not prove direct binding of HSP-16 to GFP::degron. Toxicity of GFP::Degron-To assess the cellular toxicity of the GFP::degron in vivo, we employed a temperature-inducible expression system previously used for inducible A␤ expression (15). (This approach allows recovery of transgenic animals under low transgene expression conditions, thus preventing selection against lines with high expression of a potentially toxic transgene.) We have demonstrated previously that temperature induction of A␤ expression in body wall muscle results in robust paralysis ϳ24 h after upshift. Transgenic lines were engineered in parallel to express unmodified GFP (CL2179) or GFP::degron (CL2337) when induced. Under these conditions, temperature upshift of the inducible GFP::degron strain led to paralysis of ϳ80% of the population by 24 h, whereas the inducible GFP animals were not paralyzed at this time (Fig. 6A). Immunoblot quantitation of total (formic acid-extracted) GFP levels showed that the induced CL2337 strain actually expresses less total GFP than induced CL2179, indicating that the observed paralysis is because of qualitative differences between GFP::degron and GFP (data not shown).
GFP::Degron Subcellular Distribution and Toxicity Are Altered by Overexpression of HSP-16-Co-expression of chaperone proteins can alter the toxicity of aggregation-prone proteins such as A␤ and polyglutamine repeat-containing proteins (21)(22)(23). To investigate whether this might also occur with GFP::degron, we generated stable transgenic strains expressing the hsp-16.2 coding sequence under the control of the myo-3 muscle-specific promoter, resulting in worms with constitutive expression of HSP-16.2 protein in body wall muscle. This high level constitutive expression of HSP-16.2 results in the formation of musclespecific HSP-16 inclusions visible by differential interference microscopy (DIC) in living animals. 3 These HSP-16 inclusions have not been observed when endogenous HSP-16 expression is induced (e.g. by heat shock). The myo-3/HSP-16.2 transgene was introduced by classic genetic mating into the CL2337 (GFP::degron) and CL2179 (unmodified GFP) backgrounds. Introduction of the myo-3/HSP-16 transgene into CL2337 strongly suppressed paralysis resulting from GFP::degron induction (Fig. 6B). This suppression was associated with a dramatic  1 and 2) and anti-HSP-16 (lanes 3 and 4). Note enhanced recovery of HSP-16 immunoreactivity in GFP::degron immunoprecipitates. See supplemental Fig. 1C for corresponding Ponceau S total protein stain of this immunoblot. redistribution of GFP::degron into the HSP-16.2 inclusions (Fig. 6, C  and D). Conversely, soluble GFP was excluded from the HSP-16.2 inclusions (Fig. 6, E and F). This result clearly demonstrates a differential interaction of GFP and GFP::degron with the HSP-16.2 chaperone protein, highlighting the apparent non-native character of the GFP::degron protein.

DISCUSSION
GFP::degron protein fusions (also designated GFP-u) have been used by several groups as an unstable GFP variant to monitor proteasome function (12, 24 -26) in mammalian cell culture. We have observed strong aggregation of GFP::degron upon in vivo expression in C. elegans muscle and neurons, as well as localized deposition of GFP::degron and DsRed(monomer)::degron in transiently transfected mammalian cells. These two sets of observations can be reconciled if C-terminal addition of the degron peptide promotes GFP misfolding or multimerization, resulting in a protein that is both aggregation-prone and subject to enhanced chaperone-mediated proteasomal degradation. In this model, the fate of GFP::degron (aggregation versus degradation) depends upon the balance of transgene expression and the capacity of a given cell type to degrade non-native proteins.
An alternative model is that the degron peptide acts as a specific proteasome targeting signal (e.g. analogous to a PEST domain), and the aggregates we observe occur because proteasomal function is overwhelmed by engagement with GFP::degron. We do not favor this model because of the following: 1) it does not account for the reduced solubility of GFP::degron in our biochemical assays, 2) it does not account for the specific interaction of GFP::degron with the HSP-16 chaperone protein, 3) an equivalent GFP::PEST fusion construct does not form aggregates in C. elegans muscle, 4) inhibition of proteasomal function by RNA interference of specific proteasomal subunits does not lead to aggregation of soluble GFP in C. elegans muscle (although this treatment does lead to enhanced aggregation of a GFP::Gln-35 polyglutamine repeat reporter protein) (27).
It is surprising that a relatively minimal addition of 16 residues to the C terminus of GFP can convert this normally soluble protein into a strongly aggregating one. This observation is in stark contrast to the many examples where GFP fusion proteins remain soluble or assume the cellular distribution of the partner fusion protein (28). The capacity of the degron peptide to induce GFP aggregation suggests that this peptide may have some unique structural characteristic. Overall, the degron peptide is not strongly hydrophobic, but it is predicted to form an almost perfect amphipathic ␣-helix. It is conceivable that degron addition leads to protein multimerization, perhaps driven by pairing of the hydrophobic and hydrophilic faces of the amphipathic helix. Alternatively, the C-terminal degron might directly interfere with GFP folding, resulting in misfolded (but still fluorescent) forms predisposed to aggregate.
We have shown that GFP::degron can be toxic when expressed in C. elegans muscle cells, as has been demonstrated previously for A␤ (5) and GFP::polyglutamine fusions (6). We have not observed any muscle cell changes by light or electron microscopy associated with GFP::degron expression that suggest a mechanism for this cell toxicity. However, it does not appear that this toxicity involves a classic apoptotic pathway (we see no indications of chromatin fragmentation by electron microscopy) or gross cellular damage. The large range of molecular and genetic tools available in C. elegans (e.g. suppressor screens) should enable us to identify components of GFP::degron toxicity, which we anticipate will be relevant to protein aggregation toxicity in general. Although the paralysis phenotypes induced by A␤ and GFP::degron are grossly simi-lar, these toxic proteins may still act through different or partially overlapping cellular and molecular mechanisms.
The transient association of HSP-16 with GFP::degron suggests that this chaperone protein may interact only with an initial form of GFP::degron aggregate, perhaps an oligomer. This seemingly aggregation stage-specific interaction of HSP-16 and GFP::degron may be analogous to the observation that HSP-16 does not co-localize with fully fibrillar ␤-amyloid (17). In C. elegans animals induced to express A␤, oxidative damage and gross toxicity precede amyloid formation (29), consistent with the growing body of evidence that prefibrillar oligomers may be the key toxic species for A␤ and other disease-associated proteins (30). Similarly, the initial form of GFP::degron bound by HSP-16 may be the critical toxic species of GFP::degron, not the strongly fluorescent, mature GFP::degron aggregates. The mechanism(s) by which HSP-16 overexpression suppresses GFP::degron aggregation and toxicity are currently under investigation. Of particular interest is whether the secondary or tertiary structure of GFP::degron associated with HSP-16 inclusions is different from its structure when it is deposited in perinuclear aggregates.
Interpretation of studies employing transgenic expression of specific disease-associated aggregating proteins would be enhanced if these studies included controls for the general effects of aggregating protein expression. There have been previous demonstrations of cellular toxicity resulting from exposure to nondisease-associated aggregating proteins or protein fragments, including the N-terminal domain of the E. coli HypF protein (31), the Src homology 3 domain of phosphatidylinositol 3-kinase (32), and endostatin (33). However, these studies involved exposing tissue culture cells to in vitro aggregated (fibrillar or pre-fibrillar) protein, not endogenous expression. Garcia-Mata et al. (34) have shown that fusion of the first 252 residues of membrane-associated transport factor p115 to the C terminus of GFP (GFP-250) results in an aggregating protein that forms aggresomes when expressed in COS7 cells, but they did not report on the toxicity of this fusion protein. We suggest that the GFP::degron protein is an appropriate aggregation control for transgenic C. elegans models, and possibly for transgenic models of aggregating protein diseases in general.