PROTEASOMES

A molecular census of 26Sproteasomes in intact neuronsShoh Asano,* Yoshiyuki Fukuda,* Florian Beck, Antje Aufderheide, Friedrich Förster,Radostin Danev, Wolfgang Baumeister†

The 26S proteasome is a key player in eukaryotic protein quality control and in theregulation of numerous cellular processes. Here, we describe quantitative in situ structuralstudies of this highly dynamic molecular machine in intact hippocampal neurons. Weused electron cryotomography with the Volta phase plate, which allowed high fidelity andnanometer precision localization of 26S proteasomes. We undertook a molecular censusof single- and double-capped proteasomes and assessed the conformational states ofindividual complexes. Under the conditions of the experiment—that is, in the absence ofproteotoxic stress—only 20% of the 26S proteasomes were engaged in substrateprocessing. The remainder was in the substrate-accepting ground state. These findingssuggest that in the absence of stress, the capacity of the proteasome system is notfully used.

The ubiquitin proteasome system (UPS) playsa key role in many cellular processes, in-cluding cell cycle control and proteostasis(1). The 26S proteasome executes the pro-cessive degradation of proteins marked for

destruction by the attachment of polyubiquitinchains (2, 3). Malfunctions of the UPS result inthe accumulation of damaged or misfolded pro-teins and are implicated in a number of diseases(4). The 26S proteasome is involved in manyfundamental processes in neurons (5), includingthe regulation of synaptic strength and of thepresynaptic (6) as well as postsynaptic proteome(7). However, the exact molecular mechanismsunderlying these phenomena remain to be elu-cidated, and quantitative structural studies per-formed in situ should contribute to a deeperunderstanding.Cellular electron cryotomography combines

the power of high-resolution three-dimensional(3D) imaging with the best possible structuralpreservation (8). This is particularly importantfor fragile structures such as the 26S proteasome.Electron cryotomography performed with therecently developed “Volta” phase plate (9) allowsthe identification and localization of molecularstructures in the cellular environment with highfidelity.The 26Sproteasomes could readily be identified

by visual inspection (Fig. 1A, insets). Nevertheless,for a systematic and comprehensive statisticalanalysis of the 26S proteasomes, an automatedtemplate-based search was performed (10). To en-able detection of both single-capped and double-capped 26S proteasomes, a double-capped 26Sparticle [ElectronMicroscopyData Bank (EMDB),2165] (11) with one regulatory particle (RP) re-moved computationally was low-pass filtered to aresolution of 40 Å and used as a template for an

initial search of cytosolic volumes. Particles iden-tified by cross-correlation coefficients (CCC) abovea generous cut-off value (Fig. 1B) showed an iso-tropic orientation distribution (fig. S1) and werevalidated by visual inspection. Both single- anddouble-capped particles yielded almost identicalCCCs (mean CCC ~ 0.28 for both populations)(Fig. 1B), indicating that the template-basedsearch did not bias the statistics toward one orthe other assembly state. In addition, we per-formed a search with a 20S-Cdc48 template (12);we were unable to distinguish a class populatedby such a complex.The CCC values had an approximately Gaus-

sian distribution (Fig. 1B), suggesting that almostall (>3s) 26S proteasome particles present in thetomograms were detected. In contrast, we couldnot identify 26S proteasome particles reliably incryotomograms acquiredwithout the Volta phaseplate (9). To quantify the signal quality of par-ticles acquiredwith the phase plate, we comparedthe single-capped 26S proteasomes to a high-resolution map obtained by cryoelectron micros-copy (cryo-EM) single-particle analysis [ProteinData Bank (PDB), 4cr2] (13) by Fourier cross-resolution (FCR). The resolution of the majorityof the subtomograms, each depicting an individ-ual single-capped26Sproteasomeparticle, rangedfrom 35 to 50 Å (Fig. 1C).Previous analyses of isolated proteasomes sug-

gest that double-capped proteasomes, single-capped proteasomes, uncapped proteasomes, anddifferent assembly intermediates all coexist in thecell (14, 15). However, the relative abundance ofthe different assemblies is unclear because celllysis and the subsequent purification steps maycause partial disassembly. To analyze the struc-tural heterogeneity of our in situ particle ensem-ble, we subjected it to statistical analysis bysubtomogram classification and averaging. First,we coherently aligned and averaged the sub-tomograms without using any external startingmodel to avoid template or reference bias (16)

(figs. S2 and S3). The resulting subtomogramaverage showed a single-capped 26S proteasome.However, the global average, as well as the cor-responding variance map, exhibited considerableheterogeneity at the uncapped end of the coreparticle (CP) (Fig. 2A). Both the faint density atthe CP end and the variance maps suggested thecoexistence of single- and double-capped 26Sproteasomes in our data set, consistent with invitro observations (15, 17). We used a variance-based 3D focused classification (18) approach todivide the particle set into two structurally homo-geneous classes, as well as a background bincomprising hypervariable subtomograms (fig. S4).The backgroundbin comprised 20S core particles,(dis-)assembly intermediates and, possibly, a fewremaining false positives. The particles assignedto the background bin were discarded after eachclassification step because the various molecularspecies contained in this bin were too few to yieldmeaningful averages. The main difference be-tween the two homogeneous classeswas the pres-ence or absence of the second RP (Fig. 2B). Thus,the first class depicts an average of double-capped26S proteasomes, whereas the second representsthe single-capped population. Double-capped 26Sproteasomes were far less abundant (425 par-ticles, 27%) than the single-capped complexes(1156 particles, 73%), very similar to the ratio ob-served in native gel densitometry of rat cortex(19). On average, 23 single- or double-capped 26Sproteasomes were found per tomogram, whichtranslates into a mean cytosolic concentrationof ~190 nM, a value close to the 26S proteasomesubunit concentration previously reported forSaccharomyces cerevisiae (140 to 200 nM) (20).To reveal differences in the two RPs of double-

capped 26S proteasomes, all proteasome par-ticles were subjected to in silico cutting betweenthe two b rings of the 20S core particle (Fig. 2C).The two halves of double-capped particles werethen treated as separate subtomograms. Becauseof the larger effective number of subtomograms,the resolution of the subtomogramaverage depict-ing a single-capped 26S proteasome improved to31 Å. The corresponding variance map indicateda high variability for distinct RP subunits, suchas Rpn6 and Rpn1 (Fig. 2C). Moreover, the var-iance was high in a region surrounded by Rpn1,Rpn2, Rpn10, and Rpn11, where there is no den-sity in single-particle reconstructions of purified26S proteasomes (11, 13, 21, 22). We then sep-arated the truncated particles into twomajor clas-ses, class 1 (at 27 Å resolution) and class 2 (31 Å),which were found in both single- and double-capped 26S proteasomes (Fig. 2D). The size of thetwo classes was unbalanced: Class 1 accountedfor 80% of all RPs (1367 particles) and class 2 foronly 20% (339 particles). The difference map be-tween the two classes (Fig. 2D)was in good agree-ment with the variance map, which was mostintense for Rpn1, Rpn6, and the area enclosed byRpn1, Rpn2, Rpn10, and Rpn11 (Fig. 2D).A very large cryo-EM data set of isolated S.

cerevisiae 26S proteasomes has recently revealedcoexisting conformations of purified 26S protea-somes referred to as s1, s2, and s3 (13). Substrates

SCIENCE sciencemag.org 23 JANUARY 2015 • VOL 347 ISSUE 6220 439

RESEARCH | REPORTS

Department of Molecular Structural Biology, Max-PlanckInstitute of Biochemistry, 82152 Martinsried, Germany.*These authors contributed equally to this work. †Correspondingauthor. E-mail: baumeist@biochem.mpg.de

on Novem

ber 30, 2020

http://science.sciencemag.org/

Dow

nloaded from

440 23 JANUARY 2015 • VOL 347 ISSUE 6220 sciencemag.org SCIENCE

Fig. 2. Subtomogram averaging and classifica-tion workflow. (A) Global average from all par-ticles displayed as center slice (left) and isosurfaceview (middle), as well as isosurface representationof the variance map (right; variance in violet). (B)First classification round separating single-cappedfrom double-capped 26S proteasomes.The differ-ence map between the two classes (right; differ-ence shown in red) shows the additional RP.The insilico cutting was performed along the blue dashedline between the two 20S b rings. (C) Resultingaverage displayed as slice (left) and isosurface(middle), as well as corresponding variance map(right). (D) Second classification of the cut parti-cles into two classes. Class 2 (left, light blue iso-surface) and Class 1 (middle, light green isosurface)mainly differ at Rpn1, Rpn6, and the substrate entrylocation (right, red isosurface). Scale bars for theslices, 25 nm.

RESEARCH | REPORTS

Fig. 1. Raw image and quality assessment. (A) Slice from a representative cryotomogram of acultured hippocampal neuronal cell. One single-capped (middle) and two double-capped 26Sproteasomes (left and right) are indicated by a red frame. The insets display magnified 1.7-nm Zslices through each proteasome volume. Scale bars: overview, 100 nm; insets, 25 nm. (B) Histogramdepicting the CCCs of all single-capped (blue bars) and double-capped (red bars) 26S proteasomes.The histograms were fitted with a Gaussian distribution (blue: single-capped 26S proteasomes, meanCCC~0.28; red: double-capped 26Sproteasomes,meanCCC~0.28).The dashed gray line indicates theaverage threshold (~0.12), belowwhich templatematching results have been discarded. (C) Histogramshowing the FCR of each single-capped 26S proteasome [blue in (B)] and the atomic model of theS. cerevisiae single-capped 26S proteasome (PDB, 4cr2).The mean cross resolution is at 47 Å.

on Novem

ber 30, 2020

http://science.sciencemag.org/

Dow

nloaded from

SCIENCE sciencemag.org 23 JANUARY 2015 • VOL 347 ISSUE 6220 441

RESEARCH | REPORTS

Fig. 4. Subclassification within GS and SPS. (A) Classification of GS classinto four subclasses. (Left) Variance map (top), the GS average (middle) andcorresponding down-sampled atomic model (bottom). (Center) SubclassesGS1 to GS4 displayed as isosurfaces; their relative abundances are indicated.

(Right) Differences between GS1 to GS4 and the down-sampled atomic model (red). (B) Classification of SPS class into three subclasses, SPS1 to SPS3,displayed as in (A).

Fig. 3. Rigid body fitting of atomic subunit mod-els into EM densities. (A) Fitted atomic modelsof the S. cerevisiae 26S proteasome subunits inthe GS and SPS EM densities (Fig. 2D). The sub-units Rpn9/5/6/7/3/12 are colored in differentshades of green, Rpn8/Rpn11 in light/darkmagen-ta, Rpn10 and Rpn13 in purple, Rpn1 in brown,Rpn2 in yellow, the AAA-ATPase hexamer in blue,and the CP in red (13). Red arrows indicate selectedRpns and the unassigned density in the SPS. (B)Single-capped and double-capped 26S protea-somes displayed as green (GS) and blue (SPS)isosurfaces overlaid on a slice of a representativetomogram. Scale bar, 500 nm. (C) Chart showingthe overall distribution of single-capped and double-capped 26S proteasomes (left) and of the differentstates within single-capped (middle) and double-capped (right) 26S proteasomes.

on Novem

ber 30, 2020

http://science.sciencemag.org/

Dow

nloaded from

most likely primarily bind to the most abundantlow-energy state s1; s2 mediates their tighterbinding (“commitment”) and activates the Rpn11deubiquitylating module; and s3 finally enablestranslocation of substrates into the CP (13). Tocompare the mammalian in situ classes to theS. cerevisiae in vitro conformations, atomicmod-els of the S. cerevisiae 26S proteasome subunitswere rigidly fitted into the densities of class 1 andclass 2 (Fig. 3A and fig. S5). Class 1 was most sim-ilar to s1 (substrate accepting) [root mean squaredeviation (RMSD) ~ 8.9 Å], whereas class 2 mostclosely resembled s3 (translocating) (RMSD~7.4Å)(fig. S5). In vitro, the 26S proteasome in complexwith a polyubiquitylated substrate, GFP-I27-Ub4,adopts a conformation that is essentially iden-tical to s3 and hence similar to class 2 (23). Thus,class 2 represents proteasomes that are engagedin substrate processing. Accordingly, we referto the major conformational states representedin class 1 and class 2 as “ground state” (GS) and“substrate-processing state” (SPS), respectively.Only 20% of 26S proteasomes in the analyzedneurons were in a SPS. We generated a 3D atlasof the 26S proteasome within a cell with theirindividual orientations and conformation states(Fig. 3B).Next, we analyzed whether the conformations

of the two RPs in each double-capped 26Sproteasomes correlated with one another. Thisanalysis matched the observed distribution withthe values predicted for uncorrelated states: GS-GS, 64% (measured) versus 63% (predicted); GS-SPS, 32% versus 33%; and SPS-SPS, 4% versus 4%(Fig. 3C), similar to previous single-particle cryo-EM studies (13, 23), favoring the notion that bothRPs can act independently.The GS density (Fig. 4A) showed a high var-

iance in the vicinity of the Rpn1 and Rpn13 sub-units. The corresponding subtomograms wereclassified into four classes to deconvolute theunderlying major structural differences (fig. S6).The first class (GS1, 24%) essentially did notdiffer from the density simulated with the fittedatomic model of a complex lacking the ubiquitinreceptor Rpn13 (24) (Fig. 4A). Rpn13 was also ab-sent in class GS2 (18%) and was found only in~58% of all proteasome particles, in agreementwith a dynamic binding of Rpn13 (25). Comparedwith GS1, GS2 exhibits a prominent additionaldensity associated with Rpn1, which functions as

a hub for a number of proteasome-interactingproteins (PIPs) (2). The remaining two ground-state classes, GS3 (34%) and GS4 (24%), showeddensity adjacent to Rpn2 that colocalizes withRpn13 in S. cerevisiae proteasomes (11, 13, 21, 22).This density was less prominent in GS3 but sub-stantially larger in GS4, most likely due to thepresence of PIPs.The subtomograms contributing to the SPS

were separated into three classes (Fig. 4B). Thedifferences of all three classes, when compared tothe atomicmodel, are located in a region above the“mouth” of the proteasomal ATPase Associatedwith diverse cellular Activities (AAA)–adenosinetriphosphatase (AAA-ATPase). In single-particlestudies, substrate was localized where we ob-served the additional masses in situ (23). How-ever, the additional densities in each of the threein situ classes correspond to ~150 kD to 250 kD,which is substantially larger than a typical pro-teasomal substrate or the one used in the in vitrostudies (23). In fact, one would anticipate a highdegree of structural variability of the proteasome-associated densities if they corresponded to sub-strates only, given their heterogeneous natureand different stages of processing. Most likely,the proteasome-associated densities in the SPSclasses depict mostly substrate-processing cofac-tors of the 26S proteasomes, such as deubiquityl-ating enzymes and E3 ubiquitin ligases, whichare frequently found in association with 26Sproteasomes (2).Advances in technology, such as direct detec-

tors and the contrast-enhancing Volta phaseplate openup opportunities for structural studiesin situ and visual proteomics (26). Fragile andhighly dynamic macromolecular complexes canbe studied in their functional and unperturbedcellular environments, providing quantitative in-formation about their states of assembly andconformation. A challenge for the future is tocorrelate this information in a systematic man-ner with topographic information about the cel-lular environments they inhabit.

REFERENCES AND NOTES

1. A. Hershko, A. Ciechanover, A. Varshavsky, Nat. Med. 6,1073–1081 (2000).

2. D. Finley, Annu. Rev. Biochem. 78, 477–513 (2009).3. D. Voges, P. Zwickl, W. Baumeister, Annu. Rev. Biochem.

68, 1015–1068 (1999).4. E. Leroy et al., Nature 395, 451–452 (1998).

5. H. C. Tai, E. M. Schuman, Nat. Rev. Neurosci. 9, 826–838(2008).

6. K. Willeumier, S. M. Pulst, F. E. Schweizer, J. Neurosci. 26,11333–11341 (2006).

7. M. D. Ehlers, Nat. Neurosci. 6, 231–242 (2003).8. V. Lucić, F. Förster, W. Baumeister, Annu. Rev. Biochem. 74,

833–865 (2005).9. R. Danev, B. Buijsse, M. Khoshouei, J. M. Plitzko, W. Baumeister,

Proc. Natl. Acad. Sci. U.S.A. 111, 15635–15640 (2014).10. J. Bohm et al., Proc. Natl. Acad. Sci. U.S.A. 97, 14245–14250

(2000).11. F. Beck et al., Proc. Natl. Acad. Sci. U.S.A. 109, 14870–14875

(2012).12. D. Barthelme, J. Z. Chen, J. Grabenstatter, T. A. Baker,

R. T. Sauer, Proc. Natl. Acad. Sci. U.S.A. 111, E1687–E1694(2014).

13. P. Unverdorben et al., Proc. Natl. Acad. Sci. U.S.A. 111,5544–5549 (2014).

14. T. Yoshimura et al., J. Struct. Biol. 111, 200–211 (1993).15. J. M. Peters, Z. Cejka, J. R. Harris, J. A. Kleinschmidt,

W. Baumeister, J. Mol. Biol. 234, 932–937 (1993).16. Y. Chen, S. Pfeffer, T. Hrabe, J. M. Schuller, F. Förster, J. Struct.

Biol. 182, 235–245 (2013).17. S. Bohn et al., Proc. Natl. Acad. Sci. U.S.A. 107, 20992–20997

(2010).18. Y. Chen, S. Pfeffer, J. J. Fernández, C. O. S. Sorzano, F. Förster,

Structure 22, 1528–1537 (2014).19. H. C. Tai, H. Besche, A. L. Goldberg, E. M. Schuman, Front. Mol.

Neurosci. 3, 12 (2010).20. C. G. Pack et al., Nat. Commun. 5, 3396 (2014).21. G. C. Lander et al., Nature 482, 186–191 (2012).22. K. Lasker et al., Proc. Natl. Acad. Sci. U.S.A. 109, 1380–1387

(2012).23. M. E. Matyskiela, G. C. Lander, A. Martin, Nat. Struct. Mol. Biol.

20, 781–788 (2013).24. K. Husnjak et al., Nature 453, 481–488 (2008).25. Z. Wang et al., Mol. Cell. Proteomics 7, 1639–1650 (2008).26. C. V. Robinson, A. Sali, W. Baumeister, Nature 450, 973–982

(2007).

ACKNOWLEDGMENTS

We thank Y. Chen for technical assistance and troubleshooting andP. Unverdorben for support in modeling and discussions. Theresearch leading to these results has received funding from theEuropean Commission under FP7 GA no. ERC-2012-SyG_318987-ToPAG. The work was additionally supported by the DeutscheForschungsgemeinschaft Excellence Cluster CIPSM and SFB 1035(both to W.B.) and FO 716/3-1 (to F.F.). Data availability: The cryo-EM maps for the GS and SPS were deposited into the EMDB withthe accession codes EMD-2830 and EMD-2831, respectively. FEICompany has submitted a patent for the Volta phase plate, whichis currently pending. Information on materials and methods isavailable on Science Online

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/347/6220/439/suppl/DC1Materials and MethodsSupplementary TextFigs. S1 to S6References (27–41)

12 September 2014; accepted 15 December 201410.1126/science.1261197

442 23 JANUARY 2015 • VOL 347 ISSUE 6220 sciencemag.org SCIENCE

RESEARCH | REPORTSon N

ovember 30, 2020

http://science.sciencem

ag.org/D

ownloaded from

proteasomes in intact neuronsSA molecular census of 26Shoh Asano, Yoshiyuki Fukuda, Florian Beck, Antje Aufderheide, Friedrich Förster, Radostin Danev and Wolfgang Baumeister

DOI: 10.1126/science.1261197 (6220), 439-442.347Science 

, this issue p. 439Sciencecapacity to deal with proteotoxic stress.suggests that these cells only use about 20% of their proteasomes in an unstressed state, which leaves significant sparean improved approach to electron cryotomography to look at proteasomes in intact hippocampal neurons. Their analysis

usedet al.isolated proteasomes is well established, but how intact proteasomes look within the cell is less clear. Asano ofcritical for protein quality control and for the regulation of numerous cellular processes in eukaryotic cells. The structure

proteasome is a protein machine that degrades intracellular proteins in the cytosol. The proteasome isSThe 26A detailed look at proteasomes in situ

ARTICLE TOOLS http://science.sciencemag.org/content/347/6220/439

MATERIALSSUPPLEMENTARY http://science.sciencemag.org/content/suppl/2015/01/21/347.6220.439.DC1

REFERENCES

http://science.sciencemag.org/content/347/6220/439#BIBLThis article cites 40 articles, 10 of which you can access for free

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAAS.ScienceScience, 1200 New York Avenue NW, Washington, DC 20005. The title (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Copyright © 2015, American Association for the Advancement of Science

on Novem

ber 30, 2020

http://science.sciencemag.org/

Dow

nloaded from