Real-time investigation of dynamic protein crystallization in living cells

X-ray crystallography requires sufficiently large crystals to obtain structural insights at atomic resolution, routinely obtained in vitro by time-consuming screening. Recently, successful data collection was reported from protein microcrystals grown within living cells using highly brilliant free-electron laser and third-generation synchrotron radiation. Here, we analyzed in vivo crystal growth of firefly luciferase and Green Fluorescent Protein-tagged reovirus μNS by live-cell imaging, showing that dimensions of living cells did not limit crystal size. The crystallization process is highly dynamic and occurs in different cellular compartments. In vivo protein crystallization offers exciting new possibilities for proteins that do not form crystals in vitro.

X-ray crystallography requires sufficiently large crystals to obtain structural insights at atomic resolution, routinely obtained in vitro by time-consuming screening. Recently, successful data collection was reported from protein microcrystals grown within living cells using highly brilliant free-electron laser and thirdgeneration synchrotron radiation. Here, we analyzed in vivo crystal growth of firefly luciferase and Green Fluorescent Protein-tagged reovirus lNS by live-cell imaging, showing that dimensions of living cells did not limit crystal size. The crystallization process is highly dynamic and occurs in different cellular compartments. In vivo protein crystallization offers exciting new possibilities for proteins that do not form crystals in vitro. The three-dimensional structure of a protein determines its function. Consequently, structural insights into proteins at atomic resolution are important to understand the machinery of life or to develop new specifically designed drugs for medical applications. Synchrotron-based x-ray crystallography, the preferred technique for obtaining structural information at high resolution, accounts for almost 90% of the over 107.000 protein structures deposited within the Protein Data Bank (PDB, www.pdb.org) to date. Despite the revolutionary success during the past decades, the correlation between crystal size, diffraction intensity, resolution, and radiation damage usually requires the growth of sufficiently large and well-ordered single crystals and thus emphasizes one of the major bottlenecks of this method. 1,2 The deposited energy triggers a variety of chemical reactions that finally reduce the crystalline order, leading to a quick deterioration of the diffraction data quality as a function of the applied x-ray dose, even under cryogenic conditions. 3,4 Lowering the incident fluency to reduce radiation damage in turn requires the growth of large crystals to yield sufficient diffraction intensities, which is difficult in many cases, particularly for post-translationally modified proteins and membrane proteins. 5 Conventional protein crystallization in artificial crystallization chambers requires the promotion of the nucleation process when a supersaturated sample solution is introduced to a precipitating agent in vitro. If the environmental conditions fit, the growth of large threedimensional crystals may result. This process is affected by a wide range of parameters, generating a vast multidimensional space that must be screened to meet the desired conditions yielding crystals. In modern crystallization approaches, pipetting robot-assisted sparse matrix screening approaches are routinely performed to test the potential crystallization space applying vapor diffusion and batch crystallization techniques. Although the success rate has significantly improved to date, there is still no guarantee for protein crystal formation. 6,7 For more than a century, it has been known that protein crystallization also represents a native, if somewhat rare, process within living cells. 8 Intracellular crystalline states of proteins are associated with diverse cellular functions, as observed for storage proteins in seeds, [9][10][11] insulin in pancreatic b-cells, 12 enzymes in peroxisomes, 13 and bacterial toxins. 14 Moreover, spontaneous crystal formation in vivo may also occur as a result of heterologous gene overexpression. Polyhedrin, a viral protein that usually forms a crystalline coat to protect virions against environmental challenges, 15 assembles into remarkably stable microcrystals within virusinfected insect cells. 16 Exploiting the permanent activation of the polyhedrin promotor, the exchange of the polyhedrin gene by a gene of interest in a baculovirus shuttle vector results in high local protein concentration in the baculovirus-infected insect cell, which is obviously one prerequisite for in vivo crystal formation. Thus, protein microcrystals have been discovered several times by applying the well-established baculovirus-Sf9 insect cell expression system that is frequently used to produce recombinant proteins containing post-translational modifications. 17 Mammalian cells also provide a suitable environment for heterologous protein crystallization, as recently demonstrated. [18][19][20] However, the phenomenon of in vivo crystallization was so far largely perceived as a rare and atypical behavior of proteins, preventing a systematic investigation of the intracellular crystallization process. The size of the crystal grown in vivo was previously considered to be necessarily limited by the cell's outer dimensions, 8,21 but such small crystals would harbor only low diffraction capabilities and high sensitivity to radiation damage. Thus, in vivo grown protein crystals were not considered for structural biology until recently.
This picture has significantly changed with the recent realization of novel radiation sources that produce x-rays of previously inaccessible energy and brilliance. Exploiting the "diffractionbefore-destruction" paradigm 22 by using highly brilliant x-ray free-electron laser (XFEL) pulses of a few femtoseconds duration, serial femtosecond crystallography (SFX) has already been shown to overcome resolution limits imposed by radiation damage at conventional synchrotron sources, allowing serial diffraction data collection from unprecedentedly small protein crystals down to the nanometer regime. 23,24 Tens of thousands of Bragg-diffraction snapshots from individual, randomly "oriented" crystals are recorded at room temperature (RT) and then combined into a dataset applying new data-processing tools [25][26][27] to produce interpretable electron density maps. Since each pulse destroys the individual crystal, samples need to be constantly supplied by injection in vacuum into the pulsed XFEL beam using microjet techniques. 28,29 The feasibility of this concept to elucidate protein structures at high resolution has already been demonstrated on several examples. 23,24,[30][31][32][33][34] One of the important milestones in SFX development, namely, the elucidation of the first new bioinformation by applying this approach, has been obtained using protein crystals that spontaneously grew within living baculovirus-infected Sf9 insect cells during gene over-expression. 30 In addition to the applicability of SFX techniques, we recently showed that comparable structural information on fully glycosylated and natively inhibited Trypanosoma brucei procathepsin B could be obtained from the same in vivo crystals combining a micron-sized synchrotron beam with high-precision diffractometry and a helical line scan approach. 35 Although the resolution of the diffracted synchrotron radiation was slightly reduced, which indicates the need for further methodological and technical improvement. Particularly, optimization of the sample mounting and a more focused X-ray beam are currently in discussion. 35 Both studies clearly illustrated that in vivo crystals can indeed act as suitable targets for structural biology, if the enormous potential of the highly brilliant XFEL and third-generation synchrotron radiation sources is exploited. This significantly supports and extends initial studies reporting the successful structure solution from in vivo-produced polyhedra at microfocus synchrotron beamlines. 16,36,37 The number of in vivo crystallization observations reported as a consequence of heterologous gene expression increased within the past years, 18,20,38 but crystal formation within a living cell still represents a spontaneous event that is detected by chance. A broader application of in vivo grown protein crystals as valuable targets for structural biology requires a detailed and systematic investigation of the intracellular processes involved in crystal formation. If identified, the modification of suitable biological parameters that influence crystal growth could significantly increase the chance of successful protein crystallization within living cells, comparable to multidimensional parameter screens performed in conventional crystallography. Such biological parameters could include, for example, the localization of the protein in a specific cellular compartment as well as the up or down regulation of distinct cellular pathways impacting on protein degradation or trafficking. In this context, we analysed the spontaneous crystal formation of two proteins in Sf9 insect cells, namely, firefly luciferase and a part of the avian reovirus (ARV) nonstructural protein lNS (residues 448-605) fused to Green Fluorescent Protein (GFP) (referred to as GFP-lNS). Applying live-cell imaging techniques, we obtained first real-time insights into the dynamic crystallization process in a living cell, together with new results in terms of cellular localization, stability, and diffraction capabilities of the in vivo grown protein crystals, which are discussed in this study.

A. Recombinant baculovirus generation and gene expression
The construction of the recombinant baculoviruses Bac-luc, which expresses full length luciferase from Photinus pyralis (American firefly), and Bac-GFP-lNS, which expresses GFP-lNS (448-605) from the avian reovirus S1133 strain, have previously been described. 39,40 For baculovirus-driven expression of mCherry labeled organelle marker proteins, Pex3-mCherry, 41 Pex26-mCherry, 41 and mCherry-KDEL 42 were subcloned into pFastBac1 using EcoRI and XbaI restriction enzymes. The construction of recombinant baculoviruses and subsequent generation of high-titer virus stocks were performed according to the Bac-to-Bac manual (Invitrogen). Recombinant high-titer virus stocks (P3) were used to infect 70%-confluent monolayer cultures of Sf9 insect cells (an insect cell line derived from Spodoptera frugiperda (ATCC CRL-1711, Invitrogen) with a multiplicity of infection of 0.1 p.f.u. per cell. Sf9 cells were grown in serumfree EX-CELL 420 insect cell medium (Sigma) supplemented with 100 units/ml penicillin and 100 lg/ml streptomycin at 26 C.

B. Microscopy and live cell imaging
Transiently transfected or baculovirus infected Sf9 cells were plated on glass coverslips and adherent cells were imaged using a laser confocal spinning disk microscope system based on a Nikon Ti Eclipse microscope equipped with the confocal spinning unit CSU-X1 and an Andor iXon þ EM-CCD (electron multiplier-charge coupled device) camera. The microscope was fitted with 40Â, 1.30 NA and 100Â, 1.49 NA objectives. Image acquisition was controlled with Andor Bioimaging software (Andor IQ2.1). Protein crystal growth and dynamics in living Sf9 cells were recorded by time-lapse microscopy using differential interference contrast (DIC) optics over a period of several days or, in the case of GFP-lNS, using a combination of DIC microscopy and GFP fluorescence that was elicited with a 514 nm laser. For 3D images of living cells, z-stacks were taken with images every 300 nm. Three dimensional reconstruction and animation were done in Andor IQ2.1.
Cells infected with recombinant Bac-luc and Bac-GFP-lNS have been subjected to propidium iodide (PI) staining 43 to detect dead cells in the culture and to BODIPY V R 558/568 C 12 (Life Technologies) treatment for staining of internal cellular membranes. Cells were plated on 25 mm glass coverslips at 50% confluence and infected with 5 ll high titer baculoviral P3 stocks per ml of culture medium. At 4 days p.i., cells were mounted on the live cell microscope described above and stained with 500 ng/ml propidium iodide (Invitrogen) or 10 lg/ml BODIPY V R 558/568 C 12 in culture medium at 26 C for 30 min and 10 min, respectively. The marker molecules were excited with 561 nm laser light and imaged with appropriate filter settings. 44 Staining of lysosomes and mitochondria was performed with LysoTracker Deep Red and MitoTracker Deep Red, respectively (both from Life Technologies). Cells were plated and infected as described above. At 4 days p.i., cells were mounted on the live cell microscope and stained with 40 nM LysoTracker or 7 nM MitoTracker in culture medium for 15 min at 26 C. Fluorescence was elicited with 640 nm laser light and imaged with appropriate filter settings.
For co-localization of firefly luciferase crystals with fluorescent organelle marker protein, chimeras cells were plated at 50% confluence and co-infected with identical amounts of recombinant baculovirus stocks expressing luciferase and the marker. Potential co-localization of luciferase with mCherry-KDEL (a marker for the endoplasmic reticulum (ER)) as well as Pex3-mCherry and Pex26-mCherry (peroxisome organelle markers) were analyzed 3 days and 4 days p.i., respectively. The marker proteins were excited with 561 nm laser light and imaged with appropriate filter sets.
In vivo activity of luciferase was examined by addition of the plasma membrane-permeable D-luciferin ethyl ester (Marker Gene Technologies) to cells infected with recombinant baculovirus expressing luciferase. 45 Cells were plated and infected as described above. 4 days p.i. cells were mounted on the microscope stage and imaged with DIC optics and a YFP (Yellow Fluorescent Protein) filter set for luciferin bioluminescence. During ongoing image acquisition, the culture medium was supplemented with D-luciferin ethyl ester at a final concentration of 100 lM.

C. Virus-free gene expression
The gene coding for firefly luciferase (GenBank: KC152483.1 firefly luciferase 1157-2809) was amplified from the pFastBac1 vector template (Invitrogen) by polymerase chain reaction (PCR) using primers 5 0 -GATCGGATCCATGGAAAACATGGAAAACGATGAAAATATTGTAGTT-3 0 (sense) and 5 0 -GATCCTCGAGCATCTTAGCAACTGGTTTCTTAAGGATTTCTCTAAT-3 0 (antisense). After restriction with NotI and BamHI, the PCR product was cloned into a modified pIEX4 expression plasmid (Novagen) lacking the internal start codon. Following transformation into competent DH5a Escherichia coli cells, the amplified plasmid was isolated (QIAprep Spin Miniprep Kit, Qiagen) and sequenced. Transient transfection of a monolayer culture with 0.5 Â 10 6 Sf9 insect cells per ml was performed by lipofection with ESCORT transfection reagent (Sigma). Cells were grown and maintained in serum-free medium at 27 C. Transfection of Sf9 cells with expression constructs encoding chimeric proteins, corresponding DNA cloning, and plasmid purification have essentially been performed as previously described. 46

D. GFP-lNS crystal isolation and stability tests
Baculovirus-infected Sf9 cells were harvested at 96 h p.i. and lysed by incubation in hypotonic buffer (10 mM 2-[4-(2-hydroxyethyl)piperazin-1-yl]ethanesulfonic acid (HEPES) pH 7.9, 10 mM KCl, and 5 mM MgCl 2 ) for 30 min at RT. Crystals were separated from the cell debris by centrifugation for 10 min at 200Â g and collected from the supernatant. Complete protease inhibitor (Roche) was added at recommended concentration to prevent crystal degradation. For dot blot stability analysis, 10 ll of a crystal suspension containing approximately 10 6 crystals per ml was pelleted for 5 min at 14 000Â g and resuspended in 20 ll of an appropriate test buffer. After 1 h, 2 h, and 21 h, respectively, samples were centrifuged for 5 min at 14 000Â g and 5 ll aliquots were taken from the crystal-free supernatant, followed by spotting of 1 ll of each sample on a nitrocellulose membrane. Solubilized protein from the crystals was quantified using a monoclonal anti-GFP antibody (1:1000, Roche), followed by detection with a goat antimouse IgG antibody coupled to horse radish peroxidase (HRP) (1:5000, Santa Cruz). In addition, stability of GFP-lNS crystals was analyzed by visual inspection using a Leica DMIRE2 fluorescence microscope with the crystals resuspended and incubated in appropriate test buffers for 30 min to 21 h.

E. Scanning electron microscopy (SEM)
After cell lysis, 50 ll GFP-lNS crystal suspension was centrifuged for 5 min at 14 000Â g and the hypotonic buffer was exchanged by ultra-pure water to remove excess salt and buffer components. A 2 ll crystal aliquot was spotted onto a polished silicon sample carrier and air dried for 15 min. Scanning electron microscopy was performed using an FEI Helios Nanolap microscope operated at a chamber pressure of 10 À6 bar with an electron acceleration voltage of 5 kV and a current of 0.89 pA.

F. Synchrotron diffraction test of GFP-lNS crystals
A suspension of isolated GFP-lNS crystals in hypotonic buffer with or without a supplement of 30% polyethylene glycol (PEG)400 (v/v) or 30% glycerol (v/v) was pipetted into a glass capillary with a diameter of 0.5 mm. After centrifugation for 15 min at 1000Â g, a dense crystal pellet was clearly visible at the bottom of the capillary. The capillary was sealed with crystal wax and powder diffraction was performed with a microfocus synchrotron beam of 4 Â 5 lm FWHM at the EMBL-operated P14 beamline at PETRAIII storage ring (DESY, Hamburg, Germany) with exposure times between 0.1 and 1 s, a photon flux of 10 12 ph/s, and a photon energy of 10.00 keV.

III. RESULTS AND DISCUSSION
lNS is a protein from pathogenic ARV that causes important economic losses in poultry industry. 47 It forms ordered cytosolic inclusions inside infected cells, which are suggested to constitute together with viral RNA and structural proteins the basic scaffold for so-called viral factories that drive virus replication. 39,40 So far, the structure of lNS is unknown, most likely due to the inherent difficulty in crystallizing proteins with an intrinsic tendency to aggregate, which usually prevents obtaining soluble protein in concentrations suitable for crystallization. Since the efficient auto-assembly of lNS, particularly of the C-terminal part comprising residues 448-605, is not affected by an N-terminal GFP fusion, 39 the linked fluorescence enables direct insights into the spontaneous assembly process, which might result in a crystalline state of lNS suitable for diffraction data collection.
Luciferase catalyzes the reaction of luciferin, Mg-ATP (adeonsine triphosphate), and molecular oxygen, yielding electronically excited oxyluciferin that emits visible light during relaxation to its ground state. 48 The color of this bioluminescence, which is involved in the communication of fireflies, changes by subtle structural differences in luciferase during the emission reaction which has already been elucidated applying x-ray crystallography. 49 In cells of the firefly's lantern organ, luciferase is localized within peroxisomes due to its native C-terminal "SKL" (i.e., the tripeptide serine -lysine -leucine) import signal. 50,51 These highly dynamic organelles contain more than 50 different enzymes, some of them typically assemble into regular crystals that can be both co-crystals of a mixture of enzymes or pure crystals of one enzyme depending on the cell type involved. 8,52 Thus, peroxisomes were already known to act as a suitable cellular compartment for native protein crystallization in vivo. Since light excitation is not required for luciferase bioluminescence, this enzyme is commonly used in biological research as a virtually background-free luminescence reporter. [53][54][55] In this context, we serendipitously observed spontaneous in vivo crystallization of firefly luciferase in insect cells when we used this enzyme as a model to test a new method of active enzyme purification by recruitment into lNS microspheres. 40 recombinant baculovirus that contained the gene encoding full-length firefly luciferase. The crystal morphology was similar to that previously reported after luciferase crystallization using conventional in vitro techniques. 56,57 Crystal growth continued up to $day 5 p.i., regularly exceeding the normal dimensions of the Sf9 cells (which are $15-20 lm in diameter) by many times (Figure 1(a)). The majority of cells gradually lysed thereafter, largely attributable to the ongoing viral replication process rather than to plasma membrane disruption by the growth of the in vivo crystals. This assumption is based on visual inspection of the cell culture as well as live-cell microscopy. Degradation of the plasma membrane as a result of the tension generated by the growing crystal would lead to a sudden rupture of the membrane at one of the crystal tips, which was not observed. In contrast, we detected a slow disintegration of the plasma membrane or cell blebbing followed by plasma membrane rupture around the cell body, which is characteristic for cell lysis due to viral replication. In contrast to the previous observations with Sf9 cells that produced cathepsin B crystals, 38 no free luciferase crystals floating in the medium or attached to cell remnants were detected, indicating significantly reduced crystal stability outside the intact cell compared to cathepsin B. Remarkably, individual luciferase crystals showed a dynamic degradation and re-assembly within the same living cell over the entire growth period observed by time-lapse DIC microscopy (Figure 1(b)). This phenomenon was routinely detected for a small, but significant number of luciferase crystals in the Sf9 cell cultures. The molecular and cellular basis for this unexpected dynamic behaviour of in vivo crystals, which has, to our knowledge, not been reported for crystals grown in living cells so far, will be investigated in further studies. However, a direct response to changes in the environment surrounding the crystals, e.g., in the levels of recombinant gene expression or in the nutrition state of the cell, appear to be reasonable parameters to test further.
During the progress of infection, the number of crystals continuously increased until more than 50% of the cells contained up to five crystals per cell, at day 5 p.i. When several crystals were present in the same cell, they were either individually separated or stuck together to form crystalline superstructures. Luciferase crystals were characterized by extraordinary dimensions up to 200 lm in length, depending on the respective growth state. However, the width of the crystals only varied between 1 and 3 lm (Figure 2(a)). Presenting the largest in vivo crystals reported so far, our study clearly demonstrates that the size of such crystals can indeed exceed the cells outer dimensions by several-fold, in strong contrast to prevalent suggestions in the current literature. 8,21 All previously reported in vivo crystals grown within bacteria, 14 insect cells, 16,17,58 or mammalian cells 18,20 were always restricted by the normal cell volume, without inducing alterations to cell morphology. Only our recently reported cathepsin B study provided a first indication that protein crystal length is not generally limited by the cell diameter. 38

Membrane surrounded crystals originate from peroxisomes
Despite the substantial length of the luciferase in vivo crystals, cell viability is not affected, as revealed by the plasma membrane-impermeable DNA intercalating dye PI (Figure 2(b)). PI stains the DNA in nuclei of cells with compromised plasma membrane integrity brightly red, immediately after membrane disruption. 44 Thus, the absence of PI positive nuclear signals for cells containing large luciferase crystals provided evidence for an exceptional expansion of the plasma membrane of the living cell to cover the entire crystal, as also indicated by DIC microscopy (Figure 2(a)). Applying the lipid droplet stain BODIPY V R 558/568 C 12 with a short incubation time, 59 we detected additional internal membranes within the cell body that covered the crystals next to the plasma membrane, suggesting a growth from a specific cellular compartment (Figure 2(c)). As expected due to the presence of a corresponding native C-terminal peroxisome import signal, 50  peroxisomal marker proteins Pex3 and Pex26 41 confirmed an origin of luciferase in vivo crystals from Sf9 cell peroxisomes. The fluorescence detected clearly surrounded the crystals (Figures 2(d) and S1(b)). 60 In contrast, no co-localization was detected after co-infection with an ER specific marker, mCherry-KDEL 42 (Figure S1(a)). 60 While cathepsin B that contains a native N-terminal signal sequence for rough ER import and the respective in vivo crystals were surrounded by membranes decorated with ribosomes, it was suggested to crystallize within the rough ER of Sf9 cells; 38 our data represent the first proof that crystallization of a recombinant protein can also occur in peroxisomes of living insect cells. Thus, depending on the protein and on the trafficking signals harboured in its primary sequence, different organelles may promote crystal formation. In mammalian Chinese hamster ovary (CHO) cells, successful human IgG crystallization was reported to occur also within the ER lumen after heterologous expression and translocation of the recombinant protein into the ER. 18,19 Since baculovirus replication is localized within the nucleus of the infected insect cells, 61 a direct impact of viral proteins or of the replication process itself on the spontaneous crystallization of luciferase within peroxisomes appeared unlikely. However, to directly test this, we investigated virus-free transfections of Sf9 cells with a pIEX4 vector that contained the fulllength luciferase cDNA. Strikingly, crystals comparable in size and morphology to those obtained in baculovirus-infected Sf9 cells grew within 2 days, i.e., even in the absence of virus ( Figure S2), 60 confirming that baculovirus acts only as a gene shuttle to import the gene of interest into the insect cells. However, most crystal-positive cells contained only single crystals and also the number of cells containing crystals was reduced after direct transfection, likely attributable to decreased DNA transfection efficiency compared to the active viral DNA infection/replication process.

Enzyme activity or cell lysis lead to crystal degradation
Addition of a plasma membrane-permeable luciferin ethyl ester into the surrounding medium resulted in detectable bioluminescence for several minutes in living Sf9 cells ( Figure  S3). 60 In addition to the regular cell body, the cell protrusions induced by luciferase crystals could also be clearly traced. Mg-ATP, which is required for activation of luciferin by adenylation in the first step of the reaction, 48 was here provided by the cellular environment. Yet, the luciferase crystals completely dissolved in the intact cells within 70 s after first observable light emission, as revealed by confocal and DIC microscopy ( Figure S3). 60 Degradation can be attributed to the significant conformational movements associated with the catalytic cycle which apparently destroyed the stabilizing crystal contacts. 49 The fast degradation, however, confirmed the structural and functional integrity of the luciferase protein within the in vivo crystal lattice. Unfortunately, this effect prevents the investigation of the structural dynamics of in vivo crystallized luciferase during enzymatic catalysis applying time-resolved SFX techniques. Online investigation of biomolecules in action represents one of the current challenges in XFEL science. 62,63 Luciferase in vivo crystals did not exhibit significant stability after lysis of the cell membrane. When the plasma membrane is disrupted by the viral replication, crystals dissolved within hours in the surrounding culture medium (Figure 3(a)). Lysis of the cellular membranes by addition of culture medium supplemented with 0.1% (v/v) Triton X-100 lead to much faster dissolution of the crystals (Figure 3(b)). Exchange of cell medium against crystal extraction buffer (50 mM HEPES pH 7, 50 mM KCl, 2.5 mM MgCl 2 , 2.5 mM CaCl 2 , 2% (v/v) sucrose, and 0,1% (v/v) Triton X-100) increased the stability of the luciferase crystals up to 3 h ( Figure  3(c)). This seemingly passive dissolution is in contrast to the active destruction of the crystals by activation of luciferase enzyme activity under addition of luciferin ethyl ester, where the crystals degrade within 70 s (Figure 3(d)). However, further improvement of the so far limited crystal integrity will be required to isolate sufficient amounts of intact luciferase crystals for x-ray diffraction tests and data collection at synchrotron or XFEL sources. In contrast to luciferase, cathepsin B and inosine-5 0 -monophosphate dehydrogenase (IMPDH) formed crystals within living Sf9 cells that were characterized by an extraordinary mechanical and chemical stability, which allowed the structure elucidation of cathepsin B directly from isolated in vivo crystals. 30,38 B. In vivo crystallization of GFP-lNS fusion protein in Sf9 insect cells

Nucleation and crystal growth occur in the cytosol
Production and assembly of GFP-lNS within Sf9 insect cells can be easily monitored after infection with the recombinant baculovirus due to the intrinsic GFP fluorescence In this case, GFP-lNS did not contain any native or artificially added transport signal for a specific cellular compartment. Thus, translation of the fusion protein occurred probably within the cytosol, as observed by increasing diffuse, ubiquitous fluorescence within the entire Sf9 cell starting 24 h after baculovirus infection. Accumulation of GFP molecules into several tiny dots at 77 h p.i. indicated formation of initial crystal nuclei (Figure 4). However, most crystal nuclei exhibited a transient character. Only few crystal nuclei grew into macroscopic visible structures that reached final dimensions of 10 lm in length and 3 lm in width at day 4 p.i., but without exceeding the regular cell diameter. Consequently, the morphology of the Sf9 cells was not affected by GFP-lNS in vivo crystallization, despite being much more efficient than the crystallization of luciferase. Almost all cells in the entire culture were found to be positive for GFP-lNS crystals. Remarkably, neither dynamic crystal growth nor crystal degradation, as observed for luciferase, was detected for these macroscopic GFP-lNS structures.
The mature GFP-lNS assemblies exhibited an elongated, almost needle-shaped morphology with a hexagonal cross section ( Figure 5(b)), but, in contrast to luciferase, bundled needles forming star-shaped superstructures were observed (Figures 5(a) and 5(c)). Crystals were visible in living cells showing no PI staining as well as in cells positively stained with PI ( Figure  5(d)). This result shows that general cell viability was not affected by GFP-lNS crystallization and that these crystals exhibit a much higher intrinsic stability compared to luciferase crystals.
A strong intrinsic tendency for auto-assembly represents the physiological function of the viral lNS domains in the process of viral replication factory formation, natively occurring in the cytosol of reovirus-infected cells. 39,40 As reported above for luciferase, in vivo crystallization of GFP-lNS did not depend on baculovirus infection. Spontaneous assembly formation was previously detected in primary cultures of chicken embryo fibroblasts (CEF) directly transfected with a plasmid expressing GFP-lNS. 39 The mature GFP-lNS assemblies grown in CEF were comparable in size and morphology to the structures observed after baculovirus infection of insect cells.
Another remarkable phenomenon that has not been described before is that different proteins can crystallize in the same cell ( Figure S4). 60 After co-infection with Bac-luc and Bac-GFP-lNS viral stocks, following the previously described protocol for co-infection experiments with recombinant viruses expressing marker proteins, both in vivo crystal species were simultaneously detected in some cells. These crystals showed no obvious alterations in their morphology compared to corresponding crystals individually grown in different cells, indicating no direct impact of GFP-lNS and luciferase on each other during crystal formation. This can be explained by crystallization events occurring in different cellular compartments. As shown in this study, luciferase crystals grow in peroxisomes, while GFP-lNS is suggested to crystallize within the cytosol. Thus, expression of either of the proteins did not change the cellular environment of the other.

Isolated crystals show intrinsic stability
As previously reported for cathepsin B, 30,38 isolation of in vivo grown GFP-lNS crystals was performed by membrane disruption and differential centrifugation without apparent crystal damage. Since cell lysis is associated with extreme changes in the environmental conditions of the in vivo crystals, the identification of buffers conferring optimal stability represents an important requirement to maintain maximum crystalline order after extraction. We applied dot blot techniques to enable large-scale screening of various environmental conditions, including pH value and different concentrations of cryo-protectants. Extracted GFP-lNS crystals were incubated for up to 21 h in the test buffer, followed by centrifugation and highly sensitive immunochemical detection of solubilized protein in the supernatants ( Figure S5(a)). 60 The isolated crystals were stable in high-or low-salt solutions and acidic buffers but became soluble in alkaline buffers above pH 10. Moreover, glycerol and PEG 400, both up to 40%, were identified as suitable cryo-protectants for potential single-crystal diffraction tests applying synchrotron radiation. Supplement of PEG400 increased the stability of GFP-lNS crystals, in general, even preventing the rapid dissolution at pH 10. Our screening experiments additionally revealed that Mg 2þ ions were essential for the stability of the macroscopic structure of GFP-lNS crystals outside of their natural cellular environment. Disruption of the cell membrane in buffers without Mg 2þ ions resulted in immediate diversification of the crystals ( Figure S5(b)). 60 Most noteworthy, GFP-lNS in vivo crystals withstood even dry conditions in vacuum during SEM analysis without any disruption into small fragments, as usually expected for protein crystals, revealing the high intrinsic stability (Figure 6(a)). Although SEM analysis showed roundish edges of the individual assemblies, the perfectly flat surfaces strongly supported a crystalline character of the GFP-lNS assemblies.

Powder diffraction reveals crystalline state
The formation of GFP-lNS assemblies within living cells has previously been reported, but the detected microstructures were not considered to consist of an ordered crystalline lattice so far. 39,40 However, the exceptional stability and the regular crystalline appearance of the GFP-lNS in vivo crystals (Figures 5(b) and 6(a)) determined in this study qualified them for synchrotron diffraction tests. Crystals grown in suspension culture of baculovirus-infected Sf9 cells were extracted 4 days p.i. and centrifuged to remove excess of cell remnants. Powder diffraction tests have been performed at the EMBL-operated microfocus beamline P14 of the PETRA III synchrotron source (DESY, Hamburg, Germany). The detection of typical Debeye-Scherrer rings generally verified the crystallinity of the GFP-lNS assemblies (Figure 6(b)). Amorphous material would have produced a broad and continuous background signal instead of sharp diffraction rings, which are the result of constructive interference of the incident x-rays with the electrons in the well-ordered lattice satisfying Bragg's law. However, the very low maximum resolution of around 30 Å detected for the diffracted x-rays that have not been improved so far by testing other buffer conditions, e.g., including 30% PEG 400 or 30% glycerol as cryoprotectants for potential single-crystal diffraction experiments, prevented the use of GFP-lNS in vivo crystals as suitable targets for x-ray crystallography. Assuming that lNS forms the regular crystal lattice driven by its high intrinsic assembly tendency, 39 the fusion of GFP may have disturbed crystal order by increasing the overall flexibility of the protein, thus interfering with a well-ordered crystal lattice. On the other hand, a direct impact of the extraction process and the non-native buffer conditions which might damage the in vivo crystals can also not be excluded. Consequently, in cellulo diffraction data collection represents a suitable trail to compare the diffraction data quality before and after crystal isolation. Interestingly, substitution of the GFP moiety by the C-terminal tail of lNS significantly changed the morphology of the lNS crystals into a globular shape. 39 The crystalline character and the diffraction properties of globular pure lNS assemblies are currently under investigation.

IV. CONCLUSIONS
Convincing examples, particularly the pioneering structure elucidation of insect virus polyhedra 16,21,36 and of the native B. thuringiensis toxin Cry3A, 34 but also the high-resolution structure determination of cathepsin B, 30,35 as well as the SFX diffraction tests performed with protein crystals formed in mammalian and cockroach cells, 20 already demonstrated the feasibility to implement in vivo crystallography at XFEL and third-generation synchrotron sources as a new strategy for structural biology. However, the general applicability to crystallize any kind of protein within a living cell is so far questionable, even if the number of spontaneous crystallization observations significantly increased, involving various cell types.
In the present study, we report the crystal formation of two new examples, firefly luciferase and avian reovirus lNS, upon high-level heterologous expression in living Sf9 insect cells. Together with polyhedrin from Bombyx mori cytoplasmic polyhedrosis virus, 64 cathepsin B and IMPDH from T. brucei 38 as well as calcineurin, 17 six successful in vivo crystallization events are now described. All of those examples occur during gene over-expression in these cells, indicating a more general phenomenon. Providing first evidence that recombinant proteins can crystallize in peroxisomes of insect cells, we identified the cellular compartment of crystal origin as an initial parameter that could influence successful crystal growth. Most strikingly, our live-cell imaging studies not only provided real-time insights into the spontaneous crystallization process in living cells but also clearly revealed that the normal outer cell dimensions do not limit the crystal size and that the in vivo crystallization process can be highly dynamic, disproving the previous suggestions. These important results represent the starting point for a detailed and systematic investigation of the cellular crystallization mechanisms in living insect cells, which is currently ongoing in our group. The identification of additional cellular parameters next to the compartment will enable a systematic screening of target proteins to optimize the success rate of this novel approach. The dynamics detected for luciferase in vivo crystals suggest an active involvement of cellular processes, e.g., protein and membrane transport mechanisms, in intracellular crystallization, which will be investigated and identified in future studies.
The confirmation that in vivo crystals grown within insect cells are not necessarily characterized by an extraordinary intrinsic stability outside of their natural environment, as we have previously suggested based on our experience with cathepsin B and IMPDH 38 and shown here for GFP-lNS, represents another important result of our study. This shifts diffraction data collection from in vivo crystals directly within the living cells, denoted as the "in cellulo" approach, 21 into focus, which would avoid crystal damage arising from the non-physiological environmental conditions after crystal extraction. In cellulo data collection has already been tested by injecting tiny bacterial cells into XFEL pulses 34 as well as by mounting frozen intact insect cells directly in a synchrotron beam, 21 both providing first evidence that background scattering from other cell components did not obscure the Bragg diffraction from in vivo grown protein crystals. Due to their enormous size, the luciferase crystals reported here represent highly interesting candidates for an in cellulo data collection experiment using synchrotron radiation in the future. However, particularly, the microjet-based sample injection techniques of the SFX approach need optimization, since the comparatively large insect cells require large gas focused liquid jets to avoid clogging of the nozzle, which in turn would result in high background scattering. 20 On the other hand, the systematic investigation of crystal stabilization trials directly in the living cell or the improvement of the crystal extraction procedure for instable crystals will enhance the chance to collect suitable diffraction data from in vivo crystals applying well-established SFX or serial synchrotron crystallography techniques. 30,35,65 In this context, the dot blot approach proposed in this study appears to be highly suitable for large-scale screening of appropriate buffer conditions in terms of securing maximum stability of isolated in vivo crystals.
Comparing the crystal morphology of luciferase and GFP-lNS with that of previously reported T. brucei cathepsin B 30 and IMPDH 38 indicates a preference for needle-like crystals in vivo, which renders conventional synchrotron diffraction data collection difficult due to the small size of the short crystal axis. Moreover, flow alignment of the elongated cathepsin B in vivo crystals during microjet-based sample injection increased the required amount of diffraction data to cover all crystal orientations in SFX experiments. 30 On the other hand, a preferred orientation due to the needle-like shape could help to obtain angular information in SFX experiments, particularly if fixed-target sample delivery that was recently reported for SFX techniques 66 will be improved to allow sample mounting for in a specific, well-defined orientation in the future. Full sampling of reciprocal space will be allowed by the use of a goniometer stage that rotates the normal vector of the fixed target support relative to the incident XFEL pulses, while proof-of-principle experiments indicated dramatically reduced sample consumption compared to flowing jet methods. 66 Successful protein crystallization within living cells would abolish the need for timeconsuming optimization of recombinant protein production and conventional crystallization strategies. Thus, this approach offers exciting new possibilities for proteins that do not form crystals suitable for x-ray diffraction in vitro, if the associated cellular mechanisms will be understood in more detail in the future.