Inhibition of polyglutamine protein aggregation and cell death by novel peptides identified by phage display screening.

Proteins with expanded polyglutamine domains cause eight inherited neurodegenerative diseases, including Huntington's, but the molecular mechanism(s) responsible for neuronal degeneration are not yet established. Expanded polyglutamine domain proteins possess properties that distinguish them from the same proteins with shorter glutamine repeats. Unlike proteins with short polyglutamine domains, proteins with expanded polyglutamine domains display unique protein interactions, form intracellular aggregates, and adopt a novel conformation that can be recognized by monoclonal antibodies. Any of these polyglutamine length-dependent properties could be responsible for the pathogenic effects of expanded polyglutamine proteins. To identify peptides that interfere with pathogenic polyglutamine interactions, we screened a combinatorial peptide library expressed on M13 phage pIII protein to identify peptides that preferentially bind pathologic-length polyglutamine domains. We identified six tryptophan-rich peptides that preferentially bind pathologic-length polyglutamine domain proteins. Polyglutamine-binding peptide 1 (QBP1) potently inhibits polyglutamine protein aggregation in an in vitro assay, while a scrambled sequence has no effect on aggregation. QBP1 and a tandem repeat of QBP1 also inhibit aggregation of polyglutamine-yellow fluorescent fusion protein in transfected COS-7 cells. Expression of QBP1 potently inhibits polyglutamine-induced cell death. Selective inhibition of pathologic interactions of expanded polyglutamine domains with themselves or other proteins may be a useful strategy for preventing disease onset or for slowing progression of the polyglutamine repeat diseases.

Eight inherited neurodegenerative diseases, including Huntington's disease, dentatorubral pallidoluysian atrophy, spinobulbar muscular atrophy, and spinocerebellar ataxia types 1, 2, 3, 6 and 7, are caused by expanded CAG repeats in the coding region of the disease genes (1)(2)(3). The CAG codon is translated into glutamine, and the polyglutamine domain is the only region of homology among the eight disease proteins. The length of the repeat is the critical determinant of age-of-disease onset, with repeat length greater than 40 glutamines producing neu-rodegeneration in seven of the eight diseases (1)(2)(3).
Proteins with pathologic-length polyglutamine domains display novel properties that are not present in these proteins when they contain a shorter polyglutamine domain. Length-dependent polyglutamine-protein interactions are reported for Huntington-associated protein 1, glyceraldehyde-3-phosphate dehydrogenase, leucine-rich acidic nuclear protein, vimentin, neurofilament, apopain, calmodulin, WW domain proteins, and Ras-related nuclear protein/ARA24 (4 -12). Proteins with expanded polyglutamine domains also aggregate, and aggregation is a pathologic hallmark of the polyglutamine repeat diseases (13,14). These polyglutamine length-dependent properties may arise from the ability of long polyglutamine domains to adopt unique three-dimensional conformations and serve to confer the disease proteins with a pathologic gain of function (15,16).
Lansbury proposed that during the initial stages of folding of expanded polyglutamine proteins, misfolded intermediates interact with themselves (homologous interactions) or other proteins (heterologous interactions), leading to critical cell injury (16). Supporting this hypothesis of length-dependent alteration in tertiary structure, Trottier et al. (17) identified a monoclonal antibody (1C2) that preferentially recognizes proteins with long, but not short, polyglutamine domains. Since monoclonal antibody 1C2 recognizes the unique conformation of long polyglutamine domains, we reasoned that peptides with similar polyglutamine-binding properties could be identified by screening peptide libraries. Peptides that selectively bind pathologiclength polyglutamine domains may inhibit interaction with other proteins, thereby slowing, or preventing, disease pathology. In this paper, we identify several polyglutamine-binding peptides and demonstrate the ability of one of these peptides to inhibit polyglutamine aggregation both in a novel in vitro assay and in cultured cells. We further demonstrate that expression of a tandem repeat of a polyglutamine-binding peptide in cell culture inhibits polyglutamine-induced cell death.

MATERIALS AND METHODS
Phage Display Screening-Phage display library construction and screening were performed as described (18,19). Briefly, 33-mer nucleotides were ligated to the 5Ј terminus of the pIII gene of phage M13 to generate a peptide library with 11 amino acids added to the amino terminus of the pIII protein. Individual phage libraries were screened for binding to a polyglutamine-glutathione S-transferase fusion protein with 62 glutamines (Q 62 -GST). 1 Construction of the polyglutamine-GST vectors was described previously (20). Q 62 -GST was immobilized on * This work was supported by a Beeson Physician-Faculty Scholar Award from AFAR (to J. R. B.) and a Wills Foundation Fellowship (to Y. N.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
‡ ‡ To whom correspondence should be addressed: Box 2900, Department of Medicine (Neurology), Duke University Medical Center, Durham, NC 27710. Tel.: 919-684-0054; Fax: 919-684-6514; E-mail: james.burke@duke.edu. 96-well plates at a concentration of 2.5 g/ml in 100 mM NaHCO 3 , pH 8.5. The wells were then blocked with 0.1% bovine serum albumin to decrease nonspecific binding. After blocking, the plates were washed with PBS, 0.1% Tween 20, and phage were incubated for 7 h at room temperature. Unbound phage were removed by extensive washing with PBS containing 0.1% Tween 20. Bound phage were eluted sequentially with 50 mM glycine, pH 2.0, and 100 mM ethanolamine prewarmed to 50°C. Selected phage were isolated, and DNA were sequenced. Enzyme-linked immunosorbent assays were performed to quantify bound phage using horseradish peroxidase anti-phage antibody (Amersham Pharmacia Biotech).
Thioredoxin-Polyglutamine Constructs and Protein Purification-The CAG repeat of polyglutamine-GST vectors was amplified by polymerase chain reaction and ligated into the NcoI and EcoRI site of the pThio-His B vector (Invitrogen, San Diego, CA) (20). The sequence of the clones and the length of the CAG repeats were confirmed by DNA sequencing. Thioredoxin-polyglutamine fusion protein was produced in transformed Escherichia coli DH5␣ and purified using BPER lysis buffer (Pierce) and Pro-Bond columns (Invitrogen, San Diego, CA) according to the manufacturer's instructions. Purified thioredoxin-polyglutamine fusion protein was dialyzed against PBS, pH 7.4, and concentration was determined by a modified Lowry reaction (Pierce).
Thioredoxin-Polyglutamine Protein Turbidity Assay-Turbidity assays were performed in 200-l reactions consisting of thioredoxin-polyglutamine in phosphate-buffered saline in 96-well, low protein binding plates. Plates were incubated at 4°C, and turbidity was measured at 405 nm on a Thermo Max plate reader (Molecular Devices).
Peptide Synthesis-Peptides were synthesized at the Howard Hughes Medical Institute peptide sequencing facility at Duke University with an acetylated amino terminus and protection of tryptophan with t-butoxycarbonyl during synthesis. Identical results were obtained with peptides obtained from Bio-Synthesis Inc. (Lewisville, TX). Stock solutions of peptides were prepared fresh daily by dissolving in dimethyl sulfoxide at 2.5 mM and incubated with thioredoxin-polyglutamine protein at 0.25-25 M.
Constructs for Protein Expression in Cell Culture-The CAG repeats of the Q n -green fluorescent protein vector (7) were inserted into the XhoI and EcoRI sites of the pEYFP-N1 vector (CLONTECH, Palo Alto, CA) to construct the polyglutamine-YFP (Q n -YFP) vectors. The polyglutamine-binding peptide and control constructs were prepared by ligating synthetic oligonucleotides into the XhoI and BamHI sites of the pECFP-N1 vector (CLONTECH) to obtain glutamine-binding peptide 1-CFP (QBP1-CFP), scrambled QBP1-CFP (SCR-CFP), and a computergenerated random peptide-CFP (RAN)-CFP. The insert of QBP1-CFP was duplicated by polymerase chain reaction to obtain a tandem repeat of QBP1-CFP ((QBP1) 2 -CFP). The sequence of all constructs was confirmed by DNA sequencing. The first methionine of YFP or CFP was changed to isoleucine by polymerase chain reaction-site-directed mutagenesis to eliminate translation of the fluorescent protein without the polyglutamine domain. The sizes of the fluorescent fusion proteins were confirmed by Western blotting.
Cell Culture and Fluorescence Microscopy-COS-7 cells were grown and maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum. Cells were transfected with Effectene according to the manufacturer's instructions (Qiagen GmbH, Hilden, Germany). Cells were examined 48 h after transfection using a Zeiss fluorescence microscope equipped with YFP/CFP filter sets (Omega Optical Inc., Brattleboro, VT). The percentage of cells with aggregates was calculated by dividing the number of cells with aggregates by the total number of fluorescent cells, multiplied by 100. Cell viability was examined 48 h after transfection using ethidium homodimer (Molecular Probes, Inc., Eugene, OR). Cells were incubated with 5 M ethidium homodimer for 30 min (37°C) and then examined under a fluorescence microscope equipped with a rhodamine filter set. Cells with nuclear fluorescence were counted as dead. In each experiment, at least 200 transfected cells were counted. Cells expressing RAN-CFP and Q n -YFP were used as a control to calculate the relative difference in aggregate formation and cell death. Experiments were repeated at least four times.

Identification of Peptides That Preferentially Bind
Protein with a Pathologic-length Polyglutamine Domain-We screened a peptide phage display library to identify peptides that interact with proteins containing a pathologic-length polyglutamine domain. The M13 phage display library was constructed to contain a random 11-amino acid peptide inserted at the amino terminus of the pIII capsid protein (18). The 11-mer peptide was not completely random, since a fixed amino acid was inserted in the sixth position of the peptide (X 5 -fixed-X 5 ) to decrease the vast number of possible peptides (20 11 ) and permit more thorough sampling. 2.5 ϫ 10 11 phage from each of the following fixed amino acid libraries were screened for binding to a polyglutamine-glutathione S-transferase fusion protein with 62 glutamines (Q 62 -GST): aspartate, phenylalanine, histidine, lysine, leucine, proline, and tryptophan (19,21). After four rounds of successive screening, 350 polyglutamine-binding phage clones were isolated. The selected phage were then assayed by enzyme-linked immunosorbent assay for binding to normal length (Q 19 ) or pathologic-length (Q 62 ) polyglutamine-GST. Six phage clones bound Q 62 -GST greater than Q 19 -GST (binding ratios of Q 62 -GST to Q 19 -GST: 1.23-1.66) and had their DNA sequenced to elucidate peptide sequence (Table I; QBP1-6).
In Vitro Aggregation of Thioredoxin-Polyglutamine-We then developed an in vitro aggregation assay to examine whether these polyglutamine-binding peptides inhibit polyglutamine aggregation. Polyglutamine-GST proteins are not ideal for studying polyglutamine aggregation, since the polyglutamine domain does not aggregate unless it is cleaved from the GST moiety (22,23). To circumvent this limitation, we produced thioredoxin-polyglutamine fusion proteins (thio-Q n ; where n ϭ the number of consecutive glutamine residues. Thioredoxin is highly soluble, can be expressed at high concentrations in E. coli, and is easily purified (24). The length of the expressed glutamine domains was chosen to survey a range of normal and pathologic repeat lengths (normal: 19 and 35 glutamines; pathologic: 62 and 81 glutamines) (Fig. 1A).
To monitor thioredoxin-polyglutamine protein aggregation, we developed a turbidometric assay, similar to the assays commonly used to study microtubule assembly and ␤-amyloid aggregation (25,26). Solutions of thio-Q 62 and thio-Q 81 protein increased turbidity in a polyglutamine length-, time-, and concentration-dependent manner (Fig. 1, B and C; time dependence of aggregation of thioredoxin-polyglutamine with 81 glutamines not shown). To demonstrate that turbidity was produced by aggregated thioredoxin-polyglutamine protein containing 62 or 81 glutamines (thio-Q 62 or thio-Q 81 ), we pelleted the insoluble material by centrifugation or captured it on a 0.22-m filter. The pelleted and retained material was confirmed as thioredoxin-polyglutamine fusion protein containing 62 or 81 glutamines on Western blots probed with anti-thioredoxin or anti-polyglutamine antibody (not shown). In contrast, thioredoxin-polyglutamine protein containing 19 or 35 glu-TABLE I Phage display peptides with preferential binding to Q 62 -GST The 11-amino acid inserts in the pIII protein were designed with 5 random amino acids, 1 fixed amino acid (shown here underlined), and 5 random amino acids in a X 5 -fixed-X 5 format. Each purified phage was assayed for binding to immobilized Q 62 -and Q 19 -GST by enzyme-linked immunosorbent assay. Bound phage was detected with monoclonal antibody against M13 phage (Amersham Pharmacia Biotech). Binding ratios were determined by amount of phage bound to Q 62 divided by phage bound to Q 19 . None of these clones bound bovine serum albumin (used as a blocking agent during incubation) or GST alone.
Sequence (X 5 -fixed-X 5  In this turbidity assay, thio-Q 81 aggregated faster (not shown) and at lower concentrations than thio-Q 62 (Fig. 1B). Aggregation did not occur in solutions of thioredoxin-polyglutamine containing 62 glutamines at concentrations lower than 5 M. As shown in Fig. 1C, macroscopic aggregation started following a lag period, consistent with a reaction with two kinetic components, an initial slow phase (similar to nidus formation in crystal growth), and a second rapid polymerization phase (23). Turbidity of thioredoxin-polyglutamine fusion protein containing nonpathologic glutamine repeats (19 or 35 glutamines) at concentrations ranging up to 50 M did not change after 3 weeks at 4°C.
QBP1 Inhibits Thioredoxin-Polyglutamine Aggregation-We next examined the effect of these combinatorially generated polyglutamine binding peptides on polyglutamine aggregation in vitro. We synthesized the 11-mer peptide with the greatest differential binding to pathologic-length compared with normal length polyglutamine (Q 62 /Q 19 ) (QBP1; Table I). We chose this peptide because it displays the greatest selective binding to expanded polyglutamine domain proteins and therefore might preferentially block pathologic protein interactions.
The 11-mer peptide QBP1 potently inhibited aggregation of thioredoxin-polyglutamine protein with 62 glutamines (thio-Q 62 ) in vitro (Fig. 2). QBP1 completely inhibited thio-Q 62 aggregation at a molar ratio of 3:1 (thio-Q 62 : QBP1) (Fig. 2). QBP1 was less potent at inhibiting aggregation of thio-Q 81 and required a 10-fold molar excess (1:10, thio-Q 81 :QBP1) for complete inhibition (not shown). A scrambled sequence peptide of QBP1 (SCR; Table I) and a computer-generated random 11amino acid peptide (RAN; Table I) did not inhibit aggregation (Fig. 2). The addition of QBP1 after aggregation did not reverse aggregation of thio-Q 62 (not shown). QBP1, SCR, and RAN peptides had no effect on the turbidity of the nonpathologic thioredoxin-polyglutamine proteins with 19 or 35 glutamines (thio-Q 19 or thio-Q 35 ; thio-Q 19 not shown). QBP1 Co-localizes with Polyglutamine Aggregate in Cells-We next determined whether QBP1 also inhibits polyglutamine aggregation in transfected COS-7 cells. As we previously demonstrated, COS-7 cells expressing polyglutamine domain fusion proteins are a good cellular model of the polyglutamine repeat diseases because the polyglutamine-fusion proteins mimic the native disease proteins by forming aggregates and by killing cells in a polyglutamine length-dependent pattern (7).
To determine the intracellular distribution of both polyglutamine and QBP1, we designed fusion proteins of polyglutamine with yellow fluorescent protein (Q n -YFP, where n ϭ 19, 45, 57, or 81 glutamines) and fused QBP1 with cyan fluorescent protein (QBP1-CFP) (Fig. 3A). YFP and CFP are variants of green fluorescent protein with distinct emission spectra, which enable separate detection of each fluorescent protein in double labeled cells (27). QBP1 fused to cyan fluorescent protein (QBP1-CFP) expressed in COS-7 cells remained diffusely distributed (not shown). A polyglutamine-yellow fluorescent fusion protein with 19 glutamines (Q 19 -YFP) was also diffusely distributed, and its distribution was unaffected by co-expression with QBP1-CFP (Fig. 3B, a and b). Expression of a pathologic-length polyglutamine-YFP fusion protein with 81 glutamines (Q 81 -YFP) in transfected cells formed aggregate. Co-expression of Q 81 -YFP and QBP1-CFP produced co-localization of these two fluorescent proteins in the protein aggregates (Fig. 3B, c and d). In contrast, random peptide or scrambled peptide fused to CFP (RAN-CFP and SCR-CFP) did not co-localize with the aggregate formed by Q 81 -YFP (Fig. 3B, e-h).
QBP1 Inhibits Polyglutamine Aggregate Formation in Cells-We next examined the effect of QBP1-CFP expression on Q n -YFP aggregate formation in co-transfected COS-7 cells. Diffuse fluorescence was easily distinguishable from aggregate using the fluorescence microscope (compare diffuse fluorescence of polyglutamine in Fig. 3B, a, with punctate fluorescence of aggregated protein in Fig. 3B, c, e, and g). Co-expression of QBP1-CFP reduced the percentage of cells with polyglutamine aggregates (Fig. 4). Inhibition of polyglutamine aggregation by QBP1-CFP was most pronounced with shorter pathologiclength glutamine fusion proteins (45 glutamines (Q 45 ) Ͼ 57 glutamines (Q 57 ) Ͼ 81 glutamines (Q 81 )). Compared with cells co-expressing RAN-CFP, QBP1-CFP reduced aggregation of Q 45 -YFP by 39% ( Fig. 4; p Ͻ 0.01). QBP1 also reduced aggregation of Q 57 -YFP by 26% (p Ͻ 0.01). A trend toward decreasing aggregation was seen in cells transfected with QBP1 and Q 81 -YFP, but the difference did not reach statistical significance (p ϭ 0.073). The decline in QBP1-CFP's ability to inhibit ag-  (13). Results shown are from representative experiments (n ϭ 4). Variation between duplicate wells was less than 10%.
gregation of proteins with increasingly long polyglutamine domains in cells is consistent with our in vitro data showing that QBP1 is less effective at inhibiting aggregation of thio-Q 81 compared with thio-Q 62 . To determine if duplication of the sequence of QBP1 would affect its ability to inhibit polyglutamine aggregation, we prepared a tandem repeat of QBP1 fused to CFP. (QBP1) 2 -CFP is more effective at inhibiting all lengths of polyglutamine-YFP aggregation than monomer QBP1-CFP (Fig. 4, dotted bars). SCR-CFP did not alter aggregation of cells expressing polyglutamine-CFP with 45, 57, or 81 glutamines (Q 45 -, Q 57 -, or Q 81 -YFP) (Fig. 4, hatched bars).
Polyglutamine Binding Peptide Inhibits Cell Death-The relationship between polyglutamine protein aggregation and cell death is controversial. To determine whether QBP1 or tandem-QBP1 inhibit polyglutamine-induced cell death, we assayed cell membrane permeability to ethidium homodimer. The mem-  RAN (open bar) and SCR (hatched bar), at 25 M, inhibited thio-Q 62 aggregation less than 10%. Aggregation was assayed by turbidity at 405 nm (13). Results shown are from representative experiments (n ϭ 4). Variation between duplicate wells was less than 10%. branes of living cells are impermeable to ethidium homodimer, and permeability to ethidium homodimer is a well established measure of cell death (28,29). Ethidium homodimer undergoes a 30-fold increase in fluorescence upon binding to nucleic acid, allowing easy detection with a fluorescence microscope using a rhodamine filter. QBP1 and (QBP1) 2 fused to CFP inhibit Q 57 -YFP-induced cell death (Fig. 5). As with inhibition of aggregation, (QBP1) 2 is more effective at inhibiting cell death than monomer QBP1. Similar results have also been observed with Q 45 -and Q 81 -YFP (not shown). Scrambled QBP1 fused to CFP (SCR) does not inhibit cell death (Fig. 5). DISCUSSION No therapies modify the age-of-onset of symptoms or pathologic progression of the polyglutamine repeat diseases. In this paper, we identified a novel peptide that is a therapeutic prototype because it preferentially binds pathologic-length polyglutamine domains, inhibits polyglutamine aggregation both in vitro and in cultured cells, and reduces cell death. In addition, we described an in vitro, high throughput, polyglutamine aggregation assay that can be used to identify small molecules that inhibit aggregation.
Inhibition of in vitro and intracellular polyglutamine protein aggregation by QBP1 is sequence-specific, since a scrambled version of the QBP1 peptide had no effect on aggregation. Binding to proteins with expanded-length polyglutamine domains or inhibition of polyglutamine aggregation by QBP1 is robust and is observed whether the peptide is free or fused to the M13 pIII protein or to CFP. QBP1 binds to polyglutamine whether the polyglutamine domain is fused to GST, thioredoxin, or YFP, suggesting that polyglutamine-binding peptides would interact with the expanded polyglutamine domain in disease proteins, such as huntingtin.
Our data further demonstrate that QBP1 inhibits polyglutamine-induced cell death. QBP1-CFP expression inhibited Q 57 -YFP-induced cell death by 19%, and tandem QBP1 inhibited cell death by 50%.
Despite the almost universal presence of polyglutamine aggregates in the polyglutamine repeat diseases, the role of aggregation itself in pathogenesis is controversial. Polyglutamine protein aggregates may cause neurodegeneration or form as a response to cell injury. Igarashi et al. (32) demonstrated that inhibition of aggregation by transglutaminase inhibitors partially blocked apoptotic cell death. Similarly, Chai et al. (33) found that overexpression of chaperone protein HDJ-1 in PC12 cells inhibited polyglutamine protein aggregation and toxicity. In contrast, others have found that aggregation and cell death are separate phenomena (34). Klement et al. (35) found that deletion of the self-association of ataxin 1 blocked aggregation but not cell death. Similarly, Saudou et al. showed that cell death induced by mutant huntingtin was not directly correlated with intranuclear aggregate formation (36). Whatever the function of aggregates, interaction of pathologic-length polyglutamine proteins with themselves or other proteins may disrupt critical cellular processes and destroy homeostasis. In our experiments, decreased aggregation parallels the decline in toxicity, but this does not prove that decreased toxicity is mediated by decreased aggregation. QBP1 may be exerting its effect by inhibiting interactions of polyglutamine proteins with other molecules, and the decrease in aggregation may be an epiphenomenon.
A limiting factor to identifying new therapeutic agents for these diseases is the lack of a high throughput screening assay. Aggregation of polyglutamine proteins in vitro has been previously described, but these assays either require proteolytic cleavage of polyglutamine fusion proteins or employ detection systems (such as dynamic light scattering) not readily adaptable for rapid screening (23,37). In this paper, we describe a new, simple, in vitro assay of polyglutamine aggregation that facilitates the identification of compounds that inhibit aggregation. Aggregation of thio-Q n protein in vitro faithfully recapitulates the behavior of polyglutamine proteins in human disease. As shown here, thio-Q n aggregation occurs in vitro only with repeats longer than 35; in Huntington's disease, the most common polyglutamine repeat disease, individuals develop disease only if they express a huntingtin protein with more than 36 sequential glutamines (38). In vitro, thio-Q n protein with longer pathologic-length polyglutamine domains aggregates more rapidly and at lower concentration than with shorter pathologic-length glutamine domains; similarly, in Hunting- ton's disease, polyglutamine domain length directly correlates with earlier age of onset, severity of clinical phenotype, and aggregate formation (39). The ability to identify compounds that selectively alter intracellular interactions and metabolism of pathologic-length polyglutamine domain proteins may be an effective therapeutic strategy in these diseases.