Protein crystallization in living cells

Protein storage

The dense packing of the crystalline lattice can provide a space-efficient way for permanent sequestration or temporary storage of functional proteins. Triggered by environmental changes, protein crystals dissolve, if required, resulting in high local concentrations of active protein. A frequent application of this strategy in nature is nutrient storage. In non-legume plant seeds membrane-surrounded storage organelles include ‘crystalloids’ consisting of an amorphous matrix with embedded protein packed in a lattice structure (Jiang et al., 2000), while oviparous animal species, e.g. Drosophila melanogaster (Papassideri et al., 2007), the mosquito Aedes aegypti (Snigirevskaya et al., 1997), but also frogs (Figure 1A) (Massover, 1971) or bony fish (Lange et al., 1982), crystallize yolk proteins in developing oocytes to provide a constant nutrient supply for the offspring. The viviparous cockroach Diploptera punctata follows a slightly different nourishment strategy. Crystals develop within the embryo midgut soon after ingestion of liquid milk (Figure 1B) (Ingram et al., 1977), a complex mixture of distinct lipocalin-like proteins secreted by the brood sac (Williford et al., 2004), providing a nutrient supply with 3 times the energy of an equivalent mass of dairy milk. Despite the remarkable protein heterogeneity within a single crystal the structural basis of the intracellular crystallizability of these milk proteins was recently elucidated at atomic resolution (Banerjee et al., 2016).

Figure 1:

Examples of naturally occurring protein crystals.

(A) A hexagonal crystal embedded in the mitochondrial cortex of a Rana pipiens oocyte that develops yolk platelet crystals (Republished with permission of Rockefeller University Press from Ward (1962); permission conveyed through Copyright Clearance Center, Inc.). (B) In vivo-grown Lili-Mip crystals from D. punctata enclosed inside the embryo midgut (Banerjee et al., 2016; reproduced with permission of the International Union of Crystallography). (C) MBP-1 crystals enclosed in vesicles in untreated eosinophils, purified from the peritoneal cavity of guinea pigs (Giembycz and Lindsay, 1999). (D) Granulovirus (CpGV) occlusion bodies purified from infected larvae of C. pomonella (Gati et al., 2017). (E) Rectangular platelet crystals of the Cry3A toxin from B. thuringiensis subsp. morrisoni expressed in Bt subsp. israelensis, within intact bacterial cells (right panel) and after isolation (left panel) (Sawaya et al., 2014). (F) A secretory granule of Tetrahymena thermophile, in which the crystalline core has started to expand at both ends of the granule (Turkewitz, 2004; reproduced with permission of John Wiley and Sons). (G) Transverse section through a granulovirus rod in its capsule within an infected larva of Plodia interpunctella (Reprinted from Arnott and Smith (1967) with permission from Elsevier). (H) Isolated baculoviral polyhedra from G25D AcMNPV (Coulibaly et al., 2009). (I) Crystalline alcohol oxidase in peroxisomes in methanol-grown cells of the yeast Hansenula polymorpha (Reprinted by permission from Springer Nature: Veenhuis et al., 1978). (J) Crystalline S-layer coat covering a bacterial cell (Sára and Sleytr, 2000; reproduced with permission from American Society for Microbiology). (K) Catalase-containing crystalline peroxisomal core from 5-days-old sunflower cotyledons (Heinze et al., 2000; reproduced with permission of John Wiley and Sons). (L) Paracrystalline mitochondrial inclusions in normal human skeletal muscle (Reprinted by permission from Springer Nature: Hammersen et al., 1980). (M) Hex-1 crystal in a Woronin body from Neurospora crassa (Reprinted by permission from Springer Nature: Yuan et al., 2003).

Virulence factors represent another important group of proteins that exploit the crystalline state for concentrated long-term storage in the environment, combined with protection from proteolytic degradation. Prominent examples include parasporal toxin crystals produced by ubiquitous Gram-positive spore-forming bacteria, e.g. Bacillus thuringiensis (Bt, Figure 1E) (Sawaya et al., 2014) and Lysinibacillus sphaericus (Colletier et al., 2016), or crystalline fusolin ‘spindles’ produced by insect poxviruses (Perera et al., 2010). Following ingestion by a susceptible insect, the so far stable crystals quickly dissolve in the alkaline gut. The release of the active proteins induces a multi-step intoxication process that finally allows bacterial endospores to germinate in the insect hemolymph (Sanahuja et al., 2011) or disrupts the chitin-rich peritrophic matrix of the insect to facilitate the oral poxyvirus infection (Chiu et al., 2015). As the unique activity of these microbial bioinsecticides (Schnepf et al., 1998) is usually combined with a narrow taxonomic specificity, safety for mammals, and biodegradability, virulence factors are of significant commercial interest in agriculture (Sanahuja et al., 2011). Nowadays, high-resolution structures are available from in cellulo grown crystals of poxivirus fusolin (Chiu et al., 2015), of the Bt Cry3A toxin (Sawaya et al., 2014), and of the binary L. sphaericus toxin BinAB (Colletier et al., 2016), complemented by structural information of the toxic core domains from various mature Bt toxins that were usually obtained after re-crystallization of dissolved native microcrystals (Schnepf et al., 1998; Adalat et al., 2017). However, there is evidence for the function of the associated protoxin state as an enhancer of native crystal packing and stability, provided by the first full-length protoxin structure that was recently elucidated (Evdokimov et al., 2014).

Proteins involved in regulated secretory pathways are frequently stored in a highly condensed amorphous or crystalline state in granules, ready for exocytotic release in response to an external trigger (Arvan and Castle, 1998). The most prominent and best understood example represents insulin, which has been comprehensively reviewed in terms of structure, function and regulated crystallization (Dodson and Steiner, 1998). In eosinophils, the tightly controlled crystalline self-association of a granular protein acts as a regulatory switch (Frigas and Gleich, 1986). Participating in the innate immune response against bacteria and viruses, these cells secrete highly cytotoxic proteins from their granules, e.g. the eosinophil major basic protein (MBP-1) that disrupts the membranes of pathogens (Giembycz and Lindsay, 1999). Safe intra-granular storage in host cells is only enabled by packing MBP-1 into crystalline structures (Figure 1C) that restrain toxicity, which have recently been used for diffraction tests applying SFX techniques (Soragni et al., 2015). Moreover, crystalline secretory granules can trigger a remarkable physical impulse that pushes a cell away from a potential predator, as observed for Paramecium cells belonging to the mainly unicellular eukaryotic protists (Vayssié et al., 2000). Triggered by calcium ion release, the concerted exocytosis of more than 1000 secretory granules docked at the plasma membrane rapidly propels the cells in the opposite direction. The driving force is a roughly seven-fold expansion of the crystalline cargo of the granules (Figure 1F) that is expelled as needle-like structures (Turkewitz, 2004).

Protection and stabilization

Many viruses form paracrystalline arrays in infected host cells that contribute to the ordered intracellular virion assembly, mainly by providing a stable matrix for the spatial arrangement. Although well-analyzed by EM, recent investigations applying SFX and cryo-electron tomography improved the structural insights into these arrays (Duyvesteyn et al., 2018; Ke et al., 2018). If the mechanical properties are further improved by extension of the intermolecular lattice interactions, the stability of a crystalline matrix can be exploited as a protection mechanism. This strategy is adopted by cytoplasmic (CPV) and nuclear polyhedrosis viruses (NPV), granulosis viruses, as well as entomopoxviruses, which encapsulate virions in the late stage of infection by a robust crystalline polyhedrin, granulin, or spheroidin coat, respectively, that survives harsh environmental conditions (Smith, 1976; Coulibaly et al., 2007). As the embedded virus particles remain stable and infectious in soil for many years until ingested by insect larvae, the crystalline state additionally provides an efficient delivery system between hosts by oral-fecal routes (Rohrmann, 1986). Only the specific alkaline target environment in the midgut quickly dissolves the crystalline coat and releases the infectious viral particles (Payne and Mertens, 1983).

Polyhedra produced by two completely unrelated insect viruses, baculoviruses (family Baculoviridae, NPV) and cypoviruses (family Reoviridae, CPV), represent the best investigated crystalline coat so far (Coulibaly et al., 2007, 2009; Ji et al., 2015). A well-ordered and densely packed polyhedrin lattice (Figure 1H) incorporates up to several thousands of uniformly distributed virus particles within holes (Hill et al., 1999). A comprehensive database currently consisting of three different NPV and nine different CPV polyhedrin high-resolution structures (Coulibaly et al., 2007, 2009; Ji et al., 2010, 2015; Axford et al., 2014; Ginn et al., 2015) illustrates that the overall polyhedra architecture is highly conserved during evolution, despite considerable different life cycles of baculo- and cypoviruses as well as significant variability in the sequences and individual three-dimensional (3D) structures of CPV and NPV polyhedrin (Coulibaly et al., 2007, 2009; Ji et al., 2010). A consistent overall architecture of the crystalline granulin coat that encapsulates the Cydia pomonella granulovirus (CpGV, Figure 1D and G) (Gati et al., 2017; Oberthuer et al., 2017), which is closely related to NPV, further supports a preference of the observed geometry for protection of enclosed virions (Anduleit et al., 2005; Ji et al., 2015).

Filamentous fungi and plants developed another strategy to exploit the stiffness and mechanical stability of a crystalline lattice. Here, protein crystals are used to seal pores after damage to the cell, preventing cytoplasmic bleeding (van Bel, 2003; Yuan et al., 2003). In Neurospora crassa, as in numerous ascomycetes species, the occluding function is performed by organelles named Woronin bodies that localize in the vicinity of the septal pore (Markham and Collinge, 1987). The major proteinaceous component, Hex-1, self-assembles after import to form a robust hexagonal crystalline core (Figure 1M) that is required to withstand the high intracellular turgor pressure during pore sealing (Tenney et al., 2000). In plants, a comparable sealing strategy is based on the spindle-shaped crystalloid P-protein (Knoblauch et al., 2001).

In prokaryotes, native crystalline structures are known to be involved in cell wall stabilization, compartmentalization, and DNA protection. Self-assembled two-dimensional (2D) para-crystalline protein lattices, named S-layers, form a porous coating of 5–10 nm thickness around bacteria (Figure 1J) that has been associated with various functions (Sleytr et al., 2014). In archaea, S-layers constitute the only cell wall component in most species, frequently assembling into much thicker structures (Albers and Meyer, 2011). Another function of crystalline 2D layers is to form 100–200 nm sized bacterial micro-compartments, so-called metabolosomes, that provide a specialized reaction environment for catabolic functions (Kerfeld and Erbilgin, 2015). Confinement of sequential enzymatic reactions facilitates substrate channeling and protects the cell from potentially harmful reaction intermediates (Plegaria and Kerfeld, 2017). The third example exploits a native crystallization process to protect bacterial DNA against stress (Luijsterburg et al., 2006). During the stationary phase, the genomic DNA of E. coli forms, together with RNA and certain structural and regulatory proteins, a crystalline complex in the cytoplasm that contains the histone-like Dps (DNA-binding protein of starved cells) protein as the main component (Karas et al., 2015).

Solid state catalysts

Particularly in eukaryotic cells a few active enzymes are known to be natively organized in a crystalline state. Alcohol oxidase forms pure crystals within the peroxisomes of yeast cells fed on methanol (Figure 1I), but also heterogeneous compositions additionally containing urate oxidase and glucose oxidase were observed, depending on the specific cell (Veenhuis et al., 2003). In higher plant peroxisomes, e.g. of the sunflower Helianthus annuus L., catalase is crystallized in so-called ‘cores’ (Figure 1K) (Heinze et al., 2000), while urate oxidase seems to be the sole enzyme in nucleoids of rat liver cell microbodies (Tsukada et al., 1966). Although the reason for assembling active enzymes in a crystalline state is not clear, the crystal lattice might at least not be disadvantageous for the enzymes, as long as the active site is not blocked by crystal contacts. All relevant molecules are sufficiently small to reach the catalytic center by diffusion through channels in the crystal lattice and the specific translocation into peroxisomes diminishes the need for the enzymes to find their substrates by diffusion (Doye and Poon, 2006). Moreover, the confinement to protect the cell from potentially harmful reaction intermediates could spontaneously induce the self-assembly of the peroxisomal enzymes due to the high local concentrations.

Disease-associated in cellulo crystals

If an abnormal crystalline state occurs in the cells of a patient suffering from a disease, it is often difficult to understand whether crystal formation represents the causative event of the disorder, a harmful side-effect, or a harmless byproduct (Doye and Poon, 2006), but at least two diseases are directly caused by abnormal protein crystallization. Congenital forms of cataract (Hejtmancik et al., 1995) are associated with mutations in genes encoding crystallins, the major components of the soluble lens proteins that maintain its transparency (Héon et al., 1999). Without affecting the overall protein fold, these mutations dramatically lower the protein solubility and support lattice formation, triggering the assembly of numerous prismatic crystals that result in lens opacification (Pande et al., 2001). Moreover, abnormal crystallization of a mutant hemoglobin form (HbC) in erythrocytes is known to cause a usually mild and chronic hemolytic anemia, denoted as hemoglobin C disease, which can be life-threatening for patients that are doubly heterozygous for both HbC and the sickle cell hemoglobin (HbS) (Nagel and Lawrence, 1991). The molecular basis of HbC diseases has extensively been studied (Vekilov et al., 2002; Chen et al., 2004; Feeling-Taylor et al., 2004).

For other examples, the impact of the intracellular crystalline state within the disease context is almost unknown. Hexagonal bipyramidal Charcot-Leyden crystals (CLCs) (Charcot and Robin, 1853; Leyden, 1872) assemble by auto-crystallization of galectin-10 in diverse human tissues, body fluids and secretions. If disease-related, the appearance of CLCs is considered as a hallmark of allergic, parasitic, neoplastic and inflammatory disorders (Ackerman et al., 2002). Crystal-storing histiocytosis (CSH) (Lewis et al., 2007) was associated with the accumulation of crystalline inclusions in the cytoplasm of histiocytes (Dogan et al., 2012), whereas rod-shaped hexagonal Reinke crystals are usually located within the cytoplasm of Leydig cells in the human testis (Reinke, 1896). Although the molecular nature of the crystals remains unclear (Kozina et al., 2011), the presence of Reinke crystals is pathognomonic for the identification of Leydig cell tumors and helpful in diagnosis (Jain et al., 2001). In a wide variety of neuromuscular diseases, e.g. in ‘mitochondrial myopathies’, cubic and needle shaped crystalline inclusions, which are suggested to consist of mitochondrial creatine kinase assembled by a so far unknown mechanism (Smeitink et al., 1992), represent a frequent feature in the mitochondrial intermembrane space of skeletal muscle fibers (Hurko et al., 1990). However, these inclusions also occur in healthy human tissue (Figure 1L) (Hammersen et al., 1980).

In cellulo crystallization of recombinant proteins

Different expression systems are well-established in laboratories to produce large quantities of recombinant proteins. The advantages and limitations of the individual systems have recently been discussed (Duszenko et al., 2015). As a consequence of recombinant gene expression, crystal growth within the host cells was already reported in plant cells, in chicken and mammalian cells, as well as in baculovirus-infected insect cells. In bacteria, only the crystallization of recombinant insecticidal B. thuringiensis proteins (Figure 2H) is briefly mentioned (Oeda et al., 1989), while TEM analysis indicates the crystallization of the fungal protein HEX-1 in yeast cells (Würtz et al., 2008).

Figure 2:

Selected recombinant proteins crystallized in living cells.

(A) Crystal of heterodimeric calcineurin with a human regulatory subunit and a catalytic subunit of Neurospora crassa recombinantly produced in Trichoplusia ni HighFive cells (Fan et al., 1996; reproduced with permission of John Wiley and Sons). (B) Triangular crystals of pea legumin A developing in outer endosperm cells of transformed wheat (Republished with permission of American Society of Plant Biologists from Stöger et al. (2001); permission conveyed through Copyright Clearance Center, Inc.). (C) Isolated crystal of TbIMPDH produced in Sf9 insect cells (Nass, 2013). (D) Several needle shaped crystals (up to 20 μm in length) of human IgG2 in CHO hamster cells (Hasegawa et al., 2011; © 2011 The American Society for Biochemistry and Molecular Biology). (E) Crystals of Xpa, an engineered coral fluorescent protein, produced in human HEK293 cells (Reprinted from Tsutsui et al. (2015) with permission from Elsevier). (F) Human PAK4cat (the constitutively active catalytic domain) in complex with the PAK4 inhibitor Inka1 co-crystallized in monkey COS-7 cells (Baskaran et al., 2015). (G) A crystal of firefly luciferase (Photinus pyralis) in a living Sf9 insect cell. The crystal is about 170 μm long (seven times the cell diameter). (H) Bipyramidal crystal of the 130 kDa toxin from Bacillus thuringiensis subsp. aizawai IPL7 recombinantly produced in Escherichia coli JM103 (Oeda et al., 1989; with permission from American Society for Microbiology).

Insect cells

Next to previously mentioned CPV polyhedrin expressed in a heterologous baculovirus system (Coulibaly et al., 2007), crystallization of five entirely different recombinant proteins has been reported in Sf9, Sf21, and HighFive insect cells so far, including: (i) an artificial variant of the heterodimeric phosphatase calcineurin (Figure 2A), forming up to three tetragonal, bipyramidal or cubic shaped crystals in 20–40% of all cells (Fan et al., 1996), (ii) firefly (Photinus pyralis) luciferase, forming up to five needle-shaped crystals (Figures 2G and 3) in up to 50% of all cells (Schönherr et al., 2015), (iii) inosine monophosphate dehydrogenase from T. brucei (TbIMPDH, Figure 2C), forming several needle-shaped crystals in up to 90% of all cells (Koopmann et al., 2012), (iv) glycosylated cathepsin B from T. brucei (TbCatB), forming several needle-shaped crystals (Figure 3) in approximately 70% of all cells (Koopmann et al., 2012; Redecke et al., 2013), and (v) the avian reovirus μNS protein linked to GFP (GFP-μNS), forming green fluorescent needles with a hexagonal cross section (Figure 3) in almost all cells of the culture with comparable morphology to the structures detected in CEF cells (Brandariz-Nuñez et al., 2010; Schönherr et al., 2015).

Figure 3:

Crystallization of recombinant proteins in insect cells after gene overexpression using the baculovirus insect cell system.

Left panel: Crystals of firefly luciferase from Photinus pyralis produced in Trichoplusia ni HighFive insect cells. Cells can harbor dozens of very long needles, which show a crystalline lattice in TEM studies. Middle panel: Crystals of TbCatB produced in Trichoplusia ni HighFive insect cells. Fine needles (0.4–1 μm diameter) do not penetrate through the cellular membranes. Crystals have been used to solve the TbCatB structure (Redecke et al., 2013). Right panel: The reovirus fusion protein GFP-μNS produces needle shaped crystals in Spodoptera frugiperda Sf9 insect cells (Schönherr et al., 2015). TEM images show hexagonal structures with a diameter of up to 5 μm and a crystalline lattice. Asterisks in low magnification TEM images denote crystals that have been magnified in the panels below.

Depending on the native translocation signals harbored in the sequence of the recombinant proteins, crystals occurred in the ER (TbCatB) and in peroxisomes (TbIMPDH, luciferase), next to cytosolic localization (calcineurin, GFP-μNS). For TbIMPDH crystals, an autophagosomal origin of the embedding membrane is additionally proposed (Duszenko et al., 2015), as observed for Xpa crystals in CHO cells (Tsutsui et al., 2015), but needs to be further investigated. As direct insect cell transfection also results in the growth of comparable luciferase crystals, the baculovirus proteins or the viral replication process itself seems not to be essential for intracellular crystallization, but supports an efficient gene shuttle based on the viral spread (Schönherr et al., 2015).

There is an apparent preference for needle-like structures, but the size of the crystals differs depending on the protein. GFP-μNS and calcineurin assemblies do not extend the regular cell diameter. In contrast, TbCatB, TbIMPDH, and particularly luciferase crystals with an extraordinary length of up to 200 μm clearly protrude out of the living cell, expanding the membrane without affecting cell viability. Remarkably, individual luciferase crystals showed a dynamic degradation and re-assembly within the same living cell over the entire growth period that is unique among in cellulo crystals so far, but not understood (Schönherr et al., 2015).

Diffraction of XFEL pulses at TbCatB in cellulo crystals allowed determining its high-resolution structure (Redecke et al., 2013), while diffraction tests at least confirmed the crystalline state of TbIMPDH and GFP-μNS assemblies (Koopmann et al., 2012; Schönherr et al., 2015). These experiments exploited the extraordinary mechanical and chemical stability that distinguishes the mentioned crystals after membrane disruption. Luciferase exhibits a significant exception: cell lysis leads to passive crystal dissolution within 3 h even in optimized buffers (Schönherr et al., 2015), preventing diffraction studies so far.

Data collection with isolated crystals

The majority of in cellulo crystals used for diffraction data collection so far were isolated and at least partially purified from cells. This strategy was motivated by the extraordinary intrinsic stability of the first diffraction targets, the insect virus polyhedra (Coulibaly et al., 2007), and by the concern that diffuse background scattering of residual cell material might bias the specific crystal diffraction. A multiple crystal synchrotron diffraction approach (MSX) appeared to be highly suitable for crystals with volumes between 1 and 3000 μm³. The smaller the volume the more isolated crystals usually mounted on MicroMeshes were used to collect interpretable diffraction data at microfocus synchrotron beamlines under cryo conditions. The partial datasets from the different crystals were merged to generate a final dataset with adequate redundancy. This strategy elucidated 16 structures of CPV and baculovirus polyhedra as well as the structure of entomopoxvirus fusolin and of the cockroach milk protein Lili-Mip (Table 1) (Coulibaly et al., 2007, 2009; Ji et al., 2010, 2015; Axford et al., 2014; Chiu et al., 2015; Ginn et al., 2015; Banerjee et al., 2016; Boudes et al., 2016).

If quick deterioration of the diffraction quality due to radiation damage prevents the collection of several high-resolution diffraction patterns from one crystal, serial synchrotron crystallography (SSX) can be applied. Inspired by the SFX strategy, a nylon loop or a MicroMesh loaded with thousands of isolated microcrystals is raster-scanned using a helical line-scan approach. The subset of collected detector frames that contains diffraction signals from individual crystals in random orientations is selected for further processing into a crystallographic data set. SSX was successfully applied to refine a structural model of TbCatB from isolated crystals with an average volume of 9 μm³ by combination of diffraction data from only 80 crystals (Gati et al., 2014), but resolution was restricted by radiation damage.

The application of SFX at an XFEL outruns resolution limits imposed by crystal size and radiation damage (Spence, 2017). The extremely brilliant and ultra-short X-ray pulses allow room temperature diffraction data collection even from nanometer-sized crystals before destruction by the deposited energy (Chapman et al., 2011). In 2012, the capability of SFX to generate new bioinformation was initially confirmed by elucidation of the native T. brucei pro-cathepsin B structure at 2.1 Å resolution using in cellulo crystals (Redecke et al., 2013). Based on this success, the SFX approach was subsequently applied to determine the structure of polyhedrin from CPV type 17 at 1.75 Å (Ginn et al., 2015) and of granulin from C. pomonella granulovirus (CpGV) at 2 Å and 2.56 Å (Gati et al., 2017; Oberthuer et al., 2017). Moreover, SFX tests revealed the diffraction capacity of isolated human neuraminidase crystals grown in CHO cells and of cockroach milk protein, without recording of complete datasets (Gallat et al., 2014). Recent improvements in data processing strategies and in sample injection techniques significantly reduced the number of diffraction patterns required to solve a protein structure and sample consumption (Ginn et al., 2015; Oberthuer et al., 2017). A novel sample delivery approach using micropatterned silicon chips in combination with a high-speed goniometer increased crystal hit-rates up to 70% (compared to 1–10% for liquid micro-jets), which in turn dramatically reduced the time required to collect a complete dataset of CPV18 polyhedrin to less than 10 min (Roedig et al., 2017). This highlights the capability of current SFX strategies to collect high-resolution data from smallest in cellulo crystals in short time.

In cellulo diffraction data collection

In contrast to the previous examples, firefly luciferase and Xpa crystals grown in insect and mammalian cells, respectively, dissolve quickly after disruption of the cellular membrane (Schönherr et al., 2015; Tsutsui et al., 2015). Thus, subjecting in cellulo crystals to tedious isolation and purification procedures might be detrimental. Recently, high quality diffraction data were directly collected within the living cells, denoted as in cellulo diffraction, consistently without obscuring specific crystal diffraction by background scattering. Resolution limits of the best diffraction images collected by SFX from toxin crystal-containing B. thuringiensis cells were the same as from isolated crystals (Sawaya et al., 2014). This was confirmed by synchrotron diffraction of CPV polyhedra within insect cells mounted on a MicroMesh at cryo-conditions, clearly disproving previous concerns. Differences between the unit-cell parameters and the mosaic spread together with a gain of 0.35 Å in resolution obtained by in cellulo diffraction indicate that even these robust crystals are adversely affected by isolation (Axford et al., 2014; Boudes et al., 2016, 2017). Applying the in cellulo diffraction approach, structures of Xpa (Tsutsui et al., 2015) and of the PAK4/Inka-1 complex (Baskaran et al., 2015) were determined from a single crystal at 2.9 Å and from multiple crystals at 2.94 Å, respectively, while single alcohol/methanol oxidase crystals diffracted XFEL pulses up to 6 Å resolution in intact yeast cells (Jakobi et al., 2016). These examples indicate that in cellulo diffraction is highly promising to overcome stability problems related to crystal isolation, since the living cell seems to maintain a favorable protective environment for the intracellular crystals. However, a major obstacle that currently limits the determination of entirely new structures from in cellulo collected diffraction data is the requirement for experimental phasing. While heavy atom labeling can be achieved by soaking experiments with purified in cellulo crystals, as revealed by successful de novo structure determination (Coulibaly et al., 2007, 2009; Banerjee et al., 2016; Colletier et al., 2016), intracellular labeling requires selenomethionine substitution within the crystallizing protein or the membrane transfer of heavy metal ions. First indications for a successful metal incorporation were recently reported (Boudes et al., 2016).

Native in cellulo crystallization

Following the basic dogma of X-ray crystallography, all well-folded proteins are assumed to have an intrinsic propensity to readily crystallize, which can probably be accomplished by adjusting the environmental conditions in consideration of the protein’s individual biochemical properties (McPherson and Gavira, 2014). Consequently, the crystallization tendency of cellular proteins needs to be tightly regulated to prevent unwanted crystal formation. A balance between protein synthesis and degradation usually maintains local protein concentrations below the solubility limit, rendering nucleation processes thermodynamically unfavorable (Rothman, 2010). If a specific cellular protein should adopt a native crystalline state, its intrinsic crystallization propensity is evolutionary increased, frequently combined with a local concentration by compartmentalization. The former is reflected by readily re-crystallization, reported several-fold for isolated and solubilized native in cellulo crystals (Crowfoot and Fankuchen, 1938; Hall, 1950; Yuan et al., 2003; Banerjee et al., 2016). In turn, premature auto-crystallization of the target proteins has to be prevented, which is accomplished by control of intermolecular protein interactions through precursor proteins, specific binding partners, or through environmental changes within the transport pathway towards their destination compartment.

The intrinsic optimization of natively crystallizing proteins that triggers their assembly into crystalline states usually strengthens the intermolecular interactions within the crystal lattice. Several of these proteins exhibit enormous contact interfaces that cover a significant proportion of the protein surface, as observed for BinAB (Colletier et al., 2016), HEX-1 (Yuan et al., 2003), and for insect virus polyhedra (Ji et al., 2015). In the latter example, an improved interface stabilizes a polyhedrin trimer that assembles into a tetrahedral cluster of four trimers, representing the building block of the lattice. The trimers are further cross-linked via molecular ‘arms’ formed by the protruding helix H1 in CPV polyhedrin or the projecting C-terminal loop in baculoviral polyhedrin, combined with an additional cluster stabilization by intermolecular disulfide bridges (Coulibaly et al., 2007, 2009; Ji et al., 2015). These strong crystal contacts involving the majority of the surface of each trimer are broadly conserved across all known polyhedra, despite many individual differences (Ji et al., 2015). As extended C-terminal molecular arms also interconnect domain-swapped dimers in crystalline fusolin spindles, this cross-linking mechanism has been proposed as a general intrinsic stabilization feature, at least for viral proteins (Chiu et al., 2015). Moreover, the strategy to covalently connect the crystalline matrix by a unique 3D network of disulfide bonds is also observed in fusolin spindles and in B. thuringiensis toxins, where the cysteine-rich C-terminal domain of large (135 kDa) protoxins acts as an enhancer of the native crystallization process (Evdokimov et al., 2014). Together with strong salt bridges (Bietlot et al., 1990), the intermolecular disulfide bonds improve the crystallization tendency even in other host cells (De Cosa et al., 2001). For intracellular crystallization of small 3d-Cry protoxins (65 kDa) which lack a C-terminal ‘crystallization domain’, other factors are known to be involved, including elements termed ‘crystallization’- or ‘helper’-proteins, as recently reviewed (Adalat et al., 2017). Regulated crystallization of insulin in secretory granules depends on a specific hexameric state that proinsulin adopts after ribosomal synthesis and ER import only in the presence of zinc ions. However, the proteolytic cleavage of the C peptide during secretory vesicle formation is required to trigger the self-assembly of the hexamers into crystals (Dodson and Steiner, 1998).

Other cellular control mechanisms to prevent premature crystallization are less understood. Baculoviral polyhedrin is suggested to bind importin as an aggregation inhibitor in the cytoplasm prior to transport into the nucleus (Ji et al., 2015). In CPV infected cells the pool concentration of nucleoside triphosphates that are incorporated into the crystal lattice at specific positions might regulate polyhedra assembly (Coulibaly et al., 2007; Ji et al., 2010, 2015). As shown for granulin, the crystalline capsule nucleates either on the end or at the side of the granulosis virus rod and grows until the complete capsule is formed (Arnott and Smith, 1967), but the termination of the crystallization process remains unclear. Envelopes of the viral protein p10, which is also produced in infected cells, are proposed to guide the coating process around the crystals (Van Oers and Vlak, 1997), but further investigation is required.

For several storage- and protection-associated in cellulo crystals regulated disassembly is required at highly specific conditions to fulfill their native function. An alkaline pH shift frequently acts as a trigger to distort essential crystal contacts. In viral polyhedra and bacterial BinAB toxin, tyrosine and carboxylate-mediated contacts represent the pH switches. Tyrosine acquires a negative charge after deprotonation that destroys associated hydrogen bonds and leads to repulsion between residues, breaking down the crystal lattice (Coulibaly et al., 2007; Ji et al., 2015; Colletier et al., 2016). If the environment additionally provides reducing conditions, as present in the insect gut, disulfide-based cross-links dissociate, releasing monomeric Cry protoxins for further proteolytic maturation (Soberón et al., 2007). In fusolin spindles, the alkaline pH triggers the removal of the molecular arm by proteolysis, which abolishes the stabilizing network (Chiu et al., 2015), whereas the pH shift from 5.5 to 7.4 during serum release dissolves the insulin crystals into functionally active monomers by disrupting the stabilizing zinc coordination (Dunn, 2005). Moreover, changes in yolk platelet pH during Xenopus laevis development correlate with yolk utilization (Fagotto and Maxfield, 1994).

In cellulo crystallization of recombinant proteins

Although the current knowledge is comparatively low, abnormal in cellulo crystallization is supposedly affected by other parameters than previously discussed for native crystal growth. If disease-associated mutations increase the intrinsic crystallization tendency of a cellular protein, the cell must regulate the protein level to maintain local concentrations below the nucleation threshold. This could be achieved by activation of degradation pathways or regulation of gene expression on the transcriptional or translational level. In the case of recombinant proteins, regulation of gene expression is usually not an option, leaving the cell with protein degradation as the sole regulatory pathway. However, recombinant expression systems are generally designed for high yield protein production, leading to high local concentrations if the protein is not secreted. This might act as the sole trigger required for crystal nucleation and subsequent growth, if the environmental conditions within the organelle fit.

The theory of concentration-dependent nucleation is supported by a few examples. Best insights into the crystallization process of recombinant proteins were provided by the constitutive high-level production of a human IgG clone in CHO cells engineered by genome integration and gene multiplication (Hasegawa et al., 2011). Based on the cellular phenotype, IgG synthesis and folding might have exceeded the capacity of the ER export machinery, followed by progressive accumulation of well-folded IgG in the ER lumen, until a threshold concentration was reached to nucleate needle-shaped crystals. If the rate of recombinant protein synthesis is reduced, e.g. during transient expression of human IgG in HEK293 cells, inhibition of the cargo export from the ER is required to induce comparable intracellular crystallization (Hasegawa et al., 2011). Consistent results have been obtained after transient expression of other target genes in mammalian and insect cells. Following direct transfection, nucleation of the Xpa protein was only detected in 0.1% of all transfected HEK293 cells (Tsutsui et al., 2015), and a reduced crystallization efficiency was reported after virus-free transfection of Sf9 insect cells with a pIEX4 vector containing the firefly luciferase gene (Schönherr et al., 2015). However, if a recombinant baculovirus is used for gene transfer, gene multiplication that occurs during virus replication within the infected insect cell is supposed to contribute to local protein concentrations, triggering an efficient growth of multiple crystals per cell in a significant fraction of the cultured cells (Fan et al., 1996; Koopmann et al., 2012; Redecke et al., 2013; Schönherr et al., 2015).

A direct impact of cellular mechanisms on the crystal nucleation process is currently not indicated. In IgG-producing HEK293 cells, all three arms of the unfolded protein response (UPR) were constitutively activated, as expected for cells with a high secretory activity (Hasegawa et al., 2011). This stress response strategy of the ER in higher eukaryotes and yeast is suggested to reduce protein synthesis and to up-regulate protein folding activity (Hetz, 2012), which might fail due to the high protein production rates and the already well-folded recombinant proteins that build the intracellular crystals.