Structural landscape of AAA+ ATPase motor states in the substrate-degrading human 26S proteasome reveals conformation-specific binding of TXNL1

High resolution structures of TXNL1 bound to the human 26S proteasome.

We identified TXNL1 bound to endogenous proteasomes when we further processed a previously collected dataset of the human 26S proteasome during active degradation of a FAT10-fused Eos3.2 model substrate in the presence of the NUB1 cofactor²⁵. These affinity-purified proteasomes from HEK293 cells showed sub-stoichiometric amounts of TXNL1, whose binding is sensitive to the ionic strength the solution, such that proteasomes free of endogenous TXNL1 could be prepared by high-salt washes (Extended Figure 1A). To increase TXNL1 occupancy and the resolution of our cryo-EM structures, we saturated the human proteasomes with recombinant TXNL1 purified from E. coli (Extended Figure 1B). Robust redox activity of this recombinant TXNL1 was confirmed with an insulin reduction assay, in which resolving insulin’s disulfides leads to aggregation and a quantifiable increase in sample turbidity (Extended Figure 1C).

We used time-resolved cryo-EM to investigate the conformational landscape of the substrate-processing 26S proteasome 2 min after its incubation with excess of FAT10-Eos model substrate, NUB1 cofactor, and TXNL1 (Extended Figure 2). Particles were separated into two major groups, the resting state (RS, also known as the s1 state for yeast proteasomes) with ~ 414K particles, and the processing states (PS, also referred to as non-s1 states) with ~ 335K particles (Extended Figure 2). Deep 3D classification of the processing-state proteasomes resulted in at least 6 distinct conformations with global resolutions of 3 – 3.5 Å, showing the Rpt hexamer in distinct spiral-staircase registers (Extended Figure 2 and Extended Figure 3). We refer to these processing states based on the highest substrate-engaged Rpt subunit in the staircase as PS_Rpt1, PS_Rpt5, PS_Rpt4, PS_Rpt3, PS_Rpt6 and PS_Rpt2, represented by ~19.3k, ~112.2k, ~29.5k, ~32.2k, ~24.2k and ~73.1k particles, respectively (Extended Figure 2 and Extended Figure 3). All of them show the typical characteristics of the substrate-engaged proteasome, with a shifted and rotated lid subcomplex, the Rpt4/Rpt5 coil-coil positioned near Rpn10, a co-axial alignment of the ATPase ring with the open-gated 20S CP, and substrate polypeptide threaded through the central channel. We initially focus on the most abundant and highest resolution (~3 Å) state, PS_Rpt5 (Figure 1A, Extended Figure 3). In this state, all five subunits from Rtp5 at the top to Rpt3 at the bottom of the staircase are substrate-engaged through their pore-1 loop, while Rpt4 is the non-engaged seam subunit and at low resolution, likely due to a continuous motion from the bottom to the top position. Through 3D classification focused on the ATPase hexamer of PS_Rpt5 we could resolve sub-states A and B that differ in the position of Rpt4 (Extended Figure 2), and we will concentrate our subsequent discussion of structural features on state A, PS_Rpt5A.

The extra density above Rpn11 could be unambiguously assigned to the PITH domain of TXNL1 (Figure 1A,B). It contacts Rpn2 and the VWA (von Willebrand factor type A) domain of Rpn10 primarily through ionic interactions (Figure 1C), which explains the salt-sensitivity of this interaction. The N-terminal catalytic TRX domain of TXNL1 could not be resolved (Figure 1A), likely due to its flexible linkage to the PITH domain and consequent high mobility. Aligning AlphaFold models for the full-length TXNL1 with our structures places the TRX domain near the substrate entrance of the N-ring (Extended Figure 4C), and its mobility may aid in the reductive processing of a structurally diverse array of substrates. We observed the C-terminal tail of TXNL1 draped over Rpn11’s hydrophobic catalytic groove and the Insert-1 (Ins-1) loop that plays important role in binding the C-terminus of ubiquitin and regulating access to the deubiquitination active site (Figure 1D).

TXNL1 coordinates the active site zinc of Rpn11 but does not inhibit deubiquitination

TXNL1’s interaction with the catalytic groove of Rpn11 is supported by hydrophobic interactions of the tail residues F278 and M275 (Figure 1D). Interestingly, the C-terminus of this tail reaches into Rpn11’s active site, where a conserved terminal histidine coordinates the catalytic Zn²⁺ (Figure 1D, Extended Figure 4A). The Ins-1 region of Rpn11 is a critical regulatory segment that switches its conformation depending on the state of the proteasome and the presence of ubiquitin. At least three distinct Ins-1 states have been described: an ‘open’ loop conformation in the resting-state proteasome, an active β-hairpin conformation in which Ins-1 forms a small beta sheet with the C-terminus of ubiquitin to position it for isopeptide cleavage during substrate deubiquitination, and a curled up ‘closed’ or inhibitory conformation typically found in substrate-processing states of the proteasome (Extended Figure 4B)^16,18,19,32. The specific conformation and interactions of TXNL1’s C-terminal tail is compatible only with the closed state of Ins-1, whereas a ubiquitin-bound state would lead to steric clashes, and the open state would be unable to form the necessary contacts. TXNL1’s tail may thus bind the catalytic groove of Rpn11 only on substrate-processing proteasomes when there is no deubiquitination. Previous structural and functional studies suggested that polyubiquitinated substrates first use a more distally located moiety in their ubiquitin chain to interact with a proteasomal ubiquitin receptor, before the proximal ubiquitin binds to Rpn11, followed by insertion of the substrate’s flexible initiation region into the central channel for motor engagement^11,19,33. This engagement triggers the conformational switch of the 19S RP from the resting to processing states and allows subsequent co-translocational deubiquitination^11,34. To test whether TXNL1 inhibits deubiquitination, we prepared a model substrate (Eos-I27^V15P-tail) in which Eos3.2 was fused to the titin I27 domain with a destabilizing V15P mutation. The titin domain also carried a flexible C-terminal tail that contained a single lysine residue for Rsp5-mediated polyubiquitination and was long enough to allow degradation initiation. Measurements of proteasomal substrate degradation under multiple-turnover conditions, monitored through the decay of intrinsic Eos fluorescence, showed identical rates (0.54 +/− 0.06 min⁻¹) in the absence and presence of excess TXNL1 (Extended Figure 1D). Similarly, monitoring substrate deubiquitination and peptide-product formation by SDS-PAGE with a N-terminally fluorescein-labeled Eos-I27^V15P-tail substrate revealed no effect of TXNL1 on ubiquitin-dependent degradation (Extended Figure 1E), suggesting that TXNL1’s interaction with Rpn11 does not interfere with the removal of the affinity-conferring ubiquitin chain from a substrate upon motor engagement. All purified human proteasomes were treated with ubiquitin-propargylamine (PRG), a selective inhibitor of cysteine-dependent deubiquitination, which ruled out that other human proteasome-associated DUBs substituted for Rpn11 during ubiquitin-dependent degradation in the presence of TXNL1. Importantly, there was no indication for any in vitro degradation of TXNL1 (Extended Figure 1E), despite its location above the motor entrance and previous findings about endogenous TXNL1 being degraded upon arsenic treatment of mammalian cells³¹.

TXNL1 forms low-affinity interactions with resting-state proteasomes

Interestingly, although our cryo-EM data set showed a higher number of proteasome particles in the resting state, TXNL1’s tail was not detected interacting with Rpn11 in these proteasomes, where Rpn11’s Ins-1 loop is in the open conformation (Figure 2). Moreover, TXNL1’s PITH domain adopted an ensemble of orientations between Rpn2 and Rpn10 (Figure 2, Extended Figure 5, and Extended Figure 6). We classified particles into RS.1 and RS.2 states, which showed slight variations in the lid subcomplex, but no difference for the ATPase motor (Extended Figure 6). Local refinement of the PITH domain and 3D-variability analyses for both RS.1 and RS.2 allowed us to separate two major PIT-Hdomain states that we term forward (49.8 % of particles) and backward (40.5% of particles) based on the PITH domain rotation (Figure 2A,B, Extended Figure 6), along with a smaller number of particles showing the PITH domain in intermediate positions. In the absence of C-terminal tail binding to Rpn11, the PITH domain thus appears to form ‘blurry’ interactions with the resting-state proteasome, using just a few contact points on Rpn2 and Rpn10’s VWA domain to swivel between forward and backward positions by ~ 91 °. The forward position resembled the TXNL1 conformation we observed in the processing-state proteasomes, except for the C-terminal tail that did not contact Rpn11 and was therefore not resolved (Figure 2A, Figure 1A). Remarkably, the backward conformation utilizes overlapping, yet slightly different contact points on its PITH domain, Rpn2, and Rpn10, with again no involvement of the C-terminal tail (Figure 2C). These alternative orientations and their poorly resolved intermediate states thus provide an interesting example for “fuzzy” multi-mode interactions of a cofactor with a large protein complex that allow modulating affinity and function. 3D classification of a separate resting state of particles where Rpn1 is mobile and at lower resolutions also highlighted that forward and backward conformations were populated (Extended Figure 5).

The backward conformation thereby appears to avoid any interference with deubiquitination upon substrate engagement. Indeed, for proteasomes that adopted a conformation with similarities to the previously reported E_A2 or E_B states, which were proposed to reflect substrate deubiquitination^19,20, we observed TXNL1 in the backward conformation and, in addition, detected a low-resolution density bound to Rpn11 that resembles ubiquitin and could represent co-purified ubiquitin or a UBL domain of the FAT10 substrate (Extended Figure 5). Similar to the E_B state, these proteasomes show Rpn1 at a kinked angle relative to the ATPase ring, yet the ATPase ring has no translocating polypeptide in the central channel and is in a true resting-state conformation, like E_A2. However, in contrast to the E_A2 state that unexpectedly showed two ADPs bound to Rpt6 and Rpt5 across from each other in the hexamer, we observed five subunits bound to ATP and only the seam subunit Rpt6 in an ADP-bound state (Extended Figure 5). The backward conformation of TXNL1 is hence exclusively found in resting-state proteasomes, whereas the forward conformation is populated in both resting-state and actively degrading proteasomes, with the latter showing additional interactions of TXNL1’s C-terminal tail with Rpn11’s catalytic groove and active-site Zn²⁺.

TXNL1 binds to actively degrading proteasomes with high affinity

Considering the structural findings described above, we wondered whether TXNL1 had a binding preference and formed higher-affinity interactions with actively degrading proteasomes. We therefore N-terminally labeled TXNL1 with fluoresceine-amidite (FAM) and measured its fluorescence polarization after mixing with excess human proteasome in either the absence or presence of FAT10 substrate and NUB1 cofactor. While the substrate-free, resting-state proteasome sample showed only low fluorescence polarization, actively FAT10-substrate degrading proteasomes caused a strong polarization increase (Figure 3A). Remarkably, this signal decreased as the substrate was consumed and the proteasome conformational equilibrium shifted back to the resting state, indicating the dissociation of TXNL1. Performing these polarization experiment with a TRX-domain deletion (PITH) or PITH-domain deletion (TRX) confirmed that this differential binding to human proteasomes is solely dictated by the Rpn11-interacting PITH domain (Extended Figure 7A). Furthermore, when using a 20S CP inhibitor to accumulate undigested FAT10 substrate in the degradation chamber and thereby stall substrate translocation through the 19S RP, we observed a constantly high polarization, indicative of TXNL1 tightly binding proteasomes that were trapped in the processing states (Extended Figure 7A). Titrating these stalled, substrate-engaged proteasomes allowed us to determine a K_D of ~ 35 nM for TXNL1 binding (Extended Figure 7B). By contrast, the TXNL1 titrations we performed with resting-state proteasomes for the cryo-EM structure determination let us estimate a K_D in the tens of micromolar range, at least 2 orders of magnitude higher than for the actively degrading states. Hence, TXNL1 preferentially binds the substrate-processing human proteasomes with high affinity, whereas it switches to low-affinity “fuzzy” binding and dissociates when the proteasome is in the resting conformation.

Time-resolved cryo-EM reveals conformation-specific TXNL1-binding preference

We further processed a previously collected dataset for the human 26S proteasome prepared 30 s after initiating the NUB1-mediated degradation of a FAT10-mEos3.2 substrate. Besides a large number of resting-state proteasome particles (77.3 %), we identified the substrate-degrading proteasome in the same six spiral-staircase ATPase states as observed for the 2 min time point (Extended Figure 8A). The resting-state proteasomes did not show PITH-domain density, consistent with the low concentration of the sub-stoichiometrically co-purified TXNL1 and its low-affinity binding to the resting state. For the processing-state proteasomes, we masked each Rpt subunit and used local refinements to improve the resolution of the ATPase hexamer for model building (Extended Figure 8A). The staircase states PS_Rpt1, PS_Rpt5, PS_Rpt4, PS_Rpt3, PS_Rpt6 and PS_Rpt2 were represented by ~26.7k, ~66.8k, ~16.4k, ~15.6k, ~10.7k and ~31.2k particles, respectively, and refined to moderate global resolutions of 3 – 4.5 Å (Figure 3B; Extended Figure 8A). There were no complete TXNL1 occupancies in these states, but alignments focused on the PITH domain allowed us to achieve high resolution (~ 2.8 Å) and the unambiguous assignment of TXNL1 (Figure 3B; Extended Figure 8B,C). 3D classification of each PS conformation sorted particles into TXNL1-bound and TXNL1-unbound, and we noted a clear bias of PS_Rpt2 for being TXNL1-free (Figure 3C, Extended Figure 8B). Interestingly, PS_Rpt2 differs from the other substrate-processing states by showing a larger gap between the N-terminal coiled coil of Rpt4/Rpt5 and Rpn11’s Ins-1 region (Extended Figure 4D). This larger gap allows an Ins-1 conformation with possibly higher dynamics that may destabilize its interactions with the C-terminal tail of TXNL1. TXNL1 therefore binds PS_Rpt2 similar to the resting-state proteasome, with low affinity and not involving its tail. It is tempting to speculate that the Ins-1 conformation in PS_Rpt2 is less inhibitory for deubiquitination and more readily transitions to the active hairpin state that forms a beta sheet with the C-terminus of ubiquitin. PS_Rpt2 may therefore represent a processing state that is suited for the removal of ubiquitin chains during substrate translocation or unfolding.

Overall, the structures of the degrading human 26S proteasome in the presence of sub-stoichiometric TXNL1 amounts further support our findings that TXNL1 forms high-affinity interactions only with the actively processing proteasome states, except for PS_Rpt2, in which low-affinity binding may prevent TXNL1’s interference with co-translocational deubiquitination.

Capturing an unfolding intermediate of the Eos substrate

Using the dataset for proteasomes with saturating TXNL1 at 2 min after initiating FAT10-substrate degradation, we visualized the PS_Rpt2 conformation with an unfolding intermediate of the Eos substrate above the entrance to the ATPase channel and pulled against Rpn11 (Figure 4A). N-terminal of the Eos domain we were able to model the linker to FAT10’s C-terminus and continuously all amino acids up to FAT10’s residue 154, as they span through the N-ring and the ATPase spiral staircase (Figure 4A,B). The Eos domain is partially unfolded, with its FAT10-attached N-terminal b1 strand extracted from the beta barrel and the neighboring β2-β3 and β5-β6 hairpins strongly distorted (Figure 4C). Interestingly, β5 and β6 appear to maintain their hydrogen bonds as a hairpin, despite being pulled against Rpn11 and bent to parallel the extracted β1 strand going into the ATPase pore (Figure 4E,F). We were also able to model the Eos chromophore, a cyclic imidazole group formed from residues His227, Try228, and Gly229 (Figure 4D). Due to the displacement of β1, β5, and β6 from the barrel, the environment of the chromophore is strongly distorted (Figure 4E). It is therefore likely that Eos loses its fluorescence while forming this intermediate, whose subsequent further unfolding appears to be rate-limiting for the degradation of the FAT10-Eos substrate. Indeed, in previous biochemical studies we found that the Eos fluorescence during degradation under single-turnover conditions decayed with a time constant of τ = 18 s, whereas multiple-turnover degradation showed a time constant of τ = 50 s, likely determined by unfolding of the non-fluorescent intermediated²⁵. The Eos unfolding intermediate uses hydrophobic residues in β1, β5, and β6 that are normally part of the hydrophobic core to interact with the surface of Rpn11. In particular, M172, M274, and F279 engage in hydrophobic interactions with Rpn11’s Ins-1 region and V125 of the α3 helix (Figure 4F), thus stabilizing the intermediate in a particular position on Rpn11 that overlaps with the binding site of the TXNL1 C-terminal tail (Figure 4G). Future studies with other substrates will have to address whether this surface area of Rpn11 is specifically contacted only by Eos or more generally used to pull proteins against and potentially assist in their mechanical unfolding. Although PS_Rpt2 is not the most abundant conformation of the substrate-degrading proteasome in our datasets (Extended Figure 2), it is the only one where an unfolding intermediate of the FAT10-Eos substrate could be visualized. Furthermore, the spiral staircase state with Rpt2 at the top was observed in previous structural studies of the yeast and human proteasomes with stalled substrates^18,20, suggesting that this register of the ATPase motor may play important roles during substrate unfolding or deubiquitination. Importantly, even PS_Rpt2 proteasomes without the Eos unfolding intermediate showed a major reduction in TXNL1 binding (Figure 3C), indicating that high-affinity TXNL1 tail binding is disfavored by the PS_Rpt2 conformation itself and not simply prevented by steric hindrance with the Eos unfolding intermediate pulled against Rpn11.

Asymmetric ATP hydrolysis in the Rpt hexamer

For both datasets of the human proteasome at 30 s and 2 min after substrate addition we observed non-substrate-engaged proteasomes in the typical resting-state conformation, with Rpt3 at the top of the staircase, Rpt2 at the bottom, and Rpt6 as the seam subunit bound to ADP, while all other subunits were ATP-bound. The substrate-engaged proteasomes in both datasets showed the six staircase conformations represented by vastly different particle populations (Figure 5, Extended Figure 9C). The resolutions were generally high enough to reliably determine the identities of the bound nucleotides and distinguish between ADP and ATP (Extended Figure 9B). For some of the more mobile seam subunits with lower resolution overall and for the nucleotide, we assumed ADP based on the subunit position in the staircase. PS_Rpt5 was clearly the dominant state with particle numbers in the data sets ~ 2 times higher than expected if all states were equally likely, and 4–5 times higher than three of the least populated motor states (Figure 5). Assuming that the numbers of particles in a particular conformation correlate with the lifetime of that state during ATP hydrolysis and the spiral-staircase progression in the Rpt hexamer, we postulate that the proteasomal ATPase motor functions by an asymmetric firing mechanism. This is further supported by varying distributions of nucleotide states, with processing states containing 4 ATPs and 2 ADPs, 3 ATPs and 3 ADPs, or only 2 ATPs and 4 ADPs (Figure 5). Assuming a counterclockwise progression of ATP hydrolysis, conformational changes, and nucleotide exchange events around the Rpt hexamer, we discuss the processing states in the corresponding counterclockwise order by which they may transition into each other. Although we describe discrete conformations for an almost complete cycle of ATP hydrolysis, it should be noted that we are likely missing rarer conformations and cannot assess continuous motions or very short-lived intermediates at high enough resolution for modeling, especially for the seam subunits.

The most common conformation PS_Rpt5 showed five subunits, Rpt5, Rpt1, Rpt2, Rpt6, and Rpt3, engaged with the substrate polypeptide, and Rpt4 as the disengaged seam subunit (Figure 5; Extended Figure 9A). At the top of the staircase Rpt5, Rpt1, and Rpt2 are ATP-bound, whereas Rpt6, Rpt3, and Rpt4 at the bottom have ADP in their active sites. Like for the 4D state of the substrate-engaged yeast proteasome¹⁸, this would indicate that Rpt4 is the next subunit to move to the top of the staircase and exchange ADP for ATP. However, the 4D state and the E_D2.1 state of the human proteasome²⁰ both had 4 ATP and 2 ADP bound to their Rpt subunits, suggesting that in PS_Rpt5 an additional ATP-hydrolysis event had occurred in Rpt6 without being coupled to the movement or nucleotide exchange of Rpt4.

The PS_Rpt4 state has not been previously observed in structural studies of the substrate-bound yeast or human 26S proteasomes. Interestingly, we detected only 2 ATPs, bound to Rpt4 and Rpt5, while the other 4 Rpt subunits were bound to ADP (Figure 5), which is a strong deviation from the conventional views of always having 4–5 ATP-bound and 1–2 ADP-bound or empty subunits. Again, this suggests that ATP hydrolysis can occur further upwards in the spiral staircase and ahead of the large movements for translocation and nucleotide exchange of the substrate-disengaged seam subunit. These findings likely indicate a burst phase in ATP hydrolysis, whereby additional hydrolysis events occur faster than the movement of seam subunits toward the top of the staircase, possibly because substrate translocation and staircase progression are held up by a folded substrate domain at the entrance of the ATPase motor. ADPs thus accumulate in up to 4 subunits, and their release may lead to several subunits moving from the bottom to the top of the staircase in rapid succession or even simultaneously. Our results challenge conventional models for hand-over-hand translocation that were based on structures of proteasomal or other AAA+ ATPase with stalled substrates and proposed a strict coupling of individual ATP-hydrolysis, subunit-movement, and nucleotide-exchange events¹⁸. Interestingly, the preceding PS_Rpt5 is the most abundant and likely most long-lived conformation of the hexamer, and its transition to PS_Rpt4 is linked to two ATP hydrolysis events, suggesting that the movement, nucleotide exchange, and substrate engagement of Rpt4 are slow steps during substrate processing. Consistently, Rpt4 was previously identified as particularly important for substrate degradation¹⁴, and our recent single-molecule studies revealed major unfolding defects and increased substrate release when Rpt4’s pore-1 loop was mutated³⁵. We thus propose that Rpt4 plays a critical important role for mechanical unfolding, possibly mediating a power stroke of the Rpt hexamer. The transition to the subsequent PS_Rpt3 state involves Rpt3 moving to the top of the staircase and exchanging ADP for ATP. As there are no additional hydrolysis or exchange events, this leads to a hexamer with 3 ATPs and 3 ADPs (Figure 5). Due to the lower resolution (~3.5 – 4Å), we cannot interpret the side-chain orientations for Rpt3’s pore-1 loop, yet the backbone appears positioned close to the substrate and ready to engage (Extended Figure 9A,D). Consequently, two subunits, Rpt6 and Rpt3, are disengaged, while the other four contact the substrate through their pore-1 loop.

The following PS_Rpt6 state also showed 3 ATPs and 3 ADPs bound to the ATPase hexamer, meaning that during the transition from PS_Rpt3 one hydrolysis event occurred in Rpt5, while Rpt6 exchanged ADP for ATP. Interestingly, there are visible gaps at all ADP-bound subunit interfaces, Rpt5/Rpt1, Rpt1/Rpt2, and Rpt2/Rpt6, and Rpn1 is highly mobile, likely because its anchoring Rpt1 and Rpt2 subunits are at the staircase seam and simultaneously disengaged from the substrate (Extended Figure 9A,D).

In the subsequent PS_Rpt2 state, Rpt2 and Rpt6 are ATP-bound, while the other four subunits are bound to ADP, meaning that two hydrolysis events and one nucleotide exchange occurred during the transition from PS_Rpt6. Rpt2, Rpt6, Rpt3, and Rpt4 are in contact with the substrate, whereas Rpt5 and Rpt1 are disengaged (Figure 5). Interestingly, Rpt1 is rotated up towards the top of the staircase (at level with Rpt2), with its pore-1 loop residue Y249 at ~ 20 Å distance from the substrate and potentially sequestered by interaction with Rpt2 (Extended Figure 9A,D). Hence, during the transition from PS_Rpt6, Rpt2 and Rpt1 moved together to the top, possibly due to their coupling by the associated Rpn1 subunit. Rpt5 at the bottom of the staircase is disengaged, yet very close to the substrate (~ 8 Å, Extended Figure 9A,D). A similar staircase conformation with disengaged Rpt5 and Rpt1, referred to as 1D*, was observed for the yeast proteasome¹⁸, yet assumed to be off pathway. Here we detected three states, PS_Rpt3, PS_Rpt6, and PS_Rpt2, with two disengaged subunits, and similar staircase states, like E_D0 and E_D1²⁰, were previously reported for the human proteasome. Transitions with two disengaged seam subunits may therefore be common or even an important principle for substrate translocation that includes bursts of hydrolysis and Rpt-subunit movements. In PS_Rpt1, we again observed five subunits, Rpt1, Rpt2, Rpt6, Rpt3 and Rpt4, in contact with the substrate and Rpt5 as the seam subunit, albeit at the top of the spiral staircase, ATP-bound, and ready to engage (Figure 5, Extended Figure 9A,D). This state thus immediately precedes PS_Rpt5. There are four subunits, Rpt5, Rpt1, Rpt2, and Rpt6, bound to ATP, with Rpt3 and Rpt4 ADP-bound. PS_Rpt1 is therefore directly comparable to the previously reported human E_D2.0²⁰ and yeast 5T¹⁸ states.

Cloning

TXNL1 (Trp32) was gene synthesized with the following amino acid sequence that was codon-optimized for E. coli expression. TXNL1 DNA was ligated into an expression vector with an N-terminal His-TEV-FLAG (MGSSHHHHHHSSGENLYFQGHMDYKDDDDK), so that a GHMDYKDDDDK overhang was left on the N-terminus of TXNL1. All TXNL1 mutants and truncations were generated by site-directed mutagenesis with Q5 polymerase, phosphorylation, and ligation.

TXNL1 amino acid sequence MVGVKPVGSDPDFQPELSGAGSRLAVVKFTMRGCGPCLRIAPAFSSMSNKYPQAVFLEVDVHQCQGTAATNNISATPTFLFFRNKVRIDQYQGADAVGLEEKIKQHLENDPGSNEDTDIPKGYMDLMPFINKAGCECLNESDEHGFDNCLRKDTTFLESDCDEQLLITVAFNQPVKLYSMKFQGPDNGQGPKYVKIFINLPRSMDFEEAERSEPTQALELTEDDIKEDGIVPLRYVKFQNVNSVTIFVQSNQGEEETTRISYFTFIGTPVQATNMNDFKRVVGKKGESH

The mEos3.2-titin-tail construct was generated by Gibson assembly of His-TEV mEos3.2 before titin(V15P)-tail-intein-chitin binding domain (CBD). The titin-tail has one lysine followed by a PPPY motif for Rps5 ubiquitination. The P141C, C171A mEos-titin-tail construct was generated by site directed mutagenesis.

Purification of TXNL1

TXNL1 was expressed in E.coli BL21* grown in TB medium by induction with IPTG (250 μM) when cells reached OD_600nm 0.6 – 0.8, followed by shaking (200 rpm) overnight at 16 °C. E.coli cells (from ~2 L) were resuspended in 30 mL of lysis buffer (60 mM HEPES pH 7.4, 500 mM NaCl, 5% glycerol, 0.5 mM TCEP, 20 mM imidazole) with freshly added benzonase, before sonication and clarification to remove insoluble matter. The supernatant was then passed through Ni-NTA resin (~2 mL) by gravity at room temperature (for ~ 20 mins), before extensive washing to remove contaminants (10 column volumes, sequentially with lysis buffer). TXNL1 was eluted with lysis buffer supplemented with 250 mM imidazole (~ 10 mL). Protease (His-TEV, 250 μg) was added and TXNL1 was dialyzed into 60 mM HEPES pH 7.4, 200 mM NaCl, 5% glycerol, 0.5 mM TCEP at room temperature for 2 hours, followed by overnight dialysis in fresh buffer at 4°C. Dialyzed TXNL1 was flown over equilibrated Ni-NTA resin and the flow through containing cleaved TXNL1 is collected and concentrated for gel filtration using an SD75 16/600 at 4°C. Protein is stored at −80°C in 60 mM HEPES pH7.4, 150 mM NaCl, 25 mM KCl, 10 mM MgCl, 0.5 mM TCEP as single use aliquots (typically ~350 μM). Protein concentration is estimated using 280 nm.

hs26S proteasome purification

As previously described, human 26S proteasome were purified from HTBH-Rpn11 tagged human HEK293 adapted to suspension culture^25,44. HEK293 cells were grown at 8 % CO₂, 37 °C, and 120 rpm shaking in FreeStyle^™ 293 Expression Medium with 2% (v/v) FBS. Cells were passaged twice a week at 5 × 10⁵ and newly thawed cells were grown with puromycin. Cell pellets from 4 L of culture were resuspended in 60 mM HEPES pH 7.4, 25 mM NaCl, 25 mM KCl, 5% glycerol, 10 mM MgCl₂, 5 mM ATP, and 0.5 mM TCEP, supplemented with benzonase and EDTA-free proteasome inhibitor tablets. After lysis by a dounce homogenizer (usually 15X), lysate was sonicated on ice with low amplitude (20%), clarified for 60 min at 4°C, and flowed over pre-equilibrated Pierce^™ High-Capacity Streptavidin Agarose (3 mL of resin). For preparing low salt treated hs26S proteasomes, resin was sequentially washed with lysis buffer 5X, whereas for high salt treated hs26S, resin was washed 4X with lysis buffer supplemented with 300 mM NaCl and 1X with lysis buffer without extra NaCl. Protein was eluted by TEV protease cleavage for 60 mins at room temperature and gentle agitation followed by concentration of the elution to ~ 250 μL. Samples were clarified at 21,000 x g for 15 min before fractionation using an S6 increase 10/300 column in lysis buffer. Fractions containing 26S proteasome were concentrated, and concentrations were estimated in a Bradford assay with BSA as a standard and assuming a molecular weight of 2.6 MDa for the proteasome. Typically, 10 μL ~ 4 μM aliquots were snap frozen in liquid N₂ and stored at −80°C.

Purification of Eos-Titin-tail substrate

Substrate and mutants were expressed as a His-TEV-Eos-titin I27(V15P)-tail-intein-CBD fusion in E. coli BL21* by growing cells at 37 °C to OD_600nm of 0.6, before cooling cells to 16 °C for ~1 hour and induction with IPTG at 0.25 μM. Cell were harvested after 18 hours of induction, followed by lysis with sonication (12 cycles of 70% amplitude, 10 s on and 45 s off on ice) in 50 mM HEPES, 300 mM NaCl, 5% glycerol, 10 mM MgCl₂, 20 mM imidazole with freshly added EDTA-free proteasome inhibitor tablets and benzonase. Lysate was clarified before flowing over Ni-NTA resin by gravity. Ni-NTA resin was washed extensively with lysis buffer without inhibitor tablets and benzonase. Protein was eluted with lysis buffer supplemented with 250 mM imidazole and bound to chitin resin, before washing with 5 column volumes of lysis buffer. Resin was suspended in lysis buffer overnight with the addition of 50 mM DTT and His-TEV protease. Columns were moved to room temperature for one hour, before collecting elution and flowing over freshly equilibrated Ni-NTA resin to remove uncleaved protein and His-TEV protease. The cleaned elution was concentrated using Amicon concentrator (30 kDa cutoff) before fractionation using an SD200 16/600 column in gel-filtration buffer (60 mM HEPES pH 7.4, 25 mM NaCl, 25 mM KCl, 5% glycerol, 10 mM MgCl₂, 0.5 mM TCEP). After concentrating to ~ 10 mg/mL, protein was snap frozen liquid N₂ as single-use aliquots and stored at −80°C

Additional proteins previously established

Fat10, NUB1, Sortase, TEV protease, 3C protease, His-ULP1 protease, Ubiquitin, Rsp5, His-TEV-mE1, Ubch7 and Ub-intein were expressed and purified using previously established protocols^11,25. Ubiquitin-Propargylamide (Ub-prg) was purified as previously described²⁵.

Western blot

Samples were separated by SDS-PAGE and transferred to PVDF membranes (Thermo Scientific), before blocking in 5% milk TBS-T. TXNL1 primary antibodies (15289–1-AP, Thermo Scientific) were incubated at room temperature with the membranes for 90 min in 5% milk TBS-T, before 5× 5 min washing (~10 mL in TBS-T) and incubation with secondary Rabbit-HRP antibodies (ab79773, Abcam) for 60 min. After more washes, blots were visualized using HRP after 2 min of incubation.

Sortase labelling

Fluoresceine-amidite (FAM)-labeled peptide (FAM-HHHHHHLPETG) was added N-terminally to TXNL1 and the Eos-titin-tail substrate using Sortase ligation at room temperature for 30 min. Proteins (30–100 μM) were incubated in 50 mM HEPES, pH 7.4, 100 mM NaCl, 10 mM CaCl₂, 0.5 mM TCEP with 500 μM peptide and 25 μM Sortase in 250–450 uL reaction volumes. Labelled proteins were bound to Ni-NTA, washed several times with 60 mM HEPES pH 7.4, 25 mM NaCl, 25 mM KCl, 5% glycerol, 10 mM MgCl₂, 20 mM imidazole, and eluted with the same buffer supplemented with 250 mM imidazole. Labelled proteins were concentrated and fractionated by gel filtration using an SD75 increase 10/300 in 60 mM HEPES pH 7.4, 25 mM NaCl, 25 mM KCl, 5% glycerol, 10 mM MgCl₂ to remove excess peptide. Labelled substrates were concentrated, snap frozen in liquid N₂, and stored at −80°C as 10 μL single-use aliquots. For TXNL1 constructs, 0.5 mM TCEP was included throughout the purification and in the storage buffer.

Insulin reduction assay

When reduced, the B chain of insulin will self-aggregate and precipitate, leading to increased turbidity as measured by absorbance at 650 nm. Insulin (I0516, Sigma Aldrich) at ~1.7 mM was diluted to 30 μM in reaction buffer (60 mM HEPES, pH 7.4, 150 mM NaCl, 5% glycerol with 1 mM DTT, unless otherwise stated) with TXNL1 protein. A_650nm was measured every 30 seconds for at least 2 hours at 30°C.

Substrate ubiquitination and degradation

2X reactions of His-mE1 (1 μM), Ubch7 (10 μM), Rsp5 (6 μM) and Ubiquitin (200 μM) were incubated with Eos-titin-tail substrate (5 μM) for ~30 mins at room temperature in ubiquitination reaction buffer (60 mM HEPES, pH 7.4, 25 mM NaCl, 25 mM KCl, 10 mM MgCl, 0.5 mM TCEP, 5 mM ATP). Ubiquitinated Eos-titin-tail substrate was mixed with 2X concentrated hs26S proteasome (which was treated before with 2.5 μM Ub-prg for 30 mins) in a pre-warmed 384-well plate, and degradation was measured in a BMG Labtech CLARIOstar plate reader at 30°C by monitoring the emission at 520 nm after excitation with at 500 nm. To monitor formation of peptide products by SDS-PAGE analysis, N-terminally FAM-labelled ubiquitinated Eos-titin-tail substrate was used, and samples were boiled prior to gel loading to ensure Eos was fully unfolded and no longer fluorescence.

^FAMTXNL1 proteasome binding assay

FAM-labeled TXNL1 and hs26S proteasome were incubated at 2X concentration, such that once diluted 1:1 with buffer or substrate, their final concentrations were at 50 nM for TXNL1 (unless otherwise stated) and 150 or 500 nM for hs26S proteasome. FAT10 substrate was prepared for degradation by incubating the FAT10 with Nub1 at 1:1.2 ratio for at least 15 min at 2 X concentration. Reactions were initiated by adding buffer or the pre-incubated substrate to TXNL1 with or without hs26S proteasome, and fluorescence polarization was recorded by measuring the emission at 535 nm after excitation at 480 nm in a preheated (30 °C) 384-well black plate (Costar) using a BMG Labtech CLARIOstar plate reader.

Cryo-EM Sample preparation and data collection

Human 26S proteasome (4 μM, high-salt washed, see proteasome purification method) with recombinant FLAG-TXNL1 (16 μM) was diluted 1:1 with preincubated FAT10-Eos (12 μM) and NUB1 (16 μM) for ~ 105 s before cryo-plunging. Sample buffer used was 20 mM HEPES pH 7.4, 25 mM NaCl, 25 mM KCl, 5 mM MgCl₂, 2 mM ATP, 2.5% glycerol and 0.02% NP-40. Final samples had 2 μM 26S proteasome, and 3.5 μL were applied to glow discharged (glow discharging: 25 mA, 25 seconds) UltrAufoil^® R 2/2, 200 Mesh, Au grids (Q250AR2A, Electron Microscopy Sciences). Grids were plunge frozen in liquid ethane (cooled with liquid nitrogen) using a Vitrobot (ThermoFisher) at 12°C with 3 s of blot time. Data were collected as previously described²⁵, where briefly, clipped grids were transferred to a Titan Krios transmission electron microscope operated at 300 KeV (ThermoFisher) equipped with a Gatan K3 and an energy filter (GIF quantum). Images were taken using SerielEM at a nominal magnification of x81,000 (1.048 Å pixel size) in super resolution mode with a defocus ranging from −0.5 – 1.7 μm. We collected 50 frames per shot with a total electron dose ~50 e⁻ Å⁻²s⁻¹. A total of 9,239 movies were collected for the high-salt washed proteasomes with excess TXNL1 and 2 min after addition of the NUB1/FAT10-substrate. The dataset for low-salt washed proteasomes with sub-stoichiometric amounts of TXNL1 and at 30 s after substrate addition was previously collected²⁵.

Cryo-EM data processing

Data processing was performed using CryoSparc v4.4⁴⁵ and datasets were processed separately. Movies were corrected with patch motion and patch CTF, before blob picking and particle extraction with a box size of 600, but binned by 2x. Particles were sorted by multiple rounds of 2D classification before ab initio 3D reconstruction and heterogenous refinement with 10 classes and a refinement box size set to 128. Each class was refined by homogenous refinement and high-resolution classes of 26S and 30S proteasomes were selected, pooled, and extracted with a 600-box size and no binning. All particles were refined using non-uniform refinement in C2.Particles were symmetry expanded in C2 followed by recentering, such that the proteasome regulatory particle is in the center of the box, and subsequently extracted with a box size of 340. Particles were then subjected to 3D-reconstruction to ensure their correct position and subjected to heterogenous refinement with 10 classes. Classes were pooled based on the global conformation of the RP, either belonging to resting state (RS) or processing states (PS) and refined using non-uniform refinement.

To separate processing-state proteasomes into individual states, a generous mask for the RP was generated for local refinement and aligned particles were subject to alignment-free 3D classification (10 classes, filtered at 6 Å), from which 4 dominant classes emerged. A mask for the ATPase domains was then generated for each class, followed by local refinement and another round of 3D classification (10 classes, filtered to 6 Å). Each of the 10 classes was refined with local refinement and similar states (no obvious differences) were combined, such that 6 ATPase motor conformations were achieved. ATPase 3D classification for PS_Rpt5 resulted in different classes with varying positions of the seam subunit Rpt4 and the PS _Rpt1. For PS_Rpt6 multiple poor-quality particles were included likely due to lower resolution. Therefore, another round of heterogenous refinement and local refinement of the ATPase was performed and resulted in a relatively high resolution PS_Rpt6 conformation, while the rest of the particles appears to be junk or low-resolution data. For each processing conformation an RP mask was generated, and local refinement was used to generate final unsharpened maps. An ATPase motor mask was used for processing maps from the 30 second dataset²⁵. While we describe distinct states of the 26S proteasome, it is important to note that alignments of particles and 3D reconstructions can be dominated by more rigid parts of a protein complex. In addition,within each discretestate, there is continuous motion present throughout the 26S proteasome complex, and this can vary depending on the defined conformational state described. For example, Rpn10 and Rpn1 can be at lower resolutions relative the AAA+ motor. It should be noted that due to current classification techniques particles are sorted based on their probability of belonging to each assigned state, and more discrete states might be observed when increasing particle numbers. Although several groups previously generated composite maps from individually locally refined maps^18,20, we did not use this approach. However, for guidance in model building and map representation and to aid with the anisotropy in map resolution, we used maps sharpened with deepEMhancer⁴⁶, but always validated model interpretations with unprocessed maps and locally refined maps.

For the 30-second dataset with partial occupancy of TXNL1 on Rpn11, we also generated masks that encompassed TXNL1, Rpn2, Rpn11 and Rpn10 and used local refinement to improve resolution. Using these locally refined maps, alignment-free 3D classification was used followed by local refinement of each separate class based on the presence of absence of TXNL1. Particles were pooled into two groups, with and without TXNL1, and subjected to local refinement using the same mask for TXNL1. The particle numbers for the presence and absence of TXNL1 were used to roughly calculate the ratio of TXNL1 binding to a particular processing-state conformation. However, it should be noted that numbers are estimates due to lower resolutions in certain classes and particles being sorted based on the probability of belonging to a certain class.

For the resting-state proteasome, RP-aligned particles were subjected to alignment-free 3D classification, from which 4 major classes emerged. Each class was refined by non-uniform refinement, whereby Class1 was considered junk due to broken RP, Class2 was characterized by a flexible Rpn1, and Class3 and Class4 were charactered as high resting state conformations with a shift in the position of the RP when compared. Class 2 was further processed by aligned local refinement with a mask encompassing Rpn10, Rpn2, Rpn11 and the density in between. Aligned particles were subject to alignment-free 3D classification, which generated 10 classes of varying resolutions. We chose 3 classes with interesting features to further refine. Two classes showed the PITH domain of TXNL1 in forward and backward conformations and were refined through local refinement with a whole RP mask. One conformation appeared similar to so-called ‘deubiquitination states’ E_A2 and E_B¹⁹ and showed density near Rpn11, which is likely a ubiquitin-like domain. After local refinement with a whole RP mask, we low-pass filtered the whole RP-refined map to 6 Å and docked the atomic model for the E_B state (PDB ID 6MSE), using ChimeraX⁴⁷. Local refinement with a mask encompassing Rpn11, the extra density, the coiled coil of Rpt4/5, Rpn10 and Rpn2, and subsequent 3D classification did not further improve the resolution. For Class 3 and Class4 with a slight shift in the RP, there was an ensemble of density where we would expect the PITH domain of TXNL1. We kept Class3 and Class4 separate and used local refinement to align particles using a mask encompassing Rpn2, Rpn10, Rpn11 and the ensemble of density in between those subunits. CryoSparc 3D variability and cluster analysis⁴⁸ was used to separate out particles based on the position of the PITH domain. We could characterize two major conformations, a forward and backward conformation. Those states, named RS.1 TXNL1 state 1, RS.1 TXNL1 state 2, RS.2 TXNL1 state 1, and RS.2 TXNL1 state 2, were refined by local refinement with a whole RP mask, and maps sharpened with deepEMhancer⁴⁶ along with unprocessed maps were used for representing density.

Cryo-EM model building and model visualization

Previously built atomic models for human 26S proteasomes²⁵ were used as starting models for each proteasome structure. Atomic models were generated for resting-state proteasome maps: RS.1 TXNL1 state 1, RS.1 TXNL1 state 2, RS.2 TXNL1 state 1, and RS.2 TXNL1 state 2. Briefly, models were fit using ChimeraX Fit to map and individual subunits were manually adjusted using the same function. TXNL1’s PITH domain was truncated from a model generated by AlphaFold⁴⁹. Phenix refine⁵⁰ with simulated annealing and morphing turned on for the first round was used to adjust models, followed by several rounds of iterative model building in Coot⁵¹ and real space refinement in Phenix without simulated annealing and morphing. For atomic models of the processing-state proteasomes, deepEMhancer maps were used to guide model building. For AAA+ domains, Rpts were separated into multiple parts and rigid bodies using ChimeraX Fit to map, followed by several rounds of manual adjustment in Coot. Once roughly assembled and fit, models were subjected to multiple rounds of manual adjustment in Coot and real space refinement with Phenix (refinement was typically done using unprocessed maps). The chromophore for Eos (code CR8 in Coot) was fit and adjusted in Coot, followed by Phenix real space refinement using a restraints cif file generated in Phenix. PyMol (The Molecular Graphics 981 System, Version 1.8, Schrödinger, LLC; http://www.pymol.org/), UCSF chimera⁵² and ChimeraX⁴⁷ were used to generate figures. Local resolutions were calculated using CryoSparc local resolution estimation with 0.143 from FSC curves and displayed on locally filtered maps (using a B-factor obtained during refinement) and displayed using ChimeraX with color surface.

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Andreas Martin (a.martin@berkeley.edu).

Materials availability

All constructs generated in this study are available from the lead contact upon request and completion of a Material Transfer Agreement.

Data and Code Availability:

• All data generated or analyzed during this study are included in this manuscript and the Supplementary materials. The cryo-EM density maps and corresponding atomic coordinates for human 26S proteasomes actively degrading FAT10 bound to TXNL1 can be found on the Electron Microscopy Data Bank (EMDB) under the following EMDB accession codes and under Protein Data Bank (PDB) accession codes: EMD-47719/PDB-9E8G (PS_Rpt5), EMD-47724/PDB-9E8L (PS_Rpt4), EMD-47725 /PDB-9E8N (PS_Rpt3), EMD-47723/PDB-9E8K (PS_Rpt6), EMD-47726/PDB-9E8O (PS_Rpt2+Eos), EMD-47722/PDB- 9E8J (PS_Rpt1), EMD- 47721/PDB- 9E8I (RS.1_{TXNL1 Forward}) and EMD-47720/PDB- 9E8H (RS.2_{TXNL1 Backward}), EMD-47727/PDB-9E8Q (PS_Rpt2).

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.