Atomic-resolution structures from fragmented protein crystals with the cryoEM method MicroED.

Traditionally, crystallographic analysis of macromolecules has depended on large, well-ordered crystals, which often require significant effort to obtain. Even sizable crystals sometimes suffer from pathologies that render them inappropriate for high-resolution structure determination. Here we show that fragmentation of large, imperfect crystals into microcrystals or nanocrystals can provide a simple path for high-resolution structure determination by the cryoEM method MicroED and potentially by serial femtosecond crystallography.

Introduction

Large and perfect crystals are desirable for structure determination because they yield a strong signal over background. In reality, crystals of biological material are not perfect. In the mosaic model, a real, imperfect crystal is composed of several small but well-ordered blocks[1]. These mosaic blocks have a finite size, are misaligned with respect to each other, and may be composed of unit cells with different dimensions[2]. Depending on the nature and degree of disorder between the mosaic blocks, an imperfect crystal may exhibit a plethora of pathologies, which may hamper subsequent data reduction, limit the resolution of the final model, and can prevent structure determination altogether (Supplementary figures 1–9). Many such defects primarily affect larger crystals, and small crystals may yield superior data quality where diffraction is not limited by the number of diffracting unit cells in the crystal[3].

For these reasons, methods such as serial femtosecond crystallography (SFX)[4] at an X-ray free-electron laser (XFEL) and the electron cryomicroscopy (cryoEM) method microelectron diffraction (MicroED)[5] are actively being developed. These methods can yield structures to resolutions better than 1 Å from crystals that are significantly smaller than those which are required for standard crystallography: ~10,000× smaller in volume for XFEL and ~1,000,000× smaller for MicroED. In fact, in both these techniques large crystals can prove problematic. This is because large crystals can clog up the nozzle of certain SFX sample delivery systems[6,7], while in MicroED the large electron scattering cross-section implies that absorption extinguishes diffraction when the sample is too thick[8].

Here we show that sonication, vigorous pipetting, or vortexing can be used to break large imperfect crystals into small, single-crystal fragments that are suitable for data collection and atomic structure determination (Figure 2). Delicate samples may benefit from gentler fragmentation by vortexing, while harsher methods such as pipetting and ultimately sonication are required to break more robust crystals. In many cases, the untreated crystals were mosaic, yielded diffraction patterns with multiple lattices, or were otherwise not suitable for standard crystallographic experiments (Figure 2, Supplementary figures 1–9). In the case of the amyloid-forming peptide of tau (VQIVYK), what appeared to be large crystals were in fact crystal bundles that produced low resolution powder-like diffraction. These problems are traditionally overcome by modifying the crystallization conditions to optimize crystal growth and quality a process that can be tedious and labor intensive, particularly when crystal pathologies do not become apparent until data processing. Breaking of large crystals by physical means produced fragments that appeared crystallographically homogeneous and yielded diffraction data at atomic resolution, void of the above artifacts. This was true even without further optimization of the growth conditions for crystals formed by various macromolecules of molecular weights between 0.7 kDa and 34.6 kDa, and solvent contents ranging from 30% to 60% with a range of unit cell dimensions and space groups (Table 1). MicroED has already allowed rapid structure solution from several peptide fragments that resisted solution using microfocus synchrotron sources in spite of many months of crystal optimization (Supplementary figures 2–4), and the approach could be used for samples where large crystals exist but standard crystallographic methods fail, owing to crystal imperfections.

Figure 2

Eight atomic resolution structures determined by MicroED from crystal fragments

Left column: ribbon representations of the corresponding macromolecules. The tau peptide model indicates a β-sheet generated by crystal packing. Middle column: 2mFo-DFc charge density maps contoured at 1.5 σ above the mean, showing eight residues in loop regions of the respective structures, where m is the figure of merit and D is derived from coordinate error estimates. For TGF-βm:TβRII one of the disulfide bonds is shown. Right column: SA composite omit maps contoured at 1 σ above the mean, except for the maps from lysozyme and proteinase K, which are contoured at 1.5 σ above the mean. Depressions or holes can be observed in the density of side chains of aromatic residues for tau peptide and lysozyme, and for proline residues in the proteinase K structure. Iodide and calcium ions are visible in the omit maps for xylanase and trypsin, respectively. All figures were generated using PyMol[19].

Table 1

Data processing and model refinement statistics for the reported crystal structures

Data sets were collected and processed as described in the online methods section. The dose rate did not exceed 0.01 e−/Å2/s and the mean per-crystal exposure time is given as 〈Texposure〉 for each sample. Except for tau peptide and TGF-βm:TβRII, reflections were integrated to the corners of the detector; the final resolution cutoff was determined based on CC1/2 and the stability of the refinement procedure.

	Tau peptide(PDB id: 5k7n;EMDB id: EMD-8216)	Lysozyme (PDBid: 5k7o; EMBDid: EMD-8217)	TGF-βm:TβRII(PDB id: 5ty4;EMDB id: EMD-8472)	Xylanase (PDB id:5k7p; EMBD id:EMD-8218)	Thaumatin (PDBid: 5k7q; EMBDid: EMD-8219)	Trypsin (PDB id:5k7r; EMBD id:EMD-8220)	Proteinase K(PDB id: 5k7s;EMBD id: EMD-8221)	Thermolysin (PDBid: 5k7t; EMBD id:EMD-8222)
	Data collection
Resolution1 (Å)	14.70–1.10	30.58–1.50	26.64–2.90	25.55–1.90	27.73–2.11	27.63–1.50	20.75–1.30	30.14–1.60
Number of crystals	2	7	3	4	3	10	6	4
〈Texposure〉 (s)	159.9	127.7	140.8	172.7	179.7	155.8	122.2	187.6
Molecular weight(kDa)	0.7	14.4	19.1	21.0	22.2	23.4	28.9	34.6
	Data processing
Resolution1 (Å)	14.70–1.10(1.23–1.10)	30.59–1.80(2.06–1.80)	26.64–2.90(3.07–2.90)	25.55–2.30(2.63–2.30)	27.73–2.50(2.86–2.50)	25.86–1.70(1.79–1.70)	20.75–1.60(1.64–1.60)	30.14–2.50(2.75–2.50)
Space group	C121	P43212	P212121	P212121	P41212	P212121	P43212	P6122
Unit cell
a, b, c (Å)	29.42, 4.99,37.17	76.10, 76.10,36.80	41.53, 71.33,79.51	49.10, 59.02,70.00	57.78, 57.78,149.70	53.12, 56.08,64.38	67.60, 67.60,101.36	90.75, 90.75,126.13
α, β, γ (°)	90, 111.55, 90	90, 90, 90	90, 90, 90	90, 90, 90	90, 90, 90	90, 90, 90	90, 90, 90	90, 90, 120
# total reflections1	6,185 (463)	100,693 (282)	14,911 (2,371)	38,699 (348)	51,116 (1,563)	145,833 (402)	302,892 (2,044)	224,846 (314)
# uniquereflections1	3,319 (255)	14,955 (207)	3,884 (614)	10,664 (214)	12,786 (614)	23,542 (281)	46,369 (1,005)	25,029 (173)
CC1/21	0.987 (0.639)	0.901 (0.099)	0.951 (0.255)	0.918 (0.201)	0.848 (0.089)	0.722 (0.028)	0.912 (0.056)	0.847 (0.199)
〈I/σI〉1	2.4 (1.1)	3.7 (1.1)	3.3 (0.8)	3.5 (1.2)	3.5 (2.0)	2.6 (0.4)	3.4 (0.9)	5.6 (3.6)
Completeness1	83.0 (79.4)	96.8 (91.8)	71.9 (71.3)	82.44 (74.9)	93.6 (92.2)	87.5 (56.2)	96.1 (85.8)	97.0 (96.8)
Multiplicity1	1.9 (1.8)	8.6 (6.1)	3.8 (3.9)	4.2 (3.8)	4.4 (3.9)	6.9 (3.1)	8.2 (5.7)	12.3 (12.2)
	Refinement
Rwork1 (%)	20.97 (21.04)	23.95 (32.33)	29.19 (39.51)	22.95 (35.40)	25.13 (34.08)	24.79 (38.72)	22.35 (36.33)	28.99 (34.78)
Rfree1, 2 (%)	22.28 (22.43)	28.42 (37.94)	32.80 (42.03)	26.70 (38.95)	29.45 (38.98)	28.11 (42.37)	25.46 (42.25)	30.96 (36.64)
RSCC	0.84	0.89	0.72	0.85	0.86	0.89	0.91	0.86
# residues	6	129	166	190	207	223	279	316
# protein atoms	53	1,001	1,327	1,481	1,551	1,629	2,029	2,432
# watermolecules	2	87	0	23	18	195	221	21
# ligand atoms	0	3	0	2	0	2	2	13
〈ADP〉 (Å2)
Protein	12.4	13.4	47.8	25.5	20.3	13.9	8.1	4.9
Water	17.3	14.3		19.4	13.3	14.9	13.4	4.2
Ligand		16.7		66.5		19.6	18.9	7.5
R.m.s.d. bonds (Å)	0.012	0.004	0.012	0.002	0.002	0.005	0.004	0.003
R.ms.d. angles (°)	0.770	0.609	1.573	0.496	0.462	0.739	0.663	0.509
Ramachandran(outliers, favored)(%)	0.0, 100	0.0, 97.6	2.5, 89.9	0.0, 96.3	0.0, 95.1	0.0, 96.4	0.4, 96.8	0.0, 94.9

Numbers in parentheses reflect the highest resolution shell for either data collection or refinement.

In all cases the test set comprises approximately 5% of the unique reflections, where possible chosen to match that of the deposited data for the MR search model.

Fragmentation was tested on eight proteins: lysozyme, TGF-βm:TβRII, xylanase, thaumatin, trypsin, proteinase K, thermolysin, and a segment of the protein tau (Figure 2, left column). These proteins readily form crystals that are large—too big for MicroED and SFX experiments using liquid injectors—but without crystal growth optimization sometimes exhibit pathologies. These large crystals were broken apart by mechanical means prior to preparation on cryoEM grids. Micro- or nanometer-sized crystal fragments appeared evenly distributed on the grid, and even when grids were densely populated single-crystal data sets could be collected by using the selective area aperture.

MicroED is inherently well suited to study crystal fragments. Electrons interact strongly with matter so that large crystals are not required to detect high-resolution Bragg reflections[8]. MicroED has been used to determine the structures of the enzymes lysozyme[5,9], proteinase K[10], catalase[11,12], and Ca2+-ATPase[12]. It was also recently used to solve the 1.4 Å resolution structure of the toxic NAC-core of α-synuclein, where diffraction data were collected from crystals smaller than the wavelength of visible light[13]. We previously demonstrated that crystals thinner than ~400 nm are suitable for MicroED and routinely yield atomic resolution information. The highest resolution structures reported so far by using MicroED are from four prion protein fragments determined at 1 Å resolution solved by direct phasing methods[14].

For each sample, data were collected from 2–10 crystal fragments, and data sets merged for completeness and multiplicity (Table 1). Although we have previously published two structures where a single nanocrystal was sufficient for structure determination[9,11] multi-crystal merging is generally preferred. This is partly because the grid on which the crystals are mounted limits the accessible rotation range during data collection, and partly because small, weakly scattering samples require a more intense beam to yield statistically accurate measurements of the Bragg reflections, and may be irreversibly damaged before a complete data set can be obtained. Recording narrow wedges with fewer exposures for each crystal involves a tradeoff between limiting the damage to the sample and maintaining a sufficient dose rate for the signal from high-resolution reflections to be accurately integrated. Because the resolutions obtained by MicroED are comparable or better to what was obtained by X-ray diffraction before sonication, it appears that fragmentation only broke apart the large crystals into small crystal domains but did not otherwise damage the lattice order.

The macromolecular structures in this study were phased using molecular replacement (MR) or by direct methods and refined using electron scattering factors[15]. In all cases, the refined model fits the calculated density well (Figure 2, middle column) with an overall real-space map correlation coefficient ranging from 0.72 to 0.91 (RSCC; Table 1). The resulting simulated-annealing (SA) composite-omit maps (Figure 2, right column) match their respective models very well, indicating that the data are indeed high quality and unbiased. Further, the SA omit maps for lysozyme reveal depressions or holes in the aromatic rings of amino acid side chains, while for proteinase K toroidal densities were observed even for proline residues at the 1.6 Å resolution cutoff. A well-coordinated calcium ion is clearly visible in the omit map for trypsin, as is one of the iodides in the xylanase structure. Individual atoms are visible in the density for the tau peptide, and several residues show positive 2mFo-DFc peaks for hydrogen atoms; for one residue, modeled hydrogens allowed us to unambiguously place the correct sidechain rotamer.

All structures were determined between 1.1–2.9 Å resolution. We note that the R-factors from model refinement are generally higher in MicroED than typical values obtained in X-ray crystallography at similar resolutions. Others and we have already made such observations in previously published studies[5,9,12]. We believe that the main reason is lack of adequate electron scattering factors in the crystallographic refinement software. X-rays are scattered by the electron cloud of an atom while electrons are scattered by the atomic Coulomb potential, which arises from the nucleus as well as its surrounding electrons. Current electron scattering factor tables do not adequately account for this difference and this may contribute to the residual between observed and calculated structure factor amplitudes leading to higher than normal crystallographic and free R-factors. We note that the gap between Rwork and Rfree is small indicating no concerns of overfitting (Supplementary figure 10).

Discussion

Breaking large crystals into well-formed crystal fragments has a long history in crystallography; indeed X-ray diffraction was discovered from broken up pieces of a copper sulfate crystal[16]. Sonication[17] and vortexing[18] have previously been used to prepare crystals of suitable size for XFEL measurements. By breaking up one imperfect large crystal into thousands or even millions of smaller crystallites and recombining a well-diffracting subset during data processing, we have used MicroED to determine the structure of eight different proteins to high resolution. Fragmentation does not increase the effort required for sample screening, and future work on automation is expected to reduce the time for data collection irrespective of how the sample was prepared. Importantly, as demonstrated by the breadth of size and packing of macromolecules whose crystals we investigated, this is a broadly applicable approach where large crystals are available. Fragmentation widens the scope of diffraction methods such as MicroED and SFX to include samples that do not exclusively form tiny crystals but instead form large, imperfect crystals.

Online methods

Large crystals (>500 µm along the longest edge) were grown via hanging-drop vapor diffusion at room temperature using previously established protocols. All enzymes were purchased from Sigma-Aldrich (St. Louis, MO) unless otherwise noted.

Tau peptide (VQIVYK) was dissolved in distilled water and crystals prepared by mixing tau peptide with arachidonic acid and meclocycline sulfosalicylate. Crystals formed with a 1:2 drop peptide to precipitant ratio from 65–70% ethylene glycol in Tris pH 8.5.

Lysozyme (G. gallus) crystals were prepared by equilibrating equal volumes of 50 mg/ml lysozyme in water and 1.7 M sodium chloride, 50 mM sodium acetate pH 4.7.

TGF-βm:TβRII was expressed and purified as described before[20,21]. Crystals were grown from 0.5 µl of protein (20 mg/ml), 0.25 µl mother liquor, and 0.2 µl seed stock in 100 mM HEPES/NaOH pH 7.5 and 45% MPD.

Xylanase from T. reesei (Hampton Research, Aliso Viejo, CA) was dialyzed against 10 mM bicine pH 9.0, 1 mM magnesium sulfate, 1 mM DTT and combined with a precipitant solution containing 0.3 M sodium iodide, 1.2–1.3 M ammonium sulfate, and 100 mM bicine pH 9.0[22].

Thaumatin (T. daniellii) crystals were grown from 2 µL of protein (25 mg/ml in water) and Hampton Research Index Reagent 26 (1.1 M ammonium tartate pH 7.0).

Trypsin (B. taurus) was dissolved (60 mg/ml) in 10 mg/ml benzamidine, 3 mM calcium chloride and equilibrated against 4% (w/v) PEG4000, 0.2 M lithium sulfate, 0.1 M MES pH 6.5 and 15% ethylene glycol.

Proteinase K (E. album) crystals were grown by combining 2 µL of protein solution (50 mg/ml) with 2 µL of precipitant solution (1.0–1.3 M ammonium sulfate, 0.1 M Tris pH 8.0)[10].

Thermolysin from B. thermoproteolyticus (Hampton Research, Aliso Viejo, CA) crystals were prepared by equilibrating a 160 mg/ml solution of thermolysin (45% dimethyl sulfoxide, 50 mM Tris pH 7.5 and 2.5 M cesium chloride) over 0.5 ml of water[23].

Drops containing the crystals were placed in separate microfuge tubes and suspended in crystal mother liquor. A sonicating water bath with electronic control (Elmasonic P30H, Singen, Germany) was set at its lowest power (30% at 37 kHz) for gentle agitation of the crystals in the tube. With the tube sealed, its tip was briefly submerged in the activated water bath for 0.5 s. Alternatively; crystals were fragmented by vigorously pipetting a crystal suspension in mother liquor (trypsin and thaumatin) or vortexed with 0.5 mm disruption glass beads in a 1.5 ml reaction tube for 2 s (TGF-βm:TβRII). A detailed protocol for crystal fragmentation was published online[24].

The solution containing fragmented crystals was then applied to transmission electron microscope (TEM) grids with carbon film support, and plunge-frozen in liquid ethane. Frozen grids were mounted in a Gatan Model 626 cryo-specimen holder and examined using an FEI Tecnai F20 field-emission TEM operated at an accelerating voltage of 200 kV, which corresponds to a de Broglie wavelength of 0.025 Å, and screened for crystals in over-focused diffraction mode. Where a single still shot revealed strong diffraction, data were collected as continuous rotation tilt series[9]. Individual frames were recorded on a TVIPS TemCam-F416 as 4 s exposures while the stage was rotating at 0.09°/s, except for tau peptide, which was rotated at 0.29°/s during 2 s exposures. For the macromolecular samples, the absolute tilt angle was generally <35° (<65° for tau peptide). Data sets collected from each crystal spanned between 30° and 135°, corresponding to a total dose no greater than 1.2–5.5 e−/Å2 at the given rotation rate. The selective area aperture of the TEM was used to limit the area from which data were collected, making it possible to select a single crystal fragment. The virtual detector distance varied between 730 mm and 3.6 m for the different data sets, corresponding to a maximum resolution between 0.95 Å and 2.0 Å in the detector corners at an acceleration voltage of 200 kV. The movie frames recorded in rolling shutter mode on the TVIPS TemCam-F416 CMOS camera were converted to SMV format, while preserving as much as possible of the metadata necessary for subsequent processing[25]. Detailed protocols for data collection were published recently[26].

Sweeps were corrected to account for negative pixel values[10], then indexed and integrated in MOSFLM[27] using its graphical interface iMOSFLM[28], or in XDS[29]. To further probe the scattering power of the crystallites, diffraction patterns were initially processed without imposing any resolution cutoffs other than those entailed by the geometry of the experiment. Except for the tau peptide and TGF-βm:TβRII, the high-resolution limits due to factors other than the area of the detector were instead determined case-by-case to give stable processing during refinement of the atomic model. Multi-crystal scaling and merging was performed in AIMLESS[30]. Structures were phased by molecular replacement using MOLREP[31] from the PDB entries 3j6k, 1ktz, 2dfb, 4ek0, 2ptn, 5i9s, and 2tli for lysozyme, TGF-βm:TβRII, xylanase, thaumatin, trypsin, proteinase K, and thermolysin, respectively, or by direct methods using SHELXT[32]. A free R set comprising approximately 5% of the unique reflections was copied from the deposited data for each search model. The deposited data for the xylanase and trypsin models do not define a free set; instead a new set was chosen using freerflag[33].

Maximum likelihood structure refinement was carried out in phenix.refine[34] using electron scattering factors[15]. Explicit water molecules were automatically modeled by phenix.refine and subsequently manually curated in Coot[35]. SA composite omit maps were computed from the refined models using CNS[36] with electron scattering factors[37]. For these calculations, charged species were manually removed from the phasing models, because they are not included in CNS’s electron scattering library. The SA protocol was defined to exclude ~5% of the structure at a starting temperature of 2500 K. Protocols for data analysis in MicroED were recently published[25].