Advertisement

Glycine–alanine repeats impair proper substrate unfolding by the proteasome

Martin A Hoyt, Judith Zich, Junko Takeuchi, Mingsheng Zhang, Cedric Govaerts, Philip Coffino

Author Affiliations

  1. Martin A Hoyt1,,
  2. Judith Zich1,,
  3. Junko Takeuchi1,
  4. Mingsheng Zhang1,
  5. Cedric Govaerts2 and
  6. Philip Coffino*,1
  1. 1 Department of Microbiology and Immunology, University of California, San Francisco, CA, USA
  2. 2 Department of Cellular and Molecular Pharmacology, University of California, San Francisco, CA, USA
  1. *Corresponding author. Department of Microbiology and Immunology, UCSF, 513 Parnassus Avenue, San Francisco, CA 94143‐0414, USA. Tel.: +1 415 476 1783; E-mail: Philip.Coffino{at}ucsf.edu
  1. These authors contributed equally to this work

Abstract

Proteasome ATPases unravel folded proteins. Introducing a sequence containing only glycine and alanine residues (GAr) into substrates can impair their digestion. We previously proposed that a GAr interferes with the unfolding capacity of the proteasome, leading to partial degradation of products. Here we tested that idea in several ways. Stabilizing or destabilizing a folded domain within substrate proteins changed GAr‐mediated intermediate production in the way predicted by the model. A downstream folded domain determined the sites of terminal proteolysis. The spacing between a GAr and a folded domain was critical for intermediate production. Intermediates containing a GAr did not remain associated with proteasomes, excluding models whereby retained GAr‐containing proteins halt further processing. The following model is supported: a GAr positioned within the ATPase ring reduces the efficiency of coupling between nucleotide hydrolysis and work performed on the substrate. If this impairment takes place when unfolding must be initiated, insertion pauses and proteolysis is limited to the portion of the substrate that has already entered the catalytic chamber of the proteasome.

Introduction

Proteasomes are the major neutral protease of eukaryotic cells. They constitute the effector arm of the ubiquitin–proteasome system, which first tags proteins, usually by ubiquitin chain conjugation, and then delivers them to the proteasome, where they are cleaved to peptides. Proteasomes are ATP‐dependent proteases. A feature common to these taxonomically and structurally diverse molecular machines is the sequestration of proteolytic sites within a closed chamber (Pickart and Cohen, 2004). Substrate specificity is conferred by the ubiquitin tagging system, which licenses proteins to enter the chamber. Within that closed space, destruction is relatively independent of substrate sequence. Access is through a narrow pore (<15 Å) positioned axially at either end of the barrel‐shaped chamber (Lowe et al, 1995; Groll et al, 1997). Entry thus demands tag recognition, unfolding and insertion. Unfolding is essential, as folded proteins are too bulky to traverse the entryway.

Significant progress has been made in understanding substrate recognition and proteolysis, but the intermediate processes of unfolding and insertion have substantially eluded investigation. The proteasome consists of the proteolytic 20S core particle capped at one end or at both ends by the 19S regulatory complex. This complex includes at least 18 distinct proteins, of which six are ATPases (Glickman et al, 1998). In eukaryotes, each ATPase is distinct; all are essential (Rubin et al, 1998). Hexameric ATPase rings, members of the very diverse AAA ATPase family, have the general capacity to change the conformation of their substrates (Vale, 2000), fueled by a cycle of ATP binding, hydrolysis and release. These general considerations imply that the ATPases of the proteasome are the key mediators of substrate unfolding and insertion, an inference with experimental support (Liu et al, 2002b). The hexameric ClpX ATPase of the ClpXP protease, an architecturally similar but distantly related ATP‐dependent protease of bacteria, has been shown to utilize ATP when degrading a fully unfolded protein and to consume substantially more ATP if the same protein is presented in a folded form (Kenniston et al, 2003). Experiments of this kind have demonstrated that the energy cost of selective protein destruction is similar in magnitude to that used for protein translation. How ATP‐sourced energy is produced and consumed within the proteasome is therefore important both for understanding its function and in evaluating the overall energy budget of the cell.

One way to understand the capacity and limits of a system of energy production and use is to challenge it with loads that drive it to failure. This is not easy to do with proteasomes. They are a general‐purpose protein disposal machine and must be engineered, perhaps over‐engineered, to deal with substrates of very diverse structure. One substrate that causes them to fail is the EBNA1 protein of Epstein–Barr virus (Levitskaya et al, 1995; Levitskaya et al, 1997). It contains an amino‐acid tract consisting entirely of glycine and alanine residues, termed the glycine–alanine repeat (GAr). This varies from about 60 to 300 residues in different viral isolates. The GAr acts in cis to impair EBNA1 degradation by the proteasome; the GAr is also cis‐inhibitory when transferred to various other proteins. Our laboratory recently showed that a GAr does not prevent substrate recognition by the proteasome, or impair initiation of degradation, but instead causes degradation to stall, rather than go to completion (Zhang and Coffino, 2004). Stalling can result in the production of partially processed intermediates that are trimmed at the end of the protein that first enters the proteasome (Zhang et al, 2004). Our data suggested that stalling depends on an interaction between the GAr and a nearby folded domain of the substrate. We hypothesized that the GAr interacts ineffectively with the ATPases of the proteasome; this reduces the efficiency of coupling nucleotide hydrolysis to performing work on the substrate. If the reduction of energy delivery associated with passage of the GAr through the ATPases coincides with a requirement for unfolding, this transient imbalance causes unfolding to fail. Insertion therefore pauses and proteolysis is limited to the portion of the substrate that has already entered the proteolytic chamber. We here describe several independent experimental tests that affirm specific predictions of the model.

Results

Mutations that impair the structural stability of ODC reduce halting

Mouse ornithine decarboxylase (ODC) is a native proteasome substrate with an intrinsic carboxyl‐terminal degradation tag (Ghoda et al, 1989). Ubiquitin conjugation has no role in ODC degradation (Murakami et al, 1992; Coffino, 2001). Its 37‐amino‐acid carboxyl‐terminus (C37) is a portable module sufficient for proteasome recognition and initiation of proteolysis (Ghoda et al, 1990). ODC or fusion proteins bearing the ODC C‐terminal degradation tag compete with ubiquitin‐conjugated proteins and Lys48‐linked ubiquitin chains for degradation by the proteasome (Zhang et al, 2003). Degradation of ODC is unidirectional and initiates at the degradation tag (Zhang et al, 2004). As with almost all proteasome substrates, ODC proteolysis, once initiated, goes to completion. However, inserting a module containing a 30‐residue Gly–Ala repeat and short linker sequences (the GAr) after amino acid 424 of mouse ODC, a position just N‐terminal to the degradation tag, pauses degradation, producing stable intermediates missing approximately 20, 30 or (more rarely) 45 amino acids from this ODC::GAr30 construct (Zhang and Coffino, 2004). ODC truncated after residue 424 (of the full 461 amino‐acid protein) has full enzymatic activity, and crystallographic evidence shows that the truncated protein encompasses all parts of the protein that have discernable structure (Kern et al, 1999b). Insertion after ODC residue 424 therefore positions the GAr between a well‐structured domain and the terminal C37 tag. Crystallographic data reveal a tightly folded β‐sheet domain that ends at residue 410, followed immediately by an additional nine residues in a α‐helical configuration. The remainder of ODC::GAr30 beyond this helix is strongly predicted to lack ordered structure.

We have postulated that ODC::GAr30 degradation intermediates accumulate because of an imbalance between the energy delivered by the GAr‐impaired proteasome and that needed to unfold the C‐terminal β‐sheet domain of ODC. By reducing the structural stability of this domain, it should be possible to restore a favorable energy balance during the course of ODC::GAr30 degradation. Two outcomes are possible after degradation of a protein substrate begins: complete degradation to peptides or halting. The model we are testing implies that reducing the stability of a structured domain should reduce the probability of the second outcome, resulting in a greater fraction that is fully degraded.

The C‐terminus of ODC comprises a β‐sheet domain that includes two N‐terminal β‐strands (B1 and B2, ODC residues 2–19) and one C‐terminal B18 β‐strand (residues 404–410) (Kern et al, 1999b). Upstream of B1 and B2, the N‐terminus loops around the B18 β‐strand to form a belt‐like structure with a short α‐helix packed against the C‐terminus (see Figure 1A). In this arrangement, the C‐terminus of the protein is structurally stabilized by the N‐terminus through backbone–backbone contacts in the β‐sheet and a series of other interactions, listed in Table I. In order to weaken the C‐terminus structure, the N‐terminal belt of ODC::GAr30 was partially or completely removed by a series of truncations. In addition, a hydrophobic interaction between N‐ and C‐terminal ODC residues was eliminated by replacing Leu14 with Asp (Table I). We sought mutations that would destabilize local structure, but not cause gross misfolding that might divert the protein to other (presumably ubiquitin‐dependent) pathways of proteolysis. We expressed these mutant proteins in the yeast Saccharomyces cerevisiae, a cellular milieu in which the mechanism of mouse ODC degradation in animal cells is conserved (Hoyt et al, 2003), and examined their degradation.

Figure 1.

Ribbon diagrams of interactions stabilizing β‐sheet domains. (A) Mouse ODC (PDB id: 7ODC); inset shows a global view of the monomer. The C‐terminus (residues 403–418) is shown in orange and the N‐terminus in blue. The side chains of residues involved in stabilizing interactions described in Table I are shown as solid sticks, with carbon atoms in green, oxygen in red and nitrogen in blue. Residues 30–35 are missing from the crystal structure. (B) Human DHFR (PDB id: 1DHF); inset shows a global view of the monomer. The C‐terminus is shown in orange and the ‘safety belt’ in blue. Stabilizing residues are shown as solid sticks. For N29 and G164, only the interaction backbone moieties are shown.

View this table:
Table 1. Effects of perturbations predicted to alter stability of ODC and DHFR C‐terminal β‐sheet domains on intermediate accumulation

As measured by pulse‐chase experiments, essentially all the degradation products of ODC::GAr30 were intermediates (Figure 2A). At each time point in the chase period, the sum of the intensity of the parent (full‐length) protein and the intermediates is approximately constant. Quantitation of the results of three independent pulse‐chase experiments of transformants expressing ODC::GAr30 (and two informative mutants described below) verified this assessment (Figure 2B). An 11‐amino‐acid N‐terminal truncation (Δ11ODC::GAr30) did not reproducibly alter degradation intermediate accumulation (data not shown). However, the further removal of a stabilizing hydrophobic cluster by the L14D mutation (in Δ11ODCL14D::GAr30) reduced the accumulation of degradation intermediates. With Δ11ODCL14D::GAr30, the sum of the parent and intermediate bands declined during the chase. A significant fraction of processing events therefore resulted in complete degradation (Figure 2A and B). Furthermore, a more extensive truncation of 17 amino acids (Δ17ODC::GAr30) reduced the fraction of products that appeared as intermediates similarly to the Δ11ODCL14D::GAr30 fusion protein. In summary, mutations predicted to destabilize the ODC C‐terminal β‐sheet domain reduced the ability of the GAr element to impede degradation. GAr function thus depends on the stability of the domain proximal to this element. Unfolding has been described as the rate‐limiting step in degradation of proteasome substrates (Thrower et al, 2000). Consistent with these previous findings, the half‐life of ODC::GAr30 is about twice that of Δ11ODCL14D::GAr30 or Δ17ODC::GAr30, the two mutants bearing the destabilized β‐sheet domain (Figure 2B).

Figure 2.

Mutations that disrupt ODC‐stabilizing interactions reduce the accumulation of degradation intermediates. (A) Pulse‐chase analysis of yeast cells expressing wild‐type or mutant ODC::GAr30 fusion proteins. The position of full‐length ODC::GAr30 (solid arrow) or degradation intermediates (dashed arrows) is indicated. (B) Quantitation of ODC::GAr30 turnover data from (A), presented as the mean of three independent experiments. Error bars represent the standard deviation. The values for wild‐type ODC::GAr30 (•), Δ17ODC::GAr30 (□) and Δ11ODCL14D::GAr30 (▴) are expressed as a percentage of the total of the full‐length parent fusion protein and its degradation intermediates at time zero. The changes in the wild‐type and mutant proteins are compared for the parent molecule (middle panel), its degradation intermediates (bottom panel) and the sum total of these (top panel) at each time point. (C) Pulse‐chase analysis of ODC::GAr30 fusion proteins containing the C441A mutation in the ODC degradation tag. Arrows indicate bands as in (A).

The ratio of intermediates can be summarized by the value PI, the fraction of substrates initiating degradation that end up as intermediates. PI has a value between zero (no intermediates, parent is completely degraded) and one (only intermediates). This value facilitates comparisons of diverse GAr‐containing proteins (Table I). PI was approximately 1.0 for ODC::GAr30, and was not significantly changed in Δ11ODC::GAr30. However, the further introduction of the L14D mutation in the Δ11ODCL14D::GAr30 construct reduced the value of PI to 0.63. Furthermore, the more extensive truncation of 17 amino acids (Δ17ODC::GAr30) similarly reduced PI to 0.66.

This analysis requires that the mutations do not stimulate a turnover pathway independent of the ODC degradation signal. To confirm that the proteins were solely degraded in this way, we mutated an essential residue in the degradation tag, corresponding to Cys441 of mouse ODC (Miyazaki et al, 1993). A C441A mutation was introduced into the fusion proteins, and in each case prevented degradation (Figure 2C). We conclude that these mutated fusion proteins are degraded by an ODC‐specific (and thus proteasome‐dependent and ubiquitin‐independent) degradation pathway.

Perturbations that alter the structural stability of DHFR also alter halting

Appending the C‐terminal degradation signal of ODC to diverse proteins, including dihydrofolate reductase (DHFR), converts them to proteasome substrates (Loetscher et al, 1991; Li et al, 1998; Zhang et al, 2003). To investigate whether the β‐sheet domain of DHFR has the potential to interrupt substrate processing by collaborating with a GAr, we constructed a fusion protein containing the entire human DHFR open reading frame followed by the GAr30 module and the 37‐residue ODC degradation tag, forming DHFR‐GAr30‐C37.

Pulse‐chase analysis showed that DHFR‐GAr30‐C37 generated a single stable degradation product shorter than the parent molecule (Figure 3A), in which approximately 20 residues were removed from its C‐terminus (data not shown). In contrast to ODC::GAr30, which is almost always completely processed to intermediates (PI∼1.0), DHFR‐GAr30‐C37 degradation gave rise to intermediates only about one‐third of the time (PI∼0.3). Like ODC, the C‐terminus of DHFR is a β‐strand contained in a β‐sheet domain that involves three C‐terminal β‐strands and one N‐terminal β‐strand (Prendergast et al, 1988). Here, the stability of the C‐terminal strand is enhanced by a ‘safety‐belt’ structure, where residues 157–168 loop around the last β‐strand. This segment is packed against the strand and anchored to the rest of the molecule through specific interactions (see Figure 1B and Table I). The conformation of this loop is critically dependent on the presence of Pro 160 and Pro 163. To determine whether destabilizing the C‐terminal β‐sheet domain of the DHFR moiety influenced intermediate accumulation, we followed a strategy similar to that used with ODC. Three mutations anticipated to destabilize the C‐terminal β‐sheet of DHFR were tested (Figure 1B, Table I). Mutants with a substitution of Arg36 to Ser (DHFRR36S‐GAr30‐C37), Glu172 to Ser (DHFRE172S‐GAr30‐C37) or Pro 160 and 163 to Ala (DHFRP160A/P163A‐GAr30‐C37) were constructed and their degradation in yeast cells was examined by pulse‐chase analysis. The R36S mutation had little effect on intermediate production and accumulation. However, both the E172S and P160A/P163A substitutions had marked effects (Figure 3A). The PI value for DHFR‐GAr30‐C37, 0.31, was reduced by the E172S and P160A/P163A mutations, respectively, to 0.10 and 0.07 (Table I). Pulse‐chase analysis demonstrated that all three proteins containing a C441A substitution of the critical cysteine were stable (Figure 3B).

Figure 3.

The accumulation of DHFR‐GAr30‐C37 degradation intermediates is influenced by the stability of the DHFR domain. (A) Pulse‐chase analysis of wild‐type and mutant DHFR‐GAr30‐C37 proteins. (B) Pulse‐chase analysis of fusion proteins containing the C441A mutation in the ODC degradation tag. (C) Pulse‐chase analysis of DHFR‐GAr30‐C37 in yeast cells either with (+MTX) or without (−MTX) the addition of 20 μM methotrexate. Arrows indicate bands as in Figure 2.

Because processing of DHFR‐GAr30‐C37 is poised to produce either outcome with substantial probability, this construct provides the opportunity to test the consequences of both destabilization and stabilization of the DHFR moiety. Methotrexate binds tightly to the DHFR active site and has been used extensively to impair DHFR unfolding, for example (Eilers and Schatz, 1986). We have previously shown that methotrexate can block degradation of an ODC::DHFR fusion protein (Zhang et al, 2004). Yeast cells expressing DHFR‐GAr30‐C37 were treated with methotrexate or were left untreated and subjected to pulse‐chase analysis of turnover and intermediate accumulation (Figure 3C). Treatment with the stabilizing ligand increased PI three‐fold, to 0.90 (Table I). Together, the DHFR‐destabilizing mutations and ligand‐mediated stabilization provided a bidirectional test of our hypothesis and demonstrated that GAr function is strongly influenced by the stability of an adjacent C‐terminal domain.

Position of cutting sites that generate stable intermediates not altered by extending the degradation tag

The model being tested postulates that a structured domain stops insertion of a GAr‐containing substrate and thus halts its proteolysis. This implies that the sites at which proteolysis terminates must be positioned by reference to the folded domain, not determined by the configuration of the C‐terminal degradation tag. We tested this hypothesis by duplicating the C37 degradation tag of an ODC::GAr30 fusion protein, thus adding an additional 38 residues to the C‐terminus. We expressed ODC::GAr30 fusions with either the normal or extended C‐termini in yeast transformants and performed pulse‐chase experiments to compare the size of the intermediates they produce (Figure 4). As expected, the fractionation of the two full‐length fusion proteins by SDS–PAGE was consistent with the expected differences in their molecular mass. The extension of the C‐terminus had no apparent effect on the degradation rate of the fusion protein when compared to the unextended molecule. However, the mobility of the intermediates derived from the two proteins during the chase period was not distinguishably different. This implies that cutting is both limited and positioned by structures that lie on the N‐terminal side of the sites of terminal proteolysis.

Figure 4.

The positions of cutting sites that generate stable intermediates are not altered by extending the degradation tag. Pulse‐chase analysis of yeast cells expressing ODC::GAr30 fusion proteins with wild‐type (C37) or extended (C75) degradation tags. The parent bands and common intermediates are indicated by arrows as in Figure 2.

Spacing between GAr and folded domain is critical for halting

Because the functional interaction between the GAr and a folded domain is postulated to require that they arrive simultaneously at their respective sites of action, altering the spacing between these two elements should alter the synchrony of their arrival and thus modify halting. A series of homologous substrates were constructed and expressed in yeast transformants to test this idea (Figure 5A). In the first of these, 0, 7, 13, 19 or 25 residues were inserted between ODC position 424 and the GAr30 tract. The constructs with spacer lengths of 0, 7 or 13 residues differed little from ODC::GAr30 in the extent of halting (Figure 5B and C); for each of these, most processing events produced intermediates (PI range 0.68–0.88). However, extending the spacer length to 19 or 25 residues sharply reduced the efficiency of intermediate production (PI=0.21 and 0.20). Because the folded β‐sheet domain of ODC extends to residue 410 (Kern et al, 1999a), 14 residues lie between the end of the β‐sheet domain and ODC residue 424. Therefore, the presence of 33 or 39 residues between the β‐sheet domain and the GAr30 tract reduced its function, whereas 27 or fewer did not. (As reference, in ODC::GAr30 used in the experiments described above, the β‐sheet domain‐GAr distance is 26 residues.)

Figure 5.

The effects of spacing between GAr and β‐sheet domain on production of intermediates. (A) Schematic of ODC::GAr30 constructs containing spacers of varying length between the ODC C‐terminal β‐sheet and the GAr30 domains. The number of spacer residues is indicated below each construct. (B) Pulse‐chase analysis of cells expressing constructs in (A). (C) Quantitation of turnover data for ODC[n]GAr30 fusions in (B) with spacer lengths of 0 (•), 7 (□), 13 (Δ), 19 (▪) or 25 (○) residues were analyzed as in Figure 2B.

In the above series of constructs, all structural elements were held constant except for the distance between the β‐sheet domain and the GAr30. Consequently, the distance between ODC residue 424 and the C37 degradation tag, which in the native protein immediately follows position 424, ranged from 30 to 55 residues. To determine whether this variable rather than the distance between the β‐sheet domain and the GAr is the critical parameter determining intermediate production, we compared three constructs in which the spacing between ODC position 424 and C37 (Figure 6) was identical. Moving all 25 flanking residues to the GAr C‐terminal side produced abundant intermediates, similar to the standard ODC::GAr30 fusion protein. In contrast, when the 25 residue flanking sequence was placed on the N‐terminal side of the GAr, no intermediates were detected. This result supports the inference from the experiments above that the critical variable is the spacing between the β‐sheet domain and the GAr30.

Figure 6.

Pulse‐chase analysis of ODC::GAr30 linker constructs with C‐terminal extensions of fixed size. (A) Schematic of ODC::GAr30 constructs in which the context of the Gly–Ala repeat is varied with respect to its flanking linker sequences. The length of each region in residues is indicated below each construct. (B) Pulse‐chase analysis of yeast transformants expressing the fusion proteins shown in (A). Arrows indicate bands as in Figure 2.

Intermediates are not trapped by proteasome

For the substrates described here to undergo partial proteolysis, a part of their C‐terminal end must have reached the proteolytic chamber of the proteasome 20S core particle, while a folded region of the same protein remains positioned elsewhere, likely within or contiguous to the 19S regulatory complex. We asked whether, when insertion ceases, the substrate remains in an inserted mode, persistently associated with the proteasome, or escapes instead. To address these questions, we expressed ODC::GAr30 in a yeast strain in which proteasomes had been equipped with an affinity tag to facilitate their rapid isolation. Proteasomes were fractionated from substrates and their degradation intermediates in buffer containing ATP. We first affinity purified proteasomes with a protein A tag fused to the β4 (Pre1) subunit of the 20S proteolytic core. Western blotting for 20S proteins and for the 19S component Rpt5 demonstrated that both 20S and 19S complexes were recovered with similar efficiencies in the matrix‐bound fraction—absolute recoveries were 5–10% (Figure 7). This shows that the buffer composition and fractionation conditions used effectively maintained an association between the 20S core and 19S regulatory complex. Neither 20S proteins nor Rpt5 were associated with the affinity matrix in a control strain lacking the affinity tag. To increase the sensitivity of detection for trapped intermediates, a 20‐fold excess of the affinity pulldown fractions was analyzed compared to total and supernatant fractions.

Figure 7.

Degradation intermediates do not cofractionate with proteasomes. Lysates were prepared from cells expressing ODC::GAr30 and bearing either no proteasome affinity tag or a Protein A (ProA) tag associated with either the Pre1 subunit of the 20S core particle or the Rpn11 subunit of the 19S regulatory complex. Following affinity purification, fractions designated total (T), supernatant (S) and matrix‐associated (M) were subjected to SDS–PAGE and immunoblotting; M fractions were overloaded 20‐fold compared to the T and S fractions to facilitate visualization of affinity‐fractionated, proteasome‐associated proteins. Immunoblotting was performed with antibodies specific for the FLAG epitope in the ODC::GAr30 substrate (solid arrow) and intermediates (dashed arrow), Rpt5, 20S or Sem1 proteins. ProA‐tagged Pre1 and Rpn11 proteasome subunits were detected by the direct interaction of ProA with IgG. Sem1 antibody specificity was confirmed using extracts from a sem1Δ strain as negative control; the additional Sem1‐specific bands of lower apparent molecular weight cofractionating with the proteasome were not further characterized.

We observed no intermediates of degradation (nor parent substrate) associated with the proteasomes bound to the affinity matrix via the Pre1 tag (Figure 7, Pre1‐ProA tag). ODC::GAr30 or its processed intermediates were visualized only in the total cell lysate or in the supernatant fraction following the pulldown or in the control nontagged pulldown. We also performed similar experiments using a protein A tag fused to the Rpn11 (Figure 7, Rpn11‐ProA tag) or Rpn1 (data not shown) subunits of the 19S complex, with similar results. Sem1 is a protein that has been recently identified as a component of the proteasome 19S complex (Funakoshi et al, 2004; Sone et al, 2004). However, the bulk of Sem1 is found in low molecular weight fractions (Funakoshi et al, 2004), and recombinant Sem1 readily becomes incorporated into proteasomes from Sem1‐deficient cells (Sone et al, 2004). These properties suggest Sem1 exists in exchangeable free and proteasome‐bound pools. We therefore tested Sem1 recovery as a positive control for proteasome association, using the identical pulldown and processing conditions that are used for ODC::GAr30. Sem1, in contrast to ODC::GAr30 and its derivatives, was readily visualized in pulldown fractions from Pre1‐ or Rpn11‐ProA‐tagged strains. We conclude, therefore, that ODC::GAr30 intermediates are not persistently trapped in the 20S proteasome. Gel filtration fractionation of ODC::GAr30 and its intermediates from untagged proteasomes supported the same conclusion (data not shown). Together, these results exclude persistent intermediate trapping by 19S, 20S or 26S forms of the proteasome. We also performed the converse experiment, using an internal FLAG epitope tag within ODC::GAr30 to perform the pulldown and Western blots to evaluate the recovery of proteasome proteins in association with ODC::GAr30 or its derivatives. None were detected (data not shown).

Complete degradation of a mouse ODC molecule expressed in S. cerevisiae takes about 10 min (Hoyt et al, 2003). In the pulldown experiments, approximately 45 min elapsed between cell lysis and completion of fractionation. We therefore cannot infer whether exit takes place at a rate comparable to the normal process of complete degradation or is more rapid.

Discussion

We have previously proposed (Zhang and Coffino, 2004), and here test, the following simple model: Two elements of the 19S regulatory particle are postulated to function coordinately. The first is a site impelling translocation, the ATPase ring, which binds the substrate and advances it stepwise toward the 20S core particle. The second is a site of constriction within the regulatory particle, possibly the entrance to a translocation channel through the ATPase ring. This postulated constriction impairs transit of a folded domain. The ATPase ring produces energy that performs work on the substrate, some of which is used to unravel the folded domain. The mode of energy transmission may be as simple as a lever arm motion that applies a friction drive to the polypeptide chain (Hinnerwisch et al, 2005). In the normal case, energy supply continuously exceeds need, and the substrate polypeptide continuously advances. If this balance transiently swings in the other direction, with energy need exceeding supply, insertion halts. If the substrate has penetrated sufficiently far into the core particle, proteolysis is limited by the failure to advance further. The position at which the substrate is cut reflects the furthest extent of penetration into the proteolytic chamber. The remaining substrate is released, and, if proteolysis has removed the degradation tag, partial proteolytic products persist as stable degradation intermediates.

Insertion can be stopped by a modestly stable folded structure if the capacity of the ATPase ring to transmit energy becomes impaired. The arrival of a GAr at the ATPase site is postulated to produce such impairment. This may happen because the tract of glycines and alanines (the two smallest amino acids) presents a relatively featureless face to an ATPase‐associated friction drive apparatus that normally advances the chain. This conjecture is consistent with the known substrate interaction properties of bacterial ATPases. ClpB, in collaboration with the DnaK chaperone system, solubilizes and refolds aggregated proteins. Its binding site (near the axial pore of the ATPase hexamer) prefers substrates with aromatic and basic residues (Schlieker et al, 2004). The HslU ATPase of the HslUV bacterial protease has a similar preference for aromatic and positively charged amino acids at substrate residue positions important for recognition (Burton et al, 2005).

Here, we tested this model in several ways. First, we altered the stability of an ODC or DHFR folded domain by introducing mutations expected to cause destabilization or (for DHFR) by adding a stabilizing ligand. Within ODC::GAr30, deletions were designed to destroy interactions that stabilize the β‐sheet domain proximal to the GAr and the C37 degradation tag. Using DHFR constructs, we pursued a similar strategy and obtained a similar result: mutations directed toward reducing stability reduced intermediate production. DHFR also provided the opportunity to examine the effect of domain stabilization on intermediate production. The stabilizing ligand methotrexate markedly increased the appearance of intermediates. These findings are consistent with a model of protease function described by Matouschek and co‐workers (Lee et al, 2001; Prakash et al, 2004) who showed that the proteasome unfolds and degrades substrates sequentially from an unstructured initiation site.

Alternative mechanisms through which a GAr might produce degradation intermediates are less tenable. A GAr may enhance endoproteolytic cleavage by the proteasome (Liu et al, 2002a) and thus leave the remainder of the GAr‐containing fusion protein unprocessed. Such a model seems unlikely, however, given that GAr function depends on the stability of structural domains spatially distinct from the GAr itself. Our data also exclude mechanisms whereby substrate processing initiates, but ceases when components of the proteasome encounter a GAr and become immobilized. Such a sabotage mechanism implies the stable association of intermediate with proteasome, contrary to our data showing that their association is transient.

In addition to the anticipated quantitative effect of changing domain stability, an unanticipated change in the pattern of intermediates was also seen. ODC::GAr30 produces three intermediates, missing about 20, 30 and (more rarely) 45 residues from its C‐terminus. This suggests that the insertion process overcomes a series of energy barriers, a process in which progressive clipping marks progressive subdomain unfolding events. The mutations we have investigated uniformly reduced the accumulation of the intermediate missing ∼20 residues, but had little effect on the others (Figure 2A). This is consistent with selective reduction of the most proximal energy barrier, while more distal ones are little disturbed, a result that might have been expected from the strategy of mutant design.

In a further independent test of the model, we varied the spacing between the folded domain of ODC (assumed here to end at the terminal β18 strand (Kern et al, 1999a)) and the proximal boundary of a GAr of length 30. A distance of 14, 21 or 27 amino acids between these elements led to efficient stalling of insertion. Increasing this distance to 33 amino acids strongly reduced stalling and an increase to 39 further reduced it. The distance between successive α‐carbon atoms of an extended polypeptide chain is ∼3 Å, and so these spacings of 14, 21 or 27 residues are equivalent to about 40, 60 and 80 Å. Our data imply that the minimum separation that supports functional interaction is 40 Å or less. The result is consistent with the possibility that the two functional sites in fact colocalize. This is credible: if the substrate threads through an axial pore of an ATPase annulus, the pore could be the site of restriction. Separating the two elements by 100 Å (33 residues) or more strongly impairs their capacity to collaborate. The full 19S regulatory complex seen by electron microscopy can be encompassed within a sphere about 150 Å in diameter (Walz et al, 1998). Failure of interaction at distances approaching the dimensions of the 19S complex is consistent with the requirement that both sites lie within the complex.

By using pulse‐chase data, we calculated the fraction of processed substrates that partition to intermediates, PI, and used this as the primary analytic tool for examining halting of proteolysis. PI has a simple physical interpretation: it corresponds to the probability that degradation, once initiated, will cease. This can also be expressed as a ratio of rate constants, whereby a substrate that is undergoing degradation will alternatively undergo further degradation at a rate kD or stop degradation (and likely exit the proteasome) at a rate kI. Then, PI=kI/(kI+kD). The proteasome is generally assumed to operate on most ordinary substrates with a PI that is close to zero. This is based on the perception that degradation appears to be completed once it is begun. The conclusion, however, rests largely on our failure to observe intermediates. Two considerations limit our capacity for detecting the presence or production of intermediates. The first of these is technical: SDS–PAGE, the most commonly used means to follow proteolysis, readily detects abundant protein species of discrete size, but not lower abundance populations of heterogeneous size. Therefore, only intermediates consisting of discrete molecular species, like those described here, will be observed readily. Second, degradation intermediates may include misfolded elements; these proteins may be subjected to a second round of elimination by the ubiquitin–proteasome system, thereby escaping detection. The value of PI for typical substrates therefore remains an open question.

Understanding the import machinery of the proteasome can benefit from considering better‐understood models of protein translocation, such as membrane transport systems or bacterial ATP‐dependent proteases. In principle, two mechanisms can be utilized, either a power stroke or Brownian ratchet. By analogy with other systems in which ATPases are used (Singleton et al, 2000; Kenniston et al, 2003; DeLaBarre and Brunger, 2005; Hinnerwisch et al, 2005), a power stroke model is the more likely. Consideration of the power stroke mechanism may help to illuminate why the native GAr is so long: a long GAr assures probabilistic iterative failure. If the bacterial ClpXP protease is presented with a multidomain substrate, it releases the substrate at an appreciable frequency, yielding stable intermediates that are cut between domain boundaries. A titin domain interrupts processing by ClpXP about one time in three (Kenniston et al, 2005). This significant failure rate (PI∼0.3) does not require mediation by a recognizable GAr‐like sequence. The proteasome may be a more powerful machine, so that a folded domain must be supplemented by a slippery sequence to cause a comparable rate of failure. Alternatively, the titin domain may have more stringent unfolding requirements than those we have tested in the proteasome. As discussed above, it is also possible that proteasomes have a PI that significantly exceeds zero for native folded proteins.

The Epstein–Barr virus has evolved a tract of 60–300 residues (in different isolates), which are exclusively glycines and alanines. A sequence of such length did not likely evolve to confer a global folding property, nor is it easy to imagine a specific proteasome site that requires for interaction (or impairment of interaction) the full span of a sequence of such length and invariant composition. Much more likely is a requirement for iterative impairment of many such interactions, performed as the polypeptide chain advances through the proteasome. Each completed power stroke thrust would, in this view, displace the next such interaction to a new position along the chain, a displacement equal to the power stroke length. The estimated power stroke length of other AAA ATPase motors, such as bacteriophage T7 helicase, is about 20 Å (Singleton et al, 2000), which corresponds to an extended peptide of about seven residues. Suppose that failure of a single iteration of displacement causes substrate release. A GAr of 300 residues would suffer that risk more than 40 times, converting, for example, a 90% success rate per iteration to a success rate of 1% for traversing the full sequence (PI∼0.99). When suitably juxtaposed to a folded ODC domain, a GAr as short as seven residues can cause failure of processive degradation about one time in three (PI∼0.3). A long GAr would require neither exquisite positioning nor unidirectional chain processing, and would assure that processivity is interrupted, even when the failure rate per individual iteration is low.

Our data support the conclusion that degradation intermediates are released from the proteasome, a finding also true of ClpXP (Kenniston et al, 2005). Intermediate release has several implications. First, it assures that the proteasome system will not be functionally depleted by persistent futile interactions. Second, it is consistent with a few exceptional cases in which proteasomes are used for processing rather than complete degradation (Rape and Jentsch, 2002). Converting certain precursor proteins to mature transcription factors requires them to be partially proteolyzed by proteasomes and then liberated. One such transcription factor is NFκB, the p50 subunit of which is generated from a p105 precursor by proteasomal truncation of its C‐terminal end. It is of interest that producing the p50 intermediate requires both a structured domain within the N‐terminal end of the p105 precursor and a glycine‐rich tract near the processing junction. Removing the glycine‐rich region (Orian et al, 1999) or impairing the stability of the folded domain (Lee et al, 2001) or increasing spacing between the two (Tian et al, 2005) reduces halting, biasing degradation toward completion rather than production of the p50 protein. Although the functional interactions of these elements obviously resemble those within the more artificial substrates we have studied, it remains to be determined whether the glycine‐rich region of the p105 precursor confers functional properties similar to a GAr.

Materials and methods

Yeast strains

All GAr‐containing fusion proteins were expressed from p414ADH‐based vectors (Mumberg et al, 1995) in S. cerevisiae strain MHY501 (MATα his3‐Δ200 leu2‐3,112lys2‐801 trp1‐1 ura3‐52) (Chen et al, 2002), unless otherwise noted. All yeast strains were maintained and manipulated using standard procedures (Guthrie and Fink, 1991). Strains MHY85 and MHY94, bearing C‐terminal Staphylococcus aureus protein A tags on proteasome subunits Pre1 and Rpn11, respectively, were generated by the integration of the tag sequence and the Kluyveromyces lactis TRP1 gene at their chromosomal loci in strain MHY501 using described methods (Knop et al, 1999). For this transformation, a DNA fragment encoding the TEV protease cleavage site, two protein A IgG‐binding domains, and the K. lactis TRP1 gene, and flanked by 50 bp PRE1 or RPN11 chromosomal sequences to direct integration, was amplified from the TAP tag vector pBS1479 (Puig et al, 2001) by PCR.

Plasmid construction

Details of the construction of plasmids used in this study are provided in supplementary data.

Metabolic labeling, immunoprecipitations and substrate‐trapping assay

Pulse‐chase analysis was performed as described previously (Hoyt et al, 2003). GAr30‐containing fusion proteins were immunoprecipitated using FLAG epitope tags located either within the GAr30 module itself, or at the N‐terminus of the fusion protein, using anti‐FLAG M2 affinity gel (Sigma).

Yeast strains bearing a Protein A affinity tag appended to the C‐terminus of the Pre1 (strain MHY85) or Rpn11 (MHY94) proteasome subunits were transformed with a URA3‐marked low‐copy vector (p416ADH, (Mumberg et al, 1995)) expressing ODC::GAr30 (pJ67). Strain MHY501 was used as a negative control. For cofractionation experiments, 40 ml cultures of the transformants were harvested in late exponential growth phase (OD600=∼1.5), washed with water, and resuspended in 300 μl lysis buffer (50 mM Tris–HCl pH 7.5, 150 mM NaCl, 10% glycerol, 2 mM ATP and 5 mM MgCl2). The cells were mechanically disrupted in 2‐ml screw‐top microcentrifuge tubes with 200 μl glass beads in a Mini‐beadbeater (Biospec Products) for four pulses of 30 s duration with 30 s cooling on ice between pulses. The lysates were cleared by centrifugation (18 000 g, 15 min) and 750 μg of total protein in 0.5 ml total volume was incubated with 5 μl of either rabbit IgG agarose or anti‐Flag M2 agarose at 4°C for 20 min. The affinity‐purified proteasome complexes were washed four times with 500 μl lysis buffer, resuspended in Laemmli SDS–PAGE loading buffer, and fractionated by SDS–PAGE. The fractioned proteins were transferred to nitrocellulose filters, and the filters were blocked in Tris‐buffered saline with 1% Triton X‐100 (TBS‐T) including 5% powdered low‐fat milk, and rinsed briefly with TBS‐T. Proteins were detected with mouse anti‐FLAG M2 antibody (Sigma, 1:2000 dilution) and TrueBlot anti‐mouse Ig‐horseradish peroxidase conjugates (eBioscience, 1:4000), or rabbit anti‐Rpt5 (BioMol, 1:25000), anti‐20S proteasome antibodies (gift of T Tamura, 1:2000) or anti‐yeast Sem1 antibodies (gift of M Funakoshi, 1:5000) and TrueBlot anti‐rabbit Ig‐horseradish peroxidase conjugates (1:4000). Immunoblots were developed using the Amersham ECLplus detection kit and protocol.

Data quantification

Signals from autoradiographic films were quantified using TotalLab software (Nonlinear Dynamics). Data for the calculation of PI (see Discussion) was obtained from pulse‐chase experiments with 60 min chase periods. The term kI was calculated as the accumulation of intermediates between 0 and 60 min (Int60–Int0) divided by the parent molecules lost during the same interval (P0P60). Similarly, the term kD was calculated as (P0P60)+(Int0–Int60)/(P0+Int0). PI=kI/(kI+kD).

Supplementary data

Supplementary data are available at The EMBO Journal Online.

Supplementary Information

Supplementary Materials and Methods [emboj7601058-sup-0001.pdf]

Acknowledgements

We thank Tomohiro Tamura, Daniel Finley, Kiran Madura and Minoru Funakoshi for reagents, and Fred Cohen and George Oster for helpful comments on the manuscript. This work was supported by NIH grants GM45335 and GM074760 to PC.

References