In situ structure determination by subtomogram averaging

BIROn - Birkbeck Institutional Research Online

Castaño-Díez, D. and Zanetti, Giulia (2019) In situ structure determination

by subtomogram averaging. Current Opinion in Structural Biology 58 , pp.

68-75. ISSN 0959-440X.

Downloaded from:

Usage Guidelines:

Please refer to usage guidelines at

or alternatively

In

situ

structure

determination

by

subtomogram

averaging

Daniel

Castan˜o-Dı´ez

and

Giulia

Zanetti

Cryo-tomographyandsubtomogramaveragingare

increasinglypopulartechniquesforstructuraldeterminationof macromolecularcomplexesinsitu.Theyhavethepotentialto achievehigh-resolutionviewsofnativecomplexes,together withthedetailsoftheirlocationrelativetointeracting molecules.Thesubtomogramaveraging(StA)pipelinesare well-established,withcurrentdevelopmentsaimingtooptimise eachstepbyreducingmanualinterventionanduserdecisions, followingsimilartrendsinsingle-particleapproachesthathave dramaticallyincreasedtheirpopularity.Here,wereviewthe mainstepsoftypicalStAworkflows.Wefocuson

considerationsarisingfromthefactthattheobjectsofstudyare embeddedwithinuniquecrowdedenvironments,andwe emphasisethosestepswherecarefuldecisionsneedtobe madebytheuser.

Addresses

1_{BioEMLab,CenterforCellularImagingandNanoanalytics,}

Biozentrum,UniversityofBasel,Mattenstrasse26,CH-4058,Basel, Switzerland

2_{InstituteofStructuralandMolecularBiology,BirkbeckCollege,Malet}

St.,London,WC1E7HX,UK

Correspondingauthors:

Castan˜o-Dı´ez,Daniel([email protected]), Zanetti,Giulia([email protected])

CurrentOpinioninStructuralBiology2019,58:68–75

ThisreviewcomesfromathemedissueonBiophysicaland com-putationalmethods

EditedbyLauraItzhakiandAlanLowe

https://doi.org/10.1016/j.sbi.2019.05.011

0959-440X/ã2019TheAuthors.PublishedbyElsevierLtd.Thisisan openaccessarticleundertheCCBY-NC-NDlicense( http://creative-commons.org/licenses/by-nc-nd/4.0/).

Introduction

Cryo-electron tomographyis thetechnique ofchoice

forstructuralanalysisofbiologicalcomplexes intheir

native context. Through imaging in the electron

microscopeat arange of tiltangles,a 3D

reconstruc-tion is obtained of pleiomorphic and unique

speci-mens. If many copies of a complex of interest are

present within the tomographic reconstruction,

sub-tomogramaveraging(StA)cansignificantlyincreaseits

resolution,withoutdetractingfromtheadvantagethat

thereconstructed objects are embedded within their

nativebiological context.StAis increasinglyused for

structural analysis of protein complexes in situ,

whether it be in eukaryotic cells (e.g. Nuclear pore

complexes [1,2], ribosomes [3], proteasomes [4,5],

coat protein complexes [6,7]) bacterial cells (e.g.

secretion systems [8,9,10], flagellar motors [11,12],

S-layer [13]), in purified organelles or viruses (e.g.

viral glycoproteins [14,15], viral capsids [16,17],

mitochondrial complexes [18–20], translating

ribo-somes [21]), or on reconstituted systems (e.g.

mem-brane remodelling coat complexes [22,23,24],

cyto-skeletalmotors [25]).

The typical workflow of StA is represented in

Figure 1. It starts with collection of a tilt series, and

tomogramreconstruction,andthenit involvesparticle

picking, subvolume extraction and generation of a

startingreference, uponwhich iterations ofalignment

and averaging, and eventually classification are

per-formed.Thestructuresthatareobtainedaresubjected

to postprocessing and validation before being

inter-preted.Atmanystagesthroughtheprocess,theresults

can informthe previoussteps, which can be repeated

withrefined parameters.

Processing cryo-tomographic data has necessitated

development of ad hoc protocols to deal with tilted

images and with anisotropic 3D volumes [26,27]. For

instance,defocus-dependentmodulationofimage

fre-quencies (as defined by the Contrast Transfer

Func-tion,CTF) varies acrosstilted images and within the

height of the reconstructed volume, requiring

dedi-cated CTF correction software. The limited angular

rangeto which samples can be tilted (typically60)

leads to anisotropy in the reconstructed volumes

(describedasamissing-wedgeinFourierspace),which

must be compensated during subvolume alignments

[26,27].Moreover, while thecomplexity anddiversity

ofthesamplesanalysedareusuallythereasonswhyStA

ischosenoversingleparticleanalysisorotherstructural

techniques,it alsoimplies thatdata analysis protocols

must be adapted to the object of interest. This is

especiallytrue duringparticle picking, initial

orienta-tionassessment, and in the validation and

interpreta-tionstages, asoutlinedbelow.

Here, we review the typical steps of the StA workflow

(Figure1),withanemphasisonthoseaspectsthatmust

beoptimiseddependingontheparticularcontextofthe

Preprocessing

and

tomogram

reconstruction

When reconstructing cryo-tomogramsforsubtomogram

averaging it is desirable to preserve high resolution

information asmuch as possible.To this aim,

pre-pro-cessing of tilt images includes steps that have been

optimised for high-resolution single particle cryo-EM.

Motion correction of frames within each tilt image

improves on the degrading effects of beam induced

sample motion and ofdrift. Drifttends to beworse at

hightilts,makingmotioncorrectionespeciallyimportant

incryo-tomography.Exposurefilteringtoremove

cumu-lativeelectrondamage-derivednoise[28]canbe

imple-mented on a tilt-by-tilt basis or within subframes of

every tilt.Exposurefilters workoptimallywhenpaired

withdose-symmetricdatacollectionschemes,wherethe

hightiltimages,in whichhighresolutioninformationis

corruptedbyincreasedthicknessandlowerprecisionof

CTF estimation, are effectively filtered in the

dose-compensation procedure[29].

Tomaximisehigh-resolutionsintheaveraged

subtomo-grams,itisimportantthattiltimagesarealignedwiththe

highest possible accuracy before back-projection. Tilt

geometry must be optimised by iteratively minimising

theresidual differencesbetween theactualpositions of

samplefeaturesandthosepredictedbythebest

geomet-rical tilt model [30,31]. When possible, this should be

doneusinggoldbeadsasfiducials.Incertainsamples(e.g.

cell sectionsor lamellae), it is not possible to add gold

beads and alignments must rely on image ‘patches’ to

serveasfiducials[32,33].Successinthiscasedependson

theamountof signalavailable.

Currently, obtaining high-quality tilt series alignments

suitable for high-resolution structure determination

requires a manual procedure of fiducial optimisation,

which significantly decreases the throughput. Recent

developmentsusealignmentsofsubtomogramstofurther

refinethelocalgeometryofthetiltseries,improving

high-resolution signal ([34], further discussed in the

‘Alignmentandaveraging’section).

Accurate CTF estimation and correction are important

aspectstoconsiderwhenaimingtopreserve

high-resolu-tion features. As discussed above, dose-symmetric tilt

schemes ensure that the high resolution information is

preserved in the lower tilts, where CTF estimation is

moreaccurateduetohigherSNR [29].Comparedwith

single-particle reconstruction, cryo-tomography has the

advantage that the relative defocus is known for each

voxelinthereconstruction.Thisallowstocalculatea

3D-CTF-correctedtomogrambyback-projectingimage

pix-elswheresignalhasbeenrestoredusingtheappropriate

z-height-adjusteddefocusvalues[35,36].

Tomogram

content

analysis

CurrentSingleParticlesAnalysisstudieshandlethe

loca-lisation of individual particles on micrographs as a

straightforward task. In contrast, embarking on an StA

project requires a decision on the specific strategy to

determine the approximate locations of copies of the

proteinofinterest insideeachtomogram.

In the simplest scenario, particles can be selected as

individual entities,boththroughmanualinspectionand

interaction with the tomograms [37] or through some

automated approach for pattern recognition. Such

SubtomogramaveragingCastan˜o-Dı´ez andZanetti 69

Figure1

Tomogram reconstruction

Particle picking

Creation of reference

Alignments

Averaging

Classification

Postprocessing and validation Data collection

and preprocessing

2 3

12

10

7 8

11 9

5 6

4 1

*Split dataset

*Merge dataset

Current Opinion in Structural Biology

Overviewofthetypicalsubtomogramaveragingworkflow.1.Upon datacollection,imagesaresubjectedtoframealignment,dose compensation,and–insomeprotocols–CTFcorrection,then back-projectedtoreconstructthetomogram.2.Coordinatesforparticle pickingaredefinedandsubvolumesextracted.3.Thesecanbe averagedtoobtainastartingreference.4.Thereferenceisthenused toalignsubvolumes.Atthisstagethedatasetshouldbesplitin halvestofollowgold-standardprocedures.5,6.Thefirstalignments canproviderefinedcoordinatesforsubvolumeextraction.7, 8.Alignmentsandaveragingcyclesareiterateduntilconvergence. 9.Insomeprotocols,subtomogramalignmentscanbeusedtorefine thetiltgeometryandobtainamoreaccuratetomogram

approaches range from exhaustive testing of the

tomo-gram searching local maxima of similarity to a given

template measured by cross-correlation [38], to more

sophisticated applications of deep learning [39,40], a

techniqueofartificialintelligencewherethesoftwareis

trained to mimic the performance of an expert human

operator.

However,thesignatureabilityoftomographyisthe

expo-sition of the molecular context. Frequently the spatial

locationoftheparticlesofinterestinatomogramisrelated

to an approximately identifiable geometric distribution,

inducedbyanorganisedstructureinthecellularmilieu.

Particlesmaybearrangedasdecorations,embeddedinsets,

or buildingblocks of filamentous or membranous

struc-tures,withdifferentlevelsofregularity.Particle

identifica-tioninthesecasescanbearticulatedalongthe

characteri-sationandmodellingoftheunderlyingshape(Figure2).

Parametrisationof–forexample–amicrotubuleasabent

cylinder(Figure2a),ofavesicleasanellipsoid(Figure2b)

orofamitochondrialmembraneasatriangulatedmanifold

(Figure2c)allowstoexploitthepredictedspatial

correla-tionoftheparticlesinrelationtosuchobjectsandextract

themcollectively.Hereby,themodelisusedtocreateaset

ofspatiallocations,generatedunderthedoubleconstraint

of being consistent with the support geometry and also

withinaphysicallymeaningfuldistanceamongeachother.

For a sufficiently dense distribution of positions and a

sufficiently large cropping side length, boxes extracted

aroundtheselocationswillprobablycontainactual

parti-cles,althoughthepositionoftheparticleinsidethe

sub-tomogramboxwillberandom.Thepositionsofparticles

insidetheboxescanthenbeestimatedthroughalignment

of subtomograms,a procedurethat can beaidedby the

approximate orientation that can be inferred for each

particlefromitsrelationto themodel(typicalexamples

arethenormaltoamembrane,orthedirectionoftheaxisof

afilament)(Figure3a,b[22,23]).

Asanorganisedstructuremightcontainalargenumberof

particles(rangingbetweentensandhundredsdepending

onthe particular case), computer tools for efficient

on-screenannotation of support geometriesare a powerful

resource [37]. Complementarily, the identification of

supportgeometriescanbefacilitatedthroughvolumetric

segmentationmethods[41],ideallytailoredto the

char-acteristic of cryoET data — mainly low SNR and the

presenceofmissingwedge[42,43].Inthisinsight,recent

studies have explored the suitability of deep learning

approachesfor semiautomated identificationof some of

thestructuresin thecellularenvironment thatare

typi-cally used as supporting geometries for subtomogram

extraction[39].

Alignment

and

averaging

StAseekstoreconstitutethe3Dshapeofa

macromolec-ularcompoundbyintegratingthesignalofmultiplenoisy

Figure2

(a)

(b)

(c)

Current Opinion in Structural Biology

Examplesofparticleextractionfor(a)filaments,(b)vesiclesand(c)

genericallyshapedmembranes.Inallthreecases,thepositionofa subtomogram(redcirclefilledinblack)isaccompaniedbyaninitial estimationoftheorientationoftheparticle(redbar)givenbythe respectivesupportgeometry.In(a),orangeboxesrepresenting subtomogramsareplacedalongtheaxisofamicrotubule(dataof microtubule-bounddynein–dynactincomplexesfromGrotjahnetal.

copies contained in theextracted set of subtomograms.

Knowledgeoftheexactorientationandlocationof each

particleinsideitssubtomogramboxallowstoalignallthe

particles to acommon reference,then to average them

coherently. StA finds these parameters in an iterative

fashion. Adensitymapusedas initialreferenceis

com-paredtoeachparticleinthedataset,locatingthe

combi-nation of translation and rotation that maximises the

similarityofthatparticletothereference.Thealignment

parametersoftheentiredatasetarethenusedtocreatean

averagethatwillbeusedasreferenceforanewiteration

untilconvergenceisattained.Thisgenericprocedureis

known as refinement and its implementation entails

several practicalconsiderations:

- Theinitialreferencecanbeconstructedbyaveraginga

subset of the particles. In cases where the sample

geometry allows to assign a rough initial orientation

to particles(e.g. objectson amembraneor along

fila-ments),thelow-passfilteredreferencecreatedby

aver-aging subtomograms will contain low-resolution

fea-tures that correlate with those of each particle,

greatlyfacilitatingalignments.Forexample,theinitial

referencefor particles extractedfrom thesurfaceof a

membranemightlooklikeasmoothpieceofmembrane

withrelevant curvature.

- Frequently,theparticle ofinterestdoesnotappearas

anisolatedentityinsidethesubtomogram,but

embed-dedintoahigh intensitystructure (as amembraneor

filament),orcloselysurroundedbyotherinhabitantsof

thecytosol.Thiscallsforacarefulmaskingoftheinitial

reference,thatis,providingthealgorithmwithamask

filethatexpresseswhichpixelsinthereferenceshould

contain the particle– and thus common signal in all

subtomograms – and which ones canvary itscontent

amongthedataset.

SubtomogramaveragingCastan˜o-Dı´ez andZanetti 71

Figure3

(a) (e)

(b) (f)

(c) (d) (g) (h)

Good particles Bad particles / noise

Good alignment Bad alignment

mask

edge maskedge

Picked positions

Aligned positions

Unmasked average

membrane membrane

real features outside mask noise outside mask reference bias

Current Opinion in Structural Biology

- Similaritybetweenreferenceandparticle ismeasured

throughcross-correlation(refs),withappropriate

adap-tiontoaccountforthemissingwedgeofthetomogram

and the mask imposed on the reference. Alternative

approaches based on maximum likelihood use the

similaritymeasuredforeachcombinationoftranslation

ofrotationtoassigntoitaprobabilisticweight[44,45].

Either way, the best combination of translation and

rotation parameters for a particle are determined by

exhaustivelytestingonasetofshiftsandEulerangles.

Computationally, thiscalculation constitutesthemost

demandingtask,anddifferentapproachesareavailable

toreducecomputingtime:thesetofscannedanglescan

beconstrainedtoaneighbourhoodofpreviouslyfound

alignment parameters; Fourier convolution allows to

compute all translations simultaneously for a given

orientation; similarly spherical harmonic transforms

allow the simultaneous computation of all rotations

for a given translation [46]; finally, the structure of

the operation (which reuses the same data in a large

number of computations) enables GPU

implementa-tionstoyieldimpressivespeedupfactors[47].

Recenteffortsexplore theuse ofthe alignment

param-eterscomputedonasetof subtomogramstorealignthe

tilt series using the particles themselves as fiducials,

refiningthealignmentgeneratedbygoldbeads[34].This

potentiallyleadstomoreaccuratelyalignedtiltseriesand

correspondingly better tomograms and subtomograms,

which,inturn,canberefinedtoproduce abetter

align-ment. This procedure canbe iterated till convergence,

and has been shownto produce considerable

improve-ments for samples with large numbers of copies per

tomogram.

Classification

Classification methods that havebeen reported to deal

with heterogeneity canbe separated intoi) approaches

aimed at characterising the heterogeneity of an aligned

dataset,and ii)algorithmsperformingclassificationand

alignment simultaneously. In the former group, a

fre-quentstartingpointistheexhaustivecomputationofthe

similarity of each pair of alignedsubtomograms, giving

rise to acovariance matrix.This matrixcan beused to

feedaHierarchicalClusteringmethodthatlocatesgroups

ofmutuallysimilarparticles[48],orasbasisforaPrincipal

Component Analysis [49], a widely used method for

complexity reduction that captures the spatial features

describingthelargestsourcesofstructuralvarianceinthe

dataset.Suchfeatures(eigenvolumes)appearinorderof

relevanceand canbeusedto expresseach particle asa

reducednumberofcoefficients,whichcanthenbe

clus-tered through k-means. Performance of these methods

hasbeenshowntoimprovesignificantlybyadata-driven

tuningoffrequencybandusedtomeasureparticle

simi-larity[34].

The secondfamily ofmethods isbased oncompetitive

alignment.Thealgorithmcompareseachparticlewitha

setofdifferentreferences,andassignsittothereference

thatyieldsthehighest score.Assubsequentrefinement

iterations generate gradually better references, particle

assignmentscanswapfromonereferencetootherasthe

iterations progress. This procedure is similar to

maxi-mum-likelihood-basedapproacheswhichreplaceunique

classassignmentswithprobabilityweights[50].

Inbothfamilies,maskingconstitutesacrucialdecision,as

it allows one to focus the analysis on the parts of the

compound deemed to be flexible or conformationally

different. While most methods allow to integrate this

informationasapriorifavailable,othersarespecifically

designedtoautomaticallyanalysetheregionsofmaximal

variancetogeneratedata-drivenmasksthatupdateasthe

classificationadvances[51].

Validation

Once the classification and alignment procedures have

reached convergence the final average is ready to be

interpreted: subunits and domains are assigned and,

whereavailable,atomic models arefitted or built.

Vali-dationofthestructureiscrucialbeforeanyinterpretation

canbe attempted. Manyof the validation strategies in

subtomogram averaging are shared with single particle

reconstructionmethods:forinstance,theriskofcreating

an artefactual average through overfitting can be

miti-gatedbythegold standardprocedure[52].Thisimplies

comparisonbetweenhalf datasetswhich were

indepen-dently aligned against a band-pass filtered reference.

Only information which is recovered beyond the high

frequencylimitofthereferencefiltercanbeconsidered

forthe nextalignment iteration,and eventuallydefines

theresolutionto whichtheaveragecanbeinterpreted.

ResolutioniscommonlyassessedbyFourierShell

Corre-lation(FSC)[53].TheeffectsofusingmasksontheFSC

have been described, and can be mitigated by phase

randomisation methods [54]. Often, protein and lipid

density within subtomograms continues beyond the

actualparticlestoencompassassociatedmacromolecules

untiltheperiphery of thebox.The choiceofmask will

therefore significantly affect the average resolution, as

non-correlatingfeaturesareincludedintheFSC

calcula-tion.Forthis reason, andbecauseof theinherent

flexi-bility of all biological complexes, especially when

resolvedinsitu,itisoftendesirabletomeasurethelocal

resolution [55,56]. This might be combined with local

filterstoobtainamapwhereitispossibletointerpreteach

feature at the appropriate level of detail. It remains

importantto calculatethegold-standard FSC asa

diag-nostictoolforoverfittingorothertypicalartefacts[57].

In cryo-tomography, the contextualinformation can be

exemplifiedinFigure3).Forexample,whensubunitsare

arranged in a pattern with respect to each other, and

particle picking was performed by oversampling the

underlyinggeometricalmodel(Figures2and 3a,e)then

theconvergenceofneighbouringsubtomogramstoforma

recognisable pattern is a very good validation of the

correctness of the alignments (compare Figure 3b and

f).Moreover,forembeddedobjectstypicalofStA,amask

applied during alignments may cut across continuous

density,suchasthatofamembrane.Inthiscase,correct

alignment of ‘real’ particles will produce an average

where the membrane is visible beyond the mask

(Figure 3c,d),whileincorrectalignments,or alignments

ofnoisewillproduceanaveragewherethedensitydrops

off suddenlyoutsidethemask(Figure 3g,h).

Summary

The complexityof thebiologicalsamplestypically

ana-lysed in cryo-tomography defines this technique and

constitutes its biggestpotential. At the same time, the

reasons thatmake StA an attractive approach for

struc-turaldeterminationalsodefineitscurrentlimitations,for

example,inrelationto theinefficiencyinattaininghigh

numbers of particles, or to the bias in the alignments

imposedbycontextualinformationoutsidethe‘particle’

and thedifficulty indefiningoptimal masks.

StA iscurrently relatively low-throughputwith respect

to cryo-EM single particle approaches. Nevertheless,

there is scope for improvements in automation ofStA

pipelines, especially in data collection, pre-processing

andtomographicreconstruction.StAwillgreatlybenefit

fromcomprehensivesoftwarepackagestooptimiseeach

processingstepwithouttheneedforcontinuous

conver-sion betweenformats and conventions[34,58,59].

Soft-ware which will facilitate the flexible combination of

different approaches will also significantly improve

usability and help expand the user

community—simi-larly to what has been implemented in single particle

analysis [52,60].

So far, only a few StA projects have achieved

near-atomicresolutions [16,22,34],butit isimportant to

note that novel and insightful biological information

can beretrievedfromlowerresolution studies,aslong

asappropriatevalidationproceduresareimplemented.

This is especially true for cryo-tomography and StA,

since the macromolecular structure itself only

consti-tutesoneaspectoftheresult:theinformationaboutthe

spatial relationship of the complexes of interest

between eachother, andwithother neighbouring

cel-lular features canoftenprovidemuch preciousinsight

into theirfunction.

Conflict

of

interest

statement

Nothing declared.

Acknowledgements

WewouldliketothankJoshuaHutchings,HimaniAmin,GabrielLander andDanielleGrotjahnforsharingthedatausedinthefigures.D.C.D.is supportedbytheSwissNationalScienceFoundation(SNF

205321_179041).G.Z.issupportedbyaRoyalSocietyDorothyHodgkin Fellowship(DH130048),andanAMSSpringboardaward(SBF0031030).

References

and

recommended

reading

Papersofparticularinterest,publishedwithintheperiodofreview, havebeenhighlightedas:

ofspecialinterest

ofoutstandinginterest

1.

MosalagantiSalome´ PA,MerchantS,KosinskiSS,J,PlitzkoAlbertS,JM,SchafferBaumeisterM,StrenkertW,EngelD,BD etal.:Insituarchitectureofthealgalnuclearporecomplex.Nat Commun2018,9:2361.

The authors use cryo-FIB/SEM, cryo-tomographyand subtomogram averagingtodeterminethestructureoftheNPCfromChlamydomonas cells.

2. MahamidJ,PfefferS,SchafferM,VillaE,DanevR,CuellarLK, Fo¨rsterF,HymanAA,PlitzkoJM,BaumeisterW:Visualizingthe molecularsociologyattheHeLacellnuclearperiphery. Science2016,351:969-972.

3. PfefferS,DudekJ,SchafferM,NgBG,AlbertS,PlitzkoJM, BaumeisterW,ZimmermannR,FreezeHH,EngelBDetal.:

Dissectingthemolecularorganizationofthe translocon-associatedproteincomplex.NatCommun2017,8:14516.

4. AlbertS,SchafferM,BeckF,MosalagantiS,AsanoS,ThomasHF, PlitzkoJM,BeckM,BaumeisterW,EngelBD:Proteasomes tethertotwodistinctsitesatthenuclearporecomplex.Proc NatlAcadSciUSA2017,114:13726-13731.

5. GuoQ,LehmerC,Martı´nez-Sa´nchezA,RudackT,BeckF, HartmannH,Pe´rez-BerlangaM,FrottinF,HippMS,HartlFUetal.:

InsitustructureofneuronalC9orf72Poly-GAaggregates revealsproteasomerecruitment.Cell2018,172:696-705.e12.

6.

BykovBaumeisterYS,SchafferW,EngelM,BD,DodonovaBriggsJA:SO,TheAlbertstructureS,PlitzkooftheJM,COPI coatdeterminedwithinthecell.eLife2017,6:e32493.

The authors use cryo-FIB/SEM, cryo-tomographyand subtomogram averagingtodeterminethestructureoftheCOPIcomplexfrom Chla-mydomonascells.

7. KovtunO,LenevaN,BykovYS,AriottiN,TeasdaleRD,SchafferM, EngelBD,OwenDJ,BriggsJAG,CollinsBM:Structureofthe membrane-assembledretromercoatdeterminedby cryo-electrontomography.Nature2018,561:561-564.

8. RapisardaC,CherrakY,KoogerR,SchmidtV,PellarinR,LoggerL, CascalesE,PilhoferM,DurandE,FronzesR:Insituand high-resolutioncryo-EMstructureofabacterialtypeVIsecretion systemmembranecomplex.EMBOJ2019:e100886.

9.

HornBo¨ckM,D,MedeirosPilhoferM:JM,InsituTsaoarchitecture,H-F,PenzT,function,WeissGL,andAistleitnerevolutionK, ofacontractileinjectionsystem.Science2017,357:713-717.

Theauthorsusecryo-tomographyandsubtomogramaveragingto deter-minethestructureofthetype6secretionsysteminbacterialcells. 10. ParkD,Lara-TejeroM,WaxhamMN,LiW,HuB,Gala´nJE,LiuJ:

VisualizationofthetypeIIIsecretionmediatedSalmonella-host cellinterfaceusingcryo-electrontomography.eLife2018,7.

11. KaplanM,GhosalD,SubramanianP,OikonomouCM,KjaerA, PirbadianS,OrtegaDR,BriegelA,El-NaggarMY,JensenGJ:The presenceandabsenceofperiplasmicringsinbacterial flagellarmotorscorrelateswithstatortype.eLife2019,8.

12. QinZ,TuJ,LinT,NorrisSJ,LiC,MotalebMA,LiuJ: Cryo-electrontomographyofperiplasmicflagellainBorrelia burgdorferirevealsadistinctcytoplasmicATPasecomplex. PLoSBiol2018,16:e3000050.

13. BharatTAM,Kureisaite-CizieneD,HardyGG,YuEW,DevantJM, HagenWJH,BrunYV,BriggsJAG,Lo¨weJ:Structureofthe hexagonalsurfacelayeronCaulobactercrescentuscells.Nat Microbiol2017,2:17059.

14. TranEEH,NelsonEA,BonagiriP,SimmonsJA,ShoemakerCJ, SchmaljohnCS,KobingerGP,ZeitlinL,SubramaniamS, WhiteJM:Mappingofebolavirusneutralizationbymonoclonal antibodiesintheZMappcocktailusingcryo-electron tomographyandstudiesofcellularentry.JVirol2016,

90:7618-7627.

15. RiedelC,VasishtanD,SiebertCA,WhittleC,LehmannMJ, MothesW,Gru¨newaldK:Nativestructureofaretroviral envelopeproteinanditsconformationalchangeupon interactionwiththetargetcell.JStructBiol2017,197:172-180.

16.

SchurKirkpatrickFKM,JM,ObrSachseM,HagenC,Kra¨usslichWJH,WanH-G,W,BriggsJakobiJAG:AJ, Anatomic modelofHIV-1capsid-SP1revealsstructuresregulating assemblyandmaturation.Science2016,353:506-508.

Cryo-tomographyandsubtomogramaveragingresolvesthestructureof theHIVcapsidatnear-atomicresolution.

17. WanW,KolesnikovaL,ClarkeM,KoehlerA,NodaT,BeckerS, BriggsJAG:StructureandassemblyoftheEbolavirus nucleocapsid.Nature2017,551:394-397.

18. DaviesKM,AnselmiC,WittigI,Faraldo-Go´mezJD,Ku¨hlbrandtW:

StructureoftheyeastF1Fo-ATPsynthasedimeranditsrolein shapingthemitochondrialcristae.ProcNatlAcadSciUSA 2012,109:13602-13607.

19. DaviesKM,BlumTB,Ku¨hlbrandtW:Conservedinsitu arrangementofcomplexIandIII2inmitochondrialrespiratory chainsupercomplexesofmammals,yeast,andplants.Proc NatlAcadSciUSA2018,115:3024-3029.

20. PfefferS,WoellhafMW,HerrmannJM,Fo¨rsterF:Organizationof themitochondrialtranslationmachinerystudiedinsituby cryoelectrontomography.NatCommun2015,6:6019.

21. PfefferS,BurbaumL,UnverdorbenP,PechM,ChenY, ZimmermannR,BeckmannR,Fo¨rsterF:Structureofthenative Sec61protein-conductingchannel.NatCommun2015,6:8403.

22.

HutchingsaveragingJ,ofStanchevaCOPIIassembliesV,MillerrevealsEA,ZanettihowG:coatSubtomogramorganization dictatesmembraneshape.NatCommun2018,9:4154.

Cryo-tomographyandsubtomogramaveragingresolvethestructureof theCOPIIcoatassembledonmembranesat4.9A˚ resolution.

23. DodonovaSO,AderholdP,KoppJ,GanevaI,Ro¨hlingS, HagenWJH,SinningI,WielandF,BriggsJAG:9A˚ structureofthe COPIcoatrevealsthattheArf1GTPaseoccupiestwo contrastingmolecularenvironments.eLife2017,6.

24. ZanettiG,PrinzS,DaumS,MeisterA,SchekmanR,BaciaK, BriggsJAG:ThestructureoftheCOPIItransport-vesiclecoat assembledonmembranes.eLife2013,2:e00951.

25. GrotjahnDA,ChowdhuryS,XuY,McKenneyRJ,SchroerTA, LanderGC:Cryo-electrontomographyrevealsthatdynactin recruitsateamofdyneinsforprocessivemotility.NatStruct MolBiol2018,25:203.

26. WanW,BriggsJA:Cryo-electrontomographyand subtomogramaveraging.MethEnzymol2016,579:329-367.

27. HutchingsJ,ZanettiG:Finedetailsincomplexenvironments: thepowerofcryo-electrontomography.BiochemSocTrans 2018,46:807-816.

28. GrantT,GrigorieffN:Measuringtheoptimalexposureforsingle particlecryo-EMusinga2.6A˚ reconstructionofrotavirusVP6. eLife2015,4:e06980.

29.

HagenelectronWJH,tomographyWanW,Briggstilt-schemeJAG:optimizedImplementationforhighofresolutiona cryo-subtomogramaveraging.JStructBiol2017,197:191-198.

Adosesymmetrictiltschemeisdevelopedwhichhashadahighimpact oncurrentdatacollectionprotocolsforsubtomogramaveraging.

30. KremerJR,MastronardeDN,McIntoshJR:Computer visualizationofthree-dimensionalimagedatausingIMOD.J StructBiol1996,116:71-76.

31. MastronardeDN,HeldSR:Automatedtiltseriesalignmentand tomographicreconstructioninIMOD.JStructBiol2017,

197:102-113.

32. NobleAJ,StaggSM:Automatedbatchfiducial-lesstilt-series alignmentinAppionusingProtomo.JStructBiol2015,

192:270-278.

33. Castan˜o-Dı´ez D,SchefferM,Al-AmoudiA,FrangakisAS:

Alignator:aGPUpoweredsoftwarepackageforrobust fiducial-lessalignmentofcryotilt-series.JStructBiol2010,

170:117-126.

34. HimesBA,ZhangP:emClarity:softwareforhigh-resolution cryo-electrontomographyandsubtomogramaveraging.Nat Methods2018,15:955-961.

35. KunzM,FrangakisAS:Three-dimensionalCTFcorrection improvestheresolutionofelectrontomograms.JStructBiol 2017,197:114-122.

36.

Turocorrectionnova´ B,forSchurcryo-electronFKM,WantomographyW,BriggsJAG:usingEfficientNovaCTF3D-CTF improvessubtomogramaveragingresolutionto3.4A˚.JStruct Biol2017,199:187-195.

Auser-friendlysoftwarefor3DCTFcorrectionthatimprovesthe resolu-tionofsubtomogramaveraging.

37. Castan˜o-Dı´ez D,KudryashevM,StahlbergH:Dynamocatalogue: geometricaltoolsanddatamanagementforparticlepickingin subtomogramaveragingofcryo-electrontomograms.JStruct Biol2017,197:135-144.

38. FrangakisAS,Bo¨hmJ,Fo¨rsterF,NickellS,NicastroD,TypkeD, HegerlR,BaumeisterW:Identificationofmacromolecular complexesincryoelectrontomogramsofphantomcells.Proc NatlAcadSciUSA2002,99:14153-14158.

39. ChenM,DaiW,SunSY,JonaschD,HeCY,SchmidMF,ChiuW, LudtkeSJ:Convolutionalneuralnetworksforautomated annotationofcellularcryo-electrontomograms.NatMethods 2017,14:983-985.

40. XuM,ChaiX,MuthakanaH,LiangX,YangG, Zeev-Ben-MordehaiT,XingEP:Deeplearning-basedsubdivision approachforlargescalemacromoleculesstructurerecovery fromelectroncryotomograms.Bioinformatics2017,33:i13-i22.

41. PintilieG,ChiuW:Comparisonofseggerandothermethodsfor segmentationandrigid-bodydockingofmolecular

componentsincryo-EMdensitymaps.Biopolymers2012,

97:742-760.

42. Martinez-SanchezA,GarciaI,AsanoS,LucicV,FernandezJ-J:

Robustmembranedetectionbasedontensorvotingfor electrontomography.JStructBiol2014,186:49-61.

43. LuengoI,DarrowMC,SpinkMC,SunY,DaiW,HeCY,ChiuW, PridmoreT,AshtonAW,DukeEMHetal.:SuRVoS:super-region volumesegmentationworkbench.JStructBiol2017,

198:43-53.

44. ScheresSHW,MeleroR,ValleM,CarazoJ-M:Averagingof electronsubtomogramsandrandomconicaltilt

reconstructionsthroughlikelihoodoptimization.Structure 2009,17:1563-1572.

45. BharatTAM,RussoCJ,Lo¨weJ,PassmoreLA,ScheresSHW:

Advancesinsingle-particleelectroncryomicroscopy structuredeterminationappliedtosub-tomogramaveraging. Structure2015,23:1743-1753.

46. ChenY,PfefferS,HrabeT,SchullerJM,Fo¨rsterF:Fastand accuratereference-freealignmentofsubtomograms.JStruct Biol2013,182:235-245.

47. Castan˜o-Dı´ezD:Thedynamopackagefortomographyand subtomogramaveraging:componentsforMATLAB,GPU computingandEC2Amazonwebservices.ActaCrystallogrD StructBiol2017,73:478-487.

48. XuM,BeckM,AlberF:High-throughputsubtomogram alignmentandclassificationbyFourierspaceconstrainedfast volumetricmatching.JStructBiol2012,178:152-164.

49. Fo¨rsterF,PruggnallerS,SeybertA,FrangakisAS:Classification ofcryo-electronsub-tomogramsusingconstrained correlation.JStructBiol2008,161:276-286.

basedclassificationofelectrontomographicdata.JStructBiol 2011,173:77-85.

51. ChenY,PfefferS,Ferna´ndezJJ,SorzanoCOS,Fo¨rsterF:

Autofocused3Dclassificationofcryoelectron subtomograms.Structure2014,22:1528-1537.

52. ScheresSHW:RELION:implementationofaBayesian approachtocryo-EMstructuredetermination.JStructBiol 2012,180:519-530.

53. vanHeelM,SchatzM:Fouriershellcorrelationthreshold criteria.JStructBiol2005,151:250-262.

54. ChenS,McMullanG,FaruqiAR,MurshudovGN,ShortJM, ScheresSHW,HendersonR:High-resolutionnoisesubstitution tomeasureoverfittingandvalidateresolutionin3Dstructure determinationbysingleparticleelectroncryomicroscopy. Ultramicroscopy2013,135:24-35.

55. CardoneG,HeymannJB,StevenAC:Onenumberdoesnotfit all:mappinglocalvariationsinresolutionincryo-EM reconstructions.JStructBiol2013,184:226-236.

56. KucukelbirA,SigworthFJ,TagareHD:Quantifyingthelocal resolutionofcryo-EMdensitymaps.NatMethods2014,11:63-65.

57.

PenczekmicroscopyPA:.ResolutionMethEnzymolmeasures2010,482in:73-100.molecularelectron

Acomprehensiveanduser-friendlysoftwareforsubtomogramaveragingand classification.Currentversionsofthispackageincludetoolsfortiltseries alignment,tomogramreconstructionandinspection,andparticlepicking. 58. Castan˜o-Dı´ezD,KudryashevM,ArheitM,StahlbergH:Dynamo:a

flexible,user-friendlydevelopmenttoolforsubtomogram averagingofcryo-EMdatainhigh-performancecomputing environments.JStructBiol2012,178:139-151.

59. NavarroPP,StahlbergH,Castan˜o-Dı´ez D:Protocolsfor subtomogramaveragingofmembraneproteinsinthedynamo softwarepackage.FrontMolBiosci2018,5:82.

60. delaRosa-Trevı´nJM,QuintanaA,DelCanoL,Zaldı´varA,FocheI, Gutie´rrezJ,Go´mez-BlancoJ,Burguet-CastellJ,Cuenca-AlbaJ, AbrishamiVetal.:Scipion:asoftwareframeworktoward integration,reproducibilityandvalidationin3Delectron microscopy.JStructBiol2016,195:93-99.