Initial bridges between two ribosomal subunitsare formed within 9.4 milliseconds, as studied bytime-resolved cryo-EMTanvir R. Shaikha,1,2, Aymen S. Yassina,3, Zonghuan Lub, David Barnarda, Xing Menga, Toh-Ming Lub,Terence Wagenknechta,c, and Rajendra K. Agrawala,c,2

aLaboratory of Cellular and Molecular Basis of Diseases, Division of Translational Medicine, Wadsworth Center, New York State Department of Health,Albany, NY 12201; bCenter for Integrated Electronics, Rensselaer Polytechnic Institute, Troy, NY 12180; and cDepartment of Biomedical Sciences, School ofPublic Health, State University of New York at Albany, Albany, NY 12222

Edited by Peter B. Moore, Yale University, New Haven, CT, and approved June 2, 2014 (received for review April 12, 2014)

Association of the two ribosomal subunits during the process oftranslation initiation is a crucial step of protein synthesis. The twosubunits (30S and 50S) of the bacterial 70S ribosome are heldtogether by 12 dynamic bridges involving RNA–RNA, RNA–protein,and protein–protein interactions. The process of bridge formation,such as whether all these bridges are formed simultaneously or ina sequential order, is poorly understood. To understand such pro-cesses, we have developed and implemented a class of microfluidicdevices that mix two components to completion within 0.4 ms andspray the mixture in the form of microdroplets onto an electronmicroscopy grid, yielding a minimum reaction time of 9.4 ms be-fore cryofixation. Using these devices, we have obtained cryo-EMdata corresponding to reaction times of 9.4 and 43 ms and havedetermined 3D structures of ribosomal subunit association inter-mediates. Molecular analyses of the cryo-EM maps reveal thateight intersubunit bridges (bridges B1a, B1b, B2a, B2b, B3, B7a,B7b, and B8) form within 9.4 ms, whereas the remaining fourbridges (bridges B2c, B4, B5, and B6) take longer than 43 ms toform, suggesting that bridges are formed in a stepwise fashion.Our approach can be used to characterize sequences of variousdynamic functional events on complex macromolecular assembliessuch as ribosomes.

ribosomal intersubunit bridges | millisecond time resolution cryo-EM

The ribosome is the site of protein synthesis in the cell. It isa highly complex ribonucleoprotein machine consisting of

a small and a large subunit. In prokaryotes, the small (30S)subunit consists of a 16S ribosomal RNA (rRNA, 1,542 nt) and21 ribosomal proteins (numbered S1–S21), whereas the large(50S) subunit comprises two rRNA molecules (23S [2,904 nt]and 5S [120 nt]) and 32 ribosomal proteins (numbered L1–L34with gaps). The two ribosomal subunits associate at the finalstage of the initiation step of protein synthesis, facilitated bythree initiation factors (IF1, IF2, and IF3), to yield the 70S ri-bosome (molecular weight ∼2.5 MDa). The two subunits arejoined by 12 canonical intersubunit bridges, which are formedbetween rRNA and rRNA, rRNA and protein, or protein andprotein components (1). The kinetics of ribosomal subunit as-sociation has been a subject of great interest for the last severaldecades (2); however, there have been few published reports onthe kinetics of specific structural elements during this process (3,4). Time-resolved structural studies to capture subunit assemblyintermediates on the order of seconds or minutes have been re-cently reported (5, 6). The goals of the present study are to capturestructures of the association intermediates of the two ribosomalsubunits on a millisecond timescale, to understand the process ofsubunit association, and to lay the groundwork for generalization ofthis technique to a wide variety of macromolecular systems.The challenge of time-resolved cryo-EM is to rapidly mix reac-

tants and then deposit them in a thin film of solvent within a shorttime before flash freezing the EM grid in liquid ethane. We have

found that spraying the mixture with an air atomizer can producea thin enough aqueous film if the carbon film is freshly made andplasma-cleaned before plunging (7). The mixing, reacting, andspraying steps were accomplished by means of a monolithic,microfabricated silicon device that incorporated a mixer, incuba-tion channel, and pneumatic sprayer in a single chip. At the flowrate used here of 3 μL/s per reactant, mixing was estimated to becomplete within 0.4 ms (8). This mixer-sprayer was incorporatedinto a computer-controlled plunging apparatus (9). In the currentconfiguration of the device, the mixed reactants spend 4 ms in theoutlet channel, 0.4 ms in flight from the nozzle to the grid, and 5 mson the grid while plunged into the cryogen, for a minimum reactiontime of 9.4 ms. Longer reaction times are achieved by use of ameandering path between the point of initial mixing and the noz-zle. 3D reconstructions, obtained from the cryo-EM data collectedfrom the ribosomal subunit association experiments, show missingdensities for specific intersubunit bridges, suggesting that theintersubunit bridges are formed in a sequential order.

ResultsTranslation Assay. To ascertain that the structure and function ofthe ribosomal particles remain intact after passing through the

Significance

The protein-synthesizing machinery of the cell, the ribosome, ismade up of two subunits, which in bacteria are held togetherby 12 molecular bridges. Understanding the mechanism andtimeline of bridge formation is crucial to understanding themechanism of bacterial translation initiation. To study thetimeline of bridge formation, we developed microfluidic devi-ces that mix two interacting components, spray the mixtureonto an electron microscopy grid, and flash freeze the grid fora minimum reaction time of 9.4 ms. This study shows for thefirst time to the authors’ knowledge that eight of the 12bridges form within 9.4 ms, whereas the remaining fourbridges take longer than 43 ms to form.

Author contributions: T.R.S., T.W., and R.K.A. designed research; T.R.S., A.S.Y., Z.L., D.B.,and X.M. performed research; Z.L., D.B., and T.-M.L. contributed new hardware tools; T.R.S.and R.K.A. analyzed data; and T.R.S. and R.K.A. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.1Present address: Structural Biology Programme, Central European Institute of Technol-ogy, Masaryk University, Brno, Czech Republic.

2To whom correspondence may be addressed. E-mail: shaikh@ceitec.muni.cz or Rajendra.Agrawal@health.ny.gov.

3Present address: Department of Microbiology and Immunology, Faculty of Pharmacy,Cairo University, Cairo, Egypt.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1406744111/-/DCSupplemental.

9822–9827 | PNAS | July 8, 2014 | vol. 111 | no. 27 www.pnas.org/cgi/doi/10.1073/pnas.1406744111

microfluidic devices, we had shown previously that the ribosomalsubunits sprayed through the microfluidic device retained com-petence to associate into 70S ribosomes, as demonstrated by the70S peak observed on passing the collected spray through a su-crose gradient (7). To determine whether passage through themicrofluidic device preserved the translation activity of theribosomes, we performed a cell-free translation assay on a col-lected spray mixture. We sprayed intact 70S ribosomes at threedifferent concentrations through the device and then comparedthe translation activity of the output 70S ribosomes to 70Sribosomes that were not passed through the device. In addition,to separate the effects of mixing forces from aerosol formation,in one set of experiments we turned off the nitrogen supply,which resulted in a constant stream instead of a plume ofdroplets. The sprayed 70S showed activity equivalent to the con-trol ±20% (Fig. S1A), thus demonstrating that passage of pre-associated 70S ribosomes, i.e., ribosomes formed by incubating thepurified 30S and 50S subunits, through the microfluidic devicedoes not impair translation activity. We also tested the activity ofribosomes associated from purified subunits, using the micro-fluidic device. After mixing 30S and 50S subunits and collectingthe spray in a microcentrifuge tube, the resulting ribosomesshowed the same activity ±1% compared with 70S ribosomes as-sociated on a gradient (Fig. S1B). Subunit association need nothave occurred in the mixer, necessarily, as there was no quencherto halt association after the spray was collected in the micro-centrifuge tube. However, this experiment shows that the subunitsretained competence to associate and then translate protein afterpassage through the microfluidic device.

Visualization of Association Intermediates Within 9.4 ms.We injected30S and 50S subunits that were preequilibrated at 12 mM Mg2+,a concentration that allows maximum subunit association, throughthe two microsyringe ports of the microfluidic device. From 19grids searched exhaustively, we collected 303 micrographs on film,of which 258 were deemed suitable for further analysis. Afterreference-based classification, the particle images were catego-rized into 30S, 50S, and 70S classes. Unexpectedly, some addi-tional species were seen in the form of a dimer of 50S subunits andheterotrimers with a 50S∙30S∙30S configuration (Fig. S2). Early inthe history of transmission EM of purified ribosomal subunits, 50Sdimers had been shown to form (10), and later experiments onmitochondrial ribosomes from the yeast Candida utiliz showed thata combination of high Mg2+ concentration and low deoxycholateconcentration gave rise to a dimer of large subunits (11). The50S∙30S∙30S trimer has not been reported previously, although thestructure of a 100S particle containing two 70S particles, that is,a 50S∙30S∙30S∙50S particle, has been examined by cryo-EM (12).In that case, intermediary proteins such as ribosome-modulatingfactor and hibernation-promoting factor were involved. We spec-ulate that the combination of high Mg2+, and especially the highsubunit concentration required for our experiments, drives theequilibrium toward higher-order structures (Fig. S3).From the 70S, 50S monomer, and 50S dimer classes, 17,987

particle images were selected. From the 70S class, 4,445 particleimages were selected. The presence of 50S dimers and the50S∙30S∙30S trimer yields an underdetermined kinetic scheme,so a simple metric to measure the extent of 70S formation is theratio of particles containing at least one 50S and one 30S (i.e.,70S and 50S∙30S∙30S trimers) to total 50S-containing particles,which includes 50S monomers, 70S, 50S dimers, and 50S∙30S∙30S

Fig. 1. Cryo-EM maps of the 70S ribosomes formed within 9.4 ms. (A) Control, preassociated 70S ribosomes that had been passed through the microfluidicmixer. The dashed line roughly divides the 30S and 50S ribosomal subunits. Inset to the lower left depicts the orientation of a segmented 70S ribosome (42),with 30S (yellow) and 50S (blue) subunits identified. hd, head, and pt, platform, respectively, of the 30S subunit; L1, protein L1 protuberance of the 50Ssubunit. (B) Reconstruction of the 70S-like particles, which were associated using the microfluidic mixer with a reaction time of 9.4 ms. (C) Reconstruction ofone of the three classes obtained from application of the ML3D; that is, a subclass of particles used in B. (D) Reconstruction of another subclass of the particlesused for B. Screenshots were taken with the “hide dust” function in Chimera activated. Variable intersubunit regions of interest are circled. (E–G) Magnifiedviews of the bridges B2c, B4-6 region, with densities of experimental map shown in panels B–D (solid, matching colors) superimposed on to the control map(gray wireframe). A cutting plane was applied at the far end to enhance visual clarity in the intersubunit space.

Shaikh et al. PNAS | July 8, 2014 | vol. 111 | no. 27 | 9823

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

trimers. By this measure, we obtain 4,445/17,987, or 24.7% associ-ation. Using the rate constant of 14μM−1s−1 obtained by stopped-flow and light scattering under similar experimental conditions (4),one would expect fewer 70S complexes (11.6%) from a simplebimolecular reaction.As a control for the 70S class at 9.4 ms, we determined a re-

construction from 70S particles (n = 4,391) that were preassociatedand then passed through the microfluidic mixer. Reconstruction ofthe 4,445 images classified as 70S from the 9.4-ms data set showedthat four of the 12 intersubunit bridges, bridges B2c, B4, B5, andB6, were missing or weak (Fig. 1 A, B, and E). Bridges B5 and B6share contacts (1) and are not always resolved as separate bridgesat the current resolution. In this reconstruction at 9.4 ms, the un-shared contacts are not visible either. Although bridge B6 doesappear at a lower threshold (Fig. S4A), the threshold for the iso-surface of the 9.4-ms reconstruction is lower and more permissivethan the control, and yet lacks the density corresponding to bridgesB2c, B4, B5, and B6 (Fig. S4 B and C), suggesting that none ofthese bridges are formed within 9.4 ms.In addition to the missing bridges, the resolution of the 70S-

like particles from the 9.4-ms reconstruction was poorer thanthat from the preassociated control reconstruction (33 Å versus24 Å), despite a similar number of particles. The low resolutionwas attributed to additional heterogeneity in the reassociatedparticles. Thus, we subjected the particle images to the 3Dmaximum likelihood classification (13) (ML3D). ML3D followsthe steepest gradient of its likelihood function, a composite ofparameters such as image shift, rotation, class assignment, and soon, to find its optimum (see supplemental data in ref. 13), whichcan result in entrapment in a local extremum, rather than theglobal one. To safeguard against entrapment in a local extre-mum, we ran ML3D several times, using different starting seeds.

(At the start of ML3D, classes are assigned randomly from theparticle image set.) Fig. 1 C and D shows the results of one in-stance of ML3D, and similar results were obtained from threeother runs of ML3D. The reconstruction shown in Fig. 1C showsmissing bridges comparable with the overall reconstruction inFig. 1B, whereas the reconstruction in Fig. 1D shows additionalbridge B6 (Fig. 1G). In addition, the formation of bridge B6appears to be allosterically correlated with a movement in the L1stalk region, as we observe a smearing of the L1 stalk and veryweak density corresponding to 23S rRNA helix 76 that connectsthe L1 stalk to the rest of the ribosome. The L1 stalk, which ispresent in the pretranslocation state conformation (14, 15) in allour preassociated 70S (control) maps, is known to be con-formationally variable (16), as is helix 76 (17). In each run ofML3D, there was at least one class in which the intersubunitbridges B2c, B4, B5, and B6 are missing and one class in whichadditional intersubunit bridges appeared to have formed. Notethat the initial orientation parameters assigned to each particleimage before ML3D were those that had given rise to the re-construction with bridges missing, so it is unlikely that the ap-pearance of those bridges found in mature 70S could be a resultof reference bias.

Visualization of Association Intermediates Within 43 ms. We useda microfluidic device with a reaction time of 43 ms, of which time38 ms is the median residence time, plus 5 ms for time of flightand plunging. We collected 24,472 particle images from 2,397CCD images and have performed an analysis similar to thatdescribed for the 9.4-ms data set above. From the 70S class,11,921 images were selected, corresponding to 48.7% associationusing the same metric as for the 9.4-ms data set compared withthe 37.5% expected from an association rate constant of 14 μM−1s−1.

Fig. 2. Cryo-EM maps of the 70S ribosomes formed within 43 ms. (A) Control, preassociated 70S ribosomes that had been passed through the microfluidicmixer. (B) Reconstruction of the 70S-like particles, which were associated using the microfluidic mixer with a reaction time of 43 ms. (C) Reconstruction fromone of the three ML3D classes. (D) Reconstruction from another class from the same ML3D run. Variable intersubunit regions of interest are circled. (E–G)Magnified views of the bridges B2c, B4-6 region, with densities of experimental map shown in panels B–D (solid, matching colors) superimposed onto thecontrol map (gray wireframe). A cutting plane was applied at the far end to enhance visual clarity in the intersubunit space (see Fig. S6 for more information).

9824 | www.pnas.org/cgi/doi/10.1073/pnas.1406744111 Shaikh et al.

We also selected 2,620 50S dimers, which is proportionally fewerthan at 9.4 ms: 10.7% versus 15.3%, respectively. A representativeCCD image is shown in Fig. S5.Of the 11,921 70S-like particles selected initially, we computed

a 3D reconstruction of 7,396 particles (Fig. 2B) after a secondround of manual particle-verification. As was the case with the9.4-ms data set, bridges B2c, B4, B5, and B6 were missing. Theincreased proportion of 70S-like particles from the 9.4-ms dataset suggests a buildup of a metastable intermediate.The resolution of the reconstruction was again worse (23 Å)

than that of the control, preassociated 70S (Fig. 2A) (19Å), froma comparably-sized data set with 7,705 particles (7). Furtheranalyses using ML3D-based classification (13) suggest there areat least two subpopulations in both the 9.4-ms (Fig. 1 C and D)and the 43-ms (Fig. 2 C and D) data sets. For each time, oneof the subpopulations yields a reconstruction in which the fourintersubunit bridges are missing, whereas the other subpopulationsproduce reconstructions containing one or more of those fourbridges better resembling the fully associated 70S ribosome. Thus,the two independent data sets both contain subpopulations ofparticles that contain at least one additional intersubunit bridge.

DiscussionModel for Association.Here we report, using our recently developedtime-resolved cryo-EM technology and 3D reconstruction, thatassociating ribosomal subunits form 70S-like intermediates thatlack some of the intersubunit bridges found in native 70S particles.Classification of the 70S-like images obtained for two reaction times,9.4 and 43 ms, shows subpopulations in which additional intersubunitbridges have formed. Spatially, the missing bridges cluster along the3′ domain of the small subunit. 16S RNA bridges B5 and B6 haveintersubunit contacts on the 30S subunit contributed entirely by helix44. Helix 44 has been shown to be organized late during the as-sembly of the 30S subunit and has been found to be disordered (18,19) or distorted (20–22) in structures of some of the assemblyintermediates of the small subunit from various species.Bridges B2c and B4 are also spatially near these missing helix

44 contacts. Mutants of components involved in the formationof bridge B2c are viable but are underrepresented in polysome poolsrelative to wild-type, indicative of a defect in association (23, 24).The double-stranded rRNA helix 27 of the 30S subunit, which isinvolved in the contacts of bridge B2c, requires rearrangement

Fig. 3. Depiction of the order of ribosomal intersubunit bridge formation. The inset at the top shows the 70S ribosome with two subunits (30S, yellow; 50S,blue) identified. X-ray crystallographic structures of the two ribosomal subunits [Protein Data Bank ID codes 3R8O and 3R8T (25)] were used to display theknown locations of canonical intersubunit bridges. Locations of the bridges (1) were superimposed. Bridges that are formed within 9.4 ms are outlined ingreen, whereas those take longer than 43 ms are outlined in red and purple. The bridges outlined in purple intermittently appeared during some runs ofML3D classification. Multiple contacts are involved in some bridges, and thus some contacts between bridges B5 and B6 overlap. In the lower inset, helix 44 ishighlighted in pink to indicate the directionality of bridge formation involving this helix. Chimera (44) was used for displaying the atomic models.

Shaikh et al. PNAS | July 8, 2014 | vol. 111 | no. 27 | 9825

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

from the conformation found in the isolated 30S to the confor-mation found in the associated 70S (4). Our data show thatbridge B2c is not yet formed at 43 ms, and so we suggest thathelix 27 has not yet undergone this rearrangement. Bridge B4,composed primarily of helix 34 in the 50S subunit, has not beenseen to be disordered in known 50S structures, and atomicmodels of the isolated 50S do not require any rearrangement ofhelix 34 to be fitted into the cryo-EM maps of the 70S ribosome.However, density for helix 34 is not visible in our reconstructionsof 70S at a reaction time of 9.4 ms and is only visible in a sub-population of images at 43 ms. On the basis of an atomic modelof the 70S ribosome (25) (Fig. 3), it appears that helix 34 of the50S subunit fits into a notch on the 30S subunit formed by helix 24of the 16S RNA and protein S15. Perhaps the reason that helix34’s density is absent in the 9.4- and 43-ms maps is that its initial fitis suboptimal and requires tens of milliseconds to adopt its finalconformation. Helix 34 deletion mutants are able to associate,albeit weakly (24, 26, 27), and bridge B4 is not conserved amongall phylogenetic domains (28). These biochemical observationsstrongly corroborate this study by suggesting that bridges B2c andB4 are not vital but, rather, are formed at a later stage of sub-units subunit association to help stabilize the 70S structure.We infer that a spatial pattern of the visible bridges emerges

(Fig. 3) in which bridges form along an axis (Fig. 3, lower inset).We hypothesize a model for association in which the intersubunitbridges involving helix 44 near the decoding region of the 30Ssubunit form first, whereas those involving the distal end form last.These early-forming bridges include B2a, B3, and B7b, involvingcontact points that are conserved across bacteria, eukaryotes, andarchaea, as well as mitochondria and chloroplasts (28). Next, theperipheral bridges, including bridges B1a, B1b, and B8, form.These bridges are visible at 9.4 ms but are not evolutionarilyconserved (28). Furthermore, bridges B1a and B1b undergochanges in intersubunit contacts during ribosomal ratcheting(14, 15, 29, 30), further suggesting these contacts are not crucialto hold together the 70S ribosome. The last bridges to form arebridges B2c, B4, B5, and B6. Bridges B2c and B5 (for the con-tacts not shared with B6) were absent in all the reconstructionscomputed in this study, both before and after classification. Afterclassification, we obtained reconstructions in which bridge B4 orB6 was visible (or neither), but no reconstruction at 9.4 or 43 msshowed both bridges B4 and B6 simultaneously. It is conceivablethat association occurs along at least two pathways: one pathwaywhere bridge B6 forms first and another where bridge B4 formsfirst. It is likely that the missing bridges in our 9.4- and 43-mscryo-EM data sets represent an actual sequence of bridge for-mation during the bacterial translation initiation process thatinvolves a 30S initiation complex comprising the 30S subunit,mRNA, initiator tRNA, and three initiation factors (IF1, IF2,and IF3) (31). In the 30S initiation complex, the regions involvedin the formation of missing bridges in this study would be steri-cally blocked by bulky domains 3–5 of the IF2 (32), which ap-parently realign in a subsequent step to allow full access to the50S subunit for the formation of the 70S initiation complex.However, the sequence of formation of these bridges needs to befurther investigated by collecting data at longer reaction timesand by including additional components in the reaction mixture,such as the 30S initiation complex with initiator tRNA and trans-lation initiation factors in one channel and the 50S subunit in otherreaction channels of the microfluidic devices.

Progress in Time-Resolved Cryo-EM. The results presented heredescribe the association of ribosomal subunits by time-resolvedcryo-EM, using a microfluidic mixer-sprayer that was developedin our laboratory. Among the advantages of our methodology arethat virtually any pair of reactants can be mixed, with no needfor chemical modification, such as in methods that requirephotoactivatable compounds (33); mixing is fast, completing

within 0.4 ms; interaction of the reacting macromolecules withthe carbon support film is minimized; and the minimum reactiontime is as short as 9.4 ms. However, a limitation of the techniqueis that, because of a sparsity of suitable microdroplets and the lownumbers of macromolecules per microdroplet on a typical EM grid,it is difficult to collect sufficient numbers of particles to achievehigh-resolution structures of reaction intermediates, which oftenmake up a small fraction of the total population. Further deve-lopment of this technology would be advantageous for expanding itsuse in studying millisecond-scale time-resolved steps during variousphases of translation, as well as application to other biological sys-tems. To this end, we continue to develop and test a new class ofmicrosprayers that could produce a more uniform microdropletdistribution on the EM grid to facilitate the large cryo-EM dataacquisition needed for higher-resolution reconstructions.

MethodsPreparation of Ribosomes. Ribosomes were prepared according to Blaha andcolleagues (34). Briefly, 1 L of LB broth was inoculated with 5 mL overnightculture of Escherichia coli MRE600. Cells were collected at midlog phase(o.d., 0.5–0.6) and lysed by grinding with alumina. Cellular lysate was thenlayered on 5–30% sucrose gradient in Hepes-K buffer with 4 mM MgCl2 toseparate tight-coupled (TC) 70S from loose-coupled subunits (35) (loose-coupled 30S and loose-coupled 50S). Sucrose gradients were fractionatedby a Teledyne Isco fraction collector (Teledyne Isco, Inc.), and the fractionscorresponding to (TC 70S) ribosome were collected. TC 30S and TC 50S wereprepared from TC 70S ribosomes after fractionation on 5–30% sucrosegradient in Hepes-K buffer with 0.1 mM MgCl2. Fractions corresponding toeach subunit were pooled, pelleted, quantified, and stored at −80 °C.

Cell-Free Translation Activity of Sprayed 70S. The activity of the ribosomesassociated by spraying 30S and 50S subunits mixed by the 9.4-ms device wasassayed using a cell-free translation system (PURESYSTEM Δ1 Ribosomes,Cosmo Bio Co. Ltd.), which has all the transcription and translation factorsthat would allow the coupled transcription/translation of a particular genewith the appropriate promoter; in this case, the gene for dihydrofolate re-ductase, carried on a plasmid under the T7 promoter. The incorporation ofradiolabeled S35 methionine allows the quantification of the product pro-tein after SDS/PAGE. The kit used was specially prepared in that there wereno ribosomes in the reaction mixtures supplied to allow the testing of ourassociated subunits. To test for the effect of passage through the device ofpreassociated 70S on translation activity, 70S ribosomes at three differentconcentrations (2, 3, and 4 μM) were sprayed through the device and thecollected fractions were each diluted to 0.7 μM and were used in cell-freetranslation against a control 70S of the same molar concentration that hadnot passed through the device. To assay the activity of ribosomes associatedon passage through the microfluidic mixer, 3.4 μM of each 30S and 50S wereincubated at 37 °C for 45 min and then sprayed through the device. Theoutput (regarded as 70S) was collected in a microcentrifuge tube, diluted to0.7 μM, and used in the assay. As a control, 0.7 μM of a 70S formed by as-sociation of 30S and 50S on a sucrose gradient was used.

Time-Resolved Preparation of EM Grids. A ∼5-nm-thick layer of continuouscarbon was deposited onto freshly cleaved mica using a Bal-Tec MED 020evaporator (Leica Microsystems). The carbon was floated onto R2/4 Quan-tifoil grids (Micro Tools GmbH), and the grids were glow-discharged for 30 s,using a plasma cleaner (Harrick PKC-3XG; Harrick Scientific Corp.) to renderthem hydrophilic (36). Two micromoles of each 30S and 50S subunit (fora final concentration of 1 μM after mixing) in Hepes-K buffer with 12 mMMgCl2 were preincubated at 37 °C for 45 min before being used. Subunitswere loaded into separate syringe pumps and driven through the mixer at3 μL/s per reactant. The spray was collected on a grid plunged at ∼1.0 m/s bya computer-controlled grid freezing instrument (9). Humidified nitrogen gasused to generate the spray was regulated to 50 psi.

Electron Microscopy and Digitization. Electron microscopy was performed onan FEI Tecnai F20 transmission electron microscope operating at 200 kV ata nominal 50,000× magnification (50,360× calibrated) under low-dose con-ditions. For the data set corresponding to a reaction time of 9.4 ms, from 19grids, we collected 303 micrographs on Kodak SO-163 film, of which 258were retained on the basis of quality. Micrographs were digitized on a Zeissscanner (Z/I Imaging) with a step size of 14 μm, corresponding to a pixel sizeof 2.76 Å on the object scale. For the data set corresponding to a reaction

9826 | www.pnas.org/cgi/doi/10.1073/pnas.1406744111 Shaikh et al.

time of 43 ms, from 17 grids, 2,492 CCD images were collected on a TVIPS4k×4k camera, of which 2,397 images were retained on the basis of icethickness and the quality of their power spectra. CCD images were binned bytwo, corresponding to a pixel size, including a postcolumn magnification of1.78×, of 3.38 Å on the object scale.

Image Processing. Image processing was performed using the softwarepackages EMAN2 (37), SPIDER (38), and XMIPP (39). Particles were windowedusing e2boxer.py from EMAN2. Particle images were aligned by projectionmatching (40, 41) and were categorized into subclasses of 70S, 50S, 30S, andother structures, using as references a previously reconstructed E. coli ribo-some (42) and segmented subunits thereof. Correction for the contrasttransfer function was performed by phase-flipping, which is especially usefulafter classification, when defocus groups may contain too few particles toobtain a well-defined 3D reconstruction (the protocol is available at www.wadsworth.org/spider_doc/spider/docs/techs/recon1/myproject/Docs/mr.html).

Particle images were screened individually, using classification-based verifi-cation (43). After downsampling to 64 × 64 pixels to reduce computationtime, the 3D ML3D (13) was used to further classify the particle imagesassigned to the 70S category. The number of classes, K, was varied accordingto the size of the data set, using either K = 3 or K = 4. Classes obtained fromML3D were then used for 3D reconstruction from the original images at fullresolution, and orientations were further refined. Reconstructions were vi-sualized using UCSF Chimera (44).

ACKNOWLEDGMENTS. We thank Sjors Scheres for assistance with runningML3D, ArDean Leith for help with local installation of XMIPP, and JoachimFrank for helpful discussion. The work was supported by National Institutesof Health National Center for Research Resources Grant P41 RR01219 and, inpart, by National Institutes of Health Grant R01 GM61576 (to R.K.A.) andGrant CZ.1.07/2.3.00/20.0042 (to T.R.S.) from the European Social Fund andthe state budget of the Czech Republic.

1. Yusupov MM, et al. (2001) Crystal structure of the ribosome at 5.5 A resolution. Sci-ence 292(5518):883–896.

2. Wishnia A, Boussert A, Graffe M, Dessen PH, Grunberg-Manago M (1975) Kinetics ofthe reversible association of ribosomal subunits: Stopped-flow studies of the rate lawand of the effect of Mg2+. J Mol Biol 93(4):499–515.

3. Fabbretti A, et al. (2007) The real-time path of translation factor IF3 onto and off theribosome. Mol Cell 25(2):285–296.

4. Hennelly SP, et al. (2005) A time-resolved investigation of ribosomal subunit associ-ation. J Mol Biol 346(5):1243–1258.

5. Fischer N, Konevega AL, Wintermeyer W, Rodnina MV, Stark H (2010) Ribosome dy-namics and tRNA movement by time-resolved electron cryomicroscopy. Nature466(7304):329–333.

6. Mulder AM, et al. (2010) Visualizing ribosome biogenesis: Parallel assembly pathwaysfor the 30S subunit. Science 330(6004):673–677.

7. Lu Z, et al. (2009) Monolithic microfluidic mixing-spraying devices for time-resolvedcryo-electron microscopy. J Struct Biol 168(3):388–395.

8. Lu Z, et al. (2010) Passive Microfluidic device for Sub Millisecond Mixing. Sens Ac-tuators B Chem 144(1):301–309.

9. White HD, Thirumurugan K, Walker ML, Trinick J (2003) A second generation appa-ratus for time-resolved electron cryo-microscopy using stepper motors and electro-spray. J Struct Biol 144(1-2):246–252.

10. Huxley HE, Zubay G (1960) Electron microscope observations on the structure of mi-crosomal particles from E. coli. J Mol Biol 2:10–18.

11. Vignais PV, Stevens BJ, Huet J, André J (1972) Mitoribosomes from Candida utilis.Morphological, physical, and chemical characterization of the monomer form and ofits subunits. J Cell Biol 54(3):468–492.

12. Kato T, et al. (2010) Structure of the 100S ribosome in the hibernation stage revealedby electron cryomicroscopy. Structure 18(6):719–724.

13. Scheres SHW, et al. (2007) Disentangling conformational states of macromolecules in3D-EM through likelihood optimization. Nat Methods 4(1):27–29.

14. Frank J, Agrawal RK (2000) A ratchet-like inter-subunit reorganization of the ribo-some during translocation. Nature 406(6793):318–322.

15. Frank J, Agrawal RK (2001) Ratchet-like movements between the two ribosomalsubunits: Their implications in elongation factor recognition and tRNA translocation.Cold Spring Harb Symp Quant Biol 66:67–75.

16. Agrawal RK, et al. (1999) Effect of buffer conditions on the position of tRNA on the 70S ribosome as visualized by cryoelectron microscopy. J Biol Chem 274(13):8723–8729.

17. Gao H, et al. (2003) Study of the structural dynamics of the E coli 70S ribosome usingreal-space refinement. Cell 113(6):789–801.

18. Boehringer D, O’Farrell HC, Rife JP, Ban N (2012) Structural insights into methyltransferaseKsgA function in 30S ribosomal subunit biogenesis. J Biol Chem 287(13):10453–10459.

19. Guo Q, et al. (2013) Dissecting the in vivo assembly of the 30S ribosomal subunitreveals the role of RimM and general features of the assembly process. Nucleic AcidsRes 41(4):2609–2620.

20. Datta PP, et al. (2007) Structural aspects of RbfA action during small ribosomal sub-unit assembly. Mol Cell 28(3):434–445.

21. Jomaa A, et al. (2011) Understanding ribosome assembly: The structure of in vivoassembled immature 30S subunits revealed by cryo-electron microscopy. RNA 17(4):697–709.

22. Strunk BS, et al. (2011) Ribosome assembly factors prevent premature translationinitiation by 40S assembly intermediates. Science 333(6048):1449–1453.

23. Bélanger F, Léger M, Saraiya AA, Cunningham PR, Brakier-Gingras L (2002) Functionalstudies of the 900 tetraloop capping helix 27 of 16S ribosomal RNA. J Mol Biol 320(5):979–989.

24. Liiv A, O’Connor M (2006) Mutations in the intersubunit bridge regions of 23 S rRNA.J Biol Chem 281(40):29850–29862.

25. Dunkle JA, et al. (2011) Structures of the bacterial ribosome in classical and hybridstates of tRNA binding. Science 332(6032):981–984.

26. Komoda T, et al. (2006) The A-site finger in 23 S rRNA acts as a functional attenuatorfor translocation. J Biol Chem 281(43):32303–32309.

27. Maiväli U, Remme J (2004) Definition of bases in 23S rRNA essential for ribosomalsubunit association. RNA 10(4):600–604.

28. Mears JA, et al. (2002) Modeling a minimal ribosome based on comparative sequenceanalysis. J Mol Biol 321(2):215–234.

29. Réblová K, et al. (2010) Dynamics of the base of ribosomal A-site finger revealed bymolecular dynamics simulations and Cryo-EM. Nucleic Acids Res 38(4):1325–1340.

30. Ratje AH, et al. (2010) Head swivel on the ribosome facilitates translocation by meansof intra-subunit tRNA hybrid sites. Nature 468(7324):713–716.

31. Gualerzi CO, Pon CL (1990) Initiation of mRNA translation in prokaryotes. Bio-chemistry 29(25):5881–5889.

32. Simonetti A, et al. (2008) Structure of the 30S translation initiation complex. Nature455(7211):416–420.

33. Shaikh TR, Barnard D, Meng X, Wagenknecht T (2009) Implementation of a flash-photolysis system for time-resolved cryo-electron microscopy. J Struct Biol 165(3):184–189.

34. Blaha G, et al. (2000) Preparation of functional ribosomal complexes and effect ofbuffer conditions on tRNA positions observed by cryoelectron microscopy. MethodsEnzymol 317:292–309.

35. Agrawal RK, Burma DP (1996) Sites of ribosomal RNAs involved in the subunit asso-ciation of tight and loose couple ribosomes. J Biol Chem 271(35):21285–21291.

36. Grassucci RA, Taylor DJ, Frank J (2007) Preparation of macromolecular complexes forcryo-electron microscopy. Nat Protoc 2(12):3239–3246.

37. Tang G, et al. (2007) EMAN2: An extensible image processing suite for electron mi-croscopy. J Struct Biol 157(1):38–46.

38. Frank J, et al. (1996) SPIDER and WEB: Processing and visualization of images in 3Delectron microscopy and related fields. J Struct Biol 116(1):190–199.

39. Sorzano COS, et al. (2004) XMIPP: A new generation of an open-source image pro-cessing package for electron microscopy. J Struct Biol 148(2):194–204.

40. Penczek P, Radermacher M, Frank J (1992) Three-dimensional reconstruction of singleparticles embedded in ice. Ultramicroscopy 40(1):33–53.

41. Shaikh TR, et al. (2008) SPIDER image processing for single-particle reconstruction ofbiological macromolecules from electron micrographs. Nat Protoc 3(12):1941–1974.

42. Gabashvili IS, et al. (2000) Solution structure of the E. coli 70S ribosome at 11.5 Aresolution. Cell 100(5):537–549.

43. Shaikh TR, Trujillo R, LeBarron JS, Baxter WT, Frank J (2008) Particle-verification forsingle-particle, reference-based reconstruction using multivariate data analysis andclassification. J Struct Biol 164(1):41–48.

44. Pettersen EF, et al. (2004) UCSF Chimera—a visualization system for exploratory re-search and analysis. J Comput Chem 25(13):1605–1612.

Shaikh et al. PNAS | July 8, 2014 | vol. 111 | no. 27 | 9827

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY