Design of a Molecular Support for Cryo-EM Structure Determination

Design of a molecular support for cryo-EM structure determination

Thomas G. Martina, Tanmay A. M. Bharata,b, Andreas C. Joergera,c, Xiao-chen Baia, Florian Praetoriusd, Alan R. Fershta, Hendrik Dietzd,1, and Sjors H. W. Scheresa,1

aMedical Research Council Laboratory of Molecular Biology, Cambridge Biomedical Campus, Cambridge CB2 0QH, United Kingdom; bSir William Dunn School of Pathology, University of Oxford, Oxford OX1 3RE, United Kingdom; cGerman Cancer Consortium (DKTK), Institute of Pharmaceutical Chemistry, Johann Wolfgang Goethe University, 60438 Frankfurt am Main, Germany; and dPhysik Department, Walter Schottky Institute, Technische Universität München, 85748 Garching near Munich, Germany

Edited by Fred J. Sigworth, Yale University, New Haven, CT, and approved October 13, 2016 (received for review August 2, 2016) Despite the recent rapid progress in cryo-electron microscopy angles are, therefore, determined a posteriori by image-pro- (cryo-EM), there still exist ample opportunities for improvement cessing algorithms that match the experimental projection of in sample preparation. Macromolecular complexes may disassociate every individual particle with projections of a 3D model (7). or adopt nonrandom orientations against the extended air–water However, the projection-matching procedure is ultimately ham- interface that exists for a short time before the sample is frozen. We pered by radiation damage. Because the electrons that are used designed a hollow support structure using 3D DNA origami to pro- for imaging destroy the very structures of interest (see ref. 8 for a tect complexes from the detrimental effects of cryo-EM sample recent review), one needs to limit carefully the number of elec- preparation. For a first proof-of-principle, we concentrated on the trons used for imaging. This procedure results in high levels of transcription factor p53, which binds to specific DNA sequences on experimental noise, which in turn lead to errors in the a poste- double-stranded DNA. The support structures spontaneously form monolayers of preoriented particles in a thin film of water, and offer riori determination of the viewing angles. These errors impose advantages in particle picking and sorting. By controlling the posi- severe limitations on the 3D reconstruction, in particular for tion of the binding sequence on a single helix that spans the hollow smaller complexes, because the signal-to-noise ratio in the im- support structure, we also sought to control the orientation of ages decreases with the size of the particles. If one could ex- individual p53 complexes. Although the latter did not yet yield perimentally control the orientations of each particle in the ice the desired results, the support structures did provide partial in- layer, then, in principle, one could determine structures to higher formation about the relative orientations of individual p53 com- resolution and of smaller complexes. This situation is illustrated plexes. We used this information to calculate a tomographic 3D by samples where the relative orientations of many molecules is reconstruction, and refined this structure to a final resolution of set—for example, in 2D crystals or helical assemblies of protein ∼15 Å. This structure settles an ongoing debate about the symme- molecules. In such cases, near-atomic-resolution reconstructions try of the p53 tetramer bound to DNA. were already achieved decades ago, and from much smaller molecules than currently possible with single-particle analysis (9– cryo-EM | DNA-origami | single particle analysis | structural biology | p53 12). Both developments that triggered the recent revolution in attainable resolution of cryo-EM single-particle analysis directly ryo-electron microscopy (cryo-EM) structure determination addressed this same hurdle. Better detectors led to lower levels Cof biological macromolecules is undergoing rapid progress. With the advent of efficient direct electron detectors and the Significance development of powerful algorithms for image processing, numerous structures to near-atomic resolution have been reported As the scope of macromolecular structure determination by in the past few years (1, 2). In cryo-EM single-particle analysis, cryo-electron microscopy (cryo-EM) is expanding rapidly, it is solutions of purified protein and/or nucleic acid complexes are becoming increasingly clear that many biological complexes typically applied to a thin, amorphous carbon film with micro- are too fragile to withstand the harsh conditions involved in meter-sized holes in it that is held in place by a metal grid. Excess making cryo-EM samples. We describe an original approach to liquid is then blotted away with filter paper, and the sample is protect proteins from harmful forces during cryo-EM sample rapidly plunged in liquid ethane (3, 4). This procedure ideally preparation by enclosing them inside a three-dimensional results in the formation of a film of vitreous ice that is only support structure that we designed using DNA origami tech- slightly thicker than the macromolecular complex of interest. niques. By binding the transcription cofactor p53 to a specific Keeping the frozen sample at liquid nitrogen temperatures allows DNA sequence, and by modifying the position of this sequence its insertion into the high vacuum of a transmission electron mi- in our support structure, we also sought to control the relative croscope and limits the effects of radiation damage by the elec- orientation of individual p53:DNA complexes. trons that are used for imaging (5). Images taken through the holes of the carbon film ideally contain 2D projections of many, Author contributions: T.G.M., A.R.F., H.D., and S.H.W.S. designed research; T.G.M., T.A.M.B., assumedly identical copies of the macromolecular complex of and X.-c.B. performed research; A.C.J. and F.P. contributed new reagents/analytic tools; T.G.M., T.A.M.B., X.-c.B., and S.H.W.S. analyzed data; and T.G.M., A.C.J., A.R.F., H.D., and interest, which are often called particles. Projections from dif- S.H.W.S. wrote the paper. ferent viewing directions can then be combined in a 3D re- The authors declare no conflict of interest. construction of the scattering potential of the molecule (6). If the This article is a PNAS Direct Submission. resulting map approaches a resolution of 3 Å, it allows building Freely available online through the PNAS open access option. an atomic model of the molecules, from which useful in- Data deposition: The p53 reconstruction reported in this paper has been deposited in the formation about their function may be derived. Electron Microscopy Data Bank, https://www.ebi.ac.uk/pdbe/emdb (accession no. 3453). A major hurdle in single-particle analysis is the need to re- 1To whom correspondence may be addressed. Email: [email protected] or cover the relative viewing angles of the individual particles. This [email protected]. information is lost in the experiment because every particle This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. adopts an uncontrolled orientation in the ice layer. The viewing 1073/pnas.1612720113/-/DCSupplemental.

E7456–E7463 | PNAS | Published online November 7, 2016 www.pnas.org/cgi/doi/10.1073/pnas.1612720113 Downloaded by guest on October 2, 2021 of experimental noise, and better image-processing algorithms Results PNAS PLUS led to more accurate viewing angles. Support Structure Design. Fig. 1 illustrates the design of our pro- Another complication of cryo-EM structure determination lies posed DNA-origami support structure. We designed a hollow in the way the sample is prepared (see also refs. 13 and 14). The pillar with a honey-combed motif of 82 parallel double-stranded exact physics of cryo-EM sample preparation is poorly described, DNA (dsDNA) helices with a height of 26 nm. Two parallel but several factors may negatively affect its results. First, the arrays of DNA helices create a central cavity of 13.2 × 13.6 nm blotting process itself may involve strong forces in the sample and outer dimensions of 26.4 × 32.7 nm. A dsDNA helix with a that destroy fragile protein complexes. Second, during the short central p53-specific binding sequence spans the center of the time between blotting and vitrification, the macromolecules are hollow space. Ten parallel helices at the periphery, which we will in a thin liquid film that extends for millimeters to the side, but is call the flag (Fig. 1 A and B, top left), make the structure asym- only a few hundred angstroms thick. Brownian motion will cause metric so that top and bottom views can be readily distinguished. the macromolecules to collide with the air–water interface Different aspects of this design aim to address four main ob- >1,000 times per second (15). Biological macromolecules may jectives of our support structure. unfold when they hit the air–water interface (16), or they may First, we aimed to keep the target protein in the middle of a adsorb to this interface in a nonrandom manner—for example, sufficiently thin ice layer and away from the air–water interface. by presenting their most hydrophobic patch to it. This interaction p53 binds preferentially to the sequence GGACATGTCCG- leads to an uneven distribution of viewing angles in the cryo-EM GACATGTCC. By including this sequence in the central dsDNA images, and the corresponding lack of different views may helix (Fig. 1 A and B, red), our target protein would bind to the hamper 3D reconstruction. Despite the numerous successes of approximate center of our support structure. By designing ssDNA single-particle analysis in recent years, preparing suitable cryo- overhangs (T10) on both the top and bottom faces of the pillar EM samples therefore remains a nontrivial task. Often, estab- (Fig. 1 A–C, blue), we aimed to expedite a preferred orientation of lishing optimal ice thickness and particle distribution in the ice the support in the ice layer. The ssDNA overhangs have exposed requires careful optimization of blotting and freezing conditions aromatic bases, which make them more likely to interact with the – by an experienced experimentalist, and fragile complexes have hydrophobic air water interface than the sides of the pillar, which often been observed to fall apart (17, 18). is more hydrophilic because of its exposed DNA backbones. This This work describes an attempt to address both the problems rationale was inspired by initial observations of a preferred orientation in the ice layer of a different DNA-origami object, where of the unknown viewing angles and the unfavorable conditions of BIOPHYSICS AND similar overhangs were added to prevent end-to-end polymeriza-

sample preparation from an experimental perspective. To this COMPUTATIONAL BIOLOGY end, we set out to design a support structure that simultaneously tion of the structures (41) (Fig. S1). If both faces of the pillar – exerts control over the orientations of individual particles, while would indeed interact with the air water interface at the top and also controlling the ice thickness and protecting the particles bottom of the sample, then the height of the pillar itself would – determine the thickness of the ice layer. By designing a distance of from the air water interface. We chose to use 3D DNA-origami ∼ techniques to design such a support structure. DNA-origami 260 Å between the ssDNA overhangs at the top and the bottom of the pillar, the p53 complex, with an expected size of <120 Å, allows flexible and customizable design of 3D structures at the – nanometer scale (19–21). The use of 2D arrays of DNA scaffold would then not interact with either of the air water interfaces. In to bind proteins for cryo-EM sample preparation has been pro- addition, by anchoring the protein in the middle of the ice layer, posed before (22). Many proteins naturally interact with nucleic differences in defocus height that might lead to loss of resolution in relatively thick ice layers (42) can be prevented. acids like DNA or RNA. In the first instance, to limit the number Second, by effectively enclosing the target protein inside the of technological challenges, we designed a support structure for support structure, we aimed to protect it from blotting forces, proteins that naturally bind to double-stranded DNA in a se- aggregation, or other harmful interactions. It is also conceivable quence-dependent fashion. This approach includes many pro- that interactions of the target protein with the holey carbon film, teins involved in the regulation of transcription, and we chose the with itself (in the form of aggregation), or with the air–water interface transcription factor p53 as a paradigm. may all influence optimal blotting conditions. Therefore, enclosing p53 plays a central role in the cell cycle (23) and is best known thetargetproteininsideaDNA-origami structure that is always the for its function in tumor suppression (24). The active form of × same may also have a beneficial effect on the reproducibility of op- human p53 is a homotetramer of 4 393 amino acids. Its domain timal blotting conditions for differenttargetproteins. organization consists of an intrinsically disordered N-terminal Third, even though we enclosed the target protein inside the transactivation domain, a proline-rich region, a structured DNA- support, we still aimed to minimize the overlap in projected binding domain (DBD), a tetramerization domain connected via densities of the target protein and the support, because this a flexible linker, and an intrinsically disordered C-terminal reg- density overlap would hamper alignment of the particle. The ulatory domain (25). The structures of the DBD and the tetra- central dsDNA helix traverses the middle of the hollow pillar, merization domain of human p53 have been solved by X-ray and the target protein binds to (approximately) the center of – crystallography and NMR (26 33), as have tetrameric complexes the pillar. Thereby, provided the support structure adopts the of the DBD with fragments of DNA (34, 35). The structures of intended orientation in the ice, with its open top or bottom faces full-length and truncated mutants of human p53 bound to DNA perpendicular to the electron beam, the projected density from have been analyzed by negative-stain EM and small-angle X-ray the target protein may be separated from the projected density of diffraction (36, 37), and the DBDs have the same symmetry as the support by excising smaller subimages (also see Fig. 2). found in the crystallographic studies. Other cryo-EM studies on Lastly, we also sought to exert experimental control over the mouse p53 propose a different symmetry (38, 39). Because the C orientation of our target protein, p53, inside the support struc- terminus of p53 increases unspecific binding to DNA (40), we ture. Because of the helical character of the central dsDNA, we chose to use the human truncated p531–360 construct (with a can induce different relative orientations of our target protein molecular mass of ∼160 kDa for the tetramer) in our experiments. with respect to the support structure. By translating the binding Below, we describe the rationale behind our approach, the results sequence 1 bp upward, the p53 complex will rotate 34° along the obtained with this p53 construct, and the opportunities that this axis of the dsDNA (Fig. 1D). Thereby, by making five different type of support structure offer for more general cryo-EM sample versions of our support structure, shifting the p53-binding preparation procedures. sequence one base pair at a time, we can generate different

Martin et al. PNAS | Published online November 7, 2016 | E7457 Downloaded by guest on October 2, 2021 A BC9.8 nm 6.6 nm 13.2 nm 26.4 nm 26.2 nm

13.6 nm

32.7 nm

1bp 1bp 1bp 1bp

34º 34º 34º 34º

Fig. 1. Design of the support structure. (A) Perspective view of the support structure. Each dsDNA helix is shown as a white cylinder. The position of the

specific binding sequence on the central dsDNA helix is shown in red; ssDNA overhangs (T10) are shown in blue. (B) Top view of the support structure. Inner and outer dimensions of the support structure are shown in gray. The dimensions of the asymmetric feature (flag) are shown in green. (C) Side view of the support structure. (D) Illustration of five different settings for the p53-specific binding sequence on the central dsDNA helix. A surface representation of the tetrameric DBD of p53 is shown in red. In each representation, the p53-binding site is shifted one base upward from left to right. Because of the helical nature of the dsDNA, this shift also results in a rotation of the p53 complex.

orientations that cover the entire 180° of a tomographic tilt series we were able to use the same blotting conditions as for the empty (Fig. 1D). Therefore, we also refer to the central dsDNA helix as support structures, although the grids exhibited fewer areas with the the tilt axis. The capability of rotating a single copy of our target optimal, near-crystalline arrangements of the support structures. protein complex in a controlled manner around the tilt axis turns Nevertheless, we could use the observation that support structures our design from a passive support into the nanoscale equivalent preferentially adopt top-bottom views in regions with the desired ice of a sample holder with a tilting stage. However, as we will dis- thickness to select suitable areas for data acquisition from relatively cuss in more detail below, in practice, it is difficult to achieve low-magnification overviews in the microscope (Fig. S2). Data acquisi- precise control over the orientation of p53 along the tilt axis. tion was performed for each tilt axis setting separately on an FEI Titan Krios microscope at 300 kV with a K2 Summit detector (Table 1). Synthesis and Imaging of the Support Structure. The designed DNA-origami structure was synthesized and purified with a yield The 2D Image Preprocessing. The incorporation of the target between 50% and 90% by using standard procedures (Materials protein inside a larger support structure not only provides and Methods). We first used the support structure alone (i.e., experimental information about its orientation, it also fa- without the target protein) to determine suitable freezing condi- cilitates the selection of individual particles from the mi- tions for cryo-EM grid preparation. We found conditions where crographs. We use the template-based particle selection the support structures adopt a pseudo 2D-crystalline arrangement procedure in RELION for this process (43). By using a 2D in large areas of the grid, where most structures adopt the template structure that corresponds to the top view of our intended top-bottom orientation in the ice layer (Fig. 2A). Using support structure (Fig. 2 A, Inset), we automatically selected cryo-electron tomography, we confirmed that the ice layer is in- 272,914 particles from the five different experiments and deed as thick as the designed DNA-origami support structures combined all of these particles into a single dataset. (Fig. 2B and Movie S1). Interestingly, in areas where the ice To remove inadvertently picked side views, broken support appeared thinner than the designed height, we observed no sup- structures, or other false positives (Fig. 2C) from the dataset, we port structures. In areas where the ice appeared thicker, we also selected 177,287 particles in three rounds of reference-free 2D observed support structures in side-view orientations (Fig. S2). classification. Because the support structure may adopt either a Subsequently, we made cryo-EM grids of the support structure top-up or a bottom-up orientation in the ice, we applied a mirror together with p53. We made five different samples, each with a different operation to the 53% of these particles that contributed to 2D position of the p53 binding sequence on the tilt axis. As hypothesized, class averages with a flag on the top right side of the support

E7458 | www.pnas.org/cgi/doi/10.1073/pnas.1612720113 Martin et al. Downloaded by guest on October 2, 2021 A C PNAS PLUS

DE FG BIOPHYSICS AND COMPUTATIONAL BIOLOGY

Fig. 2. Image-processing strategy. (Scale bars: 20 nm.) (A) Part of a typical micrograph. A template for automated particle picking is shown in the top right corner. (B) Tomographic side view of a hole showing a monolayer of support structures. Near the edge of the hole (indicated with an arrow), the ice gets slightly thicker. (C) Examples of 2D classes that are discarded. (D) Examples of 2D classes of intact structures with the flag in the top left or top-right. A mirror operation is applied to images with the flag on the top right. (E) The average of all intact particles with the correct orientation in ice, including the mirrored particles, is used as a template for their alignment. (F) Subimages (cyan) with a width and height of 20 nm are extracted from the aligned particles and submitted to 2D classification with a prior on the in-plane rotation. A circular mask that is applied during this process is shown in yellow. Three typesof particles are distinguished: supports without tilt axis (F, Top), supports with tilt axis but without a density for p53 (F, Middle), and supports with tilt axis and p53 density (F, Bottom). (G) Illustration of the final selection of particles, where the angle from a realignment of the p53-only subimage is compared with the angle of the entire support structure, and particles with large differences in these angles (e.g., Lower) are discarded.

structure (Fig. 2D and Table 1). The resulting set of intact tilt axis, provide information about the five parameters that define support structures was aligned to a common reference (Fig. 2E), the orientation of every p53 complex. By using only 2D alignment to allow extraction of the central part of the individual particle parameters, one can therefore calculate a tomographic 3D re- images in a smaller box. An additional round of reference-free construction of p53. The two in-plane translations of the support 2D classification was used to discard particles for which no ob- structure, combined with the known offset of the binding sequence on vious p53 density was visible (Fig. 2F). In this calculation, we the tilt axis, can be used to center the target protein. In addition, used a prior probability, or prior in short, on the in-plane rota- three Euler angles describe the relative orientations of all p53 par- tions. In the empirical Bayesian approach to image processing in ticles with respect to a common 3D frame of reference. By using a RELION (44), prior probabilities on orientational parameters reference with the dsDNA tilt axis along the y axis, the first Euler are expressed as Gaussian functions centered on the expected angle (“rot” inRELION)isexpectedtobe∼0°, because the tilt axis is value for that orientation and with a SD that expresses the un- designed to run perpendicular to the top–bottom axis of the support certainty in that expectation. In this case, we centered the prior structure. The second Euler angle (“tilt”) is determined by the tilt- on the in-plane rotations from the initial 2D alignment of the axis setting of the experiment. By defining the tilt angle for our entire support structure and used a SD of 15° to allow for errors central tilt axis position (0 bp) as 90°, the 1- or 2-bp shifted po- in those assignments. At this point, 36,837 particle images were sitions of the binding sequence on either side have expected tilt selected. A final round of particle selection was based on iden- angles of 21.4° (−2 bp), 55.7° (−1 bp), 124.3° (+1 bp), and 158.6° tifying particles that rotated too much in the p53-only re- (+2 bp). Finally, the third Euler angle (“psi”) is directly available alignment with respect to the support structure (Fig. 2G) and led from the in-plane rotation determined in the 2D alignment of the to a final dataset of 22,698 subimages containing p53. support structure. To calculate the initial tomographic p53 reconstruction, we The 3D Reconstruction. The 2D alignment parameters of the first split the selected 22,698 subimages by their original tilt axis support structures, combined with the known setting of the settings and performed five separate 2D classifications using a

Martin et al. PNAS | Published online November 7, 2016 | E7459 Downloaded by guest on October 2, 2021 Table 1. Data acquisition statistics for each tilt axis setting p53 in 2D classes Tilt axis Autopicked Intact top view Empty Supports with Supports with p53 in expected used for initial-model setting, bp Micrographs particles supports (mirrored) supports only DNA p53-like density orientation calculation

−2 604 69,214 36,727 (19,672) 11,018 15,439 10,270 5,821 2,363 −1 415 51,830 40,824 (26,886) 15,916 17,082 7,826 4,831 2,197 0 582 61,528 41,131 (19,122) 18,452 16,587 6,092 3,373 1,375 +1 576 65,942 39,303 (19,526) 14,424 15,665 9,214 5,169 1,795 +2 385 24,400 19,302 (8,988) 6,319 7,412 5,571 3,504 1,541 Total 2,562 272,914 177,287 (94,194) 66,129 72,185 38,973 22,698 9,271

prior on the in-plane rotations of 15°. From these five runs, we flexible than anticipated) will lead to less-informative priors. To selected 35 total 2D class averages with the best protein-like some extent, deviations from the anticipated tilt angle are actually features (Fig. 3A). We then used the expected Euler angles as beneficial to the reconstruction process, because they will lead to a defined above to perform a tomographic reconstruction, directly more uniform angular sampling along the tomographic tilt axis. from the 2D class averages. A preliminary map without sym- Therefore, we first performed a rotational and translational metry (Fig. 3B) indicated the presence of C2 symmetry, which realignment of the 35 selected 2D class averages, where we used was subsequently imposed (Fig. 3C). Apart from the twofold a prior with a SD of 10° on rot and psi, but tilt was left free. The rotational symmetry, the reconstructed map also showed local resulting reconstruction (Fig. 3D) was then used as an initial translational pseudosymmetry along the dsDNA axis. model in a set of 3D refinements, where we used the 9,271 p53 The observations that particles from different tilt-axis settings particles that were assigned to the selected 2D classes. Without gave rise to similar 2D class averages, and that different 2D class using rotational priors on any of the Euler angles, or when using averages were observed within a single tilt-axis dataset, indicate only a prior on rot or psi, the reconstructions looked worse (e.g., that the tilt axis is probably more flexible than anticipated. In density for the dsDNA disappeared) than when using a 5° prior principle, the statistical framework of RELION is well-suited to on both rot and psi. As expected from the observation that our model deviations from the expected orientations of each indi- tilt axis is more flexible than anticipated, imposing a prior on tilt vidual particle through the use of Gaussian priors on the trans- made the reconstruction worse (Fig. S3). When using pro- lations and the Euler angles. The SD of the Gaussian priors can gressively less informative (i.e., broader, priors on rot and psi), the be used to tune the amount of expected deviations. Larger devi- reconstructions also became worse (Fig. S4), whereas priors with ations from the expected orientations (e.g., because the attach- SDs <5° are too narrow for the angular sampling rate used in the ment of the target protein to the DNA-origami structure is more refinement. Consequently, our best refinement had a SD of 5° for

ABCD

Fig. 3. Initial model generation and final maps. (A) Class averages used for the initial model generation sorted by tilt axis setting. An illustration of the designed orientation is shown on the left, and the applied tilt angle is shown in front of the class averages. (B) Initial model reconstructed with the expected tilt angles in C1. The angles for rot and psi were all set to 0°. In RELION, the first rotation of the 3D reference object (rot) is around the z axis, which comes out of the xy plane of the figure; the second rotation (tilt) is around the new y axis, and the third rotation (psi) is around the new z axis. B, Inset shows a simplified explanation of these rotations from the point of view of the experimental particles. In that case, psi is the in-plane rotation, tilt is the rotation around the central DNA axis, and rot describes out-of-plane rocking. (C) The same model as in B with C2 symmetry imposed. The twofold symmetry axis is along the z axis and is indicated with an oval. (D) Map after realignment of the class averages with 5° priors on the rot and psi angles and unrestricted tilt angles. (E) Different views of the final map generated from 9,271 particles. The estimated resolution is ∼15 Å. Protein Data Bank model 4HJE (35) of the DBD in blue. (F) Histogram of the refined tilt angles for each of the tilt axis settings.

E7460 | www.pnas.org/cgi/doi/10.1073/pnas.1612720113 Martin et al. Downloaded by guest on October 2, 2021 the priors on psi and rot and left the tilt angle unrestrained. of p53. A further reduction in particle number came from our PNAS PLUS The resulting map, at an estimated resolution of ∼15 Å, shows the image-selection procedures: 38% of the particles that contained same domain architecture as the known crystal structure of the density for p53 had it bound in an unrealistic orientation in re- DBD of p53 bound to dsDNA (35) and shows additional densities spect to the support structure, and from these only 41% con- in both the front and the back of the complex (Fig. 3E). The tributed to the selected 2D classes of the subimages that showed histogram of the refined tilt angles confirms that the tilt axis is the best protein-like features. Second, although the low particle more flexible than anticipated. Some information in the tilt angle yield may have hampered reaching higher resolution, the se- is maintained, but angles ∼90° are somehow disfavored (Fig. 3F). lected particles still represented >18,000 asymmetric units, which should probably have yielded a higher resolution reconstruction Discussion (cf. ref. 46). One possibility could be that the central cavity of the We present an approach to sample preparation for cryo-EM support structure was too small for the defocus used. De- structure determination of biological macromolecules. Using 3D localization in the images because of defocusing may have led to DNA-origami, we designed a support structure with a defined a superposition of the signal from the support structure and the size and shape that binds specifically to a target protein of interest. target protein, which may have limited the alignment of the in- The resulting structure is an artificial molecular support that ex- dividual target proteins, and therefore the resolution of the final erts experimental control over the orientations of individual pro- reconstruction. This problem could in principle be circumvented tein molecules and protects them from aggregation or harmful by in-focus imaging through phase-plate technology (47). In interactions with the air–water interface. In addition, the support addition, future designs of larger, more stable support structures structure may facilitate the optimization of freezing conditions with a more accessible protein binding site, and with better in- and aid in the selection of suitable ice thickness for data acqui- corporation of the tilt axis may improve both particle yield and sition. Nevertheless, as discussed below, this work does not yet resolution. To create these larger supports with low defect rates, represent a ready-to-use solution for high-resolution structure de- new assembly and purification strategies will need to be con- termination of a wide range of different target proteins, but should sidered and adopted (see, for example, refs. 48 and 49). rather be considered as a proof-of-principle toward achieving this Nevertheless, at a resolution of ∼15 Å, the resulting recon- ambitious goal. struction provides useful information about how p53 binds to By choosing a target protein that naturally binds to dsDNA in dsDNA. The observation that our reconstruction shows C2 sym- a sequence-specific manner, our support structure was designed metry with an additional twofold translational component along

to act as the nanoscale equivalent of a goniometer that exerts the DNA axis, which is similar to the crystal structure of the tet- BIOPHYSICS AND

experimental control over the orientations of individual protein rameric core of p53 bound to dsDNA, indicates that p53 does not COMPUTATIONAL BIOLOGY complexes. By using five different positions of the p53 binding bind to DNA in a previously proposed arrangement with D2 sequence on a central dsDNA helix in our structure, our struc- symmetry (38, 39). Instead, the presence of extra density in the ture intended to provide five different views of a p53 tetramer front and back of the complex (Fig. 3E) provides support for an bound to the support structure. Although the prior information alternative model where the p53 tetramerization domain binds to about the five different orientations could indeed be used suc- the opposite side of the dsDNA helix from the DBD (36). cessfully to calculate an initial tomographic reconstruction of the The design of our support structure also contains several p53 tetramer bound to DNA, further refinement of the orien- useful features that may inspire future experiments. By enclosing tations of 2D class averages or individual particle images the target protein in a hollow structure, it is protected from in- revealed a distribution in tilt angles that is much broader than teractions with copies of itself or with the hydrophobic air–water one would expect from a rigid dsDNA helix, and somehow seems interface during cryo-EM sample preparation. This protection to disfavor tilt angles ∼90°. Because the complete structure of may prevent the proteins from aggregation, or from unfolding or the p53 tetramer remains unknown, it could be that the support adopting preferred orientations against this air–water interface. structure is too small to accommodate p53 in this orientation. Also, it might be that the optimization of freezing conditions will Analysis of multiple 2D class average images from our data depend more on the support structure than on the nature of the (Movie S2) revealed that the support structures are also not as protein inside, so that the optimal conditions would differ less rigid as one might need to exert precise orientation control. for different proteins compared with current methods. We in- Moreover, despite previous attempts at optimizing the indeed used similar conditions for the samples with only support corporation of the central dsDNA helix in the support structure structures and the mixture with p53, although the mixture (45), it could be that our current design (Figs. S5 and S6) still leads showed fewer regions of near-crystalline arrangements of the to incorrect attachments in a subset of the structures. Neverthe- support structures. We used approximately six times more p53 less, despite the lack of complete control over the tilt angle, the tetramers than support structure in our mixture and did not at- support structure still provided information about the center of tempt to remove unbound p53 tetramers through additional the particle, its in-plane rotation (the psi angle), and its out-of-plane purification steps. It could therefore be that interactions of un- rocking (the rot angle), and this information could be used in bound p53 tetramers with the air–water interface still had an statistical priors to improve the 3D reconstruction. The intended effect on the freezing conditions. In addition, the observation design of our support structure as a molecular goniometer that that the support structures adopt monolayers in an ice layer with provides experimental information about the orientations of in- the intended thickness is potentially useful. One might, for ex- dividual p53 complexes was, therefore, partially successful. ample, try to add support structures to standard cryo-EM sam- Our experiments also revealed several challenges that will ples with the idea of using them as “spacers” (13) to control ice need to be overcome in future studies. First of all, the final thickness. Admittedly, it remains difficult to conclude whether number of 9,271 selected p53 particle images represents a low the support structures maintain the desired ice thickness over a yield from 2,562 selected micrographs. Approximately 42% of given area or whether the monolayers just happen to form in our support structures did not incorporate the tilt axis, and 77% those areas where the ice thickness is ideal. However, because of the remaining support structures did not bind to p53. The one can easily select areas with monolayers of support structures latter was surprising, because at a concentration of 1 mM p53 from low-magnification overview images in the microscope, the tetramers and 150 nM support structures, and with a dissociation support structures do facilitate the data acquisition process. The constant of ∼50 nM for dsDNA, we had expected almost all interactions of our support structure with the air–water interface support structures to contain p53. Possibly, steric hindrance from may also have a beneficial effect on the local particle concen- the support or strain in the central dsDNA affected the binding tration. For a 300-Å-thick layer of a 150 nM solution, one would

Martin et al. PNAS | Published online November 7, 2016 | E7461 Downloaded by guest on October 2, 2021 expect only a single support structure in every micrograph. Our contribute to tackling some of the outstanding challenges in this observation of ∼100 structures per micrograph suggests that inter- rapidly changing technique. actions of the support structure with the air–water interface, possibly both at the top and the bottom, may lead to a strong local Materials and Methods enrichment in concentration. For proteins, a similar enrichment has Origami Design, Synthesis, and Purification. DNA-origami design (Dataset S1) been attributed to a sticky layer of denatured protein at the air– was performed in cadnano (Version 0.2) (58). The scaffold DNA (the 7,560-nt- water interface (15). Because our samples without protein show a long version of the M13mp18 phage genome) was prepared as described (59). similar—or even stronger—enrichment in DNA support structures, DNA staple oligonucleotides, prepared by solid-phase chemical synthesis, were ordered from Eurofins MWG. The DNA oligos for the tilt axis were ordered to unbound p53 tetramers are probably not required for the enrich- – HPLC-grade purity; all other oligos (Table S1) were ordered to high-purity, salt- ment. It is difficult to assess whether the interactions with the air free grade. The supports containing different tilt-axis settings were synthe- water interface lead to unfolding of the support structures them- sized individually in one-pot mixtures containing 50 nM scaffold DNA, 75 nM selves. We did observe partial structures (Fig. 2 A and C), but these tilt axis DNA, and 200 nM staple DNA in a 10 mM Tris buffer at pH 7.6 with

could also be explained by folding defects. It therefore remains 1 mM EDTA and 20 mM MgCl2. The mixture was incubated at 65 °C for 15 min, unclear whether the presence of denatured DNA material at the then annealed from 60 °C to 45 °C over the course of 8 h, and stored at 25 °C. air–water interface is important for the enrichment effect. Purification from excess staple oligos was performed in five rounds of mo- The experiments described here rely on the specific binding of lecular mass cutoff filtration by using 100-kDa Amicon filters (Millipore) in a a target protein that naturally binds to dsDNA in a sequence- buffer containing 20 mM Tris base and 5 mM MgCl2.Thefinalconcentrationof the support structures was adjusted to ∼250 nM. specific manner. Although several transcription factors are known to do so, the design of a more broadly applicable support P53 Expression and Purification. Production and purification of truncated p53 structure would be desirable. It would be relatively straightfor- variant (residues 1–360) lacking the C-terminal regulatory domain followed ward to include a DNA mismatch in the central axis, which could published protocols (28, 60). Briefly, the proteins were produced in Escher- include target proteins from DNA-mismatch repair pathways. ichia coli BL21(DE3) as a fusion protein with N-terminal 6× His-tag, Bacillus Alternatively, one could construct a central axis with a bubble or stearothermophilus lipoyl domain, and tobacco etch virus protease cleavage from DNA/RNA hybrids to include a range of different target site. They were then purified by using standard His-tag purification proto- proteins related to replication, transcription, or DNA repair. cols, followed by tobacco etch virus protease cleavage, heparin affinity However, to design a support that could bind to an arbitrary chromatography, and a final gel filtration step on a Superdex 200 16/60 target protein, one would probably need to combine chemical preparative gel filtration column (GE Healthcare) in 300 mM NaCl, 20 mM Tris (pH 7.5), and 5 mM DTT. Protein samples were flash-frozen in liquid modifications of some of the DNA staples within the support nitrogen and stored at −80 °C. The variant contained four stabilizing mu- with a specific tag on the target protein. For example, one could tations, M133L/V203A/N239Y/N268D (28, 61, 62), in the DBD. use commercially available biotin-labeled DNA on the support The purified p53 sample was mixed with the purified DNA-origami sample to structure to bind a counterpart to a protein tag attached to a final concentrations of ∼150 nM DNA origami and ∼4 μMp53(monomer)ina

monovalent streptavidin (50, 51). For the study of membrane 20 mM Tris buffer containing 1.5 μM DTT, 45 mM NaCl, and 4 mM MgCl2.The proteins, one might even consider designs where nanodiscs (52) mixture was incubated for 20 min at 4 °C before preparing cryo-EM grids. are attached to DNA-origami support structures or where these structures interact directly with patches of membranes (53). Such Electron Microscopy. Cryo-EM grids for the support structures alone or for the more generally applicable designs probably would retain even support structures with p53 and the five different settings of the tilt axis were μ less control over the orientation of the target protein than the prepared separately using similar procedures. Aliquots of 3 L of sample were design described in this work. Still, the support structure could incubated for 10 s on glow-discharged Quantifoil grids, blotted for 2 s, and plunge-frozen in liquid ethane by using a Vitrobot Mark 3 (FEI Company). maintain some of its other useful features, such as facilitating Grids were transferred to an FEI Titan Krios microscope that was operated at particle selection, preventing aggregation, and keeping proteins 300 kV. Images were recorded on a K2 detector using a Gatan energy filter away from harmful interfaces. Apart from the beneficial effects on (with a slit width of 20 eV). Movies with a total electron dose of ∼38 e−/Å2 cryo-EM sample preparation, large support structures that specif- were recorded in superresolution mode and subsequently downscaled to a ically bind a target protein of interest might even play a role in final pixel size of 1.76 Å in RELION. The defocus was varied between 1 and protein purification. Provided that the binding of the target protein 5 μm. Cryo-electron tomography data were collected by using described would be tight enough, their large molecular mass (∼5 MDa) procedures (63). would allow relatively straightforward separation of the target Electron micrographs were manually evaluated for astigmatism and drift, protein from smaller contaminants through the use of, for example, and 2,562 micrographs were selected for further analysis. Beam-induced motion correction was performed in UCSF MOTIONCORR (64); estimation of gel filtration of sucrose gradients. Alternatively, one could try to fix contrast transfer function parameter was performed in Gctf (65); and all the DNA supports directly on the grid surface and use the grids subsequent image-processing operations were performed in RELION-1.4 (44). themselves for on-grid affinity purification in a similar manner as was done with nitrilotriacetic acid-modified lipid monolayers (54, ACKNOWLEDGMENTS. We thank Christos Savva, Shaoxia Chen, Toby 55) or antibodies attached to carbon films (56, 57). Darling, and Jake Grimmett for technical support, and Miriana Petrovich In summary, this work provides an original approach to exert for protein purification. This work was supported by the European Molecular Biology Organisation through a long-term postdoctoral fellow- experimental control over the orientations of individual protein ship (ALTF-1229-2013 to T.G.M.) and an advanced fellowship (aALTF-778- complexes and protect them from harmful forces during cryo- 2015 to T.A.M.B.), and by European Commission Marie Skłodowska-Curie EM sample preparation. As the field of cryo-EM structure de- postdoctoral fellowships (to T.G.M. and X.-c.B.). The project was further termination keeps growing, the physics involved in its sample supported by European Research Council Starting Grant GA 256270 (to H.D.); by the Deutsche Forschungsgemeinschaft through grants provided within the preparation will become clearer, and new concepts in sample Sonderforschungsbereich SFB863, the Center for Integrated Protein Science preparation will continue to emerge. Our approach provides Munich, and the Nano Initiative Munich; and UK Medical Research Council ample possibilities for further developments and may thereby Grants MC_UP_A024_1010 (to A.R.F.) and MC_UP_A025_1013 (to S.H.W.S.).

1. Bai X-C, McMullan G, Scheres SHW (2015) How cryo-EM is revolutionizing structural 5. Taylor KA, Glaeser RM (1976) Electron microscopy of frozen hydrated biological biology. Trends Biochem Sci 40(1):49–57. specimens. J Ultrastruct Res 55(3):448–456. 2. Cheng Y, Grigorieff N, Penczek PA, Walz T (2015) A primer to single-particle cryo- 6. Frank J, Verschoor A, Boublik M (1981) Computer averaging of electron micrographs electron microscopy. Cell 161(3):438–449. of 40S ribosomal subunits. Science 214(4527):1353–1355. 3. Adrian M, Dubochet J, Lepault J, McDowall AW (1984) Cryo-electron microscopy of 7. Penczek P, Radermacher M, Frank J (1992) Three-dimensional reconstruction of single viruses. Nature 308(5954):32–36. particles embedded in ice. Ultramicroscopy 40(1):33–53. 4. Dubochet J, et al. (1988) Cryo-electron microscopy of vitrified specimens. Q Rev 8. Glaeser RM (2016) The resolution revolution: Recent advances in cryoEM. Methods in Biophys 21(2):129–228. Enzymology, ed Crowther R (Academic, New York), Vol 579, pp 19–50.

E7462 | www.pnas.org/cgi/doi/10.1073/pnas.1612720113 Martin et al. Downloaded by guest on October 2, 2021 9. Henderson R, et al. (1990) Model for the structure of bacteriorhodopsin based on 37. Melero R, et al. (2011) Electron microscopy studies on the quaternary structure of p53 PNAS PLUS high-resolution electron cryo-microscopy. J Mol Biol 213(4):899–929. reveal different binding modes for p53 tetramers in complex with DNA. Proc Natl 10. Kühlbrandt W, Wang DN, Fujiyoshi Y (1994) Atomic model of plant light-harvesting Acad Sci USA 108(2):557–562. complex by electron crystallography. Nature 367(6464):614–621. 38. Okorokov AL, et al. (2006) The structure of p53 tumour suppressor protein reveals the 11. Nogales E, Wolf SG, Downing KH (1998) Structure of the alpha beta tubulin dimer by basis for its functional plasticity. EMBO J 25(21):5191–5200. electron crystallography. Nature 391(6663):199–203. 39. Aramayo R, et al. (2011) Quaternary structure of the specific p53-DNA complex re- 12. Gonen T, et al. (2005) Lipid-protein interactions in double-layered two-dimensional veals the mechanism of p53 mutant dominance. Nucleic Acids Res 39(20):8960–8971. AQP0 crystals. Nature 438(7068):633–638. 40. Anderson ME, Woelker B, Reed M, Wang P, Tegtmeyer P (1997) Reciprocal in- 13. Glaeser RM, et al. (2016) Factors that influence the formation and stability of thin, terference between the sequence-specific core and nonspecific C-terminal DNA – cryo-EM specimens. Biophys J 110(4):749–755. binding domains of p53: Implications for regulation. Mol Cell Biol 17(11):6255 6264. 14. McDowall AW, Dobro MJ, Melanson LA, Jensen GJ (2010) Cryo-EM part A: Sample 41. Bai X-C, Martin TG, Scheres SHW, Dietz H (2012) Cryo-EM structure of a 3D DNA- – preparation and data collection. Methods in Enzymology, ed Jensen GJ (Academic, origami object. Proc Natl Acad Sci USA 109(49):20012 20017. 42. Jensen GJ (2001) Alignment error envelopes for single particle analysis. J Struct Biol New York), Vol 481, pp 63–82. 133(2-3):143–155. 15. Taylor KA, Glaeser RM (2008) Retrospective on the early development of cryoelectron 43. Scheres SHW (2015) Semi-automated selection of cryo-EM particles in RELION-1.3. microscopy of macromolecules and a prospective on opportunities for the future. J Struct Biol 189(2):114–122. J Struct Biol 163(3):214–223. 44. Scheres SHW (2012) RELION: Implementation of a Bayesian approach to cryo-EM 16. Glaeser RM, Han B-G (2016) Opinion: Hazards faced by macromolecules when con- structure determination. J Struct Biol 180(3):519–530. fined to thin aqueous films. Biophys Rep, 10.1007/s41048-016-0026-3. 45. Martin TG (2014) Functional synthetic DNA nanostructures. PhD dissertation (Tech- 17. Thompson RF, Walker M, Siebert CA, Muench SP, Ranson NA (2016) An introduction nical University of Munich, Munich). to sample preparation and imaging by cryo-electron microscopy for structural bi- 46. Fernández IS, et al. (2013) Molecular architecture of a eukaryotic translational initi- – ology. Methods 100:3 15. ation complex. Science 342(6160):1240585. 18. Stark H, Chari A (2016) Sample preparation of biological macromolecular assemblies 47. Danev R, Baumeister W (2016) Cryo-EM single particle analysis with the Volta phase for the determination of high-resolution structures by cryo-electron microscopy. plate. eLife 5:e13046. Microscopy (Oxf) 65(1):23–34. 48. Gerling T, Wagenbauer KF, Neuner AM, Dietz H (2015) Dynamic DNA devices and 19. Douglas SM, et al. (2009) Self-assembly of DNA into nanoscale three-dimensional assemblies formed by shape-complementary, non-base pairing 3D components. shapes. Nature 459(7245):414–418. Science 347(6229):1446–1452. 20. Dietz H, Douglas SM, Shih WM (2009) Folding DNA into twisted and curved nanoscale 49. Shaw A, Benson E, Högberg B (2015) Purification of functionalized DNA origami shapes. Science 325(5941):725–730. nanostructures. ACS Nano 9(5):4968–4975. 21. Castro CE, et al. (2011) A primer to scaffolded DNA origami. Nat Methods 8(3): 50. Fairhead M, et al. (2014) SpyAvidin hubs enable precise and ultrastable orthogonal 221–229. nanoassembly. J Am Chem Soc 136(35):12355–12363. 22. Malo J, et al. (2005) Engineering a 2D protein-DNA crystal. Angew Chem Int Ed Engl 51. Veggiani G, Zakeri B, Howarth M (2014) Superglue from bacteria: Unbreakable 44(20):3057–3061. bridges for protein nanotechnology. Trends Biotechnol 32(10):506–512. 23. Lane D, Levine A (2010) p53 research: The past thirty years and the next thirty years. 52. Bayburt TH, Grinkova YV, Sligar SG (2002) Self-assembly of discoidal phospholipid

Cold Spring Harb Perspect Biol 2(12):a000893. bilayer nanoparticles with membrane scaffold proteins. Nano Lett 2(8):853–856. BIOPHYSICS AND 53. Kocabey S, et al. (2015) Membrane-assisted growth of DNA origami nanostructure

24. Joerger AC, Fersht AR (2016) The p53 pathway: Origins, inactivation in cancer, and COMPUTATIONAL BIOLOGY emerging therapeutic approaches. Annu Rev Biochem 85:375–404. arrays. ACS Nano 9(4):3530–3539. 25. Joerger AC, Fersht AR (2008) Structural biology of the tumor suppressor p53. Annu 54. Kelly DF, Abeyrathne PD, Dukovski D, Walz T (2008) The Affinity Grid: A pre-fabri- Rev Biochem 77:557–582. cated EM grid for monolayer purification. J Mol Biol 382(2):423–433. 26. Cañadillas JMP, et al. (2006) Solution structure of p53 core domain: Structural basis 55. Walz T, Kelly DF, Dukovski D (2010) Cryo-EM part A: Sample preparation and data for its instability. Proc Natl Acad Sci USA 103(7):2109–2114. collection. Methods in Enzymology, ed Jensen GJ (Academic, New York), Vol 481, pp – 27. Wang Y, Rosengarth A, Luecke H (2007) Structure of the human p53 core domain in 83 107. the absence of DNA. Acta Crystallogr D Biol Crystallogr 63(Pt 3):276–281. 56. Yu G, et al. (2014) Single-step antibody-based affinity cryo-electron microscopy for – 28. Natan E, et al. (2011) Interaction of the p53 DNA-binding domain with its N-terminal imaging and structural analysis of macromolecular assemblies. J Struct Biol 187(1):1 9. – extension modulates the stability of the p53 tetramer. J Mol Biol 409(3):358–368. 57. Yu G, Li K, Jiang W (2016) Antibody-based affinity cryo-EM grid. Methods 100:16 24. 58. Douglas SM, et al. (2009) Rapid prototyping of 3D DNA-origami shapes with cadnano. 29. Cho Y, Gorina S, Jeffrey PD, Pavletich NP (1994) Crystal structure of a p53 tumor Nucleic Acids Res 37(15):5001–5006. suppressor-DNA complex: Understanding tumorigenic mutations. Science 265(5170): 59. Douglas SM, Chou JJ, Shih WM (2007) DNA-nanotube-induced alignment of mem- 346–355. brane proteins for NMR structure determination. Proc Natl Acad Sci USA 104(16): 30. Jeffrey PD, Gorina S, Pavletich NP (1995) Crystal structure of the tetramerization 6644–6648. domain of the p53 tumor suppressor at 1.7 angstroms. Science 267(5203):1498–1502. 60. Veprintsev DB, et al. (2006) Core domain interactions in full-length p53 in solution. 31. Mittl PRE, Chène P, Grütter MG (1998) Crystallization and structure solution of p53 Proc Natl Acad Sci USA 103(7):2115–2119. (residues 326-356) by molecular replacement using an NMR model as template. Acta 61. Nikolova PV, Henckel J, Lane DP, Fersht AR (1998) Semirational design of active tumor – Crystallogr D Biol Crystallogr 54(Pt 1):86 89. suppressor p53 DNA binding domain with enhanced stability. Proc Natl Acad Sci USA 32. Clore GM, et al. (1995) Refined solution structure of the oligomerization domain of 95(25):14675–14680. – the tumour suppressor p53. Nat Struct Biol 2(4):321 333. 62. Joerger AC, Allen MD, Fersht AR (2004) Crystal structure of a superstable mutant of 33. Lee W, et al. (1994) Solution structure of the tetrameric minimum transforming do- human p53 core domain. Insights into the mechanism of rescuing oncogenic muta- – main of p53. Nat Struct Biol 1(12):877 890. tions. J Biol Chem 279(2):1291–1296. 34. Kitayner M, et al. (2010) Diversity in DNA recognition by p53 revealed by crystal 63. Bharat TAM, Russo CJ, Löwe J, Passmore LA, Scheres SHW (2015) Advances in single- structures with Hoogsteen base pairs. Nat Struct Mol Biol 17(4):423–429. particle electron cryomicroscopy structure determination applied to sub-tomogram 35. Chen Y, et al. (2013) Structure of p53 binding to the BAX response element reveals averaging. Structure 23(9):1743–1753. DNA unwinding and compression to accommodate base-pair insertion. Nucleic Acids 64. Li X, et al. (2013) Electron counting and beam-induced motion correction enable near- Res 41(17):8368–8376. atomic-resolution single-particle cryo-EM. Nat Methods 10(6):584–590. 36. Tidow H, et al. (2007) Quaternary structures of tumor suppressor p53 and a specific 65. Zhang K (2016) Gctf: Real-time CTF determination and correction. J Struct Biol 193(1): p53 DNA complex. Proc Natl Acad Sci USA 104(30):12324–12329. 1–12.

Martin et al. PNAS | Published online November 7, 2016 | E7463 Downloaded by guest on October 2, 2021