A unified mechanism for proteolysis and autocatalytic activation in the 20S proteasome

Biogenesis of the 20S proteasome is tightly regulated. The N-terminal propeptides protecting the active-site threonines are autocatalytically released only on completion of assembly. However, the trigger for the self-activation and the reason for the strict conservation of threonine as the active site nucleophile remain enigmatic. Here we use mutagenesis, X-ray crystallography and biochemical assays to suggest that Lys33 initiates nucleophilic attack of the propeptide by deprotonating the Thr1 hydroxyl group and that both residues together with Asp17 are part of a catalytic triad. Substitution of Thr1 by Cys disrupts the interaction with Lys33 and inactivates the proteasome. Although a Thr1Ser mutant is active, it is less efficient compared with wild type because of the unfavourable orientation of Ser1 towards incoming substrates. This work provides insights into the basic mechanism of proteolysis and propeptide autolysis, as well as the evolutionary pressures that drove the proteasome to become a threonine protease.

T he 20S proteasome core particle (CP) is the key non-lysosomal protease of eukaryotic cells. Its seven different a and seven different b subunits assemble into four heptameric rings that are stacked on each other to form a hollow cylinder. While the inactive a subunits build the two outer rings, the b subunits form the inner rings. Only three out of the seven different b subunits, namely b1, b2 and b5, bear N-terminal proteolytic active centres, and before CP maturation these are protected by propeptides [1][2][3] . In the last stage of CP biogenesis, the prosegments are autocatalytically removed through nucleophilic attack by the active site residue Thr1 on the preceding peptide bond involving Gly(-1) 4,5 . Release of the propeptides creates a functionally active CP that cleaves proteins into short peptides.
Although the chemical nature of the substrate-binding channel and hence substrate preferences are unique to each of the distinct active b subunits 6,7 , all active sites employ an identical reaction mechanism to hydrolyse peptide bonds 2 . Nucleophilic attack of Thr1O g on the carbonyl carbon atom of the scissile peptide bond creates a first cleavage product and a covalent acyl-enzyme intermediate. Hydrolysis of this complex by the addition of a nucleophilic water molecule regenerates the enzyme and releases the second peptide fragment 8,9 . The proteasome belongs to the family of N-terminal nucleophilic (Ntn) hydrolases 10 , and the free N-terminal amine group of Thr1 was proposed to deprotonate the Thr1 hydroxyl group to generate a nucleophilic Thr1O g for peptide-bond cleavage 2,9,11 . This mechanism, however, cannot explain autocatalytic precursor processing because in the immature active sites, Thr1N is part of the peptide bond with Gly(-1), the bond that needs to be hydrolysed. An alternative candidate for deprotonating the Thr1 hydroxyl group is the side chain of Lys33 as it is within hydrogen-bonding distance to Thr1OH (2.7 Å). In principle it could function as the general base during both autocatalytic removal of the propeptide and protein substrate cleavage. Here we provide experimental evidences for this distinct view of the proteasome active-site mechanism. Data from biochemical and structural analyses of proteasome variants with mutations in the b5 propeptide and the active site strongly support the model and deliver novel insights into the structural constraints required for the autocatalytic activation of the proteasome. Furthermore, we determine the advantages of Thr over Cys or Ser as the active-site nucleophile using X-ray crystallography together with activity and inhibition assays.

Results
Inactivation of proteasome subunits by T1A mutations. Proteasome-mediated degradation of cell-cycle regulators and potentially toxic misfolded proteins is required for the viability of eukaryotic cells 8 . Inactivation of the active site Thr1 by mutation to Ala has been used to study substrate specificity and the hierarchy of the proteasome active sites 1,4,[12][13][14][15] . Yeast strains carrying the single mutations b1-T1A or b2-T1A, or both, are viable, even though one or two of the three distinct catalytic b subunits are disabled and carry remnants of their N-terminal propeptides 4 (Table 1). These results indicate that the b1 and b2 proteolytic activities are not essential for cell survival. By contrast, the T1A mutation in subunit b5 has been reported to be lethal or nearly so 1,13 . Viability is restored if the b5-T1A subunit has its propeptide (pp) deleted but expressed separately in trans (b5-T1A pp trans), although substantial phenotypic impairment remains 1,15,16 (Table 1). Our present crystallographic analysis of the b5-T1A pp trans mutant demonstrates that the mutation per se does not structurally alter the catalytic active site and that the trans-expressed b5 propeptide is not bound in the b5 substrate-binding channel ( Supplementary Fig. 1a).
The extremely weak growth of the b5-T1A mutant pp cis described by Chen and Hochstrasser 1 compared with the inviability reported by Heinemeyer et al. 13 prompted us to analyse this discrepancy. Sequencing of the plasmids, testing them in both published yeast strain backgrounds and site-directed mutagenesis revealed that the b5-T1A mutant pp cis is viable, but suffers from a marked growth defect that requires extended incubation of 4-5 days for initial colony formation (Table 1 and Supplementary Methods). We also identified an additional point mutation K81R in subunit b5 that was present in the allele used in ref. 1. This single amino-acid exchange is located at the interface of the subunits a4, b4 and b5 (Supplementary Fig. 1b) and might weakly promote CP assembly by enhancing intersubunit contacts. The slightly better growth of the b5-T1A-K81R mutant allowed us to solve the crystal structure of a yeast proteasome (yCP) with the b5-T1A mutation, which is discussed in the following section (for details see Supplementary Note 1).
Propeptide conformation and triggering of autolysis. In the final steps of proteasome biogenesis, the propeptides are autocatalytically cleaved from the mature b-subunit domains 1 . For subunit b1, this process was previously inferred to require that the propeptide residue at position (-2) of the subunit precursor occupies the S1 specificity pocket of the substratebinding channel formed by amino acid 45 (for details see Supplementary Note 2) 5 . Furthermore, it was observed that the prosegment forms an antiparallel b-sheet in the active site, and that Gly(-1) adopts a g-turn conformation, which by definition is characterized by a hydrogen bond between Leu(-2)O and Thr1NH (ref. 5). Here we again analysed the b1-T1A mutant crystallographically but in addition determined the structures of the b2-T1A single and b1-T1A-b2-T1A double mutants (Protein Data Bank (PDB) entry codes are provided in Supplementary Table 1). In subunit b1, we found that Gly(-1) indeed forms a sharp turn, which relaxes on prosegment cleavage ( Fig. 1a and Supplementary Fig. 2a). However, the g-turn conformation and the associated hydrogen bond initially proposed is for geometric and chemical reasons inappropriate and would not perfectly position the carbonyl carbon atom of Gly(-1) for nucleophilic attack by Thr1. Regarding the b2 propeptide, Thr(-2) occupies the S1 pocket but is less deeply anchored compared with Leu(-2) in b1, which might be due to the rather large b2-S1 pocket created by Gly45. Thr(-2) positions Gly(-1)O via hydrogen bonding (B2.8 Å) in a perfect trajectory for the nucleophilic attack by Thr1O g (Fig. 1b and Supplementary  Fig. 2b). Next, we examined the position of the b5 propeptide in the b5-T1A-K81R mutant. Surprisingly, Gly(-1) is completely extended and forces the histidine side chain at position (-2) to occupy the S2 instead of the S1 pocket, thereby disrupting the antiparallel b-sheet. Nonetheless, the carbonyl carbon of Gly(-1) would be ideally placed for nucleophilic attack by Thr1O g (Fig. 1c and Supplementary Fig. 2c,d). As the K81R mutation is located far from the active site (Thr1C a -Arg81C a : 24 Å), any influence on propeptide conformation can be excluded. Instead, the plasticity of the b5 S1 pocket caused by the rotational flexibility of Met45 might prevent stable accommodation of His(-2) in the S1 site and thus also promote its immediate release after autolysis.
Processing of b-subunit precursors requires deprotonation of Thr1OH; however, the general base initiating autolysis is unknown. Remarkably, eukaryotic proteasomal b5 subunits bear a His residue in position (-2) of the propeptide ( Supplementary  Fig. 3a). As histidine commonly functions as a proton shuttle in the catalytic triads of serine proteases 17 , we investigated the role of His(-2) in processing of the b5 propeptide by exchanging it for Asn, Lys, Phe and Ala. All yeast mutants were viable at 30°C, but suffered from growth defects at 37°C with the H(-2)N and H(-2)F mutants being most affected (Supplementary Fig. 3b and Table 1). In agreement, the chymotrypsin-like (ChT-L) activity of H(-2)N and H(-2)F mutant yCPs was impaired in situ and in vitro ( Supplementary Fig. 3c). Structural analyses revealed that the propeptides of all mutant yCPs shared residual 2F O -F C electron densities. Gly(-1) and Phe/Lys(-2) were visualized at low occupancy, while Ala/Asn(-2) could not be assigned. This observation indicates a mixture of processed and unprocessed b5 subunits and partially impaired autolysis 18 , thereby excluding any essential role of residue (-2) as the general base.
Next, we examined the effect of residue (-2) on the orientation of the propeptide by creating mutants that combine the T1A (K81R) mutation(s) with H(-2)L, H(-2)T or H(-2)A substitutions. Leu(-2) is encoded in the yeast b1 subunit precursor ( Supplementary Fig. 3a); Thr(-2) is generally part of Only the residues (-5) to (-1) of the b1-T1A propeptide are displayed. The major determinant of the S1 specificity pocket, residue 45, is depicted. Note the tight conformation of Gly(-1) and Ala1 before propeptide removal (G(-1) turn; cyan double arrow) compared with the relaxed, processed WT active-site Thr1 (red double arrow). The black arrow indicates the attack of Thr1O g onto the carbonyl carbon atom of Gly(-1). (b) Structural superposition of the b1-T1A propeptide and the b2-T1A propeptide highlights subtle differences in their conformations, but illustrates that Ala1 and Gly(-1) match well. Thr(-2)OH is hydrogen-bonded to Gly(-1)O (B2.8 Å; black dashed line). The major determinant of the S1 specificity pocket, residue 45, is depicted. (c) Structural superposition of the b1-T1A, the b2-T1A and the b5-T1A-K81R propeptide remnants depict their differences in conformation. While residue (-2) of the b1 and b2 prosegments fit the S1 pocket, His(-2) of the b5 propeptide occupies the S2 pocket. Nonetheless, in all mutants the carbonyl carbon atom of Gly(-1) is ideally placed for the nucleophilic attack by Thr1O g . The hydrogen bond between Thr(-2)OH and Gly(-1)O (B2.8 Å) is indicated by a black dashed line.  Supplementary Fig. 3a); and Ala(-2) was expected to fit the b5-S1 pocket without inducing conformational changes of Met45, allowing it to accommodate 'b1-like' propeptide positioning. As expected from b5-T1A mutants, the yeasts show severe growth phenotypes, with minor variations (Supplementary Fig. 4a and Table 1). We determined crystal structures of the b5-H(-2)L-T1A, b5-H(-2)T-T1A and the b5-H(-2)A-T1A-K81R mutants (Supplementary Table 1). For the b5-H(-2)A-T1A-K81R variant, only the residues Gly(-1) and Ala(-2) could be visualized, indicating that Ala(-2) leads to insufficient stabilization of the propeptide in the substratebinding channel ( Supplementary Fig. 4d). By contrast, the prosegments of the b5-H(-2)L-T1A and the b5-H(-2)T-T1A mutants were significantly better resolved in the 2F O -F C electrondensity maps yet not at full occupancy ( Supplementary Fig. 4b,c and Supplementary Table 1), suggesting that the natural propeptide bearing His(-2) is most favourable. Nevertheless, both Leu(-2) and Thr(-2) were found to occupy the S1 specificity pocket formed by Met45 (Fig. 2a,b and Supplementary Fig. 4f-h). This result proves that the naturally occurring His(-2) of the b5 propeptide does not stably fit into the S1 site. Since Gly(-1) adopts the same position in both wild-type (WT) and mutant b5 propeptides, and since in all cases its carbonyl carbon is perfectly placed for nucleophilic attack by Thr1O g (Fig. 2b), we propose that neither binding of residue (-2) to the S1 pocket nor formation of the antiparallel b-sheet is essential for autolysis of the propeptide. Next, we determined the crystal structure of a chimeric yCP having the yeast b1-propeptide replaced by its b5 counterpart 18 . Although we observed fragments of 2F O -F C electron density in the b1 active site, the data were not interpretable. Bearing in mind that in contrast to Thr(-2) in b2, Leu(-2) in subunit b1 is not conserved among species ( Supplementary Fig. 3a), we created a b2-T(-2)V proteasome mutant. As proven by the b2-T1A crystal structures, Thr(-2) hydrogen bonds to Gly(-1)O. Although this interaction was not observed for the b5-H(-2)T-T1A mutant ( Fig. 2c and Supplementary Fig. 4c,i), exchange of Thr(-2) by Val in b2, a conservative mutation regarding size but drastic with respect to polarity, was found to inhibit maturation of this subunit ( Fig. 2d and Supplementary Fig. 4e,j). Notably, the 2F O -F C electron-density map displays a different orientation for the b2 propeptide than has been observed for the b2-T1A proteasome. In particular, Val(-2) is displaced from the S1 site and Gly(-1) is severely shifted (movement of the carbonyl oxygen atom of 3.8 Å), thereby preventing nucleophilic attack of Thr1 ( Fig. 2d and Supplementary Fig. 4j,k). These results further confirm that correct positioning of the active-site residues and Gly(-1) is decisive for the maturation of the proteasome.
The active site of the proteasome. Proton shuttling from the proteasomal active site Thr1OH to Thr1NH 2 via a nucleophilic water molecule was suggested to initiate peptide-bond hydrolysis 2,9,10 . However, in the immature particle Thr1NH 2 is blocked by the propeptide and cannot activate Thr1O g . Instead, Lys33NH 2 , which is in hydrogen-bonding distance to Thr1O g (2.7 Å) in all catalytically active b subunits (Fig. 3a,b) 2,9 , was proposed to serve as the proton acceptor 19 . Owing to its likely protonation at neutral pH, however, it was assumed not to act as the general base 2,5,9 . A proposed catalytic tetrad model involving Thr1OH, Thr1NH 2 , Lys33NH 2 and Asp17O d , as well as a nucleophilic water molecule as the proton shuttle appeared to accommodate all possible views of the proteasomal active site 8,9,20 . Twenty years later, with a plethora of yCP X-ray structures in hand, we decided to re-analyse the active site of the proteasome and to resolve the uncertainty regarding the nature of the general base. Mutation of b5-Lys33 to Ala causes a strongly deleterious phenotype, and previous structural and biochemical analyses confirmed that this is caused by failure of propeptide cleavage, and consequently, lack of ChT-L activity 1,4,13 (Fig. 4a, Supplementary Fig. 3b and Table 1; for details see Supplementary Note 1). The phenotype of the b5-K33A mutant was however less pronounced than for the b5-T1A-K81R yeast (Fig. 4a). This discrepancy in growth was traced to an additional point mutation L(-49)S in the b5-propeptide of the b5-K33A mutant (see also Supplementary Note 1). Structural comparison of the b5-L(-49)S-K33A and b5-T1A-K81R active sites revealed that mutation of Lys33 to Ala creates a cavity that is filled with Thr1 and the remnant propeptide. This structural alteration destroys active-site integrity and abolishes catalytic activity of the b5 active site 4 ( Supplementary Fig. 5a). Additional proof for the key function of Lys33 was obtained from the b5-K33A mutant, with the propeptide expressed separately from the main subunit (pp trans) 15 . The Thr1 N terminus of this mutant is not blocked by the propeptide, yet its catalytic activity is reduced by B83% ( Supplementary Fig. 6b). Consistent with this, the crystal structure of the b5-K33A pp trans mutant in complex with carfilzomib only showed partial occupancy of the ligand at the b5 active sites (Supplementary Fig. 5b and Supplementary Table 1). Since no acetylation of the Thr1 N terminus was observed for the b5-K33A pp trans apo crystal structure 4,16 , the reduced reactivity towards substrates and inhibitors indicates that Lys33NH 2 , rather than Thr1NH 2 , deprotonates and activates Thr1OH. Furthermore, the crystal structure of the b5-K33A pp trans mutant without inhibitor revealed that Thr1O g strongly coordinates a well-defined water molecule (B2 Å; Fig. 3c and Supplementary Fig. 5c,d). This water hydrogen bonds also to Arg19O (B3.0 Å) and Asp17O d (B3.0 Å), and thereby presumably enables residual activity of the mutant. Remarkably, the solvent molecule occupies the position normally taken by Lys33NH 2 in the WT proteasome structure (Fig. 3c)  bond to Asp17, thereby inactivating the b5 active site 2,4 ( Supplementary Fig. 5e).
The conservative mutation of Asp17 to Asn in subunit b5 of the yCP also provokes a severe growth defect (Supplementary Note 1, Supplementary Fig. 6a and Table 1). Notably, only with the additional point mutation L(-49)S present in the b5 propeptide could we purify a small amount of the b5-D17N mutant yCP. As determined by crystallographic analysis, this mutant b5 subunit was partially processed (Table 1) but displayed impaired reactivity towards the proteasome inhibitor carfilzomib compared with the subunits b1 and b2, and with WT b5 (Supplementary Fig. 7a). In contrast to the cis-construct, expression of the b5 propeptide in trans allowed straightforward isolation and crystallization of the D17N mutant proteasome. The ChT-L activity of the b5-D17N pp in trans CP towards the canonical b5 model substrates N-succinyl-Leu-Leu-Val-Tyr-7-amino-4-methylcoumarin (Suc-LLVY-AMC) and carboxybenzyl-Gly-Gly-Leu-para-nitroanilide (Z-GGL-pNA) was severely reduced ( Supplementary Fig. 6b), confirming that Asp17 is of fundamental importance for the catalytic activity of the mature proteasome. Even though the b5-D17N pp trans yCP crystal structure appeared identical to the WT yCP ( Supplementary  Fig. 7b), the co-crystal structure with the a 0 , b 0 epoxyketone inhibitor carfilzomib visualized only partial occupancy of the ligand in the b5 active site (Supplementary Fig. 7a). This observation is consistent with a strongly reduced reactivity of b5-Thr1 and the crystal structure of the b5-D17N pp cis mutant in complex with carfilzomib. Autolysis and residual catalytic activity of the b5-D17N mutants may originate from the carbonyl group of Asn17, which albeit to a lower degree still can polarize Lys33 for the activation of Thr1. In agreement, an E17A mutant in the proteasomal b-subunit of the archaeon Thermoplasma acidophilum prevents autolysis and catalysis 21 . Strikingly, although the X-ray data on the b5-D17N mutant with the propeptide expressed in cis and in trans looked similar, there was a pronounced difference in their growth phenotypes observed (Supplementary Fig. 6a and Supplementary Fig. 7b).
On the basis of these results, we propose that CPs from all domains of life use a catalytic triad consisting of Thr1, Lys33 and Asp/Glu17 for both autocatalytic precursor processing and proteolysis (Fig. 3d). This model is also consistent with the fact that no defined water molecule is observed in the mature WT proteasomal active site that could shuttle the proton from Thr1O g to Thr1NH 2 .
To explore this active-site model further, we exchanged the conserved Asp166 residue for Asn in the yeast b5 subunit. Asp166O d is hydrogen-bonded to Thr1NH 2 via Ser129OH and Ser169OH, and therefore was proposed to be involved in catalysis 2 . The b5-D166N pp cis yeast mutant is significantly impaired in growth and its ChT-L activity is drastically reduced (Supplementary Fig. 6a,b and Table 1). X-ray data on the b5-D166N mutant indicate that the b5 propeptide is hydrolysed, but due to reorientation of Ser129OH, the interaction with Asn166O d is disrupted (Supplementary Fig. 8a). Instead, a water molecule is bound to Ser129OH and Thr1NH 2 ( Supplementary  Fig. 8b), which may enable precursor processing. The hydrogen bonds involving Ser169OH are intact and may account for residual substrate turnover. Soaking the b5-D166N crystals with carfilzomib and MG132 resulted in covalent modification of Thr1 at high occupancy ( Supplementary Fig. 8c). In the carfilzomib complex structure, Thr1O g and Thr1N incorporate into a morpholine ring structure and Ser129 adopts its WT-like orientation. In the MG132-bound state, Thr1N is unmodified, and we again observe that Ser129 is hydrogen-bonded to a water molecule instead of Asn166. Whereas Asn can to some degree replace Asp166 due to its carbonyl group in the side chain, Ala at this position was found to prevent both autolysis and catalysis 21 . These results suggest that Asp166 and Ser129 function as a proton shuttle and affect the protonation state of Thr1N during autolysis and catalysis.
Substitution of the active-site Thr1 by Cys. Mutation of Thr1 to Cys inactivates the 20S proteasome from the archaeon T. acidophilum 21 . In yeast, this mutation causes a strong growth defect ( Fig. 4a and Table 1), although the propeptide is hydrolysed, as shown here by its X-ray structure. In one of the two b5 subunits, however, we found the cleaved propeptide still bound in the substrate-binding channel (Fig. 4c). His(-2) occupies the S2 pocket like observed for the b5-T1A-K81R mutant, but in contrast to the latter, the propeptide in the T1C mutant adopts an antiparallel b-sheet conformation as known from inhibitors like MG132 (Fig. 4c-e and Supplementary  Fig. 9b). On the basis of the phenotype of the T1C mutant and the propeptide remnant identified in its active site, we suppose that autolysis is retarded and may not have been completed before crystallization. Owing to the unequal positions of the two b5 subunits within the CP in the crystal lattice, maturation and propeptide displacement may occur at different timescales in the two subunits.
Despite propeptide hydrolysis, the b5-T1C active site is catalytically inactive (Fig. 4b and Supplementary Fig. 9a). In agreement, soaking crystals with the CP inhibitors bortezomib or carfilzomib modifies only the b1 and b2 active sites, while leaving the b5-T1C proteolytic centres unmodified even though they are only partially occupied by the cleaved propeptide remnant. Moreover, the structural data reveal that the thiol group of Cys1 is rotated by 74°with respect to the hydroxyl side chain of Thr1 ( Fig. 4f and Supplementary Fig. 9b). This presumably results from the larger radius of the sulfur atom compared with oxygen. Consequently, the hydrogen bond bridging the active-site nucleophile and Lys33 in WT CPs is broken with Cys1. Notably, the 2F O -F C electron-density map of the T1C mutant also indicates that Lys33NH 2 is disordered. Together, these observations suggest that efficient peptide-bond hydrolysis requires that Lys33NH 2 hydrogen bonds to the active site nucleophile.
The benefit of Thr over Ser as the active-site nucleophile. All proteasomes strictly employ threonine as the active-site residue instead of serine. To investigate the reason for this singularity, we analysed a b5-T1S mutant, which is viable but suffers from growth defects ( Fig. 4a and Table 1). Activity assays with the b5-specific substrate Suc-LLVY-AMC demonstrated that the ChT-L activity of the T1S mutant is reduced by 40-45% compared with WT proteasomes depending on the incubation temperature ( Fig. 4b and Supplementary Fig. 9c). By contrast, turnover of the substrate Z-GGL-pNA, used to monitor ChT-L activity in situ but in a less quantitative fashion, is not detectably impaired ( Supplementary Fig. 9a). Crystal structure analysis of the b5-T1S mutant confirmed precursor processing (Fig. 4g), and ligand-complex structures with bortezomib and carfilzomib unambiguously corroborated the reactivity of Ser1 (Fig. 5).
However, the apo crystal structure revealed that Ser1O g is turned away from the substrate-binding channel (Fig. 4g). Compared with Thr1O g in WT CP structures, Ser1O g is rotated by 60°. This renders it unavailable for direct nucleophilic attack onto incoming substrates and first requires its reorientation, which is expected to delay substrate turnover. Because both conformations of Ser1O g are hydrogen-bonded to Lys33NH 2 (Fig. 4h), the relay system is capable of hydrolysing peptide substrates, albeit at lower rates compared with Thr1. The active-site residue Thr1 is fixed in its position, as its methyl group is engaged in hydrophobic interactions with Thr3 and Ala46 (Fig. 4h). Consequently, the hydroxyl group of Thr1 requires no reorientation before substrate cleavage and is thus more catalytically efficient than Ser1. In agreement, at an elevated growing temperature of 37°C the T1S mutant is unable to grow (Fig. 4a). In vitro, the mutant proteasome is less susceptible to proteasome inhibition by bortezomib (3.7-fold) and carfilzomib (1.8-fold; Fig. 5). Nevertheless, inhibitor complex structures indicate identical binding modes compared with the WT yCP structures, with the same inhibitors 22,23 . Notably, the affinity of the tetrapeptide carfilzomib is less impaired, as it is better stabilized in the substrate-binding channel than the dipeptide bortezomib, which lacks a defined P3 site and has only a few interactions with the surrounding protein. Hence, the mean residence time of carfilzomib at the active site is prolonged and the probability to covalently react with Ser1 is increased. Considered together, these results provide a plausible explanation for the invariance of threonine as the active-site nucleophile in proteasomes in all three domains of life, as well as in proteasome-like proteases such as HslV (ref. 24).

Discussion
The 20S proteasome CP is the major non-lysosomal protease in eukaryotic cells, and its assembly is highly organized. The b-subunit propeptides, particularly that of b5, are key factors that help drive proper assembly of the CP complex 1 . In addition, they prevent irreversible inactivation of the Thr1 N terminus by N-acetylation 4,15,16 . By contrast, the prosegments of b subunits are dispensable for archaeal proteasome assembly, at least when heterologously expressed in Escherichia coli 25 . In eukaryotes, deletion of or failure to cleave the b1 and b2 propeptides is well tolerated 5,13-16 . However, removal of the b5 prosegment or any interference with its cleavage causes severe phenotypic defects 1,13 . These observations highlight the unique function and importance of the b5 propeptide as well as the b5 active site for maturation and function of the eukaryotic CP.
Here we have described the atomic structures of various b5-T1A mutants, which allowed for the first time visualization of the residual b5 propeptide. Depending on the (-2) residue we observed various propeptide conformations, but Gly(-1) is in all structures perfectly located for the nucleophilic attack by Thr1O g , although it does not adopt the tight turn observed for the prosegment of subunit b1. From these data we conclude that only the positioning of Gly(-1) and Thr1 as well as the integrity of the proteasomal active site are required for autolysis. In this regard, inappropriate N-acetylation of the Thr1 N terminus cannot be removed by Thr1O g due to the rotational freedom and flexibility of the acetyl group. The propeptide needs some anchoring in the substrate-binding channel to properly position Gly(-1), but this seems to be independent of the orientation of residue (-2).
Autolytic activation of the CP constitutes one of the final steps of proteasome biogenesis 26 , but the trigger for propeptide cleavage had remained enigmatic. On the basis of the numerous CP:ligand complexes solved during the past 18 years and in the current study, we provide a revised interpretation of proteasome active-site architecture. We propose a catalytic triad for the active site of the CP consisting of residues Thr1, Lys33 and Asp/Glu17, which are conserved among all proteolytically active eukaryotic, bacterial and archaeal proteasome subunits. Lys33NH 2 is expected to act as the proton acceptor during autocatalytic removal of the propeptides 19 , as well as during substrate proteolysis, while Asp17O d orients Lys33NH 2 and makes it more prone to protonation by raising its pK a (hydrogen bond distance: Lys33NH 3 þ -Asp17O d : 2.9 Å). Analogously to the proteasome, a Thr-Lys-Asp triad is also found in L-asparaginase 27 . Thus, specific protein surroundings can significantly alter the chemical properties of amino acids such as Lys to function as an acid-base catalyst 28 .
In this new view of the proteasomal active site, the positively charged Thr1NH 3 þ -terminus hydrogen bonds to the amide nitrogen of incoming peptide substrates and stabilizes as well as activates them for the endoproteolytic cleavage by Thr1O g (Fig. 3d). Consistent with this model, the positively charged Thr1 N terminus is engaged in hydrogen bonds with inhibitory compounds like fellutamide B (ref. 29), a-ketoamides 30 , homobelactosin C (ref. 31) and salinosporamide A (ref. 32). Furthermore, opening of the b-lactone compound omuralide 2 by Thr1 creates a C3-hydroxyl group, whose proton originates from Thr1NH 3 þ . The resulting uncharged Thr1NH 2 is hydrogenbridged to the C3-OH group. In agreement, acetylation of the Thr1 N terminus irreversibly blocks hydrolytic activity 15,16 , and binding of substrates is prevented for steric reasons. By acting as a proton donor during catalysis, the Thr1 N terminus may also favour cleavage of substrate peptide bonds (Fig. 3d). In all proteases, collapse of the tetrahedral transition state results in selective breakage of the substrate amide bond, while the covalent interaction between the substrate and the enzyme persists. Cleavage of the scissile peptide bond requires protonation of the emerging free amine, and in the proteasome, the Thr1 amine group is likely to assume this function. Analogously, Thr1NH 3 þ might promote the bivalent reaction mode of epoxyketone inhibitors by protonating the epoxide moiety to create a positively charged trivalent oxygen atom that is subsequently nucleophilically attacked by Thr1NH 2 .
During autolysis the Thr1 N terminus is engaged in a hydroxyoxazolidine ring intermediate (Fig. 3d), which is unstable and short-lived. Breakdown of this tetrahedral transition state releases the Thr1 N terminus that is protonated by aspartic acid 166 via Ser129OH to yield Thr1NH 3 þ . The residues Ser129 and Asp166 are expected to increase the pK a value of Thr1N, thereby favouring its charged state. Consistent with playing an essential role in proton shuttling, the mutation D166A prevents autolysis of the archaeal CP 21 and the exchange D166N impairs catalytic activity of the yeast CP about 60%. The mutation D166N lowers the pK a of Thr1N, which is thus more likely to exist in the uncharged deprotonated state (Thr1NH 2 ). This renders the N terminus less suitable to stabilize substrates and to protonate the first cleavage product during catalysis, although it favours its M45 S1 S1 V31 S129 T1 A49 (L1) G47 M45 V31 S129 T1 A49 (L1) G47 S1 S1 β5:bortezomib β5-T1S:bortezomib β5:bortezomib β5-T1S:bortezomib T1 S1 S1 P1 P2 S1 S1 ability to act as a nucleophile. This interpretation agrees with the strongly reduced catalytic activity of the b5-D166N mutant on the one hand, and the ability to react readily with carfilzomib on the other. Hence, the proteasome can be viewed as having a second triad that is essential for efficient proteolysis. While Lys33NH 2 and Asp17O d are required to deprotonate the Thr1 hydroxyl side chain, Ser129OH and Asp166OH serve to protonate the N-terminal amine group of Thr1. In accord with the proposed Thr1-Lys33-Asp17 catalytic triad, crystallographic data on the proteolytically inactive b5-T1C mutant demonstrate that the interaction of Lys33NH 2 and Cys1 is broken. Consequently, efficient substrate turnover or covalent modification by ligands is prevented. However, owing to Cys being a strong nucleophile, the propeptide can still be cleaved off over time. While only one single turnover is necessary for autolysis, continuous enzymatic activity is required for significant and detectable substrate hydrolysis. Notably, in the Ntn hydrolase penicillin acylase, substitution of the catalytic N-terminal Ser residue by Cys also inactivates the enzyme but still enables precursor processing 33 .
To investigate why the CP specifically employs threonine as its active-site residue, we used a b5-T1S mutant of the yCP and characterized it biochemically and structurally. Activity assays with the b5-T1S mutant revealed reduced turnover of Suc-LLVY-AMC. We also observed slightly lower affinity of the b5-T1S mutant yCP for the Food and Drug Administration-approved proteasome inhibitors bortezomib and carfilzomib. Structural analyses support these findings with the T1S mutant and provide an explanation for the strict use of Thr residues in proteasomes. Thr1 is well anchored in the active site by hydrophobic interactions of its C g methyl group with Ala46 (C b ), Lys33 (carbon side chain) and Thr3 (C g ). Notably, proteolytically active proteasome subunits from archaea, yeast and mammals, including constitutive, immuno-and thymoproteasome subunits, either encode Thr or Ile at position 3, indicating the importance of the C g for fixing the position of the nucleophilic Thr1. In contrast to Thr1, the hydroxyl group of Ser1 occupies the position of the Thr1 methyl side chain in the WT enzyme, which requires its reorientation relative to the substrate to allow cleavage (Fig. 4g,h). Notably, in the threonine aspartase Taspase1, mutation of the active-site Thr234 to Ser also places the side chain in the position of the methyl group of Thr234 in the WT, thereby reducing catalytic activity 34 . Similarly, although the serine mutant is active, threonine is more efficient in the context of the proteasome active site. The greater suitability of threonine for the proteasome active site, which has been noted in biochemical as well as in kinetic studies 35 , constitutes a likely reason for the conservation of the Thr1 residue in all proteasomes from bacteria to eukaryotes.

Methods
Yeast mutagenesis. Site-directed mutagenesis was performed by standard techniques using oligonucleotides listed in Supplementary Table 2. The pre2/doa3 (b5) mutant alleles in the centromeric, TRP1-or LEU2-marked shuttle vectors YCplac22 and pRS315, respectively, were verified by sequencing and subsequently introduced into the yeast strains MHY784 (ref. 1) or YWH20 (ref. 13), which express WT PRE2 from a URA3-marked plasmid. Counter-selection against the URA3 marker with 5-fluoroorotic acid yielded strains expressing only the mutant forms of b5.
The strain producing a processed b5-T1A variant and the b5 propeptide in trans is a derivative of YWH212 (ref. 15). It carries an additional deletion of the NAT1 gene to avoid N-acetylation of Ala1; this strain exhibits extremely slow growth rates and served for crystallographic analysis only. All strains used in this study are listed in Supplementary Table 3.
Purification of yeast proteasomes. Yeast strains were grown in 18-l cultures at 30°C in YPD into early stationary phase, and the yCPs were purified according to published procedures 36 . In brief, 120 g yeast cells were solubilized in 150 ml of 50 mM KH 2 PO 4 /K 2 HPO 4 buffer (pH 7.5) and disrupted with a French press. Cell debris were removed by centrifugation for 30 min at 21,000 r.p.m. (4°C). The resulting supernatant was filtered and ammonium sulfate (saturated solution) was added to a final concentration of 30% (v/v). This solution was loaded on a Phenyl Sepharose 6 Fast Flow column (GE Healthcare) pre-equilibrated with 1 M ammonium sulfate in 20 mM KH 2 PO 4 /K 2 HPO 4 (pH 7.5). CPs were eluted by applying a linear gradient from 1 to 0 M ammonium sulfate. Proteasomecontaining fractions were pooled and loaded onto a hydroxyapatite column (Bio-Rad) equilibrated with 20 mM KH 2 PO 4 /K 2 HPO 4 (pH 7.5). Elution of the CPs was achieved by increasing the phosphate buffer concentration from 20 to 500 mM. Anion-exchange chromatogaphy (Resource Q column (GE Healthcare), elution gradient from 0 to 500 mM sodium chloride in 20 mM Tris-HCl (pH 7.5)) and subsequent size-exclusion chromatography (Superose 6 10/300 GL (GE Healthcare), 20 mM Tris-HCl (pH 7.5) and 150 mM NaCl) resulted in pure CPs for crystallization and activity assays.
Fluorescence-based activity assay. ChT-L (b5) activity of CPs was monitored by fluorescence spectroscopy using the model substrate Suc-LLVY-AMC. Purified yCPs (66 nM in 100 mM Tris-HCl, pH 7.5) were incubated with 300 mM substrate for 1 h at room temperature or 37°C. The reactions were stopped by diluting samples 1:10 in 20 mM Tris-HCl, pH 7.5. AMC fluorophores released by proteasomal activity were measured in triplicate with a Varian Cary Eclipse Fluorescence Spectrophotometer (Agilent Technologies) at lexc ¼ 360 nm and lem ¼ 460 nm.
Inhibition assays. Purified yCPs were mixed with dimethylsulfoxide as a control or serial dilutions of inhibitor and incubated for 45 min at room temperature. A final concentration of yCP of 66 nM was used. After addition of the peptide substrate Suc-LLVY-AMC to a final concentration of 200 mM and incubation for 1 h at room temperature, the reaction was stopped by diluting the samples 1:10 in 20 mM Tris-HCl, pH 7.5. AMC fluorophores released by residual proteasomal activity were measured in triplicate at lexc ¼ 360 nm and lem ¼ 460 nm. Relative fluorescence units were normalized to the dimethylsulfoxide-treated control. The calculated residual activities were plotted against the logarithm of the applied inhibitor concentration and fitted with GraphPad Prism 5. The IC50 value, the ligand concentration that leads to 50% inhibition of the enzymatic activity, was deduced from the fitted data.
Diffraction data were collected at the beamline X06SA at the Paul Scherrer Institute, SLS, Villigen, Switzerland (l ¼ 1.0 Å). Evaluation of reflection intensities and data reduction were performed with the programme package XDS 38 . Molecular replacement was carried out with the coordinates of the yeast 20S proteasome (PDB entry code: 5CZ4) by rigid body refinements (REFMAC5; ref. 39). MAIN 40 and COOT 41 were used to build models. TLS (Translation/ Libration/Screw) refinements finally yielded excellent R work and R free , as well as root mean squared deviation bond and angle values. The coordinates, proven to have good stereochemistry from the Ramachandran plots, were deposited in the RCSB Protein Data Bank (Supplementary Table 1).
The coordinates for the yeast 20S proteasome deposited under the entry code 1RYP do not represent the WT yCP but the double-mutant b5-K33R b1-T1A. At the time of deposition (in 1997), these data were the best available on the yCP. As yCP structure determination has become routine today, and structure refinement procedures have significantly improved, we here provide coordinates for the WT yCP at 2.3 Å resolution (PDB entry code: 5CZ4). Furthermore, the structures of most mutant yCPs described in this work were determined in their apo and ligandbound states. For mutants with proteolytically inactive b5 subunits, the best crystallographic data obtained are given. For ligands or propeptide segments that were only partially defined in the 2F O -F C electron-density map the occupancy was reduced (for details see Supplementary Table 1).