A humanized yeast proteasome identifies unique binding modes of inhibitors for the immunosubunit β5i

Abstract

Inhibition of the immunoproteasome subunit β5i alleviates autoimmune diseases in preclinical studies and represents a promising new anti‐inflammatory therapy. However, the lack of structural data on the human immunoproteasome still hampers drug design. Here, we systematically determined the potency of seven α' β' epoxyketone inhibitors with varying N‐caps and P3‐stereochemistry for mouse/human β5c/β5i and found pronounced differences in their subunit and species selectivity. Using X‐ray crystallography, the compounds were analyzed for their modes of binding to chimeric yeast proteasomes that incorporate key parts of human β5c, human β5i or mouse β5i and the neighboring β6 subunit. The structural data reveal exceptional conformations for the most selective human β5i inhibitors and highlight subtle structural differences as the major reason for the observed species selectivity. Altogether, the presented results validate the humanized yeast proteasome as a powerful tool for structure‐based development of β5i inhibitors with potential clinical applications.

Introduction

Protein breakdown by the ubiquitin‐proteasome system ensures cell survival and proliferation. The 20S proteasome core particle (CP) constitutes the heart of this non‐lysosomal protein degradation pathway. In mammals, three different CP types are known: the constitutive proteasome (cCP), the immunoproteasome (iCP), and the thymoproteasome (tCP) (Murata et al, 2007; Groettrup et al, 2010). All of them are built of seven different α‐ and seven different β‐subunits that are stacked in four heptameric rings around a central pore. Each type of CP incorporates a distinct set of catalytically active β1, β2 and β5 subunits (cCP: β1c, β2c and β5c; iCP: β1i, β2i and β5i; tCP: β1i, β2i and β5t) (Groll et al, 1997; Unno et al, 2002; Murata et al, 2007; Huber et al, 2012; Harshbarger et al, 2015). These subunits employ the same mechanism of peptide bond hydrolysis (Huber et al, 2012, 2016), but display different substrate specificities. Because of its pivotal biological role, the proteasome is an attractive drug target. Blockage of the β5 active sites causes cell cycle arrest, an effect that led to the approval of the proteasome inhibitors bortezomib, carfilzomib, and ixazomib for the treatment of multiple myeloma (Appendix Fig S1) (Huber & Groll, 2012; Muz et al, 2016). More recently, selective targeting of the iCP‐subunit β5i, which promotes class I antigen processing and pathogen clearance as well as the production of proinflammatory cytokines, has been shown to suppress various autoimmune diseases in mouse models (Muchamuel et al, 2009; Ichikawa et al, 2011; Basler et al, 2014; Ettari et al, 2016). Structural data on the mouse cCP and iCP in complex with ONX 0914 (formerly PR‐957) (Huber et al, 2012), the first β5i selective inhibitor (Muchamuel et al, 2009), provided insights into its mode of binding and enabled the structure‐guided design of compounds specific for β5c (Xin et al, 2016) and of ligands with improved β5i‐selectivity (Fig 1) (de Bruin et al, 2014). The latter ones unify structural features of the α', β' epoxyketone inhibitors ONX 0914 and PR‐924, another β5i‐specific compound (Parlati et al, 2009; Singh et al, 2011; Niewerth et al, 2014). While ONX 0914 inhibits both the mouse and the human β5i subunit to the same extent (Muchamuel et al, 2009; Huber et al, 2012), PR‐924 was reported to be more selective for human β5i (Niewerth et al, 2014). However, structural data on the yeast 20S proteasome (yCP) in complex with PR‐924 (de Bruin et al, 2014) could not explain the molecular reason for the enhanced β5i‐selectivity and the species specificity of PR‐924. Due to the easy purification of proteasomes from erythrocytes, cCP crystal structures from cow (Unno et al, 2002), mouse (Huber et al, 2012), and humans (Harshbarger et al, 2015) are available to date. By contrast, iCPs are less amenable to isolation because of their cell‐type‐specific and cytokine‐inducible expression pattern. We therefore used mutagenesis to create yeast/human and yeast/mouse chimeric proteasomes. These mutant particles served to structurally investigate the subunit and species selectivity of PR‐924 and related epoxyketone compounds. Compared to wild‐type (WT) yeast, X‐ray structures of β5c/β6 and β5i/β6 humanized yCPs visualize distinct and exceptional binding modes for PR‐924 and its analogues. These results provide insights into the molecular mechanisms of these inhibitors and deliver explanations for their high selectivity toward human β5i. Furthermore, specific guidelines for the design of human β5i‐ligands can be inferred with respect to the stereochemistry of the peptide backbone and the N‐cap size. These data now set the basis for further optimization of iCP‐specific ligands for future clinical applications.

Figure 1. Chemical structures of α', β' epoxyketone immunoproteasome inhibitors analyzed in this study.

The syntheses of the ligands PR‐924, 9, 14, 16, 17, and 18 have been described previously (de Bruin et al, 2014). All ligands share the peptide scaffold shown in the upper left corner with P1, P2 and P3 side chains and an N‐cap; they are equipped with a l‐Ala (stereochemistry (S)) or d‐Ala (stereochemistry (R)) residue at P3 (red) and either an N‐acylmorpholine or a 3‐methyl‐1H‐indene group (green). Furthermore, compounds differ in their P2 residue (either methoxytyrosine or tryptophane; blue). Inhibitor 9, also termed LU‐015i, is endowed with an exceptional cyclohexyl residue (gray), which is associated with increased selectivity for human β5i.

Results

Immunoproteasome inhibitors

The proteasome inhibitor bortezomib, a dipeptide boronic acid (Appendix Fig S1), equally targets the subunits β5c and β5i (Demo et al, 2007). Despite its off‐target activity toward serine proteases, which causes severe side effects (Arastu‐Kapur et al, 2011), bortezomib was granted approval by the FDA for cancer treatment. Carfilzomib, a second‐generation compound based on the natural product epoxomicin, only targets the small class of N‐terminal nucleophilic hydrolases such as the proteasome. By a two‐step reaction mechanism, the epoxyketone inhibitor carfilzomib covalently and irreversibly modifies the active site Thr1O^γ and Thr1N of predominantly β5c and β5i proteasome subunits (Groll et al, 2000; Demo et al, 2007). Screening of diverse peptide libraries (Harris et al, 2001; Nazif & Bogyo, 2001; Britton et al, 2009; Blackburn et al, 2010; Huber et al, 2015a) led to the development of subunit‐selective inhibitors. The prototypes of β5i‐specific epoxyketones are ONX 0914 (Muchamuel et al, 2009) and PR‐924 (Parlati et al, 2009) (Fig 1). As shown in our activity assays with genuine human as well as mouse cCP and iCP, both compounds are potent inhibitors of human β5i (IC₅₀ ≤ 0.051 μM) with up to 20‐ and 255‐fold selectivity over human β5c, respectively (Fig 2, Appendix Table S2 and Appendix Fig S2). Notably, the superior selectivity of PR‐924 is not based on enhanced hβ5i inhibition but worse hβ5c targeting compared to ONX 0914 (IC₅₀ hβ5c (ONX 0914): 0.487 μM; hβ5c (PR‐924): 13.020 μM; Fig 2B). The chemical structures of ONX 0914 and PR‐924 differ in their P2 residue (methoxytyrosine versus tryptophane), in the stereochemistry of their P3‐Ala residue (l versus d), and in their N‐cap (morpholine versus 3‐methyl‐1H‐indene; Fig 1). First, we addressed the question which of these structural differences gives rise to the improved β5i selectivity of PR‐924, and secondly, we aimed to investigate why PR‐924 is seven times less potent for mouse β5i (IC₅₀ 0.353 μM) than for human β5i (IC₅₀ 0.051 μM)—a phenomenon that is not observed for ONX 0914 (IC₅₀ hβ5i: 0.024 μM versus mβ5i: 0.040 μM; Fig 2B).

Figure 2. Inhibition of the chymotrypsin‐like activities of yeast, human, and mouse proteasomes by ONX 0914, PR‐924, 9, 14, 16, 17, and 18.

1. Inhibition of human β5c and β5i by ONX 0914 and PR‐924. IC₅₀ values are deduced from the fitted data.
2. IC₅₀ values (μM) for all compounds toward the yeast, human, and mouse β5 subunits. For inhibition curves, see Appendix Figure S2. The ratio IC₅₀(β5c)/IC₅₀(β5i), a measure of β5i selectivity, is given. Large values indicate high selectivity for β5i.

Data information: Genuine WT proteasomes were used for all inhibition assays. IC₅₀ values (μM) were determined in triplicates with the fluorogenic substrate Suc‐LLVY‐AMC using an enzyme concentration of 13 nM. Mean standard deviations are indicated as bars and given in Appendix Table S2.

Since available structural data on yeast and mouse CPs failed to provide plausible explanations for the subunit and species inhibition profiles of PR‐924, we created yeast/mammalian chimeric proteasomes and analyzed their X‐ray structures in complex with a set of inhibitors, including PR‐924 and ONX 0914 as well as previously described derivatives of both (compounds 9, 14, 16, 17 and 18, Fig 1). For clarity reasons, compound numbering was taken over from the first report of these ligands (de Bruin et al, 2014).

Construction of chimeric yeast/human proteasomes

In an initial attempt, we tried to express the full‐length human β5c and β5i as well as mouse β5i subunits in a yeast pre2Δ knockout strain under the control of the PRE2 promotor and terminator sequences (for details see Appendix Supplementary Methods). Despite the strictly conserved fold of proteasome subunits across species, this genetic setup was lethal, even after replacement of the genuine β5c/β5i propeptide by that of yβ5 (Fig 3A). Most likely the mammalian β5 subunits are unable to adopt their correct three‐dimensional structure or fail to incorporate into the yeast proteasome due to species‐specific differences in the subunit interface contacts.

Figure 3. Expression of mammalian β5 proteasome subunits in yeast.

1. Schematic representation of yeast and mammalian β5 subunits and their propeptides. Secondary structure elements, helices (H) and sheets (S), are numbered. The full‐length human hβ5i (yellow) and β5c (red) as well as the mouse mβ5i (green) subunits cannot substitute the endogenous yeast yβ5 subunit (gray). Expression of the mammalian subunits with their natural propeptide (pp; colored) or with the yβ5 propeptide (pp; gray) is lethal in a yeast pre2Δ knockout strain (for details, see Appendix Supplementary Methods).
2. Chimeric β5 subunits are schematically illustrated according to panel (A). The N‐terminal part of the subunits, which forms the primed and non‐primed substrate binding pockets and the active site surroundings, was taken from human β5i and the C‐terminus from yeast. A construct featuring residues 1–173 from human β5i was still lethal. Replacement of the residues 1–139 of yβ5 by the human β5i counterpart (aa 1–138; see Appendix Supplementary Note), however, was compatible with yeast survival. Based on this observation, analogous fragments of the human β5c and mouse β5i subunits were incorporated into the yβ5 subunit to create the respective hβ5c and mβ5i chimeras.
3. Subunit β6 significantly contributes to the S3 and S4 pockets of the ChT‐L active site. In order to mimic the entire mammalian β5/6 substrate binding channel in yeast, we mutated five residues of the yβ6 subunit as well (G89S, K90R, H98Y, T99N, and E120Q). These mutations correspond to both the human and the mouse β6 sequence and create two mammalian segments aa 87–101 and aa 108–121 (blue; see also Appendix Supplementary Note).
4. The mouse‐specific point mutation V31M was inserted into the hβ5i/hβ6 chimera to investigate its impact on inhibitor binding.
5. Serial dilutions of yeast cells were spotted on YPD plates and grown for 2 days either at 30°C or 37°C. Mutant yeasts are significantly retarded in growth and show increased temperature sensitivity.
6. Surface illustration of the yCP. The β5 and β6 subunits are depicted in dark gray, with the residue ranges that correspond to mammalian β5 and β6 subunits in the chimeric proteasomes highlighted in gold and blue, respectively. The β5' and β6' subunits of the second β‐ring are hidden on the back of the 20S proteasome.

Next, we replaced the endogenous yβ5 by various chimeric hβ5i/yβ5 subunits that feature the yβ5 propeptide, N‐terminal parts of human β5i, and C‐terminal segments of yβ5 (Fig 3B). Among the viable mutant yeasts, we selected that one that carried the largest fragment of β5i (amino acids (aa) 1–138; see also Appendix Supplementary Note). Importantly, this residue range encodes both primed and non‐primed pockets of the substrate binding channel. The mutant strain suffered from a pronounced growth defect, from increased temperature sensitivity as well as significantly reduced chymotrypsin‐like (ChT‐L; β5) proteasome activity (Fig 3E and Appendix Fig S3A). Crystallographic analysis (for X‐ray data see Appendix Table S1) however revealed that the chimeric β5 subunits were properly incorporated into the CP and fully matured. Despite extensive mutagenesis, superposition with the mouse iCP and WT yCP coordinates proved the preservation of the overall fold of the chimeric β5 subunits (root‐mean‐square deviation (r.m.s.d.) for mouse β5i: C^α 0.751 Å; for yeast yβ5: C^α 0.209 Å; Appendix Fig S4A). Furthermore, the structures displayed Met45 in the same peculiar “open” conformation that has previously been discovered for the mouse iCP in the apo state (Huber et al, 2012) (Appendix Fig S4B). Hence, our results indicate that the reorientation of Met45 is inherent to iCPs irrespective of the species and that the molecular switch for this conformational change is encoded within the β5i residues 1–138.

After having addressed subunit β5i, we extended our mutagenesis to the neighboring β6 subunit in order to create an entire human β5i/β6 substrate binding channel for X‐ray inhibitor binding studies. In total, five amino acids of yβ6 that mediate important interactions with ligands bound to the β5 active site were mutated (Fig 3C). These residues further aggravated the growth defect of the respective yeast strain (Pre7^mut; Fig 3E). Compared to WT yCPs, purified hβ5i/hβ6 chimeric proteasomes showed normal β1 and β2 but 90% reduced β5 activity (Appendix Fig S3B). In accordance, the β5 subunit was less sensitive to epoxyketone inhibitors (Appendix Fig S3C). X‐ray data on the hβ5i/hβ6 chimeric proteasome however proved that the modified β6 residues take the same place as in the human or mouse β6 subunit (Huber et al, 2012; Harshbarger et al, 2015) and that all crucial β5 residues including Met45 are arranged as in the natural mouse iCP (Fig 4A). In analogy to the hβ5i/hβ6 chimera, mouse mβ5i/mβ6 and human hβ5c/hβ6 proteasomes were constructed (aa 1–138; Fig 3B). The respective yeasts as well as the purified CPs suffered from the same defects observed for the hβ5i/hβ6 chimera (Fig 3E and Appendix Fig S3). Although the mutant CPs are not equivalent to endogenous WT yCP, cCP, and iCP with respect to catalytic activity, their structural congruence is striking. While in the i‐chimeras, Met45 is in the open conformation, the crystal structure of the hβ5c/hβ6 chimera visualizes the typical β5c “closed” orientation of Met45 (Fig 4B–D). This structural equivalence validates the designed chimeric proteasomes for X‐ray diffraction‐based ligand binding studies.

Figure 4. Crystal structures of chimeric human proteasomes in their apo state.

1. Superposition of the β5i/β6 active sites of genuine mouse iCP and the hβ5i/hβ6 human/yeast chimera illustrates that Met45 adopts the same “open” conformation in both structures. Amino acid side chains important for substrate binding are depicted as sticks. The catalytic Thr1 is colored in black. Met31, the single amino acid difference between mouse and human β5i active sites, is highlighted in green.
2. Superposition of the genuine mouse and the human chimeric β5c/β6 active sites according to panel (A) proves the proper orientation of Met45 in the “closed” conformation.
3. Superposition of the β5c/β6 and β5i/β6 active sites of the natural mouse cCP and iCP according to panel (A). Note the distinct conformations of Met45 (Huber et al, 2012).
4. Superposition of the hβ5c/hβ6 and the hβ5i/hβ6 chimeric proteasomes confirms the distinct positions of Met45 analogously to the natural mouse subunits shown in panel (C).

Data information: This figure contains previously published coordinates (Huber et al, 2012): mouse cCP (PDB accession code: 3UNE) and mouse iCP (PDB accession code: 3UNH).

Enhanced selectivity of PR‐924 and its analogues for human β5i versus β5c

Crystals of the hβ5c/hβ6 and the hβ5i/hβ6 chimeras were soaked with the prevalent proteasome inhibitors ONX 0914, bortezomib, and carfilzomib. These compounds were chosen, because structural data on their binding mode to the WT yCP (Groll et al, 2006; Huber et al, 2015b), the human (Harshbarger et al, 2015) and mouse cCP (Huber et al, 2012) as well as the mouse iCP (Huber et al, 2012) are available for comparison. Despite the reduced activity of the chimeric proteasomes, all ligands covalently modified the mutant β5 active sites at high occupancy (Fig EV1A and B, Appendix Table S3). Furthermore, the coordinates revealed uniform modes of binding for all three inhibitors to WT yCP, chimeric CPs, and available WT mammalian CP structures (r.m.s.d. < 0.65 Å; Fig EV1C) (Huber et al, 2012; Harshbarger et al, 2015). For ONX 0914, we previously noticed distinct orientations of the N‐cap in the mouse constitutive proteasome and the immunoproteasome crystal structures (Huber et al, 2012). Remarkably, these distinct conformations are now also observed with the humanized yeast structures (Fig EV1C).

Figure EV1. ONX 0914, bortezomib, and carfilzomib bound to human chimeric proteasomes.

1. The 2F_O‐F_C electron density maps for ONX 0914, bortezomib, and carfilzomib covalently bound to the β5 active site Thr1 of the hβ5c/hβ6 chimera are shown as blue meshes contoured at 1σ (according to Fig 5A). The N‐caps of carfilzomib and ONX 0914 are less well defined.
2. The 2F_O‐F_C electron density maps for ONX 0914, bortezomib, and carfilzomib linked to the β5 active site Thr1 of the hβ5i/hβ6 chimeric proteasomes are depicted according to panel (A). The P4 site and the N‐cap of carfilzomib are poorly defined.
3. Superpositions of ONX 0914, bortezomib, and carfilzomib bound to β5 subunits of the WT mouse iCP and cCP, WT human cCP, the human chimeric proteasomes, and the WT yCP illustrate their similar binding modes. Note that the structures are rotated by 45° compared to the panels (A) and (B).

Data information: Panel (C) contains previously published coordinates with the following PDB accession codes: WT mouse iCP:ONX 0914, 3UNH (Huber et al, 2012); WT mouse cCP:ONX 0914, 3UNE (Huber et al, 2012); WT yCP:ONX 0914, 4QWX (Huber et al, 2015b); WT yCP:bortezomib, 4QVL (Huber et al, 2015b); WT yCP:carfilzomib, 4QW4 (Huber et al, 2015b); WT human cCP:carfilzomib, 4R67 (Harshbarger et al, 2015).

Next, complex structures of the chimeric human proteasomes with PR‐924 and derivatives thereof were determined (Figs 5 and EV2A and B). For the constitutive hβ5c/hβ6 chimera, we observed similar results as with WT yCP (de Bruin et al, 2014; and Appendix Fig S5). Compounds exclusively built of l amino acids (ONX 0914, 14) bind in an extended manner to the hβ5c/hβ6 chimera regardless of their type of N‐cap. Ligands that are furnished with a P3‐d‐Ala and the 3‐methyl‐1H‐indene cap (PR‐924 and 18) adopt a linear orientation as well, but those combining a P3‐d‐Ala amino acid and a morpholine N‐cap (16 and 17) are bent, with the N‐cap occupying a pocket outside of the natural substrate binding channel, which we termed S3* (Figs 5 and EV2A). Notably, ligands with d stereochemistry at P3 are generally less potent for yβ5 and human β5c than those with natural l stereochemistry at P3 (ONX 0914 and 14; Fig 2B).

Figure 5. Ligand complex structures of human chimeric proteasomes reveal distinct binding modes for β5c and β5i.

1. Complex structures of the human β5c/β6 (left) and β5i/β6 (right) chimeric proteasomes with PR‐924. The ligand binds in a linear (l) manner to the β5c/β6 substrate binding channel, but in a kinked (k) conformation to the β5i/β6 active site. The 2F_O‐F_C electron density maps for the ligands bound to Thr1 of the chimeric β5 active sites are shown as blue meshes contoured at 1σ. The inhibitor and Thr1 have been omitted for phasing.
2. Comparison of the conformations of ONX 0914, PR‐924, 14, 16, 17, and 18 bound to the chimeric hβ5/hβ6 substrate binding channels. The l amino acid compounds ONX 0914 and 14 display a linear (l) binding mode to the hβ5c/hβ6 (left) and the hβ5i/hβ6 (right) chimeras, whereas the ligands 16 and 17 (P3‐d‐Ala and morpholine cap) adopt a kinked (k) conformation in both β5 substrate binding channels. Remarkably, the orientation of PR‐924 and its analogue 18 (P3‐d‐Ala and 3‐methyl‐1H‐indene cap) in the hβ5c/hβ6 chimera (and in WT yβ5; see Appendix Figure S5) significantly differs from that observed with the hβ5i/hβ6 chimera. Note that the inhibitors are rotated by about 45° compared to panel (A).

Figure EV2. Crystal structures of human chimeric proteasomes in complex with derivatives of ONX 0914 and PR‐924.

1. The 2F_O‐F_C electron density maps for the ligands 14, 16, 17, and 18 linked to the β5 active site Thr1 of the hβ5c/hβ6 chimeric proteasomes are shown according to Fig 5A. Compound 9 could not be trapped, presumably due to its low affinity (IC₅₀ > 200 μM; Fig 2B).
2. The 2F_O‐F_C electron density maps for the compounds 9, 14, 16, 17, and 18 bound to the β5 active site Thr1 of the hβ5i/hβ6 chimeric proteasomes are depicted according to Fig 5A.

In stark contrast, the chimeric immuno‐hβ5i/hβ6 proteasome binds all inhibitors with a P3‐d amino acid—irrespective of their type of N‐cap—in a kinked conformation (Figs 5 and EV2B). This observation agrees with modeling results for PR‐924 (Sharma et al, 2012).

In summary, the compounds 14, 16, and 17 dock in an identical manner to both chimeras, while PR‐924 and 18 significantly differ in their binding mode to hβ5c versus hβ5i. These results also provide an explanation for the varying potencies of the ligands. As shown by our inhibition assays with endogenous human cCP and iCP, all derivatives of ONX 0914 and PR‐924 are highly effective human β5i inhibitors (IC₅₀ ≤ 0.374 μM, Fig 2B). However, with respect to β5c targeting PR‐924, 9 and 18 are the worst, thereby resulting in highest selectivity ratios of ≥ 155 (Fig 2B). Obviously, the linear binding mode of ligands with non‐natural P3‐d stereochemistry and a 3‐methyl‐1H‐indene cap bound to hβ5c is energetically strongly disfavored compared to the kinked orientation in the hβ5i subunit. This is remarkable, as ligands that adopt the bent conformation cannot be stabilized by Asp114 from the neighboring subunit β6, which usually hydrogen bonds to the peptide backbone of linearly arranged inhibitors (Huber et al, 2015a).

Reducing the size of the N‐cap from a 3‐methyl‐1H‐indene to a morpholine moiety (compounds 16 and 17) leads to bent ligand binding modes in both hβ5c and hβ5i and comparable inhibition strengths for both active sites. We thus conclude that the S3* pocket of β5c impedes binding of inhibitors with P3‐d amino acids and bulky N‐caps such as the 3‐methyl‐1H‐indene moiety for steric and energetic reasons. From the structural data, no obvious barrier can be inferred but hydrogen bonds at the interface of the subunits β5 and β6 involving Ser28 (yβ5, mβ5c, mβ5i, hβ5c)/Ala28 (hβ5i), Thr30 (yβ5, mβ5c, hβ5c)/Arg30 (mβ5i, hβ5i), and Glu120 (yβ6)/Gln120 (hβ6, mβ6) might play crucial roles. We also determined a crystal structure of the hβ5i chimera (with entire subunit β6 from yeast) in complex with PR‐924 and found that the ligand takes the same position as in the hβ5i/hβ6 chimera. These data indicate that solely amino acid exchanges in subunit β5i are responsible for triggering the kinked binding mode of PR‐924 (Appendix Fig S6). Taken together, the N‐cap, which originally served to improve solubility and pharmacokinetic parameters of peptide inhibitors, now becomes a dominant factor for subunit selectivity of β5i ligands with d stereochemistry at the P3 position.

Structural basis for species selectivity of P3‐d‐Ala 3‐methyl‐1H‐indene‐capped inhibitors

As derived from our activity assays with genuine mouse and human cCP as well as iCP, no pronounced species selectivity is observed for ONX 0914 (Muchamuel et al, 2009; Huber et al, 2012) and its closest structural analogue, compound 14 (Fig 2B). By contrast, PR‐924, 9, and 18 show at least a sevenfold preference for human versus mouse β5i (Fig 2B). Furthermore, all three compounds discriminate strongly between hβ5c and hβ5i (factor ≥ 155), while this effect is less pronounced for mβ5c and mβ5i (factor ≤ 16; Fig 2B). In order to analyze the molecular reason for this species selectivity, we used the mouse mβ5i/mβ6 chimeric proteasome. As expected, ONX 0914, bortezomib, and carfilzomib as well as PR‐924, 9, 14, 16, 17, and 18 bind to the mβ5i/mβ6 chimeric proteasome similar as has been observed for the hβ5i/hβ6 construct (Figs 6 and EV3). As noted for the hβ5i/hβ6 chimeric proteasomes, compounds with a non‐natural d amino acid at the P3 site position the N‐cap close to the P1 site. In the case of mβ5i, the N‐cap approaches Met31 of the S1 pocket, which in human and mouse β5c as well as in human β5i is a Val residue (Appendix Fig S7). While the tiny Val31 is compatible with the kinked binding mode of P3‐d‐Ala inhibitors, the larger Met31 imposes a steric barrier to such ligands. This hindrance increases with the size of the P1 site and of the N‐cap. The small and flexible N‐acylmorpholine group is easily accommodated, but the bulky and rigid 3‐methyl‐1H‐indene cap clashes with Met31. Hence, only compounds endowed with the latter N‐cap show impaired affinity for the mouse β5i subunit. We therefore mutated Val31 to methionine in the hβ5i/hβ6 chimeric proteasome (Fig 3D) and determined the respective ligand complex structures (Figs 6 and EV4). In support of our hypothesis, the X‐ray data revealed that the morpholine‐capped compounds 16 and 17 were unaffected by Met31 (Fig EV5), while the more bulky 3‐methyl‐1H‐indene caps of PR‐924 and 18 were rotated by 180° and shifted up to 2.3 Å (Fig 6C). Notably, the N‐caps of PR‐924 and 18 are positioned as in the mβ5i/mβ6 mutant.

Figure 6. The impact of Met31 on ligand binding to the β5i active site.

1. Complex structures of the mouse chimeric proteasome (left) and the hβ5i‐V31M/hβ6 mutant (right) with PR‐924 visualize the kinked (k) orientation of the inhibitor. The 2F_O‐F_C electron density maps for the ligands bound to Thr1 of the chimeric β5 active sites are depicted according to Fig 5A. Steric hindrance of Met31 (green) with the bulky 3‐methyl‐1H‐indene cap is indicated by a black double arrow.
2. Superposition of inhibitors bound to the mouse chimeric proteasome (left) and the hβ5i‐V31M/hβ6 mutant (right). Compounds display the same binding modes (kinked (k) or linear (l)) to the mβ5i/mβ6 active site as to the corresponding human β5i/β6 subunit (Fig 5) and to the hβ5i‐V31M/hβ6 mutant. Note that the inhibitors are rotated by about 45° compared to panel (A).
3. Superposition of chimeric mammalian and WT yeast β5 active sites bound to the P3‐d‐Ala compounds PR‐924 and 18. The nucleophilic Thr1 is depicted as well as the S1 pocket forming residues 31 and 45. Residue 31 impacts on the position of the 3‐methyl‐1H‐indene cap of PR‐924 and 18. Due to steric hindrance, the N‐cap is shifted in the mouse and in the hβ5i‐V31M/hβ6 chimera compared to the hβ5i/hβ6 mutant proteasome crystal structures (black double arrow). Note that the structures are rotated by 90° compared to panel (B).

Figure EV3. Crystal structures of the mouse β5i/β6 chimeric proteasome in complex with ONX 0914, bortezomib, carfilzomib, and derivatives of ONX 0914 as well as PR‐924.

1. The 2F_O‐F_C electron density maps for the ligands bound to the β5 active site Thr1 of the mβ5i/mβ6 chimeric proteasome are depicted according to Fig 5A. The mβ5i‐specific Met31 side chain is colored in green. Steric hindrance of Met31 and compound 18 is indicated by a black double arrow.
2. Structural superpositions of ONX 0914, bortezomib, and carfilzomib bound to β5 subunits of the WT mouse iCP and cCP, WT human cCP, the human, and the mouse chimeric proteasomes as well as the WT yCP illustrate their similar binding modes.

Data information: Panel (B) contains previously published coordinates with the following PDB accession codes: WT mouse iCP:ONX 0914, 3UNH (Huber et al, 2012); WT mouse cCP:ONX 0914, 3UNE (Huber et al, 2012); WT yCP:ONX 0914, 4QWX (Huber et al, 2015b); WT yCP:bortezomib, 4QVL (Huber et al, 2015b); WT yCP:carfilzomib, 4QW4 (Huber et al, 2015b); WT human cCP:carfilzomib, 4R67 (Harshbarger et al, 2015).

Figure EV4. Crystal structures of the hβ5i‐V31M/hβ6 mutant chimeric proteasome in complex with the compounds 16, 17, and 18.

The 2F_O‐F_C electron density maps for the ligands are shown according to Fig 5A. Met31, a hallmark of the mouse β5i subunit, has been incorporated in the human β5i/β6 chimera and is colored in green. Steric hindrance of Met31 and compound 18 is indicated by a black double arrow.

Figure EV5. Comparison of ligand binding modes to chimeric proteasomes.

The orientation of ONX 0914 as well as 14, 16, and 17 is identical in all β5 substrate binding channels, including those of chimeric proteasomes, WT yCP, WT mouse cCP and iCP. Met31 has no significant influence on the position of the N‐cap in the compounds ONX 0914, 14, 16, and 17. However, the morpholine N‐cap of 16 and 17 adopts a tilted position in the hβ5i/hβ6 chimeric proteasomes. This figure contains previously published coordinates (Huber et al, 2012, 2015b): yCP:ONX 0914 (PDB accession code: 4QWX), mouse cCP:ONX 0914 (PDB accession code: 3UNB), and mouse iCP:ONX 0914 (PDB accession code: 3UNF).

Taken together, a combination of activity assays, yeast genetics, and X‐ray crystallography elucidated how a single amino acid can significantly affect ligand binding and why compounds featuring a non‐natural d‐Ala at P3 and a 3‐methyl‐1H‐indene cap are more potent and more selective for human β5i than for the mouse counterpart. As a result of the various crystal structures, we here provide for the first time experimental evidence that proteasome inhibitors can adopt different conformations in distinct eukaryotic CPs. This observation is of particular importance for the development of CP‐type selective inhibitors and clearly limits the use of WT yeast as a model system. On the other hand, the data presented here strongly vote for yeast as an excellent tool box that can be easily genetically modified to create mammalian proteasome models for advanced structure‐based drug design efforts.

Discussion

By systematically investigating the inhibition potencies of various epoxyketones for mouse and human β5c and β5i, we noticed that the compounds tested significantly differ in their β5i selectivity (Fig 2). Since large amounts of human iCPs for structural analyses of all these ligands are difficult to obtain, we aimed at integrating the human ChT‐L active site of cCP and iCP into the yeast 20S proteasome. Purification and high‐throughput crystallization of mutant yCPs are well established, and baker's yeast in addition allows for rapid point mutagenesis to evaluate the impact of individual amino acids on ligand binding. However, all attempts to replace the endogenous yβ5 by human or mouse β5 subunits as a whole was lethal to yeast. We therefore created chimeric proteasomes that feature either the entire human β5c/β6, human β5i/β6, or mouse β5i/β6 substrate binding channel. The respective yeast strains were viable, but suffered from growth retardation and increased temperature sensitivity suggesting proteasome activity defects (Fig 3E). In fact, purified chimeric proteasomes showed WT‐like β1 and β2 activities but were almost deprived of ChT‐L activity and less sensitive to β5‐inhibition (Appendix Fig S3B and C). Altered or disrupted inter‐ and intra‐subunit contacts as well as impaired dynamics caused by subtle structural changes below the resolution limits might restrict reactivity of the ChT‐L active site toward substrates and inhibitors. Thus, with respect to activity, the chimeric proteasomes are not equivalent to endogenous WT particles and candidate drugs have to be evaluated for their inhibition strength and subunit selectivity in enzymatic assays with genuine iCP and cCP.

However, the chimeric CPs proved to be valuable structural models. X‐ray crystallographic data revealed that the mutant β5 subunits structurally conform with the genuine mouse β5 entities and that the orientations of Met45 correspond to those observed for the natural mouse CPs (Huber et al, 2012) (Fig 4A and B). Numerous inhibitor complex structures finally revealed that compounds furnished with a large 3‐methyl‐1H‐indene cap and a P3‐d amino acid adopt distinct conformations in the β5c‐chimeric proteasome compared to the β5i mutant one (Fig 7). This observation provides an explanation for the β5i‐selectivity of these inhibitors and validates the chimeric proteasomes for sophisticated structure‐based drug development.

Figure 7. Overview of key inhibitors and their distinct binding modes to the subunits yβ5, hβ5c, mβ5i, and hβ5i.

ONX 0914 binds in a linear mode to all active sites. Changing the stereochemistry at the P3 site from l‐Ala (S) to d‐Ala (R) leads to a kinked conformation in the β5 substrate binding channels (compound 16). Substitution of the morpholine cap of 16 by a larger 3‐metyhl‐1H‐indene moiety yields PR‐924 and results in different inhibitor conformations at the yβ5/hβ5c and the mammalian β5i active sites. Hereby, the selectivity for hβ5i over hβ5c is drastically increased (IC₅₀ ratio hβ5c/hβ5i: 255) compared to ONX 0914 (IC₅₀ ratio hβ5c/hβ5i: 20) and compound 16 (IC₅₀ ratio hβ5c/hβ5i: 5.2, Fig 2). Residue Met31 of mβ5i sterically hinders the kinked binding mode of PR‐924, thereby creating sevenfold selectivity for human β5i.

In a second approach, we created a mouse chimeric proteasome in order to investigate the species selectivity of PR‐924 and its closest analogues 9 and 18. Mouse and human β5i primary sequences differ in only 17 positions. By X‐ray analysis, we found that Met31 of the mouse β5i subunit sterically hinders binding of inhibitors featuring a P3‐d amino acid in combination with a 3‐methyl‐1H‐indene cap. In contrast, the corresponding human β5i residue Val31 is compatible with such compounds, thereby enhancing selectivity for human β5i at least sevenfold. Notably, this species selectivity effect largely depends on the size of the N‐cap. The substrate binding pockets and the active site residues are highly conserved among mouse and human, but residues outside the natural substrate binding channel, for example, the identified S3* pocket, are subjected to higher mutagenesis rates, thereby creating species‐specific differences such as Val/Met31. Furthermore, P1 side chains larger than Phe may be incompatible with a P3‐d amino acid and a large N‐cap because of mutual sterical hindrance. The non‐natural d stereochemistry at P3 and the associated unique binding mode render the P3‐d‐Ala compounds PR‐924, 9 and 18 exclusive β5i inhibitors (de Bruin et al, 2014), a fact that may reduce cytotoxicity in clinical applications (Parlati et al, 2009). Moreover, the d‐Ala amino acid may prolong the resistance to proteases as well as improve the metabolic stability and therapeutic efficacy. However, the resistance of the murine iCP against these compounds requires the use of laboratory animals other than mice for preclinical studies. Species that encode Val31, for example, rats but also apes, might be suitable alternatives (Appendix Fig S7).

In summary, our study describes the creation of a humanized yCP variant that unifies all striking structural features of the mammalian β5i subunit. Proof of principle of the exceptional value of this chimeric CP for the future development of human iCP‐specific inhibitors has been provided by structural analyses with PR‐924 and its analogues. Certainly, these results will further support structure‐guided design and optimization of CP‐type‐specific ligands.

Materials and Methods

Yeast strain construction

See Appendix Supplementary Methods.

Yeast growth tests

Overnight cultures of WT (WCG4a) and mutant yeast strains were grown in rich (YPD) medium at 30°C. Their optical density was determined at 600 nm, and appropriate serial dilutions in YPD were prepared. Total cell numbers of 10,000, 1,000, 100, and 10 were spotted on YPD plates and grown for several days at 30°C or 37°C to evaluate proliferation rates by colony size as well as survival rates by colony numbers.

Mammalian proteasomes

Human and mouse constitutive and immunoproteasomes were purchased from Boston Biochem (USA).

Purification of yeast proteasomes

Yeast strains were grown in 18 l YPD cultures at 30°C into early stationary phase. Cells were harvested by centrifugation for 15 min at 5,000 g and frozen at −20°C until further use. Mutant yCPs were purified according to published procedures (Gallastegui & Groll, 2012). In brief, 120 g yeast cells were solubilized in 150 ml of 50 mM KH₂PO₄/K₂HPO₄ buffer (pH 7.5) and disrupted with a French press. Cell debris were removed by centrifugation for 30 min at 41,000 g (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 KH₂PO₄/K₂HPO₄ (pH 7.5). CPs were eluted by applying a linear gradient from 1 to 0 M ammonium sulfate. Proteasome containing fractions were pooled and loaded onto a Resource Q column (GE Healthcare). Upon gradient elution (0–500 mM sodium chloride in 20 mM Tris–HCl pH 7.5), CPs were subjected to size exclusion chromatography (Superose 6 10/300 GL (GE Healthcare), 20 mM Tris–HCl pH 7.5, and 150 mM NaCl).

Proteasome substrates and inhibitors

Peptide substrates (Bachem) and inhibitors were stored as 50–100 mM stock solutions in DMSO at −20°C. Bortezomib, carfilzomib, and ONX 0914 were purchased from Selleckchem. The chemical synthesis of PR‐924 and the compounds 9, 14, 16, 17, and 18 was described previously (de Bruin et al, 2014).

Fluorescence‐based activity assay

Proteasome activity was determined by fluorescence spectroscopy using the model substrates carboxybenzyl‐Leu‐Leu‐Glu‐7‐amino‐4‐methylcoumarin (Z‐LLE‐AMC; β1), tert‐butyloxycarbonyl‐Leu‐Arg‐Arg‐AMC (Boc‐LRR‐AMC; β2), and N‐succinyl‐Leu‐Leu‐Val‐Tyr‐7‐amino‐4‐methylcoumarin (Suc‐LLVY‐AMC; β5). Purified yCPs (13 nM in 100 mM Tris–HCl, pH 7.5) were incubated with 200 μM substrate for 1 h at room temperature. The reactions were slowed down 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 λ_exc = 360 nm and λ_em = 460 nm.

IC₅₀ determination

Initial point measurements were carried out with 200 μM of inhibitor. Only compounds that showed significant inhibition at this concentration were further evaluated. CPs (final concentration: 13.2 nM) were mixed with serial dilutions of inhibitor or with DMSO as a control and incubated for 1 h at room temperature. After addition of the peptide substrate Suc‐LLVY‐AMC (final concentration of 200 μM) and incubation for 1 h at room temperature, proteolysis was slowed down by diluting the samples 1:10 in 20 mM Tris–HCl, pH 7.5. The AMC molecules released by residual proteasomal activity were measured in triplicate with a Varian Cary Eclipse Fluorescence Spectrophotometer (Agilent Technologies) at λ_exc = 360 nm and λ_em = 460 nm. Relative fluorescence units were normalized to the DMSO 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.

Note: IC₅₀ values depend on time and enzyme concentration and hence are not absolute; only relative tendencies within the same experimental settings are inferable.

Crystallization and structure determination

Mutant yCPs were crystallized as previously described for wild‐type 20S proteasomes (Groll & Huber, 2005; Gallastegui & Groll, 2012). Crystals were grown at 20°C using the hanging drop vapor diffusion method. Drops contained a 1:1 mixture of protein (40 mg/ml) and reservoir solution (25 mM magnesium acetate, 100 mM 2‐(N‐morpholino)ethanesulfonic acid (MES) pH 6.8 and 9–13% (v/v) 2‐methyl‐2,4‐pentanediol (MPD)). Crystals were cryoprotected by addition of 5 μl cryobuffer (20 mM magnesium acetate, 100 mM MES, pH 6.8 and 30% (v/v) MPD). Inhibitor complex structures were obtained by incubating crystals in 5 μl cryobuffer supplemented with inhibitors at a final concentration of 1.5 mM for at least 8 h.

Diffraction data were collected at the beamline X06SA at the Paul Scherrer Institute, SLS, Villigen, Switzerland (λ = 1.0 Å). Evaluation of reflection intensities and data reduction was performed with the program package XDS (Kabsch, 2010). Molecular replacement was carried out with the coordinates of the yeast 20S proteasome (PDB entry code: 5CZ4; Huber et al, 2016) by rigid body refinements (REFMAC5; Vagin et al, 2004). MAIN (Turk, 2013) and COOT (Emsley et al, 2010) were used to build models. TLS refinements finally yielded excellent R_crys and R_free as well as r.m.s.d. bond and angle values. The coordinates, proven to have good stereochemistry from the Ramachandran plots, were deposited in the RCSB Protein Data Bank (Appendix Table S1).

Figures were prepared with PyMOL (DeLano, 2002). For the calculation of all electron density maps, the inhibitor and Thr1 have been omitted for phasing.

Data availability

Coordinates and structure factors have been deposited in the RCSB Protein Data Bank under the following accession codes (for details see Appendix Table S1):

5L52, 5L54, 5L55, 5L5A, 5LTT, 5L5B, 5L5D, 5L5E, 5L5F, 5L5H, 5L5I, 5L5J, 5L5O, 5L5P, 5L5Q, 5L5W, 5L5X, 5L5Y, 5L5Z, 5L60, 5L61, 5L62, 5L63, 5L64, 5L6B, 5L65, 5L66, 5L67, 5L68, 5L69, 5L6A, 5L6C, 5L5R, 5L5S, 5L5T, 5L5U, 5L5V

Author contributions

WH created yeast proteasome mutants; EMH performed yeast growth and proteasome activity/inhibition assays; GdB synthesized proteasome ligands under the supervision of HSO; EMH and MG collected and analyzed X‐ray data; EMH and MG wrote the manuscript.

Conflict of interest

The authors declare that they have no conflict of interest.

Supporting information

Appendix

Expanded View Figures PDF

Review Process File

The EMBO Journal (2016) 35: 2602–2613