*2.3. Structural Insights into the Mechanisms Underlying PAC3*/*PAC4-Dependent* α*4-*α*5-*α*6 Assembly*

To investigate the structural mechanisms underlying the chaperone-dependent formation of the α4-α5-α6 subcomplex, we built a three-dimensional model of the putative quintet complex comprised of PAC3, PAC4, α4, α5 and α6, using previously-reported crystallographic data. Crystal structures for the PAC3 homodimer [16], domain-swapped PAC4 homodimer [18], and 20S proteasome [21] have been published. In addition, we newly determined a 0.96-Å high-resolution trigonal structure of the PAC3 homodimer (Figure S2a). The overall structure of the trigonal form was very similar to that of the tetragonal structure we have previously published, except for a loop comprised of residues 51–61 (Figure S2b), suggesting that it is mobile. Loop flexibility was also observed in the corresponding segment of the yeast ortholog Pba3 in its heterodimer with Pba4 [16] (Figure S3). Our nuclear magnetic resonance (NMR) relaxation data from the human PAC3 homodimer confirmed that the loop is indeed mobile and disordered in solution (Figure S4).

In the quintet-complex models, in addition to the interactions between PAC3-PAC4 and α5, based on the crystal structure of the yeast counterparts [16], the assembly chaperone interacted with the neighboring α4 and α6 subunits (Figure 3a). When the PAC3-4/α4/α5/α6 quintet complex model was superimposed onto the crystal structure of 20S CP, PAC3 and PAC4 make steric hindrance with β6 and β5, respectively, which possibly triggers the release of PAC3-4 from the α-ring upon binding of the β subunits onto the α-ring. A complex model, model A, based on the 2.00-Å PAC3 structure showed that the mobile loop was turned toward the solvent. Another model based on the 0.96-Å structure, model B, showed that the corresponding loop contacts α6. Apart from interactions involving this mobile loop, intermolecular contacts between the PAC3-PAC4 heterodimer and the proteasomal subunits are almost identical in the two models. Therefore, in the rest of this paper, we discuss the structural basis of the PAC3/PAC4-dependent α4-α5-α6 subunit assembly using model B.

**Figure 3.** Three-dimensional model of the quintet complex comprising PAC3, PAC4, α4, α5, and α6. (**a**) Complex model A, based on the 2.00-Å PAC3 structure. (**b**) Complex model B, based on the newly-determined 0.96-Å structure. The positions of the N- and C-termini are indicted. Overall and close-up views between PAC3 and α6 of the quintet-complex models are shown in the upper and lower parts of the figure, respectively. Putative α6-binding residues of PAC3, Ser55 and Val61 (see also Figure 4b), are highlighted in red in both models to highlight the conformational differences of the loop between the two models.

In this model, the interaction of α5 with the PAC3-PAC4 heterodimer is mediated by Gln70, Glu72, and Lys104 in PAC3, Arg48 in PAC4, and Glu95, His99, Tyr103, and Asp129 in α5 through electrostatic interactions and hydrogen bonds (Figure 4). Our pull-down data indicated that scPAC3/4 interacted most strongly with α5 and weakly with α4 and α6 (Figure 1). The model predicted additional

interactions between PAC3 and α6, and between PAC4 and α4 (Figure 4b,c). Specifically, Ser55, Lys80, and Asn81 of PAC3 form hydrogen bonds or electrostatic interactions with Ser110, Asp94, and Arg96 of α6, respectively. There are also predicted hydrophobic interactions of Val61, Phe85, and Val77 in PAC3 with Phe87, Phe97, and Leu93 in α6. Additionally, Asp70 and Arg85 of PAC4 have electrostatic interactions with Arg117 and Glu99 of α4, respectively (Figure 4c). Gln81 and Ile61 of PAC4 form hydrogen bonds and hydrophobic interactions with Ser93 and Val98 of α4, respectively.

To validate our docking model, we performed mutational experiments, especially focusing on the interaction between PAC3 and α6, which were specifically observed in the human proteins as compared with yeast counterparts. We constructed an scPAC3/PAC4 mutant in which putative α6-binding residues, Val77 and Lys80, of PAC3 are replaced with Ser and Ala, respectively. As expected, our mutational analysis indicated that mutations of Val77 and Lys80 of PAC3 impaired interaction with α6 but not with α4 and α5 (Figure 2b), confirming the validity of our docking model.

**Figure 4.** Predicted interaction interfaces between the PAC3-PAC4 heterodimer and proteasomal α4-α5-α6 subunits. (**a**) PAC3-α5. (**b**) PAC3-α6. (**c**) PAC4-α4 or α5 interfaces. Residues involved in the interactions are shown as stick representations. Potential hydrogen bonds and non-polar interactions are indicated as black and yellow dotted lines, respectively.

Although the overall structures of human PAC3 and PAC4 are similar to those of yeast Pba3 and Pba4 (RMSD = 1.9–2.1Å and 1.9–2.2 Å) respectively, their amino acid sequence similarities are low (PAC3 versus Pba3 11.0%; PAC4 versus Pba4 14.6%). The α-subunit contacting residues of human PAC3 and PAC4, as predicted by the model, are not well-conserved in the yeast orthologs Pba3 and Pba4 (Figure S5). Nevertheless, our model predicts that the complementarity at the interaction interfaces between the PAC3-PAC4 heterodimer and the proteasomal α4-α5-α6 subunits can be conserved in the yeast counterparts with a few exceptions. Perhaps the best example is the replacement of electrostatic interactions between Glu72 of PAC3 and His99 of α5 by non-polar contacts between Ala105 of Pba3 and Gln114 of α5. Therefore, despite the low sequence similarity, the overall interaction modes of the matchmaking chaperones with the proteasomal subunits appear to be conserved between humans and yeast. It is plausible that the conformational flexibility of the mobile 51–61 loop of PAC3, which carries the α6-contacting residues, contributes to the interaction adjustability.

In summary, we produced structural insights into the functional mechanisms of the PAC3-PAC4 heterodimer as a molecular matchmaker underpinning the α4-α5-α6 subcomplex during α-ring formation. These findings offer potential new approaches to the design of inhibitors against the protein-protein interactions involved in proteasome biogenesis.

#### **3. Methods**

#### *3.1. Sample Preparation*

Human proteasome α6 short isoform and α7 subunits were produced and purified as previously described [22,23]. Genes encoding the proteasome α1 and α4 subunits were subcloned into *Nde*I

and *Sal*I sites in pET28b, and the α2 gene was inserted into the pRSFDuet-1 vector using *Nde*I and *Xho*I restriction enzyme sites (Merck Millipore, Burlington, MA, USA). As for α4, 3xFLAG sequence (DYKDHDGDYKDHDIDYKDDDDK) was added at the N-terminus. The α3 and α5 genes were subcloned into the *Bam*HI and *Xho*I or *Sal*I sites of modified pCold-I and pCold-GST vectors (TaKaRa Bio Inc., Kusatsu, Japan), respectively, in which a factor Xa cleavage site was replaced with that of TEV protease. The PAC3 and PAC4 genes were subcloned into *Nde*I and *Xho*I sites in pET28b, in which the C-terminus of PAC4 was connected to the N-terminus of PAC3 through a (GGGS)4 liner. Standard polymerase chain reaction method was used to generate a V77S/K80A PAC3 mutant. *Escherichia coli* BL21-CodonPlus (DE3)-RIL (Agilent Technologies, Santa Clara, CA, USA) was used for all recombinant protein expression.

For the expression of recombinant proteins, the *E. coli* cells were grown in LB medium containing kanamycin or ampicillin. Briefly, the recombinant proteins were purified from the soluble fractions, except for α2, which was purified from the inclusion bodies and refolded using standard dilution methods.

Purification of these recombinant proteins was performed using affinity chromatography with Anti-FLAG M2 Affinity gel (Sigma-Aldrich, St. Louis, MO, USA), Ni+-charged Chelating Sepharose, or Glutathione Sepharose 4B, anion-exchange chromatography with RESOURCE Q resin, and size exclusion chromatography with Superdex 75 pg or 200 pg resins (GE Healthcare, Chicago, IL, USA). For NMR analyses, the PAC3 homodimer was expressed in *E. coli* cells which were grown in M9 minimal medium containing [13C]glucose (2.0 g/L) and/or [15N]NH4Cl (1.0 g/L), and purified using a previously-described protocol [12].

#### *3.2. Pull-Down Experiments*

The 3xFLAG-tagged α4, GST-fused proteasome α5-subunit (GST-α5), non-tagged or His6-tagged forms of proteasome α1, α2, α3, α6, and α7 subunits, and scPAC3/4 were used in the pull-down assays. For immobilization, 20 μg of His6-tagged scPAC3/4 or GST-α5 was applied to Ni2+-charged Chelating Sepharose or Glutathione Sepharose 4B (GE Healthcare) resins, respectively. The His6-scPAC3/4-immobilized resins were incubated with 50 μg of α1–α7 subunits for 2 h at 4 ◦C in an incubation buffer (20 mM Tris-HCl (pH 8.0) and 150 mM NaCl). For α5, the GST-α5-immobilized resins were incubated with 50 μg of α1–α4, α6, and α7 in the presence and absence of 50 μg of scPAC3/4 as described above. Since α7 makes a stable complex with α6 [19,20], the pull-down experiments containing α7 were performed separately. The resins were washed four times with the incubation buffer, which contains 60 mM imidazole in the His6-tag pull-down assays. Proteins bound to the Chelating or Glutathione Sepharose resins were eluted using 20 mM Tris-HCl (pH 8.0)/500 mM imidazole or 50 mM Tris-HCl (pH 8.0)/10 mM reduced glutathione, respectively, and analyzed by SDS-PAGE, stained with CBB.

#### *3.3. Crystallization, X-ray Data Collection, and Structure Determination*

For crystallization, purified non-tagged PAC3 homodimer was produced at a concentration of 8.0 mg/mL in 20 mM Tris-HCl (pH 7.5) and 150 mM NaCl. Crystals were obtained in a buffer containing 30% PEG2000 monomethyl ether and 0.1 M potassium thiocyanate with incubation at 20 ◦C for three to four days. Crystals were transferred into the reservoir solution and flash-cooled in liquid nitrogen. Diffraction intensities were integrated using XDS [24] and data scaling was carried out using AIMLESS [25]. The crystals of PAC3 belonged to space group *P*3121 and diffracted up to a resolution of 0.96 Å.

The trigonal structure of PAC3 was solved by the molecular replacement method using MOLREP [26], using the previously-reported tetragonal structure (PDB code 2Z5E) [16] as a search model. Automated model building and manual model fitting to electron density maps were performed using ARP/wARP [27] and COOT [28], respectively. Model refinement was carried out using REFMAC5 [29], and structure validation was conducted using MolProbity [30]. The data collection and refinement statistics of the PAC3 homodimer are summarized in Table S1. The molecular graphics were prepared using PyMOL (Schrödinger, New York, NY, USA).

#### *3.4. Computer-Aided Model Building*

The quintet-complex model comprising PAC3, PAC4, α4, α5, and α6 was created by several rounds of superimpositions using the coordinates of the human PAC3 homodimer (PDB codes, 2Z5E [16], and 6JPT (from this study)), the human PAC4 homodimer (PDB code: 5WTQ) [18], the human 20S proteasome (5LE5) [21], the yeast Pba3-Pba4 heterodimer (2Z5B) [16], and the yeast Pba3-Pba4/α5 ternary complex (2Z5C) [16]. The human PAC3-PAC4 heterodimer was created by superimposition of PAC3 and PAC4 monomers onto yeast Pba3 and Pba4, respectively. The resulting PAC3-PAC4 model was superimposed onto the yeast Pba3-Pba4 structure complexed with α5. Finally, to make a quintet complex model, the PAC3-PAC4-α5 (yeast) model was superimposed onto the human α5 subunit of the 20S proteasome. Subsequent protonation and energy minimization was performed using the CHARMm force field with the Discovery Studio program suite [31] (BIOVIA, San Diego, CA, USA).

#### *3.5. NMR Spectroscopy*

13C- and 15N-labeled non-tagged PAC3 homodimer (0.3 mM) and 15N-labeled non-tagged PAC3 homodimer (0.1 mM), dissolved in PBS (pH 6.8) containing 10% D2O (*v*/*v*), 1 mM EDTA, and 0.01% NaN3, were used for spectral assignment and relaxation experiments. All NMR data were acquired at 303 K using DMX-500, AVANCE-500, and AVANCE-800 spectrometers equipped with a 5-mm triple-resonance cryogenic probe (Bruker, Billerica, MA, USA). The NMR data were processed using TOPSPIN (Bruker) and NMRPipe [32]. Conventional 3D NMR experiments [33] were carried out for chemical shift assignments of the heteronuclear single-quantum correlation (HSQC) peaks originating from the PAC3 homodimer. Spectral assignments were carried out using SPARKY [34] and CCPNMR [35] software. 15N relaxation parameters, *T*1, *T*2, and 15N-1H heteronuclear nuclear Overhauser effect (NOE) were obtained at 303 K using an AVANCE-800 spectrometer and analyzed using the Protein Dynamics software in the Dynamics Center (Bruker).

#### *3.6. Accession Numbers*

The coordinates and structural factors of the crystal structure of the PAC3 homodimer have been deposited in the Protein Data Bank under accession number 6JPT. Backbone 1H and 15N chemical shift data of the PAC3 homodimer have been deposited in the Biological Magnetic Resonance Data Bank under accession number 27844.

#### **Supplementary Materials:** Supplementary materials can be found at http://www.mdpi.com/1422-0067/20/9/2231/s1.

**Author Contributions:** T.S. and K.K. conceived and designed the study; T.S. and E.K. performed protein design and preparation. T.S. performed the pull-down and crystallographic experiments; M.Y.-U. and K.O. performed the NMR experiment; K.T. contributed the materials and reagents; T.S. and K.K. mainly wrote the manuscript.

**Funding:** This work was supported in part by the Grants-in-Aid for Scientific Research (Grant Numbers JP26460051 to E.K. and JP25102008, JP15H02491 to K.K.) and Nanotechnology Platform Program (Molecule and Material Synthesis) from the Ministry of Education, Culture, Sports, Science and Technology, Japan. This work was also supported by the Joint Research by Exploratory Research Center on Life and Living Systems (ExCELLS) (ExCELLS program No. 18-402).

**Acknowledgments:** We thank Kumiko Hattori, Kiyomi Senda and Yukiko Isono for their help in the preparation of recombinant proteins. We also thank Hirokazu Yagi (Nagoya City University) for his useful discussion. The diffraction data set was collected at Osaka University using BL44XU at SPring-8 and Nagoya University using BL2S1 at Aichi Synchrotron Radiation Center (Japan). We acknowledge the synchrotron beamline staff and Institute of Drug Discovery Science at Nagoya City University for providing the data collection and computational facilities.

**Conflicts of Interest:** The authors declare that they have no competing financial interests.
