1
High-Resolution Protein Structure Determination by Serial Femtosecond Crystallography
Sébastien Boutet
1
, Lukas Lomb
2,3
, Garth J. Williams
1
, Thomas R.M. Barends
2,3
, Andrew
Aquila
4
, R. Bruce Doak
5
, Uwe Weierstall
5
, Daniel P. DePonte
4
, Jan Steinbrener
2,3
, Robert L.
Shoeman
2,3
, Marc Messerschmidt
1
, Anton Barty
4
, Thomas A. White
4
, Stephan Kassemeyer
2,3
,
Richard A. Kirian
5
, M. Marvin Seibert
1
, Paul A. Montanez
1
, Chris Kenney
6
, Ryan Herbst
6
, Philip
Hart
6
, Jack Pines
6
, Gunther Haller
6
, Sol M. Gruner
7,8
, Hugh T. Philipp
7
, Mark W. Tate
7
,
Marianne Hromalik
9
, Lucas J. Koerner
10
, Niels van Bakel
11
, John Morse
12
, Wilfred Ghonsalves
1
,
David Arnlund
13
, Michael J. Bogan
14
, Carl Caleman
4
, Raimund Fromme
15
, Christina Y.
Hampton
14
, Mark S. Hunter
15
, Linda Johansson
13
, Gergely Katona
13
, Christopher Kupitz
15
,
Mengning Liang
4
, Andrew V. Martin
4
, Karol Nass
16
, Lars Redecke
17
, Francesco Stellato
4
,
Nicusor Timneanu
18
, Dingjie Wang
5
, Nadia A. Zatsepin
5
, Donald Schafer
1
, James Defever
1
,
Richard Neutze
13
, Petra Fromme
15
, John C.H. Spence
5
, Henry N. Chapman
4,16
and Ilme
Schlichting
2,3
1. Linac Coherent Light Source, LCLS, SLAC National Accelerator Laboratory, 2575
Sand Hill Road, Menlo Park, California 94025, USA.
2. Max-Planck-Institut für Medizinische Forschung, Jahnstrasse 29, 69120 Heidelberg,
Germany.
3. Max Planck Advanced Study Group, Center for Free-Electron Laser Science, Notkestrasse 85,
22607 Hamburg, Germany.
4. Center for Free-Electron Laser Science, DESY, Notkestrasse 85, 22607 Hamburg, Germany.
5. Department of Physics, Arizona State University, Tempe, Arizona 85287, USA.
6. Particle Physics and Astrophysics, SLAC National Accelerator Laboratory, 2575
Sand Hill Road, Menlo Park, California 94025, USA.
7. Department of Physics, Laboratory of Atomic and Solid State Physics, Cornell University,
Ithaca, NY 14853, USA.
8. Wilson Laboratory, Cornell University, CHESS, Ithaca, NY 14853, USA.
9. Electrical and Computer Engineering, SUNY Oswego, Oswego, NY 13126, USA.
10. The Johns Hopkins University Applied Physics Laboratory, 11100 Johns Hopkins Road,
Laurel, MD 20723, USA.
11. Nikhef, National Institute for Subatomic Physics, Science Park 105, 1098 XG, Amsterdam,
The Netherlands.
12. European Synchrotron Radiation Facility, 38043 Grenoble Cedex, France.
2
13. Department of Chemistry and Molecular Biology, University of Gothenburg, SE-405 30
Gothenburg, Sweden.
14. PULSE Institute, SLAC National Accelerator Laboratory, 2575 Sand Hill Road, Menlo Park,
California 94025, USA.
15. Department of Chemistry and Biochemistry, Arizona State University, Tempe, Arizona
85287-1604, USA.
16. University of Hamburg, Luruper Chaussee 149, 22761 Hamburg, Germany.
17. Joint Laboratory for Structural Biology of Infection and Inflammation, Institute of
Biochemistry and Molecular Biology, University of Hamburg, and Institute of Biochemistry,
University of Lübeck, at DESY, Hamburg, Germany.
18. Laboratory of Molecular Biophysics, Department of Cell and Molecular Biology, Uppsala
University, Husargatan 3 (Box 596), SE-751 24 Uppsala, Sweden.
Abstract
Structure determination of proteins and other macromolecules has historically required
the growth of high-quality crystals sufficiently large to diffract x-rays efficiently while
withstanding radiation damage. Here we apply serial femtosecond crystallography (SFX)
using an x-ray free-electron laser (XFEL) to obtain high resolution structural information
from microcrystals (<1×1×3 μm
3
) of the well-characterized model protein lysozyme. The
agreement with synchrotron data demonstrates the immediate relevance of SFX for
analyzing the structure of the large group of difficult-to-crystallize molecules.
Elucidating macromolecular structures by x-ray crystallography is an important step in the quest
to understand the chemical mechanisms underlying biological function. Although facilitated
greatly by synchrotron x-ray sources, the method is limited by crystal quality and radiation
damage (1). Crystal size and radiation damage are inherently linked, as reducing radiation
damage requires lowering the incident fluence. This in turn calls for large crystals that yield
sufficient diffraction intensities while reducing the dose to individual molecules in the crystal.
Unfortunately, growing well-ordered large crystals can be difficult in many cases, particularly
for large macromolecular assemblies and membrane proteins. In contrast, micron-sized crystals
are frequently observed. Although diffraction data of small crystals can be collected using micro-
focus synchrotron beamlines, this remains a challenging approach due to the rapid damage
suffered by these small crystals (1).
Serial femtosecond crystallography (SFX) using x-ray free-electron laser (XFEL) radiation is an
emerging method for 3D structure determination using crystals ranging from a few micrometers
to a few hundred nanometers in size and potentially even smaller. This method relies upon x-ray
pulses that are both sufficiently intense to produce high quality diffraction while of short enough
3
duration to terminate before the onset of significant radiation damage (2-4). X-ray pulses of only
70 femtoseconds duration terminate before any chemical damage processes have time to occur,
leaving primarily ionization and X-ray induced thermal motion as the main sources of radiation
damage (2-4). SFX therefore promises to break the correlation between sample size, damage and
resolution in structural biology. In SFX, a liquid microjet is used to introduce fully hydrated
randomly oriented crystals into the single-pulse XFEL beam (5-8), as illustrated in Fig. 1. A
recent low-resolution proof-of-principle demonstration of SFX performed at the Linac Coherent
Light Source (LCLS) (9) using crystals of photosystem I ranging in size from 200 nm to 2 µm
produced interpretable electron density maps (6). Other demonstration experiments using
crystals grown in-vivo (7) as well as in the lipidic sponge phase for membrane proteins (8) were
recently published. However, in all these cases, the x-ray energy of 1.8 keV (6.9 Å) limited the
resolution of the collected data to approximately 8 Å. Data collection to a resolution better than 2
Å became possible with the recent commissioning of the LCLS Coherent X-ray Imaging (CXI)
instrument (10). The CXI instrument provides hard x-ray pulses suitable for high-resolution
crystallography and is equipped with Cornell-SLAC Pixel Array Detectors (CSPADs) consisting
of 64 tiles of 192 × 185 pixels each, arranged as shown in Fig. 1 and Figs. S1 and S2. The
CSPAD supports the 120 Hz readout rate required to measure each x-ray pulse from LCLS (11).
Here we describe SFX experiments performed at CXI analyzing the structure of hen egg white
lysozyme (HEWL) as a model system using microcrystals of approximately 1×1×3 μm
3
(4,11).
HEWL is an extremely well-characterized protein that crystallizes easily. It was the first enzyme
to have its structure determined by x-ray diffraction (12), and has since been thoroughly
characterized to very high resolution (13). Lysozyme has served as a model system for many
investigations, including radiation damage studies. This makes it an ideal system for the
development of the SFX technique. Microcrystals of HEWL in random orientation were exposed
to single 9.4 keV (1.32 Å) x-ray pulses of 5 fs or 40 fs duration focused to 10 μm
2
at the
interaction point (Fig. 1). The average 40 fs pulse energy at the sample was 600 µJ/pulse,
corresponding to an average dose of 33 MGy deposited in each crystal. This dose level
represents the classical limit for damage using cryogenically-cooled crystals (14), . The average
5 fs pulse energy was 53 µJ. The SFX-derived data were compared to low-dose datasets
collected at room temperature using similarly prepared larger crystals (11). This benchmarks the
technique with a well-characterized model system.
We collected approximately 1.5 million individual “snap-shot” diffraction patterns for 40 fs
duration pulses at the LCLS repetition rate of 120 Hz using the CSPAD. About 4.5 % of the
patterns were classified as crystal hits, 18.4 % of which were indexed and integrated with the
CrystFEL software (15) showing excellent statistics to 1.9 Å resolution (see Table 1 and Table
S1). In addition, 2 million diffraction patterns were collected using x-ray pulses of 5 fs duration,
with a 2.0 % hit rate and a 26.3 % indexing rate, yielding 10,575 indexed patterns. The structure,
partially shown in Fig. 2A, was determined by molecular replacement (using PDB entry 1VDS)
and using the 40 fs SFX data. No significant differences were observed in an F
obs
[40 fs] –
F
obs
[synchrotron] difference electron density map (Fig. 2B). The electron density map shows
features that were not part of the model (different conformations of amino acids, water mole-
cules) and show no discernable signs of radiation damage. Also, when the data were phased with
molecular replacement using the turkey lysozyme structure as a search model (PDB code 1LJN),
the differences between the two proteins were immediately obvious from the maps (Fig. S3).
4
Even though the underlying radiation damage processes differ due to the different time scales of
the experiments using an XFEL and a synchrotron/rotating anode (femtoseconds vs.
seconds/hours), no features related to radiation damage are observed in difference maps
calculated between the SFX and the low-dose synchrotron data (Fig. 2B). In addition to local
structural changes, metrics like I/I
0
and the Wilson B-factor are most often used to characterize
global radiation damage in protein crystallography (17). I/I
0
is not applicable to the SFX data.
However, the Wilson-B factors of both SFX data sets show values typical for room temperature
data sets and do not differ significantly from those obtained from synchrotron and rotating anode
data sets collected with different doses, using similarly grown larger crystals kept at room
temperature and fully immersed in solution (11) (Table 1 and S1). The R-factors calculated
between all collected data sets do not show a dose dependent increase (Figure S4). However,
higher R-factors are observed for the SFX data, indicating a systematic difference. This is not
caused by non-convergence of the Monte Carlo integration since scaling the 40 fs and 5 fs data
together does not affect the scaling behavior (not shown). Besides non-isomorphism, possible
explanations for this difference could include suboptimal treatment of weak reflections, the
difficulties associated with processing still diffraction images and other SFX-specific steps in the
method.SFX is an emerging technique, and data processing algorithms, detectors and data
collection methods are under continuous development.
5
Fig. 1: Experimental geometry for serial femtosecond crystallography at the Coherent X-ray
Imaging instrument. Single pulse diffraction patterns from single crystals flowing in a liquid jet
are recorded on a CSPAD at the 120 Hz repetition rate of LCLS. Each pulse was focused at the
interaction point using 9.4 keV x-rays. The sample-to-detector distance (z) was 93 mm.
Fig. 2: A: Final, refined 2mF
obs
-DF
calc
(1.5σ) electron density map (16) of lysozyme at 1.9 Å
resolution calculated from 40 fs pulse data. B: F
obs
[40 fs]-F
obs
[synchrotron] difference Fourier
map, contoured at +3 σ (green) and -3 σ (red). No interpretable features are apparent. The
synchrotron dataset was collected with a radiation dose of 24 kGy.
![Fig. 2: A: Final, refined 2mFobs-DFcalc (1.5σ) electron density map (16) of lysozyme at 1.9 Å resolution calculated from 40 fs pulse data. B: Fobs[40 fs]-Fobs[synchrotron] difference Fourier map, contoured at +3 σ (green) and -3 σ (red). No interpretable features are apparent. The synchrotron dataset was collected with a radiation dose of 24 kGy.](/figures/fig-2-a-final-refined-2mfobs-dfcalc-1-5s-electron-density-dzps6ydx.webp)
