Amyloid conformation-dependent disaggregation revealed by a reconstituted yeast prion system

Motomasa Tanaka (  motomasa.tanaka@riken.jp ) RIKEN Center for Brain Science https://orcid.org/0000-0002-2994-7703 Yoshiko Nakagawa School of Life Science and Technology, Tokyo Institute of Technology Howard C.-H. Shen RIKEN Center for Brain Science Shinju Sugiyama School of Life Science and Technology, Tokyo Institute of Technology https://orcid.org/0000-00031822-0098 Yuri Tomabechi RIKEN Elena Krayukhina Research Department, U-Medico Inc. Kenji Okamoto RIKEN Takeshi Yokoyama RIKEN Center for Biosystems Dynamics Research Mikako Shirouzu RIKEN Center for Biosystems Dynamics Research https://orcid.org/0000-0002-7997-2149 Susumu Uchiyama Graduate School of Engineering, Osaka University Megumi Inaba School of Life Science and Technology, Tokyo Institute of Technology Tatsuya Niwa Tokyo Institute of Technology Yasushi Sako RIKEN Hideki Taguchi Tokyo Institute of Technology


Introduction
A number of neurodegenerative diseases, including Alzheimer's, Huntington's, and prion diseases, are elicited by misfolding of causal proteins and subsequent formation of b-sheet-rich brillar protein aggregates called amyloid 1 . Therefore, disaggregation of amyloid into monomeric proteins potentially provides a means for intervening the intractable diseases 2 . However, disaggregation could result in production of new amyloid fragments (seeds), which can enhance successive accumulation of amyloid and thereby exacerbate disease pathophysiology. In fact, disaggregation-mediated production of seeds may contribute to cell-to-cell transmission of amyloid, which is universal and at least partly account for disease mechanisms in many neurodegenerative diseases 3,4 . These notions highlight the importance of our better understanding of amyloid disaggregation process.
Mechanistically, propagation of protein aggregates or amyloids is achieved by the two key processes, template-dependent extension (growth) and disaggregation of amyloid 5,6 . Until now, many efforts have delineated molecular mechanisms of nucleation and amyloid growth processes 7 . In contrast, however, details of disaggregation mechanism have remained elusive, although the rate of amyloid disaggregation provides greater impacts on e ciency of amyloid replication and ultimately cellular phenotypes, such as prion strain phenotypes, than that of amyloid growth 8 . Notably, our poor understanding of amyloid disaggregation stems from the lack of technology and experimental systems that allow us to thoroughly investigate the process of amyloid disaggregation in vitro, as compared with the disaggregation mechanism of amorphous protein aggregates 9,10 .
To solve this fundamental problem in amyloid biology, yeast prions have been employed to dissect molecular mechanisms of amyloid disaggregation process [11][12][13][14] . Previously, the hexametric AAA + ATPase Hsp104 was shown to play crucial roles in amyloid formation and disaggregation of Sup35NM translation termination factor and protein determinant of yeast prion [PSI + ] state) and is su cient for amyloid formation in vitro 16 . However, it has long been controversial whether Hsp104 alone is su cient 15,17,18 or additional cellular factors are required to disaggregation of Sup35NM amyloid 9,10,19,20 though in vivo studies suggest that molecular chaperones such as Ssa1 (Hsp70) and Sis1 (Hsp40) play critical roles in propagation and maintenance of [PSI + ] yeast [21][22][23][24][25] . More importantly, it has been unclear how Hsp104 cooperates with other cellular factors, if any, to regulate the disaggregation process of Sup35 amyloid, and whether amyloid disaggregation mechanisms are similar or different between distinct strain conformations, which are widely observed and crucial for manifestation of distinct disease phenotypes in many neurodegenerative diseases 3 . Therefore, our understanding of amyloid disaggregation mechanism is highly limited.
To address these key questions, we developed a highly robust in vitro reconstitution system with three recombinant chaperones, Hsp104, Ssa1, and Sis1. This system recapitulated the yeast prion propagation in vitro, including de novo amyloid formation and amyloid disaggregation. Notably, we showed that all the three chaperones are essential for e cient disaggregation of Sup35NM amyloid. Next, in order to dissect disaggregation mechanisms, it is critical but has been challenging to simultaneously monitor and visualize dynamics of distinct chaperones in real time during the process of amyloid disaggregation. Therefore, we established a platform of single-molecule, multi-wavelength total internal re ectance uorescence (TIRF) imaging analysis, coupled with the reconstitution system, and simultaneously observed dynamics of Hsp104, Ssa1 and Sis1 in real-time during disaggregation of Sup35NM amyloid. Remarkably, we identi ed and characterized two, distinct amyloid conformation-dependent modes of disaggregation, fragmentation and dissolution, in which we found chaperone dynamics are markedly different. These ndings will provide a basis for developing novel therapeutic strategies for neurodegenerative diseases in which amyloid propagation is involved.
First, we established a method of accelerating formation and disaggregation of Sup35NM amyloid brils in vitro by optimizing the methods of puri cation and quality control of three chaperones (Hsp104, Ssa1, and Sis1) and the biochemical assays conditions (Supplementary Fig. 1). Notably, our analytical ultracentrifugation (AUC) analysis showed that Hsp104 exclusively formed a hexamer ( Supplementary   Fig. 1b) 26 When Sup35NM monomer was incubated with the three chaperones, the intensity of thio avin T (ThT) uorescence showed a remarkably rapid increase, indicating acceleration of amyloid formation (Fig. 1a). However, amyloid formation was not observed when Hsp104 was replaced with Hsp104KT, an ATPase-de cient mutant 27 , with Ssa1 and Sis1, indicating acceleration of amyloid formation is ATPaseactivity dependent. Amyloid formation of Sup35NM in the presence of only Ssa1 and Sis1 was inhibited, relative to no chaperones, consistent with the previous reports 28, 29 .
Next, we examined how Sup35NM brils grow during the acceleration process by atomic-force microscopy (AFM) (Fig. 1b). The number of small brils increased during 30 min between the time points 15 min and 45 min. After the ThT uorescence intensity reached the plateau point, the length of Sup35NM brils that formed in the presence of Hsp104/Ssa1/Sis1 was considerably shorter than that of the spontaneously formed brils (Fig. 1c). These results suggest that amyloid fragmentation occurred during the acceleration process (Fig. 1c). To gain further insights into the acceleration process, we investigated interactions between Alexa488-labeled Sup35NM monomer and the chaperones by AUC with a uorescence detector ( Supplementary Fig. 1d) 30 . Sup35NM was dominantly monomeric as evidenced by the peak at 1.5 S, as reported previously 31 . A new peak appeared at 16 S by the addition of Hsp104 in the presence of ATPγS, an ATP analog which slowly hydrolyzes ATP ( Supplementary Fig. 1d, left), but not ADP ( Supplementary Fig. 1d, right). This result indicates ATP state-dependent interaction of Sup35NM monomer with Hsp104.
We next attempted to develop a robust disaggregation assay using Sup35NM amyloid by optimizing various reaction conditions such as the concentrations of chaperones (Supplementary Fig. 1e-g). We used Sc4 amyloid, which has an amyloid core region spanning residue 1-42 and induces strong [PSI + ] phenotypes [32][33][34] . Our ThT assay and sedimentation assay showed that Sc4 amyloid was e ciently disaggregated by Hsp104, Ssa1, and Sis1 ( Fig. 1d and Supplementary Fig. 1h, i), but neither Hsp104 alone nor Ssa1/Sis1 showed disaggregation, suggesting all the three chaperones are essential for amyloid disaggregation. When Hsp104 was titrated in the ThT assay, Hsp104 showed concentrationdependent disaggregation activity ( Supplementary Fig. 1e), which is consistent with the in vivo studies showing Hsp104 concentration-dependent propagation of [PSI + ] state 14 . Similar to the acceleration assay, disaggregation potential was also abolished by replacing Hsp104 with Hsp104KT, indicating that Hsp104 disaggregates amyloid brils in an ATPase-activity dependent manner (Fig. 1d). AFM images corroborated the small size of brils by disaggregation in the presence of Hsp104 wild-type (WT)/Ssa1/Sis1, but not Hsp104 KT/Ssa1/Sis1 (Fig. 1e). To investigate whether our chaperone system can also disaggregate in vivo-derived Sup35 prions, our recombinant chaperones were incubated with lysates of [PSI + (Sc4)] yeast which was generated by infection of non-prion [psi − ] yeast with Sc4 amyloid 32 . Importantly, semi-denaturing detergent agarose gel electrophoresis (SDD-AGE) analysis 35 revealed that Sup35p migrated to the position of monomeric Sup35 in the presence of Hsp104/Ssa1/Sis1 (Fig. 1f). Taken together, these results demonstrate that our chaperone system can e ciently disaggregate both in vivoand in vitro-derived Sup35 amyloid. Ssa1 and Sis1 bind to Sup35NM amyloid in an ATP-activity dependent manner.
To investigate how Hsp104, Ssa1, and Sis1 cooperate to regulate the disaggregation activity for Sc4 amyloid, we used TIRF microscopy and observed the disaggregation process of uorescently-labeled Sc4 amyloid, which was partially tethered to a glass surface by biotinylated Sup35NM (Fig. 2a). Neither Hsp104 alone nor Hsp104/Ssa1 had effects on amyloid disaggregation, whereas Hsp104, Ssa1, and Sis1 together showed dramatic disaggregation within 300 sec in our conditions ( Fig. 2b and Supplementary Video 1). However, a combination of Hsp104KT, Ssa1, and Sis1 showed no effect, consistent with the biochemical assays. During the chaperone-mediated disaggregation, the number of nick in brils substantially increased with incubation time (Fig. 2b), and Sc4 amyloid mostly disappeared within 300 sec. This result revealed that Hsp104/Ssa1/Sis1 e ciently fragments Sc4 amyloid and yields a number of ber fragments (seeds) during the disaggregation process.
We next examined how Ssa1 and Sis1 impacted the binding of Hsp104 to Sc4 amyloid. To address this question, we performed simultaneous dual-color TIRF imaging analysis to observe uorescently-labeled SNAP-tagged Hsp104 with unlabeled Ssa1 and Sis1 during disaggregation. Hsp104-SNAP549 showed ATPase and disaggregation activities indistinguishable from untagged Hsp104 ( Supplementary Fig. 2a,  b). We used a low concentration of Hsp104-SNAP549 (40 nM) for this TIRF experiment to reduce the uorescence background, which is required for single-molecule observations. Hsp104 showed repeated binding and dissociation on Sc4 amyloid ( Fig. 2c and Supplementary Video 2). To evaluate the effects of Ssa1/Sis1 on Hsp104 binding to Sc4 amyloid, we counted the number of Hsp104-SNAP549 binding events to Sc4 amyloid in the presence or absence of Ssa1/Sis1. Strikingly, the binding of Hsp104 to Sc4 amyloid was approximately 50-fold higher in the presence of Ssa1/Sis1, whereas either Ssa1 or Sis1 alone did not induce the binding of Hsp104 to Sc4 amyloid (Fig. 2d).
The binding of all the three chaperones induced fragmentation of Sc4 amyloid, which prevented us from determining a binding a nity of Hsp104 to the amyloid in the presence of Ssa1/Sis1. Therefore, we used a Hsp104trap mutant 36 , which cannot hydrolyze ATP and is therefore unable to disaggregate Sc4 amyloid ( Supplementary Fig. 2c). As expected 36 , we found that Sup35NM monomer is not released from Hsp104trap but from Hsp104 WT ( Supplementary Fig. 2d, e). Remarkably, however, Hsp104trap showed repeated binding and dissociation on Sc4 amyloid ( Supplementary Fig. 3a, b), which allowed us to evaluate a binding a nity of Hsp104 to Sc4 amyloid by excluding the effects caused by amyloid disaggregation. The binding a nity of Hsp104trap to Sc4 amyloid was increased by the presence of Ssa1/Sis1, yielding a K d value of 81 nM compared with 14 µM in the absence of Ssa1/Sis1. Therefore, Ssa1/Sis1 enhanced the binding of Hsp104 to amyloid by more than 170-fold ( Supplementary Fig. 2f). We next investigated whether Ssa1-Alexa488 interacts with Sc4 amyloid in the presence or absence of Sis1. Ssa1-Alexa488 showed disaggregation activity indistinguishable from unlabeled Ssa1 ( Supplementary Fig. 4a), and Ssa1-Alexa488 alone showed no interaction with the Sc4 amyloid. In contrast, Ssa1-Alexa488 with Sis1 was rst associated with Sc4 amyloid as patches at distinct sites, then the patches gradually spread throughout the brils (Fig. 2e). Remarkably, in the presence of ADP or ATPγS, Ssa1-Alexa488 with Sis1 did not bind to Sc4 amyloid (Fig. 2f). Therefore, Ssa1 interacted with Sc4 amyloid through its ATP hydrolysis. Furthermore, we found that Sis1-Cy3 binding to Sc4 amyloid requires Ssa1 ( Supplementary Fig. 4c, d). These ndings demonstrate that both Ssa1 and Sis1 cooperatively bind to Sc4 amyloid in an ATP-activity dependent manner.
To explore the binding order of Ssa1 and Sis1 to Sc4 amyloid, we performed three-color TIRF imaging analysis to simultaneously observe the binding of Ssa1-Alexa488 with Sis1-Cy3 to STELLA650-labeled Sc4 amyloid in real time (Fig. 3a). The uorescence intensities of both Ssa1 and Sis1 along Sc4 brils were plotted at each time point (Fig. 3b). Sis1 and Ssa1 were accumulated at the same sites at ~ 100 sec, and their uorescence intensities increased until ~ 800 sec. To investigate the kinetics of Ssa1 and Sis1 binding, the uorescence intensities of Ssa1 and Sis1 in each peak in Fig. 3b were plotted as a function of time (Fig. 3c). We determined the binding rates of Ssa1 and Sis1 at each accumulation site in Sc4 amyloid (Fig. 3d) and found a high correlation between the two rates (Slope = 0.93, R 2 = 0.83), indicating that Ssa1 and Sis1 almost simultaneously bind to Sc4 amyloid (Fig. 3d).
To investigate the relationships between Hsp104 and Ssa1/Sis1 for their binding to Sc4 amyloid, we next performed three-color TIRF imaging analysis in real time, using Hsp104-SNAP549, Ssa1-Alexa488, unlabeled Sis1 and STELLA650-labeled Sc4 amyloid (Fig. 3e). Binding of Hsp104-SNAP549 was detected (at ~ 200 sec) at the sites where Ssa1-Alexa488 had already been bound (at ~ 100 sec). In addition, Sc4 amyloid was fragmented (at ~ 900 sec) at the sites where Hsp104 was recruited (Fig. 3f). These results indicate that Ssa1 and Sis1 prime fragmentation sites and play important roles in recruiting Hsp104 to the speci c sites. We found the difference in the arrival time between Hsp104-SNAP549 and Ssa1-Alexa488 was ~ 73 sec (Fig. 3g). This ordered chaperone binding suggests that Ssa1/Sis1 may rst remodel the Sc4 amyloid conformation and the altered structure is targeted by Hsp104.
Repeated binding of Hsp104 to the same site is crucial for amyloid fragmentation.
We clearly detected appearance of nicks (i.e. fragmentation) in the Sc4 amyloid during the incubation with the three chaperones. However, we also found non-fragmented sites in Sc4 amyloid, even though the binding of Hsp104 and Ssa1/Sis1 to the amyloid was observed (Fig. 3e, right). What factors differentiate fragmentation from non-fragmentation within the same bril? To determine binding parameters of the chaperones required for e cient amyloid fragmentation, we took an advantage of the single-molecule analysis by TIRF, and compared the intensity pro les of Sc4 amyloid, Ssa1 and Hsp104 between fragmented and non-fragmented sites within the same bril ( Fig. 4a, b). To observe a single molecule of Hsp104, we used 1 µM of unlabeled Hsp104 including 30 nM of Hsp104-SNAP549. We measured the uorescence intensities of Sc4 amyloid, Hsp104, and Ssa1 in a 3 × 3 pixel area to examine the chaperone dynamics at fragmented and non-fragmented sites. At the fragmented sites, the uorescence intensity of Sc4 amyloid rapidly dropped after Ssa1's and subsequently Hsp104's binding (Fig. 4b, top). In contrast, at non-fragmented sites, the uorescence intensity of Sc4 amyloid gradually declined (Fig. 4b, bottom). Importantly, this decline was not caused by photobleaching (Fig. 4b, black), and thus we de ned this mode of disaggregation as dissolution. We found that the rst arrival time of Ssa1 was not different between fragmented and non-fragmented sites, whereas the cumulative intensity of Ssa1 at fragmented sites was higher than that at non-fragmented sites (Fig. 4c, e). Therefore, the accumulation, but not the rapid binding, of Ssa1 is associated with e cient fragmentation of Sc4 amyloid (Fig. 4e). The difference in arrival time between Ssa1 and Hsp104 at fragmented sites was approximately 50 sec, which was signi cantly shorter than that at non-fragmented sites (Fig. 4d). The cumulative intensities of Ssa1 and Hsp104 at fragmented sites were signi cantly higher and lower than those at non-fragmented sites, respectively (Fig. 4e, f). These results indicate that accumulation of Ssa1, but not Hsp104, is favorable to fragmentation of Sc4 amyloid. Consistently, when we counted the number of Hsp104 appearance on Sc4 amyloid in a 3 × 3 pixel area (i.e. the number of repeated binding of Hsp104 to the same site) from 0 to 600 sec, we found the number at fragmented sites was signi cantly larger than that at non-fragmented sites (Fig. 4g). Together, these results suggest that transient, repeated binding of Hsp104 to the same site of Sc4 amyloid is crucial for e cient amyloid fragmentation.
Two distinct modes of chaperone-mediated disaggregation were observed between Sc4 and Sc37 amyloids.
Amyloid conformation is a critical determinant of prion strain phenotype, which is largely regulated by amyloid disaggregation rate, rather than amyloid growth rate 8 . Therefore, to investigate the possibility of amyloid conformation-dependent disaggregation, we performed three-color TIRF imaging analysis and examined the binding of Hsp104-SNAP549 and Ssa1-Alexa488 to Sc37 amyloid, which has a longer amyloid core region (residues 1-72) 33,34 and induces weak [PSI + ] phenotypes. In contrast to the previous result of Ssa1/Sis1-independent disaggregation 18 , Sc37 amyloid was disaggregated by the three chaperones, but with less e ciently than Sc4 amyloid (Fig. 5a, b and Supplementary Fig. 6a). We found that the cumulative number of nick within Sc37 amyloid brils was signi cantly smaller than that of Sc4 amyloid (Fig. 5c). Importantly, however, the uorescence intensity of Sc37 amyloid disappeared gradually, indicating uniform disaggregation (i.e. dissolution) throughout the brils (Fig. 5b), as observed in nonfragmented sites of Sc4 amyloid (Fig. 4b, bottom). Analysis of the intensity pro le in a 3 × 3 pixel area showed that Ssa1 recruits Hsp104 to Sc37 amyloid, in a manner similar to Sc4 amyloid ( Fig. 6a, b); the rst arrival time of Ssa1 and the time difference of the rst arrival time between Ssa1 and Hsp104 was almost indistinguishable between Sc4 and Sc37 amyloid (Fig. 6c, d). Interestingly, however, amounts of both Ssa1 and Hsp104 accumulated on Sc37 amyloid were remarkably larger than those on Sc4 amyloid ( Fig. 6e, f). In addition, we observed a smaller number of appearance of Hsp104 on Sc37 amyloid than on Sc4 amyloid (Fig. 6g). This result shows that the number of repeated binding of Hsp104 to the same site was reduced for Sc37 amyloid, similar to what we found at non-fragmented sites in Sc4 amyloid (Fig. 4g). These observations suggest that the enhanced binding of both Ssa1 and Hsp104 to Sc37 amyloid may result in a "locked" conformation of Hsp104 within the amyloid, which may prevent Hsp104's repeated binding and following extraction of polypeptides from the amyloid, leading to less e cient amyloid fragmentation.
To gain further insight, we investigated a dwell time of Hsp104-SNAP549 at fragmented or nonfragmented sites on each Sc4 and Sc37 amyloid, using Hsp104-SNAP549 (30 nM) and unlabeled Hsp104 (1 µM) with Ssa1 and Sis1 ( Supplementary Fig. 5a-c). The majority of the dwell time of Hsp104 on Sc4 amyloid was 3-30 sec at fragmented sites, whereas a fraction of a long dwell time (30-300 sec) was larger at the non-fragmented sites in Sc4 amyloid. This result implies that the shorter dwell time of Hsp104 at fragmented sites is simply due to the dissociation of Hsp104 from Sc4 amyloid by amyloid fragmentation (Supplementary Fig. 5c). In contrast, the majority of dwell time of Hsp104 on Sc37 amyloid was longer (30-300 sec), suggesting that Hsp104 stayed on Sc37 amyloid due to its lower e ciency in fragmentating the amyloid or its intrinsic high a nity to Sc37 amyloid structures. In order to address it, we used Hsp104trap, as our nding that Hsp104trap lacking a disaggregation activity repeatedly binds to the same site of amyloid allowed us to determine a dwell time by excluding the effects caused by amyloid disaggregation (Supplementary Fig. 3b). As a result, a dwell time of Hsp104trap on Sc37 amyloid was longer than that on Sc4 amyloid ( Supplementary Fig. 3c), indicating an intrinsic higher a nity of Hsp104 to Sc37 amyloid conformation. Furthermore, we hypothesized that if Hsp104 repeatedly binds to the same site of Sc4 amyloid due to multiple rounds of extraction of Sup35NM monomers from the amyloid in order to complete a fragmentation process, Hsp104trap will not show such repeated binding to Sc4 amyloid due to a lack of its disaggregation activity. As predicted, we found that the number of Hsp104trap binding to Sc4 amyloid in a 3 × 3 pixel area in 600 sec was small and approximately similar to that of Hsp104trap on Sc37 amyloid (Supplementary Fig. 3d) and also that of Hsp104 WT on Sc4 non-fragmented sites (Fig. 4g). These results together with the longer dwell time of Hsp104trap on Sc37 amyloid ( Supplementary Fig. 3c) indicate that an intrinsic higher a nity of Hsp104 to Sc37 amyloid than that to Sc4 amyloid prevents repeated binding of Hsp104 to the same site of amyloid, resulting in less e cient fragmentation.
Differences in amyloid structure determine chaperone binding and amyloid disaggregation modes.
Previous studies showed that the core of Sc37 amyloid is located around residues 2-72, whereas that of Sc4 amyloid around residues 2-42 31,33,34 . Therefore, it was surprising that even though the exposed amino acid region is shorter in Sc37 amyloid, a substantially larger amount of both Ssa1 and Hsp104 bound to Sc37 amyloid than Sc4 amyloid (Fig. 6e, f). Since Hsp104 is recruited to Sup35NM amyloid by Ssa1 (Fig. 3f), the enhanced binding of Ssa1 to Sc37 amyloid structure (Fig. 6e) is likely to be a causal factor for the enhanced accumulation and longer dwell time of Hsp104 on Sc37 amyloid ( Fig. 6f and Supplementary Fig. 3c, 5c right). To investigate it, we acquired super-resolution microscopy images, using uorescently labeled Ssa1 in Sc4 or Sc37 amyloid. Sc4 amyloid showed wavy and helical structures, whereas Sc37 amyloid exhibited relatively straight morphology (Fig. 6h), which was consistent with independent observations by electron microscopy (Supplementary Fig. 6b). Remarkably, Ssa1 was accumulated preferentially at the speci c sites where the uorescent intensity of Sc4 amyloid is high. In contrast, Ssa1 uniformly bound to Sc37 amyloid throughout the brils, consistent with the enhanced accumulation of Ssa1 on Sc37 amyloid by TIRF observation (Fig. 6e). These results suggest that reduction of Ssa1 concentration may decrease the binding of Sc37 amyloid, altering the disaggregation mode from dissolution to fragmentation. When we used a low concentration of Ssa1 in the disaggregation assay with Sc37 amyloid by TIRF, amounts of Ssa1 and Hsp104 on Sc37 amyloid were signi cantly reduced ( Supplementary Fig. 7f, g), as expected. Importantly, however, neither the frequency of fragmentation ( Supplementary Fig. 7a-c) nor the number of repeated binding of Hsp104 increased ( Supplementary Fig. 7h). Rather, the disaggregation (dissolution) activity was simply decreased ( Supplementary Fig. 7d, e). These results indicate that the tertiary or quaternary structure of amyloid is more critical in determining a disaggregation mode than the numbers of chaperones residing on amyloid.
Together, we surmise that the difference in amyloid structure between Sc4 and Sc37 prion strain conformations is responsible for their marked difference in amyloid-chaperone interactions and underlies the remarkable difference in disaggregation mode between fragmentation and dissolution.

Discussion
We developed a robust in vitro reconstitution system, in which both of the de novo amyloid formation and disaggregation were recapitulated in vitro. Our data demonstrated that all the three chaperones, Hsp104, Ssa1, and Sis1 are essential for e cient disaggregation of Sup35NM amyloid. More importantly, the reconstituted yeast prion system allowed us to dissect the amyloid disaggregation process. Under our TIRF conditions, Sis1 and Ssa1 simultaneously bound to Sup35NM amyloid. After the ~ 73 sec interval, Hsp104 was recruited to the speci c sites in Sc4 amyloid where Ssa1/Sis1 had already bound. This result, together with the previous report that Hsp70 can unfold protein aggregates 37,38 , suggests that Ssa1/Sis1 rst remodel amyloid structures and then the loosened polypeptide region may be easily captured and extracted by Hsp104. Once a Sup35NM monomer is extracted from amyloid, newly created breaking regions in the Sc4 amyloid will be exposed and thereby easily targeted again by Hsp104. Such multiple rounds of monomer extraction may account for the observed transient, repeated binding of Hsp104 to the same site of amyloid, followed by fragmentation of Sc4 amyloid (Fig. 7). In addition to the fragmentation, our TIRF imaging analysis revealed another mode of amyloid disaggregationdissolution. Interestingly, the fragmentation was favorable to Sc4 amyloid while the dissolution was a predominant disaggregation process of Sc37 amyloid. Therefore, our work revealed prion strainconformation dependent, distinct modes of amyloid disaggregation.
Our results resolved the long-standing, puzzling question whether Hsp104 alone is su cient or additional factors are required for amyloid disaggregation 9,10,15,17−20 . Our ndings support the previous in vitro study 19 and a number of in vivo studies that described the necessity of additional factors [21][22][23][24] . Furthermore, it has been a decade-long mystery that Hsp104 is not easily identi ed as a Sup35 prionassociated protein in vivo despite its major role in disaggregation. Our yeast prion reconstitution system revealed that Hsp104 binds transiently to the Sc4 amyloid, whereas Ssa1 associates with the Sc4 amyloid in a more stable capacity (Fig. 7), thus explaining the di culty of detecting Hsp104, but not Ssa1, in prior in vivo studies 39 . Importantly, TIRF imaging analysis allowed us to directly compare physical parameters of chaperones dynamics between fragmented and non-fragmented sites within the same amyloid, and also between Sc4 amyloid and Sc37 amyloid. The binding of Ssa1/Sis1 to speci c sites of Sc4 amyloid and following repeated binding of Hsp104 induces fragmentation (Fig. 1d, 2b), whereas more stable interactions of Hsp104 in non-fragmented sites of Sc4 or in Sc37 amyloid results in dissolution (Fig. 4b bottom, 5a and b). The intrinsic higher a nity of Hsp104 to Sc37 amyloid ( Supplementary Fig. 3c) may induce non-productive binding of Hsp104 to the amyloid, impairing its multiple rounds of extraction of polypeptides from Sc37 amyloid that is required for fragmentation. The detailed mechanism that determines such distinct disaggregation modes and kinetics requires further investigation. Nonetheless, our data show that variation of amyloid conformations constitutes the physical foundation to dictate the differences in chaperone binding, disaggregation mode, and prion strain phenotype.
This study provided unexpected, novel ndings which were not easily understood in the previous studies, but might be speci c to the disaggregation process of amyloid in general or that of Sup35NM amyloid. First, we found that Sis1 and Ssa1 almost simultaneously bound to Sup35NM amyloid, though the previous studies indicate Hsp40 rst binds to a substrate, then recruits Hsp70 40,41 . Second, an ADPbound Hsp70 is known to be a substrate-binding form 40,41 , but neither ADP-nor ATPγS-bound Ssa1 with Sis1 interacted with Sup35NM amyloid. Rather, only ATP-bound Ssa1 together with Sis1 was able to bind to Sup35NM amyloid, indicating that the ATPase activity (i.e. ATP hydrolysis) of Ssa1 is essential for binding of Ssa1 to the amyloid.
Notably, the prion conformation-dependent, distinct modes of disaggregation showed in this study provides a molecular explanation for the previous ndings where the underlying mechanisms had remained elusive. First, the preference of fragmentation (to dissolution) of Sc4 amyloid leads to an increased number of Sc4 ber fragments. Therefore, the preferred fragmentation process would be responsible for the observed larger number and shorter size of propagons in strong [PSI + (Sc4)] strains than that of weak [PSI + (Sc37)] strains 8,32 . Second, previous studies indicated that Hsp104 is more immobile in weak [PSI + ] strains than that in strong [PSI + ] strains, suggesting that more Hsp104 proteins interact with weak [PSI + ] prions 42 . Consistently, the immunoprecipitation assay revealed that more Hsp104 and Ssa1 proteins bind to "weak" [PSI + ] prions than "strong" [PSI + ] prions 43 . These results might had been paradoxical if one assumes that interactions of more Hsp104 and Ssa1 proteins with Sup35NM amyloid are advantageous to disaggregation. However, our results indicate that enhanced binding of the chaperones with weak [PSI + ] prions (Fig. 6e, f and Supplementary Fig. 5c) underlie the preference for dissolution, rather than fragmentation, which yields less propagons and elicits weak [PSI + ] strain phenotypes.
Given the potential application of Hsp104-mediated disaggregation for neurodegenerative diseases 2 , the mechanistic insights in this study provide broad implications beyond the yeast system. We suggest that selective enhancement of the dissolution process, but not fragmentation, provide a means for therapeutic intervention of pathological protein aggregates or amyloid, because enhanced fragmentation events, thereby increased ber fragments (seeds), may accelerate amyloid formation and exacerbate disease pathophysiology. Therefore, the distinct modes of amyloid disaggregation that we uncovered in this study will provide a novel direction for the development of therapeutic intervention. Furthermore, the robust disaggregation assay for amyloid brils in this study will help diagnose or classify disease (strain) phenotypes at the molecular level by evaluating susceptibility of in vivo amyloid to the disaggregation machinery.

Plasmid construction and protein puri cation
For expression and puri cation of tagged Sup35NM proteins in Escherichia coli, we followed the previous protocols 31 . Brie y, Sup35NM with a 7xHis or 7xHis-Cys tag were overexpressed in E. coli Rosetta (DE3) cells (Nonagen) in LB media and puri ed by Ni-NTA a nity chromatography (GE healthcare) and anionexchange chromatography under the denatured condition 31,34,44 . The buffer exchange to acetonitrile/water was performed using HPLC (Hitachi). The sample was lyophilized and stored at -20 o C until use.
cDNAs of Hsp104, Sis1 and their mutants were cloned into pET28a vector including N-terminally His-MBP-tag 45 . A plasmid of Hsp104-SNAP was generated by addition of a SNAP-tag to the C-terminus in the expression vector. cDNA of Sis1 with a C-terminal cysteine was introduced into the pET28a plasmid.
Hsp104 and Sis1 were overexpressed in E. coli Rosetta (DE3) in LB media and puri ed by Ni-NTA a nity chromatography (GE Healthcare), as previously described with modi cations 46 . Brie y, the eluent from Ni-NTA resin was puri ed by amylose-resin a nity chromatography (BioLabs). After digestion of a His-MBP-tag by the TEV enzyme, Hsp104 and Sis1 was puri ed by anion-exchange chromatography using HiTrap Q 6 ml column (GE Healthcare) with the gradient system. To remove any remaining His-tagged proteins, the eluent was subjected to the 2nd Ni-NTA gravity ow puri cation and the ow-through was obtained. Only pure fractions (>95%) were pooled, and concentration and buffer exchange were performed. Expression and puri cation of Ssa1 with a with a C-terminal 7xHis tag was followed by the previous report 47,48 . Brie y, Ssa1 was puri ed by Ni-NTA a nity (GE Healthcare) and HiTrap Q (GE Healthcare) columns with the gradient system. Only pure fractions (>95%) were pooled, and concentration and buffer exchange were performed. Purity of all the proteins were rigorously checked throughput the puri cation procedures by Coomassie Brilliant Blue staining and the absorbance at 280 nm. Quality control of Hsp104 hexamer was performed by AUC as described later. All of the proteins were aliquoted into a small volume, ash frozen by liquid nitrogen and stored at -80°C. Typically, we used up the proteins within 1-2 months to prevent a decrease of their activities.
As a plasmid for expressing TEV, pRK793 was purchased from Addgene (RRID:Addgene_8827) 49 . A TEV protease with a 7xHis tag was overexpressed in E. coli Rosetta (DE3) in LB media and was puri ed by Ni-NTA, as previously reported 45 . The pure fractions of the eluent from Ni-NTA a nity chromatography was used for digestion of the tagged proteins. A plasmid for expressing ClpP was kindly provided by Dr. Axel Mogk (Universität Heidelberg). ClpP was overexpressed as a C-terminally 6xHis tagged protein in E. coli Rosetta (DE3) in LB media and puri ed by Ni-NTA a nity chromatography (GE Healthcare), as previously described 50 .
Fluorescence and chemical labeling of proteins For uorescence labeling of Sup35NM, Sup35NM with a 7xHis-tag-Cys tag was mixed with a 10-fold excess of STELLA650-maleimide (GORYO Chemical) or Alexa-488-maleimide (ThermoFisher Scienti c). An excess amount of uorescence dye was removed by passing the protein through a Bio-Gel P-6 Gel (Biorad #154130). Labeling e ciencies were spectrophotometrically determined (Hitachi U-3900), using the absorbances at 646 nm (extinction coe cient: 110,000 M -1 cm -1 ) for STELLA-650-maleimide, at 493 nm (extinction coe cient: 72,000 M -1 cm -1 ) for Alexa-488-maleimide, combined with the absorbance at 280 nm (extinction coe cient: 29,800 M -1 cm -1 ) for Sup35NM protein. For biotin labeling of Sup35NM, Sup35NM including an Avi-tag was overexpressed in E. coli Rosetta (DE3) that contains a pBirA plasmid in LB media. After optical density of the culture reaches ~0.6, d-biotin (Nacalai, the nal concentration of 50 mM) and IPTG (the nal concentration of 0.6 mM) were added for biotin labeling. Biotinylated Sup35NM was puri ed as described above, and the incorporation of biotin and elimination of nonbiotinylated Sup35NM were con rmed by anti-Biotin antibody (Jackson Immuno Research Laboratories, Inc.).
For uorescence labeling of a SNAP tag, Hsp104 with a SNAP-tag was mixed with a 2-fold excess of SNAP-Surface549 (Biolabs) together with 1 mM DTT for 1 hour in dark. An excess dye was removed by passing of the protein through a P6 column as above. Labeling e ciencies were spectrophotometrically ATPase activity assay An ATPase activity of 0.25 mM Hsp104 and 1 mM Ssa1 was measured using a commercially available kit (Innova) in the ATPase buffer (40 mM Hepes-KOH pH 7.4, 150 mM KCl, 20 mM MgCl 2 , 2 mM DTT) including 1 mM ATP. The assay was followed by manufacturer's protocol and the previous report 52 .

Disaggregation assay
The Sc4 or Sc37 amyloid was freshly prepared by mixing 5uM Sup35NM with 5% (mol/mol) of corresponding Sc4 or Sc37 seeds, respectively, as previously reported 16,32 . The disaggregation assay of Sc4 or Sc37 amyloid was typically performed in the presence of 1 mM Hsp104, 2 μM Ssa1, 2 μM Sis1 with ATP or an ATP regeneration system (10 mM creatine phosphate and 0.1 mg/ml creatine kinase) in the disaggregation buffer (25 mM Hepes-KOH (pH 7.5), 150 mM potassium acetate, 10 mM magnesium acetate, 2 mM DTT) at 30 o C . Hsp104KT (K218T, K620T), an ATPase-de cient mutant of Hsp104, was used as a negative control. The extent of amyloid disaggregation was measured by thio avin T uorescence, sedimentation assay, atomic force microscopy (AFM). The thio avin T (ThT) assay was followed by the previous report 31  Excitation/emission wavelengths for anisotropy measurements of Sup35NM-Alexa488 was 490/530 nm.

TIRF data analysis
Acquired image sequences were converted to 16-bit TIFF les using NIS Elements (Nikon Instruments Inc.) and all images was analyzed by ImageJ (NIH, Bethesda, MD). All images were automatically aligned using micropattern images by a plugin, Template Matching and Slice Alignment 56 . Background uorescence was subtracted using the background subtraction tool (rolling ball radius: 50 pixel) in ImageJ. We calculated the fragmentation frequency of Sc4 and Sc37 amyloids in Fig. 5c and Supplementary Fig. 7c by manually counting fragmentation events. The number of fragmentation events was divided by the initial length of amyloid.
The number of Hsp104 binding on Sc4 amyloid in Fig. 2d was counted per 600 sec and the number was subtracted by the number of control experiments without Sc4 amyloid. For binding of Ssa1-Alexa488 or Sis1-Cy3, Sc4 amyloid-STELLA650 was traced over time, and the traces were used to determine the corresponding uorescence pro les in other channels. Mean uorescence intensities of Ssa1 and Sis1 were normalized to the length of Sc4 amyloid ( Fig. 2f and Supplementary Fig. 4d). The same method of the trace analysis was used for our preparation of spatiotemporal pro les by the Plot Pro le tool. The binding rates of Ssa1 and Sis1 to Sc4 amyloid in Fig. 3d were determined by tting the uorescence data with a single exponential function with x offset constant using IgorPro (Wavemetrics , OR, USA).
For binding of Hsp104-SNAP549 and Ssa1-Alexa488, Sc4 amyloid-STELLA650 was traced over time using the plot pro le tool. Then, the trace was used to determine the corresponding uorescence pro les in other channels. The time of the rst appearance of Hsp104-SNAP549 (T Hsp104 arrival ) and Ssa1-Alexa488 (T Ssa1 arrival ) in Fig. 3g was determined by analysis of the time when uorescence reached 10% of the maximum intensity in a 3x3 pixel area on amyloid. Distribution of the interval times of Hsp104 and Ssa1 was calculated as T Hsp104 arrival -T Ssa1 arrival .
To determine fragmented and non-fragmented sites on amyloid, the spots in a 3x3 pixel area where amyloid was nally fragmented or not fragmented were traced, and uorescence intensities were measured. The same pixel regions were also used to determine the corresponding uorescence pro les in other channels. To subtract background signals, uorescence intensities at the spots in a 3x3 pixel area that did not contain amyloid were traced. Fragmented sites were de ned as the sites showing a more than 70% decrease of the initial uorescence intensity at 600 sec, while that of non-fragmented (dissolution) sites showed over 30% of the initial uorescence intensity at 600 sec. The rst appearance time of Ssa1 was determined when uorescence intensity reached 10% of the maximum in a 3x3 pixel area (Fig. 4c, 6c). The time of the rst appearance of Hsp104-SNAP549 (T Hsp104 arrival ) and Ssa1-Alexa488 (T Ssa1 arrival ) was determined when uorescence intensity reached 10% of the maximum intensity in a 3x3 pixel area on amyloid (Fig. 4d, 6d). Cumulative intensity of Hsp104 and Ssa1 uorescence was integrated from 0 to 600 sec (Fig. 4e, 4f, 6e, 6f, Supplementary Fig.7f, g). The number of Hsp104 appearance on amyloid in a 3x3 pixel area, which represents the number of Hsp104 binding to the same site, was counted from 0 to 600 sec (Fig. 4g, 6g and Supplementary Fig. 3d, 7h). Dwell time of Hsp104 was determined as the period of time when uorescence intensity was over 100 a.u..

Negative staining with electron microscopy
Sc4 or Sc37 (5 mM Sup35NM monomer) amyloid was prepared as described above. Sc4 or Sc37 amyloid was centrifuged (10,000 g, 10 min, 4 o C) to eliminate small aggregates or monomers. We removed 90% of the supernatant and resuspend the pellet fraction with the same volume of disaggregation buffer. A 3 ml volume of the samples was applied to the carbon-coated holy microgrid (EM Japan). Subsequently, the applied sample was negatively stained with 2% (w/v) uranyl acetate three times. Images were recorded on a Falcon II direct electron detector using a Tecnai TF20 transmission electron microscope (Thermo Fisher Scienti c) at the acceleration voltage of 200kV and at the magni cation of 62,000.

Confocal microscopy
All coverslips and labeled proteins (Sc4 or Sc37 amyloid-STELLA650 and Ssa1-Alexa488) were prepared as described by TIRF analysis. First, uorescence images of Sc4 or Sc37 amyloid-STELLA650 immobilized on cover glass was acquired. Next, 2 mM Ssa1-Alexa488 and 0.5 mM Sis1 was owed into the ow-cell and then uorescence images at the identical place were acquired. A Leica TCS SP8 confocal microscope equipped with 638 nm diode and 488 nm OPSL laser (Leica Microsystems, Wetzlar, Germany) was used to collect z-stacks of the amyloid images, and these images were processed by deconvolution using the Leica Application Suite X (LAS X) software (Leica Microsystems, Wetzlar, Germany

Declarations
Competing Interests statement: The authors declare no competing interests.     Repeated binding of Hsp104 to the same site is more frequent in fragmented sites than non-fragmented sites of Sc4 amyloid. a, Representative time-lapse TIRF images showing interactions between Sc4 amyloid-STELLA650, Hsp104-SNAP549 and Ssa1-Alexa488, after 1 μM Hsp104 including 30 nM Hsp104-SNAP549 (green), 2 μM Ssa1-Alexa488 (blue) and 0.5 μM Sis1 were mixed and owed. Images were acquired every 500 msec. White and yellow arrow heads show the sites where Sc4 amyloid was fragmented and non-fragmented (dissolution), respectively. White and yellow boxes indicate the fragmented and non-fragmented sites shown in Fig. 4b. Scale bar, 3 μm. b, Representative normalized intensity pro les of Sc4 amyloid-STELLA650 (red), Hsp104-SNAP549 (green) and Ssa1-Alexa488 uorescence (blue) in a 3x3 pixel area at the site where Sc4 amyloid was fragmented (white box) or not fragmented (yellow box) in the merged image in Fig. 4a. Notably, due to the mixture of uorescently labeled and unlabeled Hsp104 proteins, a decrease of the uorescence signal intensity of Sc4 amyloid was observed even though Hsp104 did not apparently bind (Fig. 4b, bottom).   A model of two distinct, strain conformation-dependent modes of amyloid disaggregation. In Sc4 amyloid, Ssa1 and Sis1 bind to the amyloid, then recruit Hsp104. Following repeated binding of Hsp104 to the same site of amyloid results in "fragmentation" of Sc4 amyloid. In Sc37 amyloid, enhanced accumulation of both Ssa1 and Sis1 recruits more Hsp104 proteins and induces "dissolution".