• automation;
  • crystal growth;
  • high-throughput;
  • macromolecular assemblies;
  • methods of crystallization;
  • microgravity;
  • nucleation;
  • nucleic acids;
  • protein crystallization;
  • virus


  1. Top of page
  2. Abstract
  3. Introduction
  4. The time of physiology and chemistry (1840–1934)
  5. The birth of biocrystallogenesis as a science (1934–1990)
  6. Crystallogenesis in the era of technologies and structural genomics (1990–2013)
  7. The status in 2013 and perspectives
  8. Acknowledgements
  9. References
  10. Supporting Information

Protein crystallization has been known since 1840 and can prove to be straightforward but, in most cases, it constitutes a real bottleneck. This stimulated the birth of the biocrystallogenesis field with both ‘practical’ and ‘basic’ science aims. In the early years of biochemistry, crystallization was a tool for the preparation of biological substances. Today, biocrystallogenesis aims to provide efficient methods for crystal fabrication and a means to optimize crystal quality for X-ray crystallography. The historical development of crystallization methods for structural biology occurred first in conjunction with that of biochemical and genetic methods for macromolecule production, then with the development of structure determination methodologies and, recently, with routine access to synchrotron X-ray sources. Previously, the identification of conditions that sustain crystal growth occurred mostly empirically but, in recent decades, this has moved progressively towards more rationality as a result of a deeper understanding of the physical chemistry of protein crystal growth and the use of idea-driven screening and high-throughput procedures. Protein and nucleic acid engineering procedures to facilitate crystallization, as well as crystallization methods in gelled-media or by counter-diffusion, represent recent important achievements, although the underlying concepts are old. The new nanotechnologies have brought a significant improvement in the practice of protein crystallization. Today, the increasing number of crystal structures deposited in the Protein Data Bank could mean that crystallization is no longer a bottleneck. This is not the case, however, because structural biology projects always become more challenging and thereby require adapted methods to enable the growth of the appropriate crystals, notably macromolecular assemblages.




aminoacyl-tRNA synthetase (e.g. AspRS for aspartyl-tRNA synthetase, PheRS for phenylalanyl-tRNA synthetase, etc.)


atomic force microscopy


dynamic light scattering


hen egg-white


International Conference on the Crystallization of Biological Macromolecules


isoelectric focusing


polyacrylamide gel electrophoresis


Protein Data Bank


small-angle neutron scattering


small-angle X-ray scattering


turkey egg-white


tobacco mosaic virus


  1. Top of page
  2. Abstract
  3. Introduction
  4. The time of physiology and chemistry (1840–1934)
  5. The birth of biocrystallogenesis as a science (1934–1990)
  6. Crystallogenesis in the era of technologies and structural genomics (1990–2013)
  7. The status in 2013 and perspectives
  8. Acknowledgements
  9. References
  10. Supporting Information

The art of crystallization dates back to antiquity and, for a long time, primarily comprised the growth of salt crystals by evaporation procedures. Protein crystallization is much more recent and appeared in the first half of 19th Century, with an initial publication in 1840 on the observation of crystallites in blood preparations [1], which in fact were haemoglobin crystals. Over the years, the diversity of crystallized proteins has expanded, although crystallization often occurred by chance and using empirical procedures. For approximately one century, crystallization was used as a means of protein purification and characterization by biochemists and physiologists. The situation changed when X-ray crystallography entered biology in 1934 after the first X-ray photograph of a protein crystal was taken [2]. Improvements in crystallization procedures and the fabrication of crystals suitable for structure determination arose in parallel with advances in X-ray crystallographic methods and the ambition of structural biologists who were seeking to image the macromolecular components of living organisms. This became possible as a result of interdisciplinary efforts merging biochemistry/molecular biology, chemistry, physics and engineering, which gradually transformed the field of protein crystallization into a scientific discipline of its own. I have named this discipline ‘crystallogenesis’ [3], where the aim is to understand and control crystal growth and quality; note that a German version, ‘Krystallogenese’, was already proposed in the 19th Century by different individuals, such as Preyer [4]. The literature on biocrystallogenesis is manifold. The present review restricts itself to a few introductory references on historical [3, 5-7] as well as on methodological and physicochemical [8-16] aspects and to a selection of most significant research articles and focused reviews. More citations on facts listed in Tables and Figures are provided in Data S1 to S10. Additional bibliographic sources, particularly books, reviews and International Conference on the Crystallization of Biological Macromolecules (ICCBM) Proceedings, are given in Data S11.

The present review is divided into three sections describing how biocrystallogenesis emerged and became a mature field, as well as how it became seminal for modern structural biology. They cover: (a) the period of physiological and colloidal chemistry before the birth of protein X-ray crystallography; (b) the early years of structural biology when conventional methods of protein crystallization were established; and (c) the years of more recent technologies and structural genomics. The conclusion outlines perspectives and sketches a few applications beyond the field of structural biology (e.g. in medicine).

The time of physiology and chemistry (1840–1934)

  1. Top of page
  2. Abstract
  3. Introduction
  4. The time of physiology and chemistry (1840–1934)
  5. The birth of biocrystallogenesis as a science (1934–1990)
  6. Crystallogenesis in the era of technologies and structural genomics (1990–2013)
  7. The status in 2013 and perspectives
  8. Acknowledgements
  9. References
  10. Supporting Information

In the 19th and early 20th Centuries, knowledge on proteins was elusive and the name ‘protein’ (coined by Berzelius in 1838) was not of universal use in biology and chemistry. Terms such as ‘Proteid/Eiweisskörper’ substances, ‘albumineous’ material or ‘colloids’ were often employed for these mysterious substances. However, during this period, a few visionary physiologists, chemists and physico-chemists established the cornerstones of modern biology, notably structural biology, when they worked out protocols leading to the production of crystalline proteins. The basic methods of protein crystallization were established and the essential physico-chemical properties of proteins discovered.

Crystallinity of haemoglobin and plant globulins

In 1840, Friedrich Ludwig Hünefeld published a book entitled Der Chemismus in der thierischen Organisation (Chemical Properties in the Animal Organization) in which he reported (p. 160 and 161) how he accidentally discovered the formation of crystalline material in samples of earthworm blood held under two glass slides and occasionally observed small plate-like crystals in desiccated swine or human blood samples [1]. These were crystals of ‘haemoglobin’, a name coined 1864 by Felix Hoppe-Seyler for the ‘colorant substance of blood’ [17]. In the following years, and likely even before Hünefeld, many scientists observed haemoglobin crystals when examining various animal tissues or animal blood (e.g. Julius Budge, Otto Funke, Albert von Kölliker, Karl G Lehmann, Franz Leydig and Karl Reichert) but except Funke did not investigate further the properties of these crystals [4].

In 1855, Theodor Hartig discovered a second family of crystalline proteids in the gluten flour ‘Klebermehl’ from the Bertholletia excelsa Brazil nut [18]. Soon, ‘crystalloids’, as they were named, of globulins were described by several authors in extracts of other plant seeds (e.g. from Avena, Camelia, Crocus, Croton and Ricinus), notably by Heinrich Ritthausen [19] and mostly by Thomas B. Osborne [20] who knew and extended Ritthausen's work. By 1889, when Osborne started his thorough biochemical work on plant globulins, his main interest was to prepare pure specimens of globulins by employing all of the available methods at the time (particularly crystallization) to ensure homogeneity of the preparations. As a result, he obtained crystals of several globulins (two examples are provided in Fig. 1A) and assigned them specific designations; for example, ‘excelsin’ for the globulin from Brazil nut (an allergen presently known as the Ber e 2 protein) [21] and ‘edestin’ or ‘avenalin’ for those from hemp seeds or oat kernels [20]. In 1907, Osborne published a monograph summarizing his investigations (revised in 1924) in which he described procedures to obtain crystals that are based, amongst others, on protein extractions from warm salt solutions (40–60 °C) followed by slow cooling to room temperature; for further details, see the online version of the original 1924 publication [22].


Figure 1. Animal haemoglobins and plant globulins, comprising the first animal and vegetable proteins that were crystallized. (A) Crystals of B. excelsa exelsin from the Brazil nut (left) and of Avena sativa avenalin from oat kernels (right) [20]. (B, C) Haemoglobin crystals of the Tasmanian wolf: (B) photographs of α-oxyhaemoglobin showing groups of plates in parallel growth (left) and of β-oxyhaemoglobin showing small dodecahedral crystals (right); (C) schematized drawing of the above crystals emphasizing their prismatic (left) and dodecahedral (right) habits [25] .

Download figure to PowerPoint

Deliberate protein crystallizations in the 19th and early 20th Centuries

After the seminal findings by Hünefeld and Hartig, many other physiological chemists and botanists tried to deliberately produce crystals of haemoglobin and plant globulins using more controlled protocols [6]. Thus, in 1851, Funke described how to grow human haemoglobin crystals by successively diluting red blood cells with solvents such as pure water, alcohol or ether, followed by slow evaporation of the solvent from the protein solution [23]. This was the first use of organic solvents in protein crystallization. In 1871, the English-born physiologist William T. Preyer, Professor at University of Jena, published a book entitled Die Blutkrystalle (The Crystals of Blood), reviewing the features of haemoglobin crystals from ~ 50 species of mammals, birds, reptiles and fishes [4]. Franz Hofmeister entered the theater of crystal science in 1890 when he crystallized hen egg-white (HEW) albumin [24].

The interest in haemoglobin crystals did not decline in the 20th Century and was first highlighted in 1909 when the physiologist Edward T. Reichert, together with the mineralogist Amos P. Brown, published an impressive treatise on the preparation, physiology and geometrical characterization of haemoglobin crystals from several hundreds animals, including extinct species such as the Tasmanian wolf (Thylacyanus cynocephalus) [25] (Fig. 1B,C). The crystallization of other proteins was also actively pursued in the first half of the 20th Century. As was common practice in chemistry, crystallization became a powerful step in purification protocols. Examples are the crystallization of animal and plant globulins (e.g. various serum albumins and canavalin from jack beans), the crystallization of a plant lectin (concanavalin A), the crystallization of several enzymes (carboxypeptidase, catalase, chymotrypsin, ribonuclease, pepsin, trypsin, urease, etc.), the crystallization of the diphtheria toxin, and the crystallization of the polypetidic hormone insulin [6]. All of the investigators noted the importance of salts, organic solvents, pH and/or temperature for crystallization. Progressively, they took advantage of the new ideas of Hofmeister on salt effects (especially the ‘salting in’ and ‘salting out’ phenomena) to reach supersaturation and discovered the crucial role of metal ions in protein crystallization, notably for insulin crystallization where Zn2+ ions are indispensable [26].

Precursors that impacted upon the field of protein crystallization

The biographical notes of pioneers and the highlights of their achievements in crystal science are summarized in Table 1. Besides Hünefeld, Funke and Hartig who opened the field, the inspired physico-chemical contributions of Hofmeister and Ostwald deserve particular attention, although they were of indirect influence on early crystallization investigations. Hofmeister was the first individual to systematically study the effects of salts on protein stability and solubility [27]. He is the father of what is presently known as the Hofmeister lyotropic salt series, which ranks the relative influence of ions on the physical behaviour of proteins [28]. These salt effects (with NH4+ having the strongest effect with respect to decreasing solubility) turned out to be critical for understanding protein crystallization [29, 30]. Ostwald established the rules for time dependent phase changes in chemical mixtures (solid–liquid transitions) and discovered the phenomenon of ripening [31] that has found recent applications in macromolecular crystallization [32].

Table 1. Early pioneers that impacted upon the emerging science of crystallogenesis. For references, see text and Data S1.
PioneersHighlightsBiographical notes
  1. a

    His son Wolfgang Ostwald (1883–1943) was the initiator of colloid chemistry and biochemistry.

Friedrich L. Hünefeld1840: First observation of crystals in blood samples (haemoglobin)German MD and chemist, b.1799 Müncheberg – †1882 Greifswald. Was active at Greifswald University (Professor of Chemistry and Mineralogy). Was with Berzelius in 1827
Otto Funke1851: First deliberate crystallization of haemoglobin (Blutfarbstoffes) by evaporationGerman MD and chemist, b.1828 Chemnitz – †1879 Freiburg. Was active at Leipzig, then Freiburg University (Professor of Medicine, then Physiological Chemistry)
Theodor Hartig1855: Crystalline particles from extracts of the Brazil nut (a storage protein known as B. excelsa excelsin)German forestry biologist and botanist, b.1805 Dillenburg – †1880 Brauschweig. Was active at the German Forestry Organization
Franz Hofmeister1888: Different salts can be placed in a regular order with respect to their salting-out effect on proteins (ranking now known as the ‘Hofmeister series’ or ‘lyotropic series’) 1890: First crystals of HEW albuminBohemian-German MD, physiologist, chemist and pharmacologist, b.1850 Prague – †1922 Würzburg. Worked at Prague University until 1896 (Professor of Pharmacology), succeeded Hoppe-Seyler 1896 at Strasbourg University, left for Würzburg in 1919
Wilhelm Ostwalda1897: Phenomenon of ripening describing the change of an inhomogeneous structure over time (called Ostwald ripening). Applies to proteins, where large crystals can grow at the expense of small onesBaltic-German physical chemist and philosopher, b.1853 Riga – †1932 Grossbothen. Educated in Tartu; worked at Riga (1881–87) then Leipzig University (Professor of Chemistry and Philosophy). Nobel Prize in Chemistry in 1909
Edward T. Reichert and Amos P. Brown1909: Publication of an impressive opus on the solubility, crystallization and crystal characterization (shape, angles, etc.) of haemoglobins from ~ 100 mammalian species and a few Batrachia, birds, fishes and reptilesAmerican MD from the Medical Department of Pennsylviana University, b.1855 – †1931. Educated in Berlin, Leipzig and Geneva; worked mainly at University of Pennsylvania (Professor of Physiology)
American mineralogist, b.1864 Germantown – †1918 Philadelphia. Was head Professor of Department of Mineralogy and Geology, University of Pennsylvania, Philadelphia, PA
James B. Sumner1919: Crystals of Canavalis ensiformis concanavalin A & B (jack bean) 1926: First crystallization of an enzyme, urease from jack bean. Despite skepticism he claimed that the crystalline enzyme is a proteinAmerican chemist and biochemist, b.1887 Canton, MA – †1955 Buffalo, NY. Graduated from Harvard University; most research at Cornell University, Ithaca, NY (Professor of Biochemistry). Nobel Prize in Chemistry in 1946
John H. Northrop1930: Pepsin in crystalline form. Northrop was visionary in realizing that a crystalline form of a protein is not in itself a criterion of purityAmerican biochemist, b.1891 Yonkers, NY – †1987 Wickenburg, AZ. Main work at Rockfeller Institute in New York, NY, and Princeton, NJ. Nobel Prize in Chemistry in 1946
1931–33: Crystallization of trypsin and chymotrypsin
Wendel M. Stanley1930–40: Use of chemical methods, including crystallization, for isolation of active substances from viruses that are harmful to plants. In 1935, isolated tobacco mosaic virus in crystalline formAmerican biochemist and virologist, b.1891 Ridgeville, IN – †1987 Salamanca, Spain. Main work at the Rockfeller Institute in Princeton, NJ; after 1948 at University of California, Berkeley, CA (Professor of Biochemistry). Nobel Prize in Chemistry in 1946
Arda A. Green1931–32: Seminal papers on the solubility of horse haemoglobin as a function of pH, ionic strength and temperatureAmerican protein chemist and biochemist, b.1899 Prospect, PA – †1958 Baltimore. Many prominent scientists worked under A. Green (e.g. Krebs, 1992 Nobel Prize) or were associated with her (e.g. the Cori's, 1947 Nobel Prize). Posthumous Garvan Medal awarded to notable women chemists
1956: Crystallization of luciferase (her last contribution)
John D. Bernal and1934: First X-ray diffraction pattern of a protein crystal (pepsin)British crystallographer, b.1901 Nenagh, Ireland – †1971 London. Mentor of D. Hodgkin at Cambridge University (Professor of Physics); 1937: moved to Birkbeck College, London (Professor of Crystallography)
Dorothy (Crowfoot) Hodgkin British chemist and protein crystallographer, b.1910 Cairo – †1994 Ilmington. Educated in Oxford; was in Cambridge with J. Bernal and held a post at Sommerville College, Oxford, until 1977. Nobel Prize in Chemistry in 1964

Reichert and Brown aimed to correlate the classification of animal species with their evolution on the basis of the morphology of their haemoglobin crystals [25]. Today, this appears naive but, by 1909, the idea underlying their work was in some way visionary because, in present biology, evolution is accounted for by protein sequences and three-dimensional structures. In a more crystallographic perspective, they were the first individuals to thoroughly describe polymorphism in protein crystals, which is now amply demonstrated.

The motivation of James B. Sumner was different. In 1917, when he was at Cornell University and had heavy teaching obligations, he decided to accomplish something of real importance during his spare time. This was the risky project of purifying an enzyme. Fortunately, using urease from the jack bean, he opted for a good experimental model. Two years later, he obtained crystals of the lectin concanavalin, which is abundant in the jack bean. It took him an additional 7 years to find the appropriate recipe to prepare crystalline urease. The clue to success was the extraction of the enzyme from the protein bulk with 30% alcohol [33]. However, his 1926 paper, in which he reported that solutions of dissolved urease crystals possess ‘to an extraordinary degree the ability to decompose urea into ammonium carbonate’ [33] generated skepticism and his conclusion was rejected by the renowned German organic chemist Willstätter (1915 Nobel Prize in Chemistry), who was convinced that the catalytic activity of enzymes is a result of organic compounds copurified or adsorbed on carrier proteins [5]. This forced Sumner to provide stronger arguments and stimulated John H. Northrop to study the crystallization of swine pepsin for which he had strong biochemical evidence of its protein nature [34]. Despite intensive efforts, no putative catalytic entity could be separated from either urease or pepsin. The controversy was resolved when Northrop developed better quantitative tools to purify, characterize and crystallize proteins, and thereby generalized the concept of catalytic proteins to pepsin, trypsin and chymotrypsin [35]. By 1937, Sumner closed the debate with a decisive publication on catalase from beef liver showing that its catalytic activity requires both the protein and an iron porphyrin group [36]. Both Sumner and Northrop received the Nobel Prize in Chemistry in 1946 for these biochemistry-focused contributions [5]. They shared the Nobel Prize with Wendel M. Stanley, who was the first to have prepared a crystalline virus [37], although he did not immediately realize the implications of his finding as he was on a quest to prepare protein constituents of tobacco mosaic virus (TMV) [38].

More influential from the viewpoint of crystal science was Arda A. Green with her seminal papers on the physical chemistry of proteins completed in a continuation of the early observations of Hofmeister on the solubility of horse carboxy- and oxyhaemoglobin as a function of the concentration of various salts, pH and temperature [39, 40]. Accordingly, she deduced an empirical relationship between protein solubility and ionic strength

log S = β – Ksμ

where S is solubility and μ is ionic strength, Ks is the salting-out constant considered to be independent of pH and temeperature, and β is a protein-, pH- and temperature-dependent constant). Interestingly, she noted decreasing values of Ks correlated with the ranking of the salts in the Hofmeister series. Arda A. Green was active in many other domains of protein science while working with the most famous American biochemists and this explains why her work on protein solubility did not receive the recognition that it deserves, although it did influence the two Cori's and Krebs, and all three were awarded a Nobel Prize, and later impacted decisively upon the whole field of protein crystallization.

Considerations on protein crystallization in the epoch of physiology and chemistry

In this epoch, crystallization was a tool for protein purification and was instrumental to demonstrate that the catalytic activity of enzyme resides within the protein itself. In the 19th Century, most crystalline proteins were of plant origin and only a few animal proteins (haemoglobins and albumins) were characterized as pure substances. Crystals were obtained in the microscale range by desiccation/evaporation procedures of crude biological materials, mainly from extracts treated by water, alcohol, hot acetic acid or salts to solubilize their ‘albumineous’ entities. Scaling-up procedures represented a challenge that was first tackled by Preyer with haemoglobin [4] and pursued by the biochemists in the early 20th Century, who significantly enriched the repertoire of crystalline proteins. All of these proteins were easily available and had rather robust structures, a feature not known at the time. In retrospect, one can wonder why the early investigators were not intrigued by the fact that proteins considered as colloidal substances with an elusive structure can be crystallized. Being physiologists and biochemists, it is fortunate that they were not refrained by the rules of classical crystallography, which claim that crystals are formed by strictly identical entities, although, today, it is well established that macromolecular crystals can encompass proteins with disordered domains.

The paradigm change in the field occurred in 1934 when John Desmond Bernal and Dorothy Crowfoot (Hodgkin), two prominent figures in British science, reported the first diffraction pattern of a protein crystal [2]. This closed the epoch of chemistry and physiology in biocrystallogenesis and marked the beginning of structural biology.

The birth of biocrystallogenesis as a science (1934–1990)

  1. Top of page
  2. Abstract
  3. Introduction
  4. The time of physiology and chemistry (1840–1934)
  5. The birth of biocrystallogenesis as a science (1934–1990)
  6. Crystallogenesis in the era of technologies and structural genomics (1990–2013)
  7. The status in 2013 and perspectives
  8. Acknowledgements
  9. References
  10. Supporting Information

Growing crystals was not the major concern for the pioneers of structural biology who were busy establishing methods for structure determination. They used proteins available in large amounts and easy to crystallize with the bulk methods worked out by the biochemists (see above). Once the first protein structures were solved in the 1950/60s (Table 2), researchers became more ambitious and enrolled in objective-focused projects. The supply of interesting proteins became a limiting factor and the fabrication of crystals turned out to be a major bottleneck. Studies aiming to understand the functioning of enzymes and the mechanisms of protein synthesis were probably the first emblematic objective-focused projects that stimulated worldwide interdisciplinary efforts to overcome this bottleneck (e.g. for understanding tRNA biology) [7]). Furthermore, deciphering protein synthesis enlarged the problem of protein crystallization to nucleic acids [41] and nucleoprotein complexes [42]. Crystallization of membrane proteins was the other great challenge [43].

Table 2. Landmarks in macromolecule crystallization leading to three-dimensional (3D) structures. For references, see text and Data S2
Macromolecular class SubclassExample of crystallized entities Name, origin, (year)a3D structure (year)a and PDB codeb
  1. a

    Prime publication(s) of crystallization or structure.

  2. b

    PBD codes can correspond to a refined structure deposited after the prime publication.

  3. c

    Crystal structure not yet in PDB, although NMR structure has been solved.

  4. d

    PBD codes can also correspond to a structure of different taxonomic origin than the first crystallized entity.

  5. e

    Structure at 1.2 Å resolution solved from crystals grown under microgravity in gel with data collected at room temperature.

  6. f

    Example of a structure of an A-DNA decamer at 2 Å resolution.

  7. g

    Fibre-diffraction structure at 3.5 Å resolution.

  8. h

    First X-ray structure of a ribosome (i.e. that of Thermus thermophilus) at moderate resolution (5.5 Å).

GlobinsHaemoglobin, human (1840)(1963) – 4hhb
PhytoglobulinsExcelsin, Brazil nut (1855)(2007) – 2lvfc
EnzymesHEW lysozyme (1890)(1965) – 1lyz
Urease, jack been (1926)(2012) –4h9m
Pepsin, pig (1929)(1990) – 1pep
Catalase, beef liver (1937)(1985) – 7cat
HormonesInsulin, rabbit (1926)(1969) – 4insd
ToxinsErabutoxin, sea snake (1971)(1989) – 5ebx
AntibodiesIntact IgG, human (1969)(1973) – 7fab
Membrane proteinsPorin, Escherichia coli (1980)(1995) – 1opf
Sweet tasting proteinsThaumatin, Thaumatococcus daniellii (1975)(2002) – 1kwne
Nucleic acids
tRNAstRNAPhe, Saccharomyces cerevisiae (1968)(1974) – 1tn2
DNA fragmentsSynthetic DNA duplexes (1988)(1989) – 2d13f
Supramolecular assemblies
VirusesPlant virus, TMV (1935)(1986) – 1vtmg
Enzyme:RNA complexesAspRS:tRNAasp, S. cerevisiae (1980)(1991) – 1asy
Membrane embedded assembliesPhotosynthetic reaction center Rhodopseudomonas viridis (1982)(1986) – 1prc
Protein:DNA complexesNucleosome, Xenopus laevis (1984)(1998) – 1aoi
Ribosomes70S, Bacillus stearothermophilus (1980)(2001) – 1giyh

Widening and exploring the crystallization parameter space

Crystallization processes are multiparametric phenomena and therefore the primordial duty of experimenters is to properly choose the parameters leading to best crystal growth. In the field of protein crystallization, early investigators were not always aware of this fact and often obtained crystals by chance, although it was soon noted that some factors were of importance, such as the solubility of the protein, the type of salts used to induce supersaturation, the temperature, the need for metal ions, and the source and amount of the protein. Nevertheless, many crystallographers considered protein crystallization as an art where magic skills are essential for success. This idea remained popular for some time, especially because the amount of material available for crystallization purposes was often limited. This prevented systematic studies aiming to understand the global or specific effects brought by the known parameters affecting protein crystallization [10] (Table 3).

Table 3. Parameters affecting protein crystallization. For references, see text and Data S3.
Main parametersaCommentsa
  1. a

    For details, see text and crystallization databases (e.g. Most of the crystallization parameters were known in 1990.

  2. b

    Only a few crystallization parameters were explicitly characterized before 1934.

  3. c

    Note the few additional parameters that were identified after 1990.

  4. d

    Jeffamines are polyetheramines based on either propylene oxide (PO), ethylene oxide (EO) or mixed PO/EO backbones with terminal amino groups.

  5. e

    Poloxamers are amphiphilic non-ionic multiblock polymers.

  6. f

    Polysaccharides include alginic acids, chitosans, pectins and dextrins.

  7. g

    The gravity force is g (on Earth, the standard acceleration due to g is 9.81 m.s−2).

Chemical and biochemical
MacromoleculeCan be considered as the most important parameter
PurityPurity and homogeneity essential but not absolute prerequisites
ConcentrationMostly in the range 5–20 mg·mL−1 (but examples at 1 mg·mL−1 or > 60 mg·mL−1)
Deliberate modificationChemical modification of amino acids, fragmentation into structural domains
Crystallants> 40 Single compounds and ~ 40 associations of two or more compounds
Saltsb22 Compounds, with ammonium sulfate at rank 1
Organic moleculesb13 Compounds, with 2-methyl-2,4-pentanediol at rank 1
Polymers10 Families, notably poly(ethylene glycol) (first use in 1976), Jeffaminesc,d(1992), poloxamersc,e(2009), miscelleanous polymersc(2010), polysaccharidesc,f (2011) with poly(ethylene glycol) 6000 at rank 1
Ionic liquidsImidazolium-based compounds (first use in 2007) after a precursory finding in 1999 on the properties of ethylammonium nitrate
Buffer and pHbHigh success rate in the pH 6–8 range and near pI of proteins
SupersaturationbControls nucleation (number of crystals)
LigandsModify macromolecules properties (importance of stoichiometry)
AdditivesMetal ionsb; other ions; miscellaneous small compounds (in mm range)
Detergents> 50 Potentially useful detergents for membrane proteins; can be useful for ‘soluble’ proteins
PurityBeneficial effects of conformational purity; solid impurities (dust particles)
TemperaturebTested in the range 4–60 °C; temperature-dependent solubility; temperature fluctuations
TimeMinutes to years for nucleation; can modify properties of macromolecules
Pressure (up to 220 MPa)Affects solubility and nucleation (first tested in 1990)
Magnetic field (up to 10 T)Diminishes convection, can orient crystals (first tested in 1997)
Electric field (up to 10 kV·cm−1)Affects nucleation rate (first tested in 1999)
Gravitational fieldg
Earth gravity (1 g)Importance of convection & sedimentation at 1 g
Microgravity (10−6 g range)In space shuttles, stations or satellites; frequent g-variations during flights
Hypergravity (> 1000 g)In ultracentrifuges (from 1000 to 40000 g) (first tested in 1936)
Flow and motionHampers or enhances crystal growth (combined effects of convection and diffusion)
MinimizedIn gelled or viscous media (microgravity mimicry), first tested in 1954
Enhanced/amplifiedBy deliberate stirring (first explicitely tested in 2002)
Vibrations and soundscMostly uncontrolled; also deliberate sonocrystallization proceduresc
Laser irradiationcTriggers nucleation by cavitation effect (first tested in 2006)
Geometry of set-upsInfluences crystallization kinetics (equilibration) (see text)
Volume and geometry of samplesAffects physico-chemical properties of sample media (see text)
Contact surfacescHeterogeneous nucleation and deliberate epitaxy)
MacromoleculeCan be considered as the most important parameter
StateHomogeneity; purity; presence of natural contaminants
OriginExtremophiles versus mesophiles and difficulty with Eukarya
In vivo modificationModification of amino acids/nucleotides; enzymatic fragmentation
Genetic variantsCrystallizability affected by mutations (e.g. disruption of crystal contacts; conformational changes in the protein)

When projects became more ambitious, the poor success rate in crystallization attempts led a few pioneers to develop methods better adapted to the requirements of nascent structural biology. The aim was to produce the rather large crystals needed at the time for diffraction measurements with limited amounts of protein material [44] and, importantly, to enable an exploration of the huge crystallization parameter space (Table 3). Handling the diversity of parameters then became another motivation to devise new crystallization procedures. Thus, in the 1980s, ~ 90 different crystallants were tested, with ammonium sulfate and poly(ethylene glycol) 6000 ranking at the first places [9].

From conventional and forgotten methods to project-driven approaches

Batch and dialysis methods were commonly employed to obtain protein crystals for X-ray crystallography (Table 4). In conventional batch methods, supersaturated protein solutions containing all the required ingredients are left undisturbed in sealed vessels. However, the success of crystallization, notably in terms of number and size of grown crystals, is dependent on the level of supersaturation at time zero, which should be chosen and tuned appropriately. Accordingly, conditions can easily be varied by temperature changes or the addition of small aliquots of chemicals in the experimental vessels. Alternatively, sealed crystallization chambers can be opened to allow concentration changes by evaporation. An attractive variation of the conventional batch method is a sequential extraction procedure by ammonium sulfate, which applies temperature gradients on protein solutions at high ionic strength [45]. It was validated with several proteins and employed for crystallizing E. coli MetRS [46]. Similarly, in dialysis methods, modification or exchange of the solutions in which the dialysis bags are immersed allows tuning of the experimental conditions. However, the main drawback of both methods is the large volume of samples (in the millilitre range) and, consequently, the large amounts of material (> 10 mg) required for each assay.

Table 4. Early crystallization methods and their variants with examples of deliberate crystallizations for X-ray crystallography. For crystallization data, see For references, see text and Data S4.
MethodRemarks, Cell type (sample volume)Early applications, Year (macromolecule)
Batch methods
ConventionalVials (mL range)1971 (sea snake erabutoxin)
Jakoby variantApplicable to protein samples of ≥ 4 mg1971 (proteolyzed E. coli MetRS)
MicrobatchDrops under oil (≤ 2 μL)1990 (e.g. lysozyme)
ConventionalDialysis bags (> 1 mL)1959 (e.g. yeast cytochrome b2)
MicrodiffusionZeppezauer cells (≤ 100 μL)1968 (e.g. lysozyme)
Heavy-walled capillary cells (≤ 100 μL)1970 (e.g. aldolase)
Meso and micro methodsLagerkvist cells (~ 50 μL)1972 (S. cerevisiae LysRS)
Cambridge cells (4–350 μL)1973 (B. stearothermophilus TyrRS)
Microcaps (< 50 μL)1985 (E. coli enterotoxin)
Double-dialysis Cambridge buttons1989 (Staphylococcus aureus delta toxin)
Interface diffusion
Conventional free interface methodPasteur pipettes and other types of glass tubes (diameter < 6 mm)1972 (validated with several proteins, e.g. cytochromes)
Liquid-bridge variantDroplets of protein sample and mother liquor connected by a liquid-bridge1974 (Chlorobium limicola bacteriochlorophyll-protein)
Hybrid diffusion/dialysis methodCapillaries submitted to temperature pulses1975 (Lactobacillus casei thymidylate synthetase)
Sitting dropPlates/trays with 6–100 drops (2–40 μL)1968 (S. cerevisiae tRNAPhe)
Hanging dropTissue culture plates with 24 wells (2–20 μL)1971 (carp albumins)
Sandwiched dropDrops between two glass plates1994 (bacterial cytochrome C-552)
Capillary apparatusSample in a capillary (≤ 1 μL)1988 (ribosome)

The advent of molecular biology and the first successes of X-ray crystallography stimulated biologists and crystallographers to embark on ambitious projects. This was a driving force to devise adapted crystallization methods. A initial breakthrough with an immediate impact on structural biology came in 1968 with the invention of user-friendly vapour-diffusion methods; first, in a sitting drop version for the crystallization of tRNAs [47] that rapidly evolved in a number of variants, notably hanging and sandwiched drops displayed in various experimental arrangements. The method is based on the equilibration of a drop with the protein to be crystallized and all ingredients for crystallization against a reservoir containing the crystallant at a higher concentration than in the drop. Equilibration proceeds by diffusion of the volatile species (e.g. water in most cases, although it can be organic solvents or ammonia always present in ammonium sulfate) until the vapour pressure of the drop equals that of the reservoir, which is accompanied by a volume decrease in the drop and an increase of the protein concentration that can enter in the supersaturated phase during which crystallization can occur. The method can operate in a reverse regime if the initial concentration of the crystallant in the reservoir is lower than that in the drop. In that case, water exchange occurs from the reservoir to the drop. The reverse vapour-diffusion method was discovered fortuitously in the course of an attempt to gently dissolve a tRNAAsp precipitate in a drop (provoked by spermine) by adjunction of water in the reservoir that was followed by the appearance of a new crystal form of this tRNA [48]. Vapour-diffusion was followed in the early 1970s by improved dialysis methods with the invention of new dialysis arrangements (e.g. Cambridge buttons, Lagerkvist cells, capillaries), initially promoted by projects on aminoacyl-tRNA synthetase (aaRS) crystallization [49, 50]. At the same time, the free interface diffusion method became popular [51] and generated related methods, such as a liquid-bridge variant [52] and a hybrid method combining dialysis and diffusion in capillaries. The latter method is attractive because it allows a decrease in the number of crystal nuclei and an increase in crystal size by temperature pulses [53]. Of practical interest was the miniaturization of the batch method to the microlitre level. In that case, experiments are conducted in sitting drops under oil to prevent evaporation and to keep volume constant [54]. A common characteristic of these methods is a significant reduction of the volume of individual assays that decreased by ~ 100-fold (from the millilitre- down to the 2–50-μL range), thereby allowing a more extensive screening of the parameter space with limited amounts of macromolecules.

Two methods that were forgotten for a long time and that have recently been rediscovered are worth mentioning at this point. The first is protein crystallization by centrifugation. This was already used in 1936 to crystallize the coat protein of TMV [55] and its physical basis was investigated in some depth in the 1970/80s with the growth of catalase crystals of various sources in a preparative ultracentrifuge at Institute of Crystallography in Moscow [56, 57]. Even though the original Russian publications were translated into English, they were overlooked by most western scientists, despite the fact that the centrifuge-grown crystals led to the first structure of a catalase solved in collaboration with the Rossmann laboratory [58]. The method was rejuvenated and miniaturized in 1992 with the crystallization of the Trichoderma resei aspartic proteinase [59] and was thoroughly reinvestigated in 2008 with the crystallization of a panel of model proteins and a RNA plant virus at hypergravity levels between 1000 and 22000 g [60]. The underlying idea of the method is to create by centrifugation a gradient of protein concentration in the crystallization vessel that encompasses a supersaturated region favourable for nucleation. A similar rationale underlies a hybrid dialysis method where an increase of protein concentration occurs by electrophoresis [61]. The second forgotten method, first published in 1954 [62], is protein crystallization in a gelled medium where convection is reduced. This diffusion-dependent method, validated by the crystallization of human serum albumin (dimer form) in gelatin, was rediscovered 34 years later [63] in the frame of microgravity projects. Other methods only marginally exploited are crystal growth under levitation [64] and at high pressure [65]. Interestingly, in all of these atypical procedures, the parameter ‘diffusion’ and hence its counterpart ‘convection’ are on the forefront (see implications below).

Attempts to control and understand the crystallization process of biomacromolecules

Before the first interdisciplinary conference on protein crystallization in 1986 (ICCBM1), attempts to understand the physico-chemical basis of protein crystal growth were extremely scarce [66]. For example, Schlichtkrull concluded that, after initial nucleation, subsequent nucleations occur mainly on the surface of existing beef insulin crystals [67] and Bunn distinguished between amorphous and crystalline material when measuring the solubility of calf rennin [68]. The situation changed radically when physicists outside of biology entered the game and tried to adapt the theoretical background of small molecule crystal growth to the protein field [69-71]. This trend was also fostered by the first protein crystallizations in microgravity [72-74]. As a result, data accumulated rapidly and significant information was obtained with model proteins (lysozyme, canavalin, concanavalin A) on precrystallization [75-77], nucleation [78-83], growth [78, 79, 81, 84] and cessation of growth [79, 80, 83]. For example, nucleation rate and final lysozyme crystal size were found to depend upon the rate at which critical supersaturation is approached [81].

The establishment and exploration of phase diagrams represented important trends (Fig. 2A). Initial investigations were conducted on nucleic acid crystals grown by the vapour-diffusion micromethod; first of yeast tRNAPhe [85] and, subsequently, of DNA fragments [86]. The combined effects of Mg2+ and spermine concentrations on crystal quality were explored and, in the case of tRNA, this allowed the identification of a crystal polymorph diffracting at high resolution [85]. In the protein field, initial investigations were conducted on lysozyme by material-consuming batch methods (1–80 mg of protein per measurement), showing the rate-limiting attachment of protein molecules on growing crystals, the preferential growth of large crystals from moderately saturated protein solutions [87] and temperature-dependent solubility accompanied by negative or positive crystallization heats for tetragonal or orthorhombic polymorphs [66]. Because high amounts of material were needed refrained to explore phase diagrams, this encouraged the development of user-friendly micromethods based on vapour-diffusion in 10-μL drops and of sensitive microassays for measurement of protein concentration. This allowed systematic studies with Arthrobacter glucose isomerase, jack bean concanavalin A and HEW lysozyme. Thus, glucose isomerase crystallization was found to be pH-dependent over pH range 5.5–6.5 [88]. With concanavalin A, solubility decreased when salt concentration increased in accordance with the empirical Green's law (see above) and increased with temperature (Fig. 2B). Moreover, crystal morphology was found to be temperature dependent [82]. Importantly, as found for HEW lysozyme, the main effects of salts on protein solubility were a result of anions ranked in the reverse order of the Hofmeister series (SCN > NO3− > Cl− > citrate2− > acetate ~ H2PO4 > SO42−) [29]. On the other hand, the discovery of the peculiar effects of ammonium sulfate at high concentration was unexpected and was beneficial for the crystallization of the yeast AspRS:tRNAAsp complex [42, 89, 90].


Figure 2. Phase diagrams. (A) Theoretical 2D phase diagram showing how macromolecule concentration relates to crystallant concentration; the diagram is multidimensional and can encompass additional dimensions covering physical parameters (see list provided in Table 3). The diagram shows how the solubility curve separates the undersaturated region with the three zones of the supersaturated region and also how parameters vary in a crystallization assay (from the undersaturated region toward the nucleation zone following trajectory a, until a nucleus forms in b that will grow following trajectory c until the crystal/solute equilibrium is reached in d) [14]. (B) Part of the phase diagram of concanavalin A from jack bean showing the solubility of the lectin as a function of ammonium sulfate concentration and temperature [82].

Download figure to PowerPoint

In summary, many different crystal forms were observed when exploring the parameter space of crystallization. A few examples from the author's laboratory are shown in Fig. 3. The important conclusion to emphasize at this point was the absence of a positive correlation between apparent perfection of crystal habits and high diffraction quality.


Figure 3. A gallery of crystals illustrating shape and habit variability, as well as growth pathologies, as observed under different experimental conditions. (A–H) Crystals of model proteins: (A–D) Diversity of lysozyme crystals grown with NaCl as the crystallant [from HEW: (A) microcrystalline precipitate, (B) twinned embedded crystals, (C) classical tetragonal habit obtained at high pressure (50 MPa); from TEW: (D) hexagonal prisms obtained in agarose gel under 75 MPa pressure (length increased and width diminished)]. (E–G) Example of three habits of jack bean concanavalin A crystals found in a phase diagram screening solubility as a function of ammonium sulfate concentration (0.4–2.0 m), pH (5.0–7.0) and temperature (4–40 °C): (E) the typical form grown under almost all conditions, (F) round-shaped crystals grown especially at 12 °C and (G) small crystals growing out of the fracture of a large crystal by 2D nucleation, as occasionally observed in 10 μL sitting drops. (H) Tetragonal bipyramidal crystals of T. daniellii thaumatin grown in free interface diffusion reactors after 10 days of microgravity at 20 °C with 1.6 m Na tartrate as the crystallant [USML-2 (United States Microgravity Laboratory-2) mission in October 1995; note the increased number of smaller crystals at the crystallant entrance of the crystallization chamber at the right side and the gravity vector from right to left]. (I–P) Crystals of key partners in translation: (I) An orthorhombic yeast tRNAAsp crystal that was useful for structure determination. (J–L) Crystals of yeast and T. thermophilus AspRS: (J) tetragonal crystals from the yeast enzyme showing growth defects together with brush-like spherulitic needle bunches and (K, L) gorgeous crystals of the bacterial enzyme from T. thermophilus grown (K) under microgravity or (L) on earth from a microcrystalline precipitate by Ostwald ripening. (M) Crystals of yeast initiator tRNAMet with growth pathology not suitable for X-ray analysis. (N–P) Crystalline diversity in yeast tRNAAsp:AspRS complex crystals grown in the presence of a high concentration of ammonium sulfate, showing a great sensitivity of enzyme purity and RNA/protein stoichiometry: (N) spherulite-like bodies observed with heterogeneous AspRS and a slight stoichiometric excess of tRNA (spherulites are circular bodies composed of thin crystalline and divergent needles/fibres), (O) polymorphism in the same crystallization drop showing cubic and orthorhombic crystal habits and (P) orthorhombic P212121 polymorphs diffracting up to 2.7 Å. For references, see text and S5.

Download figure to PowerPoint

From another viewpoint, the breakthroughs brought by light microscopy, electron microscopy [and later by atomic force microscopy (AFM), see below] and dynamic light scattering (DLS), either to visualize and quantify protein crystal growth processes or as a tool for crystallization diagnostics, were important. Thus, monodispersity of protein solutions under precrystallization conditions, as monitored by DLS, was shown to be a good indicator of crystallizability [75-77]. Also of fundamental importance were investigations on lysozyme crystallization that monitored the size and shape distribution of small aggregates appearing during pre-nucleation and kinetic features characterizing the growth and cessation of growth phases [79]. These were concluded later for non-uniform growth over time accompanied by imperfections on fast-growing faces [80] and growth by lattice defects at low supersaturation and two-dimensional (2D) nucleation at high supersaturation [83]. On the other hand, the time-dependent pH changes that can occur in vapour-diffusion set-ups [91] and the dramatic variations in water equilibration rates when varying temperature and initial drop volume [92] confirmed the importance of kinetic effects in protein crystallization.

Towards better and optimized crystallization strategies

The initial efforts towards rationality in protein crystal growth and the many observations gathered during empirical practice of crystallization in the 1970s and 1980s led to new concepts (notably on purity) and technologies for apprehending protein crystal growth, to the search for optimization strategies, and to proposals regarding improved crystallization strategies that were developed in the 1990s (see below). The fact that many proteins remained recalcitrant to crystallize also stimulated work on the physical chemistry of protein crystallization and the search of biology-based strategies.

A reasonable assumption made by investigators working with proteins recalcitrant to crystallize was to conjecture that evolution has shaped more stable proteins in organisms adapted to live under extreme conditions. The idea was validated with a thermophilic TyrRS [50] that yielded better crystals than the mesophilic counterparts. The same is true for thermophilic ribosomes [93]. Rationalization of the concept of purity was another accomplishment. It was based on personal observations (e.g. the presence of microheterogeneities in tRNA and protein samples) [94] and data from literature (e.g. beneficial effects of purification on both crystal growth and crystal quality) [95-97]. Altogether, this led to a refined definition of what is really protein purity, namely chemical and conformational homogeneity, an absence of protein and small molecule contaminants, and stability over time. Considerations about purity gave also a refined view on the nature and importance of impurities (isoforms or denatured/aggregated versions of the protein of interest, foreign protein material, small molecule contaminants) in protein samples that could affect crystallization. Striking examples concern contaminants present in poly(ethylene glycol), especially phosphate or sulfate anions [accounting for the growth of Eco elongation factor polymorphs depending on the brand of poly(ethylene glycol) used] [98] or aldehydes and peroxides that were shown to affect the crystallization of rabbit muscle phosphoglucomutase [99]. In this context, a crystallization method combining purification and protein conditioning in crystallization media [100] is worth mentioning.

Accordingly, it was conjectured that the intrinsic flexibility of peptides and many proteins would be detrimental to their crystallization. A remedy would be to stabilize the unstable structures with other macromolecular partners. The idea was validated with the crystallization of antibody:antigen complexes, with initial proof-of-concept experiments using lysozyme as the antigen [101]. The relative ease to prepare monoclonal antibodies permitted a rapid generalization of the strategy with, for example, the crystallization of neuraminidase from influenza virus [102] or of the human angiotensin II peptide [103] in complex with specific Fab fragments derived from monoclonal antibodies. Today, cocrystallization strategies have many applications in structural biology (see below).

Among thermodynamic parameters, temperature and time [104] were identified first as being important for protein crystallization. Both affect protein conformation and, consequently, solubility, as well as crystal habits (Fig. 3) and growth mechanisms. Similar effects are brought about by pressure [65] and pH changes [104]. On the other hand, nonreproducibility remained a major drawback and pointed to the primordial role of the geometry and size of set-ups (both crystallization chambers/drops and reservoirs) that affect equilibration kinetics and modulate the balance between convective and diffusive mass transport during crystal growth, as well as the extent of crystal floating or sedimenting in the mother liquor. Furthermore, experimental evidence indicated heterogeneous and epitaxial nucleation brought about by contact of proteins with solid surfaces [105], with such phenomena even occurring on the surface of growing crystals (Fig. 3G). The fact that diffusion is favoured under microgravity (and convection disfavoured), as well as the expectation of better crystals when grown in this environment, was the main justification of crystallization programs in weightlessness. Initial experiments showing growth of larger lysozyme crystals [72] were the start of a race for the access to microgravity [106], which generated both controversial debate [107] and a search for an alternate means to favour diffusive mass transport on earth. This line of thinking was first suggested in 1988 by Robert and Lefaucheux, who grew lysozyme and porcine trypsin crystals in gelled media [63], and was largely exploited in the 1990s with studies of protein crystallization by counter-diffusion or under magnetic- and electric-fields (Table 3; see also below). On the other hand, seeding procedures were recognized as practical means to optimize crystallization as soon as initial crystalline material becomes available. They have been shown to trigger new nucleations or to enlarge the size of crystals [108].

Because of the impressive number of parameters affecting protein crystal growth and crystal quality, which largely exceed that involved in small molecule crystal growth (Table 2), it became rapidly evident that identifying the appropriate crystallization conditions could not occur by systematic screening of the parameter space. The need to understand the hierarchy of parameters and their relationships became essential. This was not an easy task because this hierarchy is dependent on the class of macromolecules. Emblematic examples are the detergents essential for membrane protein crystallization [43, 109] but not required for soluble proteins, although they can have beneficial effects [110], and the polyamines that are only essential for tRNA crystallization [111]. To overcome difficulties, statistical methods were invented. The first comprised an incomplete factorial method validated with B. stearothermophilus TrpRS that aimed to find correlations between crystallization parameters and crudely estimated crystal quality [112, 113]. It was followed in the 1990s by sparse-matrix methods (see below). In parallel, robotic systems were proposed to facilitate the practice of crystallization and to achieve better reproducibility [54, 114].

Summary before entering the era of structural genomics

The cooperation between biologists and physicists in the 1980s with respect to crystal growth, as illustrated by the first ICCBM Conferences, provided insights into the mechanistic aspects of protein crystal growth. On the other hand, crystallization was no longer restricted to isolated proteins and now also concerned protein assemblies, nucleic acids and nucleic acid:protein complexes. Highlights were the miniaturization of conventional batch and dialysis crystallization and also the invention of vapour-diffusion methods. Vapour-diffusion methods were rapidly adopted by structural biologists because of their versatility, although drawbacks were soon intuitively recognized. They rely on the fluctuating geometry of the crystallizing drops and the dynamic nature of the vapour-diffusion process leading to a decrease of protein concentration and a concomittant increase of impurities in the crystallizing media, accompanied by an enhanced poisoning of the growing crystals, as first suggested by Wayne Anderson [115]. Because physico-chemical conditions in crystallization drops are not well defined, this might explain the large number of irreproducible results. Batch methods that are more static and easier to implement remained popular, especially in their miniaturized versions under oil. Despite the remaining poor understanding of many aspects of protein crystallization, the sound theoretical basis that emerged in the period between 1934 and 1990 opened new routes for more rational and efficient biocrystallogenesis, which were successfully explored in the era of structural genomics.

Crystallogenesis in the era of technologies and structural genomics (1990–2013)

  1. Top of page
  2. Abstract
  3. Introduction
  4. The time of physiology and chemistry (1840–1934)
  5. The birth of biocrystallogenesis as a science (1934–1990)
  6. Crystallogenesis in the era of technologies and structural genomics (1990–2013)
  7. The status in 2013 and perspectives
  8. Acknowledgements
  9. References
  10. Supporting Information

Crystallogenesis always benefited from the interplay between science and technology. This trend became especially prevalent after 1990 when the new biotechnologies provided tools for the preparation of any type of protein or nucleic acid present in nature and, when new instrumentations and robotic systems became accessible, for more efficient crystal growth and faster crystal analysis [116]. The initial fundamental work with HEW lysozyme and a few other model proteins was pursued, and extended to membrane proteins and a large panel of other proteins and macromolecular assemblies of high biological value. In parallel, ideas originating from fundamental work were translated into applications useful for the growth of crystals for structural biology. Altogether, this led to a paradigm change with deep impact upon the field. New strategies for qualitative and quantitative evaluation of the different steps of the crystal growth process were proposed and specific crystallizability features were discovered. The need for large crystals declined with the easy access to second-generation synchrotron X-ray sources, a trend that even applies for modern neutron crystallography. Automation progressively became essential in crystallogenesis and, recently, the nascent nanotechnologies found many applications in structural biology. Last but not least, the biocrystallogenesis field received support from Space Agencies that fostered microgravity research and were particularly interested in protein crystallization. Altogether, during the period between 1990 and 2013, the field benefited from dramatic advances in analytical and gene technologies and was nourished by a constant interplay between fundamental and practical focused research. For simplicity, these two aspects are discussed separately.

Fundamental crystallogenesis

The effects exerted by physical and chemical parameters on macromolecular crystal growth and many related questions have been investigated in depth by a variety of approaches [117, 118]. Exploration of parameter-space in phase diagrams was first on the forefront for the selection of parameters leading to protein crystallization. Imaging growth processes, scrutinizing crystal anatomies and, important from the viewpoint of structural biology, comparing X-ray structures solved from crystals grown under different conditions, represented other challenges. In the late 1990s, investigations on atypical physical, chemical and method-related parameters that might affect crystallization became more prevalent and led to alternative crystallization methods. This was first the case for microgravity and related factors and, more recently, for light, ionic liquids, new additives, the volume of crystallizing samples and the geometry of set-ups. Other goals were to control nucleation and to uncouple it from growth.

Solubility and phase diagrams

Because of the multiparametric nature of the crystallization process, protein phase diagrams are multidimensional and therefore can only be partly explored. Their landscapes represent the solubility behaviour of proteins under crystallization conditions and can be considered as footprints characterizing individual proteins or group of proteins. Thus, the proper handling of these parameters could be used to initiate and control crystallization. Based on empirical rules derived from Arda A. Green's work [39, 40] and precursory theoretical thoughts on protein solubility, it was expected that some general rules could govern protein solubility and thus predict crystallization. However, as a result of a poor understanding of the crystallization process, only qualitative rules could be expected at the time. Thus, the pH-dependent solubility of proteins, with a minimum at the isoelectric point (pI), where the average charge is zero, is accounted for by the zwitterionic nature of proteins. Similarly, the salt-dependent solubility relies on the ionic interactions that salts can make with proteins. This occurs especially at high ionic strength as reflected by salting-out (i.e. decrease in solubility when the salt concentration increases) and at the less frequent opposite and poorly understood salting-in (i.e. increase in solubility) phenomena. Understanding how the many other factors listed in Table 3 affect protein crystallization remained essentially unknown, notably the gravity-related factors convection and diffusion, which are well explored for the crystal growth of conventional molecules but not for that of proteins [69, 119].

To obtain insight into these unexplored issues, systematic studies were initiated in the early 1990s, first on the effects of Hofmeister salt concentrations, pH and temperature on protein solubility and crystal growth, and later on those of a variety of additional parameters, either chemical [organic crystallants such as poly(ethylene glycol)] or physical (pressure, convection, diffusion, light, etc.). Experiments were conducted not only with the standard models, but also with proteins of interest for structural biology. Thus, besides exploring crystallizability of HEW lysozyme [120-123], partial phase diagrams were established, amongst others, for a collagenase [124], two membrane proteins (bovine cytochrome c, Rhodobacter sphaeroides photoreaction centre) [125, 126], a carboxypeptidase [127], S. cerevisiae AspRS [128] and even plant viruses [129, 130]. Thus, with lysozyme and whatever the pH, increasing pressure resulted in greater numbers of crystals, as well as a transition from the initial tetragonal to the orthorhombic polymorph [122]. On the other hand, huge temperature effects on solubility were observed with most investigated proteins [131, 132]. In the majority of cases, a normal temperature-dependence was observed (increase of solubility with temperature, as for lysozyme, trypsin and insulin) but retrograde solubilities were suggested as well (decreased solubility, as for catalase and glucose isomerase) [131]. The situation was paradoxical with T. daniellii thaumatin known to crystallize with Na tartrate because the temperature-dependence of its solubility depends on the chirality of the tartrate ion (i.e. normal with l-tartrate and retrograde with d-tartrate) [133]. This observation is of importance because it reconciles the contradictory results obtained with crystals grown from solutions of racemic Na d,l-tartrate. Other dramatic temperature effects were observed with S. cerevisiae tRNAPhe, notably a transition between three different growth mechanisms within a narrow range of only 5 °C as seen on AFM images [134]. Regarding the effects of Hofmeister salts on solubility, these differ globally for acidic and basic proteins and, in the case of individual proteins, they depend on the acidic (pH < pI) and basic (pH > pI) state of the protein [135], as well as the kosmotropic and chaotropic nature of the salts (making strong or weak, respectively, water interactions in the solvent shell around the protein).

In the case of the poly(ethylene glycol), often associated with salts, the situation becomes more complex because liquid–liquid phase separations are frequently observed, with consequences on protein solubility [136, 137]. Thus, with the extremelly soluble Aspergillus flavus urate oxidase, a poly(ethylene glycol)-induced depletion potential in the protein solution could be demonstrated by small-angle X-ray scattering (SAXS) measurements and validated by theory [137]. It was also shown that the liquid–liquid phase separation precedes and slows down crystallization [138]. Globally, poly(ethylene glycol) modifies phase diagrams and favours the attractive intermolecular interactions needed for crystallization. This offers the possibility to control crystallization by varying the size and concentration of the poly(ethylene glycol) in crystallization media.

Of practical interest were light scattering studies [both SAXS and small-angle neutron scattering (SANS)] (Table 5) that established a correlation between the second virial coefficient B22 and solubility [139, 140]. This coefficient characterizes the nature and the strength of the interactions between protein particles in solution and thus provides essential information on crystallizability. If B22 is positive, the overall intrections are repulsive. By contrast, if B22 is negative, the interactions are globally attractive, which favours crystallization, a conclusion that received theoretical support [141]. From the viewpoint of phase diagrams, the existence of a metastable liquid–liquid immiscibility region was predicted in which small liquid droplets with a high protein concentration form before nucleation proceeds. This region corresponds to the ‘crystallization window’ (–8 × 10−4 < B22 < −2 × 10−4 mL·mol−1·g−2), as proposed by George and Wilson [139]. A refinement of this concept proposes a ‘kinetic crystallization window’, independent of the shape and conformation of the protein [142]. It is characterized by a kinetic coefficient, ζc, defined as the ratio between the diffusion rate of the protein in solution and its surface integration rate (based on the kinetics of protein surface self-assembly at the air/water interface as evaluated by surface tension measurements). Formation of single crystals is kinetically favoured in the range 1 < ζc < 8 where both diffusion and integration rates are comparable. This criterion has been succesfuly verified for several proteins [142].

Table 5. Diagnostic tools for protein homogeneity, crystallizability and crystal quality. For references, see text and Data S7.
ToolType of information (year of early inputs)
  1. a

    Classical and scanning electron microscopy, transmission electron microscopy, etc.

  2. b

    X-ray, UV and correlation fluorescence spectroscopies, etc.

  3. c

    Mach–Zehnder, Michelson and dual polarization interferometry, etc.

  4. d

    Experimental and virtual bioinformatic predictions, etc.

  5. e

    Classical light microscopy and advanced methods, such as laser confocal differential interference contrast miscroscopy, second-order nonlinear optical imaging of chiral crystals and ultrahigh resolution optical coherence tomography.

AFMGrowth mechanisms (1992); growth pathologies (1992)
CalorimetryThermodynamics of crystal growth; stabilization of proteins by additives (1996)
DLSScreening homogeneity protein homogeneity under precrysrallization conditions (1978); detection of nucleation (1978)
Electron microscopy toolsaVisualization of lattice defects and 2D nucleation (1990); in situ detection of crystalline phase in biological samples (2002); sample-quality analysis of membrane proteins (2003)
Fluorescence spectroscopiesbDetection of salts in crystals (1997), of protein aggregation in solution (2009)
InterferometrycQuantitative mapping of solution properties (solute concentration, convection, etc.) around growing crystals (1993)
Informatic predictionsdIncomplete factorial and sampling methods (1979); database screening (2003)
Sequence-based crystallizability prediction (2006); nucleation prediction (2012)
pI (2004)
Mass spectrometryContent of macromolecules in crystals and detection of bound or contaminating small molecules within crystals (2000)
Optical light microscopieseCrystal habit (1840); protein crystal detection in crystallization media with precipitates (2010, 2012); measure of growth velocities on individual elementary steps (2012)
PAGE and IEFSequence size homogeneity (1982); crystallization screening (2001)
Raman microscopyQuality control of crystals with derivatized proteins (2008)
SANSTime resolved diagnostic of the crystallization process (2008); protein fate in precrystallization (1994) and supersaturated solutions (1995)
SAXSCrystallization screening (1995), following crystallization process (1998); detection of crystallization artefacts (2010)
Surface plasmon resonanceFor identifying compounds that bind to target proteins (2012)
ThermophoresisSearch of macromolecule solubility on a thermal gradient device (1998) and crystallization screening (1999)
X-ray topographyVisualizing crystal perfection (1996)
Nucleation and growth

In the 1990s, the focus was to crystallize recalcitrant proteins and to enhance quality of crystals not suitable for structural work. This necessitated fundamental research and was influenced by space-crystallization programmes. Indeed, theory claimed that a number of gravity-dependent phenomena that prevent or perturb crystal growth on earth are minimized in weightlessness (e.g. sedimentation, mass transfer, concentration gradients and convective currents). The logical consequence is an enhanced quality of space-grown protein crystals. The expectation received support from the early space-crystallization experiments, thereby justifying ground-based research aiming to obtain deeper insight into the mechanisms of protein crystallization and to optimize the forthcoming microgravity missions. This also stimulated new research lines aiming to simulate microgravity conditions on earth and to develop alternative methods of crystallization (see below) (Table 6). The main results are summarized below.

Table 6. New advanced and old rejuvenated methods of protein crystallization. For references, see text and Data S7.
MethodComments on instrumentation and outputsProof-of-concept (year)
ContainerlessElectrostatic levitation method (air/liquid contacts) for vapour-diffusion; nice crystals1990
Batch method with floating drops (5–100 μL) under two oil layers (liquid/liquid contacts); fewer crystals1990
EpitaxyMany recent advances: epitaxial growth can occur in vapour-diffusion set-ups, on minerals, on lipid or protein layers, on etched surfaces1988
Flow-cellCrystallization under well-defined conditions (e.g. for either quiescent or forced convection), growth under constant protein concentration)1986
Hybrid methods combining
Microgravity and gelsX-ray topographs indicate more ordered thaumatin crystals than the earth control1999
Gel and oilCan be operated automatically in microbatch technology; improves the gel acupuncture method; reduces growth rate2002
Magnetic field and levitationObservation of new phenomena for crystallization and dissolution processes2008
Microgravity and counter-diffusionHigh quality crystals of a lectin grown in Gel-Tube R crystallization kit flown in Russian Service Module and crystals of several proteins grown in the dedicated Granada box operated in ESA FOTON mission2008
Gel and laser pulsesEnhancement of nucleation at very low supersaturation2013
Induced nucleation by
Continuous lightCrystallization by Xe-lamp irradiation or by photon pressure produced by a continuous wave laser2006
Laser light pulsesCavitation effects essential for induction of nucleation; allowed crystallization of many proteins, including membrane proteins; nucleation can be induced at very low supersaturation at gel-solution interfaces2003
Natural or modified surfacesModified glass or mica surfaces, porous surfaces, organic fibres, etc.2000
Langmuir–Blodgett technologyUse of template protein film for growth of microcrystals2002
TemperatureFirst conducted in a thermonucleator (with local supersaturation control); adapted for vapour-diffusion, batch and multiwell microbatch with T-gradient1992
UltrasoundNucleation of lysozyme after short ultrasonic irradiation (100 kHz and 100 W); reduction of metastable zone and crystal growth at lower supersaturation2006
MicrogravityBatch, dialysis, vapour-diffusion, free interface diffusion in advanced protein crystallization facility (APCF) and protein crystallization diagnostic facility (PCDF), counter-diffusion in Granada box; convection minimized but frequent perturbations by g-jitters1984
Microgravity features (e.g. reduced convection and favored diffusion, crystal orientation), simulated in/by
‘Ceiling’ geometryA ‘seed’ crystal attached to the top of a growth cell continues to grow in a diffusion-limited regime; sedimenting microcrystals do not perturb the growing crystal2009
Counter-diffusione.g. Granada box; first known as gel acupuncture method; at present generalized use of capillaries, works with membrane proteins1993
Electric fieldAdapted to microbatch or vapour-diffusion; control of nucleation rate and better quality of HEW lysozyme crystals1999
Gelled mediaClassical devices; mass transport restricted to diffusion1954
High pressureBatch reactors; control of solubility and crystallization1990
HypergravityOperated in batch vessel in ultracentrifuge; metastable starting conditions become supersatureted during centrifugation1936
Magnetic fieldLatest advances in superconducting magnets that provide quasi-microgravity conditions: improvement of crystal quality (resolution, B-factor) observed1997
MicrofluidicFree interface diffusion, nanobatch, counter-diffusion, formulation chips; variety of chips available (e.g. for visual crystal inspection, initial X-ray screening and high-throughput data collection)2002
LevitationCrystals obtained under ultrasonic levitation grow at higher rates are fewer and have better shape and larger size2012
Reverse vapour-diffusionOperated in any classical vapour-diffusion system; requires gentle drop volume increase by vapour-diffusion to dissolve protein precipitates (rediscovered in 1995)1977
Stirring/vibration/flowRotatory shaker or mechanical vibrator; improvement of resolution and mosaicity of crystals grown in stirring mode2002

Although it was known that nucleation occurs at much higher supersaturation than growth and that, once a nucleus is formed, growth follows spontaneously, little was known in the 1990s about its exact mechanism in the protein world, except that it should depend exponentially on supersaturation and should occur preferentially on solid surfaces [70]. The reality of heterogeneous nucleation of proteins induced by substances, such as contaminating dust or other solid/colloidal particles, was rapidly confirmed by experiments [105]. It took more time to unravel the nucleation process itself because two decades of intensive work were needed [143-146] before a comprehensive two-step mechanism emerged [147]. One reason for this is that researchers applied classical nucleation theory to solution crystallization without taking into account differences between theoretical predictions and experimental results [148]. Thus, according to the two-step model, crystalline protein nuclei appear inside pre-existing metastable clusters, which consist of dense liquid and are suspended in the solution. Such small-size nuclei have been visualized by AFM [149]. At high supersaturation, the nuclei are generated in the spinodal regime where the nucleation barrier is negligible. The solution–crystal spinodal helps to clarify the role of heterogeneous substrates in nucleation and the selection of crystalline polymorphs. These ideas provide powerful tools for the control of the nucleation process by varying the solution thermodynamic parameters [147]. It is essential to note that this two-step model worked out for proteins appears to apply for all crystallization processes occurring in nature and industry [148]. This new nucleation scenario could explain specific effects observed with poly(ethylene glycol) where liquid–liquid phase separations are often observed, as well as with various substances or solid supports known as heterologous crystallization nucleants [105, 150]. In the particular case of human hair, which can act as a heterologous nucleant, it was shown by confocal fluorescent microscopy that the protein is concentrated on the nucleating surface, with a substantial accumulation of protein on the sharp edges of the hairs cuticles [151].

Controlling nucleation has practical applications. A simple solution consists of removing uncontrolled solid nucleants by filtration [152]. This can also be achieved by counter-diffusion methods [153] or by application of an electric field [154, 155]. Another simple solution is to eliminate poor quality crystals appearing after nucleation by manual selection [156]. An alternative possibility would be to stimulate protein nucleation by temperature or ultrasounds, as demonstrated for small molecules [157]. This was achieved in the 1990s for temperature [158] and, more recently, for ultrasound waves [159] (Table 6).

Growth of protein crystals highly depends on supersaturation and on the presence of impurities in the solute. It is also sensitive to crystal-dependent parameters, such as structure and defects of crystal faces, as well as on the bonds established between growth units. At low supersaturation, crystals predominantly grow by screw dislocations propagating in a helical path around lattice defects. At higher supersaturation, they grow by 2D island formation from 2D clusters/nuclei that form randomly on flat regions on crystal faces. These mechanisms were predicted by theory in the small molecule field [70] and were explicitly visualized by AFM images for proteins (Fig. 4A) [160], RNA [134] and viruses [160]. AFM revealed also mesoscopic defects, such as stacking faults, point defects, vacancies at surface protein layers on crystal faces and other statistical misalignments [161]. They originate from perturbed growth conditions, which are unavoidable because crystal growth is accompanied by a decrease in supersaturation in the mother liquor. This effect is particularly prevalent with tRNAPhe crystals that show a dynamic change in growth morphologies induced by even minute temperature changes [134]. Incorporation of impurities or microcrystals can further affect crystal growth [161, 162] and harm the production of high-quality crystals assumed to grow at the lowest supersaturation and with a constant growth regime. Uncontrolled growth conditions likely account for nonreproducibility of diffraction properties.


Figure 4. Visualizing microscopic crystal morphology in AFM and X-ray topographic images. (A) AFM images of (a, b) yeast tRNAPhe crystals seen at two temperatures [134] and (c) T. thermophilus AspRS crystals. (a) Dislocation hillocks on tRNAPhe crystals are formed at 15 °C by multiple right-handed (left of image), single left-handed (centre of image) and double right-handed screw dislocations (right of image). (b) Growth by 2D nucleation at 13 °C showing growth and coalescence of islands and expansions of stacks. Formation of a hole caused by incorporation of foreign particles during the growth of additional layers is shown in the bottom centre of the image. (c) AspRS growth proceeds by screw dislocation mechanism, as seen on the (100) crystal face. (B) X-ray topographs on (d) TEW lysozyme and (e) T. daniellii thaumatin crystals [212]. (d) Optical view and schematic drawing of TEW lysozyme crystals (left), reflection profiles of crystals grown from solution (top left) and in gel (top right) (notice the same full width at half maximum of 6.5 arcsec of the two crystals), and topographs taken at the top of the reflection profiles plotted for solution and gel grown crystals (bottom left and right, respectively). (e) The same images for thaumatin crystals grown from solution and in gel. For references, see text and Data S6.

Download figure to PowerPoint

Convective solution flow, mass transport and concentration gradients play essential roles in crystal growth. According to theory, crystal quality is usually better under diffusion-limited growth where a depleted zone of the solution surrounds the growing crystal [70, 157]. However, because of convective fluid motions, the depletion zone is hardly maintained around a crystal on earth, which might explain why protein crystals should be of higher quality when grown under microgravity conditions. To explore these issues, interferometric studies were undertaken under earth-gravity and reduced space-gravity. The first data on protein crystals were obtained in 1993 [163, 164] and were followed by a series of investigations using Michelson or Mach–Zehnder interferometry that quantitatively characterized concentration gradients, depletion zones and diffusion boundary layers around growing and dissolving protein crystals [161, 165-168]. From all these studies, it was possible to propose kinetic models of growth and to realize that quasi-stable depletion zones form around growing crystals in space and, consequently, that best conditions for crystal growth occur under microgravity and that vapour-diffusion geometry does not provide spatially stable crystal position or fluid conditions for optimized growth under a diffusive regime. This last conclusion is in line with other observations made about difficulties encountered in vapour-diffusion methods, as a result of drop size and shape, geometry of the crystallization set-up and associated evaporation kinetics that all control the output of crystallization trials [92, 169-171], and accounts for the nonreproducibility likely explained by uncontrolled physics inside droplets.

A few comments about Ostwald ripening and membrane protein crystallization are worth noting. Ripening concerns the fate of precipitates occuring at a high supersaturation that occasionally transform into large crystals. In the macromolecular field, the phenomenon was first explicitly described in 1996 for Thermus thermophilus AspRS [32] (Fig. 3), although it has been occasionally seen by many protein crystal growers. A recent study shed some light on the mechanism. Using a combination of DLS, optical microscopy and microfluidics, it could be shown that a dense amorphous phase constituted by precrystalline protein clusters displays classical Ostwald ripening growth kinetics but deviates from this trend after nucleation of the crystal phase. It was concluded that this behaviour arises from a metastable relationship between the clusters and the ordered solid phase [172].

Regarding the mechanism underlying membrane protein crystallization, although it likely follows general rules demonstrated for soluble proteins, it presents specific features as a result of the intricate interaction networks created under crystallization conditions by the detergents, amphiphiles, crystallants and hydrophobic membrane proteins. Thus, using SANS, it was shown that optimization of micelle size and shape for crystallization requires specific combinations of detergent, amphiphile and crystallant [173]. It was also shown that poly(ethylene glycol), often included in crystallization media for membrane proteins, favours solution conditions where the stability of the liquid phase changes from stable to unstable [174]. A great breakthrough came in 1996 when Ehud Landau and Jurg Rosenbusch replaced the micellar crystallization media with lipidic cubic phases [175]. These are gel-like lipid–water systems comprising lipidic compartments interpenetrated by aqueous channnels that were discovered in the 1960s by Vittorio Luzzati [176]. Recent data indicate that nucleation of bacteriorhodopsin crystals occurs in such media following local rearrangement of the highly-curved lipidic cubic phase into a lamellar structure mimicking the native membrane in which the crystals will grow in a constrained environment surrounded by lamellar structures [177]. This mechanism leads to an absence of dislocations and the generation of new crystalline layers at numerous locations, as well as to voids and block boundaries. The characteristic macroscopic lengthscale of these defects suggests that the crystals grow by attachment of single molecules to the nuclei [177]. At present, the in cubo method is widely used [178] and applications for soluble proteins are expected. Recently, the method was extended to other mesophases in the lipid (monoolein)/water diagrams and led to a user-friendly fast screening technology [179].

Microgravity and related issues

A further step towards understanding protein crystallization consisted of an evaluation of the parameters governing mass transport and dynamic flow during the process. Viscosity and gravity are the major parameter accounting for convection/diffusion and buoyancy-induced phenomena. Their effects are well known with respect to the crystal growth of conventional molecules but were only thoroughly studied in the case of proteins after the first protein crystallization in microgravity [72] and the claim that reduced convection under such conditions should favour crystal quality [106]. It took approximately two decades to convincingly validate the expectation [180]. Overall, space-grown crystals grow larger, and have more regular external morphology and better internal order with reduced mosaic spread [181, 182], although contradictory results have been reported [14]. Resolution is sometimes overwhelmingly improved, as for space-grown paralbumin crystals that diffract at 0.9 Å, whereas the earth controls are not suitable for diffraction analysis [183]. In a few cases, when real ground controls were available, space-grown crystals gave more accurate structures (e.g. obtained with better defined initial electron density maps) [180, 184, 185].

Access to space is not easy and, already in 1988, a first solution to simulate microgravity effects was proposed consisting of crystallization in gelled media [63]. This possibility was followed by proposals advertising protein crystallization by counter-diffusion [186], under magnetic [187] or electric [188] fields, and, most promisingly, under microfluidic conditions [189] (Table 6). Preferential orientations of crystals were observed under a magnetic field [190] and numerical predictions revealed the damping of convection by magnetization [191]. It was also realized that some atypical methods could reproduce any potential beneficial features of the microgravity environment, such as crystallization in containerless systems [192], under hypergravity or at high pressure.

In all of these methods, mass transport and fluid movements are affected, as accounted by the dimensionless Grashof GrN number [193], a classical predictor in fluid mechanics, which, in the present case, evaluates how buoyancy and viscosity forces affect proteins in their liquid crystallization media according to:

GrN ≈ buoyancy forces/viscosity forces ≈ L3 α Δc g ν−2

where L is the characteristic length of the system in which a protein is immersed (e.g. the diameter of a sphere in which the protein can move), α Δc is a density gradient dependent on the concentration of the protein, g is the gravity force, and ν is the viscosity of the fluid. It can be easily seen that the same GrN value can characterize both microgravity and earth conditions provided that the low gravity force (g) in space is balanced by adapted geometrical characteristics of the crystallization device (L) and viscosity forces (ν) on earth. This typically occurs in gelled media and under counter-diffusion and microfluidic conditions.

A posteriori, the usefulness of these methods for structural biology is demonstrated by the increasing number of Protein Data Bank (PDB) entries of structures solved with crystals grown by these atypical procedures. In a few proof-of-concept cases, it was shown that the quality of the X-ray structures solved from diffraction data originating from crystals grown under conditions simulating microgravity conditions are improved. For thaumatin, the crystals grown in agarose gel diffracted to a previously unachieved resolution and yielded a structure at 1.2 Å resolution computed from diffraction data collected at room temperature [194]. In the case of magnetic field, a comparison of HEW lysozyme structures of 0-T and 10-T crystals revealed only limited overall structural changes but demonstrated significant fluctuations at a few residues, improvement in crystal perfection and increased diffracted intensities leading to a higher resolution [195]. Interestingly, for earth- and space-grown HEW lysozyme crystals grown in the advanced protein crystallization facility apparatus, counter-diffusion crystallization even improved the resolution of the tetragonal crystals from 1.40 to 0.94 Å [196].

The dimensionless Reynold ReN number [193] quantifies the relative importance of inertial and viscosity forces in fluid dynamics according to:

ReN ≈ inertial forces/viscosity forces ≈ L v

where L is the characteristic length of the system, v is the mean velocity of the protein and ν is the kinematic viscosity of the crystallization medium (ν = μ/ρ; where μ is the dynamic viscosity and ρ is the density of the fluid). Evaluation of Reynold numbers was used to find optimal stirring conditions for HEW lysozyme crystallization [197]. Note that the stirring crystallization method is widely used in the small molecule field for growing high-quality crystals and, in the present case, it was shown that intermittent flow and low ReN values contribute to the growth of a smaller number of larger crystals [197]. Finally, crystallization of HEW lysozyme was also analyzed in quiescent and forced-convection environments [198].

Other theories and simulations predict that shear flow could enhance or, conversely, suppress the nucleation of crystals from solution. These ideas were tested in droplets held on a hydrophobic substrate in an enclosed environment and in a quasi-uniform constant electric field that induces a rotational flow with a maximum rate at the droplet top [199]. The likely mechanism of the rotational flow involves adsorption of the protein and amphiphilic buffer molecules on the air–water interface and their redistribution in the electric field, leading to non-uniform surface tension of the droplet and surface tension-driven flow.

Thermodynamic considerations on protein crystallization

Although thermodynamic approaches are appropriate to describe solubility, phase separation and crystal growth processes, they were only scarcely used in the field of protein crystallization. A first phenomenological approach in 1996 with HEW lysozyme found agreement between values of crystallization enthalpies determined by calorimetry and by analysis of van't Hoff solubility plots [200]. It was followed by a few other studies [201]. However, a fruitful paradigm change occurred when researchers tried to understand the enthalpic and entropic contributions to the Gibbs free energy of crystallization (ΔG°cryst = ΔH°cryst – TΔS°cryst) from the viewpoint of chemistry; in other words, taking into account the contribution during crystallization of intermolecular bond formation between protein and solvent. Thermodynamic data were gathered for several proteins (apoferritin, haemoglobin C, insulin, lysozyme) and showed that their crystallization is dominated by entropic phenomena [202-204]. Thus, the solvent structure, together with the trapping and release of water molecules, is essential in the crystallization of these proteins. This implies structural rearrangements in protein and solvent, mimicking by some aspects that which occurs in macromolecular recognition phenomena. These facts have important consequences for protein crystallization because, by engineering ΔS°cryst, it becomes possible to find thermodynamically favoured crystallization conditions. The idea was exploited under two versions: either by protein surface engineering to favour intermolecular interactions [205] or by calorometry-based selection of additives for their propensity to stabilize protein structures in crystallization solutions [206]. A few proof-of-concept cases of crystal structures of protein variants showing modified crystal packing contacts provided strong support for these approaches [205] (for applications, see below).

Anatomy and quality of protein crystals

Optical microscopy images show a variety of crystal habits, with some exhibiting perfect shape and symmetry. However, at higher resolution, as seen by AFM, crystal faces are not flat but have rough surfaces with growth-dependent morphologies comprising frequent imperfections and level differences reaching up to 1000 Å and even more. This raises the important question of the impact that growth conditions and growth-induced defects can have on the internal order of crystals, as ultimately reflected by diffraction properties. X-ray topography is an appropriate technique to answer such concerns. It was used for the first time with protein crystals in 1996 [207, 208]. The method informs about the spread of mosaic blocks, and detects imperfections and variations in the internal order within a crystal [209-211]. Typical X-ray topographs obtained from TEW lysozyme and T. daniellii thaumatin crystals grown under two different growth conditions are shown in Fig. 4B. They clearly show, especially for TEW lysozyme, more homogeneous images for the gel-grown crystals, demonstrating that the gel improves crystal quality [212]. Over the years, the technique has been refined and applied to an increasing number of proteins. The most recent studies have characterized individual domains in HEW lysozyme crystals (with homogeneous diffraction qualities) [213], as well as the presence of loop and curve shaped dislocations in such crystals [214]. Information on the internal structure of protein crystals may be useful to aid in the improvement of crystal growth techniques [213] and may guide femtosecond laser processing of gel-grown crystals for diffraction data collection on the most perfect crystal domain [215].

Summary and main conclusions

The science of biocrystallogenesis has made considerable progress in the period between 1990 and 2013, in great part through the combined efforts of biochemists, biophysicists, protein crystallographers and scientists from the small molecule crystal growth community. Thus, as anticipated, it was convincingly demonstrated that the general rules of crystal growth apply to the protein field. In the case of nucleation, a novel two-step mechanism was proposed by Peter Vekilov and coworkers that could be generalized for all crystallization processes, as reported by experts of the crystal science of conventional molecules [148]. Altogether, a better understanding of the physical chemistry of proteins in the different zones of phase diagrams (Fig. 2A) emerged and, in the case of membrane proteins, understanding how lipidic cubic phases sustain their crystallization was an important achievement. From the standpoint of structural biology, it was realized that crystal growth under diffusive conditions enhances the quality of protein crystals, which is reflected by the better quality of the crystallographic models of macromolecules. To reach these conclusions, a panel of analytical and diagnostic tools (listed with their characteristic features in Table 5) were of operational importance. These tools were adapted to the specific requirements of protein crystallization, in particular for measurements on microsamples (down to the microlitre-scale for DLS) [216] or on small and fragile crystals. The fact that some of these tools are used by the practitioners of crystallization in structural biology laboratories is rewarding, especially with respect to DLS presently being widely used as a diagnostic tool regarding protein quality and crystallizability [217, 218].

Other offspring of the interplay between crystal science and technology were proposals followed by validations of new crystallization methods and an update of more conventional methods (Table 6). Most of the novelties exploit atypical parameters (well known for the crystallization of conventional molecules but not yet explicitely assayed with proteins, such as temperature, pressure, stirring) or are based on emerging new technologies (e.g. the femtosecond laser and nanotechnologies). Here also, it is satisfying that practitioners became progressively convinced of the usefulness of several of the new crystallization methods. This is especially the case for counter-diffusion, and gel and microfluidic based-methods (partly inspired by the microgravity programmes that created so many controversies among structural biologists), and is reflected by the increasing number of structures in the PDB solved from crystals obtained by these methods.

Practical crystallogenesis

This section discusses the changes in the practice of protein crystallization in the period between 1990 and 2013 and shows how the knowledge gained from basic research has benefited structural genomics. Different topics developed synergically (e.g. purity, screening, structure engineering, high-throughput and automation, nanotechnology-based methods, optimization) but, for simplicity, they are discussed separately. Taken together, a series of strategies for facilitating and/or enhancing protein crystallization could be defined (Table 7) and were succesfully employed.

Table 7. A large and diversified panel of crystallization strategies. For references, see text and Data S1.
Early quotation and yearStrategyProof-of-concept (year)
1971: Creating more compact/less flexible structures
Limited proteolysis1971
Removal of floppy protein extensions or fragmentation in structural modules1994
Chemically synthesized RNA domains1995
1973: Protein as a variable
Various methods using homologous proteins with potential better crystallizability (e.g. from thermophiles, etc.); screening alternate intrinsic protein characteristics1973
1981: Optimization
By seeding procedures1981
By automation1990
By controlled drop size variations2001
By controlled temperature variations2005
By solubility screening2005
By advanced DLS methods (e.g. aggregate size, drop volume., etc.)2008
By Thermofluor method (estimation of protein thermal stability)2011
1983: Cocrystallization with chaperones for soluble and membrane protein crystallization
Antibody-assisted (antibody fragments selected by hybridoma or phage display)1983
Ankyrin-assisted (ankyrins selected by ribosome display)2004
1988: Robotics/automation
Use of laboratory robotics to help crystallization1987/8
First dedicated system for microbatch crystallization1990
Automation of all procedures, in particular for high-throughput crystallogenesis2000
1991: Sparse-matrix sampling
A plethora of commercially available crystallization screens1991
1991: Protein engineering
Site-directed mutagenesis for modifying structural properties (e.g. stability)1991/2
Surface entropy reduction mutagenesis2001
Site-directed mutagenesis for modifying physical properties (e.g. solubility)2012
1992: Uncoupling nucleation and growth
Attempts to control nucleation by a variety of novel methods1992
1994: Cocrystallization with chaperones for RNA crystallization
With designed protein modules1994
With designed RNA module1998
With antibodies2011
1998: Heterologous cocrytallization
Easier crystallization if partners originate from different organisms1991
For RNA structure determination1998
Predicting likelihood of crystallization

Identifying the crystallization conditions of a protein target can be challenging and explains why researchers have tried to relate sample properties with crystallizability. This was achieved by exploring the vast ensemble of data available on macromolular structures and crystallization features. Although predicting exact crystallization condition remains a dream, important guidelines for practitioners originated from these studies. Thus, predictors of crystallizability were proposed, with the most emblematic being the second virial coefficient B22 characterizing undersaturated protein solutions [217, 218]. Also of potential utility is the kinetic coefficient ζc, a predictor reflecting competition between protein volume transport and protein surface integration within single crystals or amorphous aggregates [142]. Other predictors of crystallization likelihood are based on sequence features and intrinsic physico-chemical properties of the target proteins (pI, melting temperature, hydrophobicity, flexibility, etc.) [219-222] or on an analysis of experimentally characterized phase diagrams [223]. For example, analysis of crystallization data in the PDB revealed a significant relationship between the calculated pI of successfully crystallized proteins and the reported pH at which they were crystallized, thus providing information for the optimal choice of range and distribution of the pH sampling in crystallization trials [220]. Another analysis of the PDB indicated that protein crystals favour some space groups over others and suggested that symmetric proteins, such as homodimers, would crystallize more readily on average than asymmetric monomeric proteins [224]. This idea was validated experimentally and led to the crystallization of bacteriophage T4 lysozyme after creating by mutagenesis an artificial homodimer [225]. A recent attractive tool for crystallization prediction combines experimentally characterized physico-chemical features and sequence-derived data from target proteins [226]. Note that most criteria of crystallizability are correlated with the fact that the target should have an enhanced structural stability, as amply confirmed by many successful crystallization projects based on this idea (see below).

Biotechnological tools for macromolecule purification and crystallization purposes

It is common sense, although not always taken into consideration, that the macromolecule itself is an important parameter, if not the most important one, for crystallization, as explicitly discussed for proteins [227] and nucleic acids [228]. This emphasizes the importance of purification and macromolecule modification procedures in crystallogenesis. In the protein field, advances towards efficient protein expression and purification for crystallography are well covered in the literature [229-231]. However, although DNA recombinant methods present many advantages, a few drawbacks detrimental to crystallization should be noted, such as uncontrolled overexpression leading to inclusion bodies (particles containing protein aggregates), precipitated and/or denatured proteins, proteolytic degradations, incomplete post-translational modifications, and so on. Solving these problems can be time-consuming and costly. Lowering the overexpression level represents a possible remedy that decreases the amount of inclusion bodies. The alternate technology that eliminates most of these drawbacks is cell-free in vitro protein synthesis. Presently, only a few crystallography groups have employed this technology to prepare soluble proteins [232, 233] and, recently, membrane proteins [234]. In the case of RNAs as well, specific drawbacks have to be overcome for the preparation of homogeneous samples for crystallization. They rely on the structural and conformational diversity of RNA molecules and their susceptibility to enzymatic or chemical hydrolytic cleavages. Although pure RNA samples can be prepared from crude biological material, enzymatic and chemical synthesis are presently favoured and in case of large RNAs, enzymatic synthesis using T7 RNA polymerase is the only possible technology [235]. Note that, for some RNAs, annealing methods are required to assume conformational homogeneity [236].

Many methods were used for engineering protein or nucleic acid variants with enhanced structural stability favouring crystallization [14]. Limited proteolysis is probably the simplest one, as already employed in the 1970s [46] and recently rejuvenated in a version where trace amounts of protease are added/seeded in situ to crystallization assays [237, 238]. Other methods that aim to refine physico-chemical properties of proteins were used to specifically favour packing contacts. They consist of changing surface residues on the targets, either by DNA recombinant technology [205, 239, 240], or by chemical modification [241], in particular by reductive methylation of lysine residues [242]. These new methods are based on important precursory observations, such as the change of a single amino acid that created a packing contact enabling the crystallization of a human ferritin [243], the application of the concept of entropy-driven crystal growth of proteins [205], or the idea that intermolecular contacts can favour or disfavour crystallization and therefore should be created or eliminated. Also of great potential is the DARPin technology based on the natural ankyrin repeat-protein fold with randomized surface residues allowing specific binding to virtually any target protein [244, 245], thus allowing chaperone-assisted crystallization.

Producing stable homogenous samples of membrane proteins for crystallization is particularly challenging and, as for soluble proteins, screening large numbers of target proteins is common practice. A new strategy has recently been proposed that involves the use of green fluorescent protein fusion constructs and screening procedures based on expression level, detergent solubilization yield and homogeneity, as determined by high-throughput and automated chromatographies [246]. Notably, antibody-assisted crystallization, introduced by 1983 for soluble proteins [247], applies also to membrane proteins [248].

Screening crystallization parameters

The idea of using condition screens for the crystallization of proteins was proposed in 1991 with the sparse-matrix method [249]. In its original version, the method used a set of 50 conditions statistically chosen in a crystallization database to screen the crystallization of a target protein. After validation of the method, a rapid release of new screens was observed, as illustrated by screens of general use based, for example, on alternate polymeric crystallants [250] and screens specifically designed, for example, for RNAs [235, 251], protein assemblies [252] or membrane proteins [253, 254]. Today, a large panoply of crystallization kits is available, either for initial screening or for optimization [255]. However, many screens are redundant and making a good choice can be delicate, especially for challenging projects when the amount of macromolecular entities is limited and the number of required trials before success is large. A new database for the comparison of crystallization screens could be useful for a rational choice of the adequate screen [256].

Because the parameter-space for crystallization is quasi-unlimited, there was always a quest to find new compounds that sustain or improve protein crystallization. This quest was pursued in the period 1990–2013 and led to the identification of several classes of new crystallants, such as Jeffamines, ionic liquids, poloxamers, polysaccharides and other polymers commercially available, as well as of new detergents. Among them (Table 3), ionic liquids are particularly appealing because of the many potential interactions that they may establish with proteins. Thus, in a precursory work on lysozyme crystallization published in 1999, it was suggested that the liquid organic salt ethylammonium nitrate could be of interest for protein crystallography [257]. It took several years and more systematic crystallization studies, however, before the concept could be firmly established [258-260]. On the other hand, the catalogue of additives that can be of potential use in crystallization trials constantly enlarges [255], as well as the possible buffers and salt combinations. This creates a huge combinatorial diversity of crystallization conditions that will even augment if the parameter temperature is included in the screens. Several condition-screening strategies aiming to restrict the number of trials either consist of the use of mixes of properly chosen crystallants and/or additives [255] or optimization of the choice of additives or the buffer formulation by calorimetric approaches [206, 261]. Also of practical interest are the positive effects on crystallization of heterogeneous nucleants introduced on purpose in crystallization experiments, particularly fragments of hairs [151], which have their efficacy enhanced when included in sparse-matrix or high-throughput screens [262, 263].

It should be noted that most compounds within the screens (except salts) were found empirically and that their mechanisms of action are not well understood, especially for the small additive molecules. This is not satisfactory and does not facilitate the design of efficient new screens. A few recent dedicated studies have provided some answers with respect to this issue. A first case study investigated the thermodynamic effects of acetone on insulin crystallization and concluded that acetone displaces water molecules on the surface of the insulin molecules [204]. Another additive widely used in protein and nucleic acid crystallization, 2-methyl-2,4-pentanediol, was found bound to proteins in many crystal structures. Similarly, it could be concluded that binding is accompanied by the displacement of water molecules and promotes stabilization of the protein molecules, thereby enhancing crystallizability [264]. Interestingly, the calorimetric approaches, discussed above, arrived at the same conclusion [206, 261]. This is in line with the working hypothesis tested by Alex McPherson, according to which additives promote crystallization by enhancing intermolecular contacts between proteins or by removing such contacts between proteins or solvent [241].

Purity and impurities

Purity was a hot topic all along the history of biocrystallogenesis [94] and its versatile importance is constantly emphasized by new publications [265-268]. For example, commercial HEW lysozyme used in nucleation studies was shown to contain significant populations of large pre-assembled lysozyme clusters that result in a deterioration of the quality of macroscopic crystals [265]. At the other extreme, lipidic cubic phase-based crystallizations appear to be more robust than crystallizations conducted in more classical detergent environments because up to 50% of impurities are tolerated in the case of the R. sphaeroides photosynthetic reaction centre crystals grown in cubic phases [267]. Similarly, it was shown that highly contaminated samples of a recombinant Eco protein yielded reproducibly crystals diffracting at high resolution [266]. Given such data, it is understandable that bulk crystallization from impure protein batches remains an open issue [268].

From the standpoint of applications, it would be important to understand how impurities can exert either detrimental or beneficial effects on crystal growth. When studying the surface morphology of Bence-Jones protein crystals, it was shown that impurities adsorbed on the crystalline surface form an impurity adsorption layer that prevents further growth of the crystal: by growth–dissolution–growth cycles, impurities can be removed and growth can resume [269]. In another study, the role of the rate of supersaturation was highlighted. Thus, when impurity adsorption on crystal surface is delayed, crystal growth is enhanced and a ‘purifying’ effect takes place. By contrast, when impurity desorption is delayed, crystal ‘poisoning’ occurs [270]. This would imply that vibrations, stirring or forced flow during crystallization [271] could protect from detrimental impurity effects.

Automation and high-throughput

Robotic crystallization systems are efficient, tireless and accurate, and can carry out experiments using drop samples of very small volume (1 μL in most cases, nanolitres in some). They can perform enormous numbers of trials using remarkably small amounts of biological sample. Many of the robotic systems reproduce procedures currently used for manual experiments, such as sitting and hanging drops. They are affordable and well implemented in academic laboratories [272]. In recent years, and as boosted by the large Structural Genomics Consortia and Platforms, entire integrated systems have been developed to accelerate all steps of the crystallization process. Besides automation of the crystallization trials and their monitoring, screening of recombinant protein expression [273], protein purification for crystallization [274], protein stability [275], image analysis [276], seeding [277] and other optimization procedures [278], ligand soaking [279], crystal harvesting [280], and crystal mounting [281] have also been automated. Moreover, integrated systems have been installed near to synchrotron sources enabling in situ diffraction analyses [282]. In summary, automated crystallization by sparse-matrix methods and screening techniques to optimize protein homogeneity and crystal quality improved dramatically and revolutionized the crystallogenesis field in the last decade.

It was noted, already one decade ago, that high-throughput screening of crystallization conditions does not necessarily produce reproducible results when carried out in different laboratories, demonstrating that some important features before crystallization trials are not under control [283]. This explains the recent efforts aiming to automatize and standardize the preparation and handling of samples. It would also be timely to share worldwide the huge amount of data generated by the automated high-throughput crystallization systems with the objective of extracting useful predictive information. Being aware of this need, a group a structural biologists and bioinformaticians convened to develop a crystallization ontology [284].

Towards nanocrystallogenesis

Scaling-down of crystallization methods was a continuing goal both for practical and theoretical reasons, aiming at low sample consumption and especially for the provision of growth conditions favouring crystal quality. The challenge was to invent miniaturized crystallization devices based either on conventional methods (batch, vapour-diffusion, etc.) or on alternative methods that were shown to favour crystal quality (counter-diffusion, under stirring, etc.) [197, 285, 286]. A breakthrough was the demonstration in 2002 of the feasibility of growing protein crystals in volumes as small as 1 nL [287]. The same year saw also the entry of the microfluidic technology in the protein crystallization field [189]. At the same time, synchrotron technologies made significant advances (see below) and offered the possibility of collecting diffraction data on small crystals [288].

The first microfluidic chip on the market was based on the free interface diffusion technique [189]. It consists of a complex integrated fluidic circuit including two networks of channels: one for liquid handling and a second serving as actuation valves. The chip was dedicated to high-throughput screening and was designed to test 48 crystallization conditions with < 10 μL of sample solution in total. This chip was modified for the establishment of precipitation diagrams useful for crystallization screening [289] and for fine tuning supersaturation in combining free interface diffusion with vapour-diffusion [290]. A great step in miniaturization was the possibility to generate complex mixtures of reagents in 5-nL reactors [289].

Batch crystallization was implemented in a microfluidic system in 2003 [291]. In this case, the chip design was extremely simple and consisted of inlets for protein, buffer and crystallant solutions, and a microfluidic channel in which 10-nL droplets are prepared by mixing these solutions in various ratios. This device allows formulation of thousands of nanodrops, which are carried by a flow of inert oil. The nanodrops can be stored on the chip and the crystals appearing therein can be easily analyzed by X-ray diffraction [292]. Based on the nanodrop approach, a more complex system was designed for basic research purposes. It is able to formulate droplets and to flow them to storage chambers where they can be concentrated or diluted by water permeation through the chamber walls. This PhaseChip was designed to establish phase diagrams with total control over supersaturation, nucleation and growth kinetics in each individual drop [293]. This technology evolved for measuring nucleation rates [294] and for manipulating temperature and concentrations in phase diagrams [295].

Subsequently, counter-diffusion features were successfully reproduced in microchannels with the production of crystalline material ranging from single crystals to larger monocrystals along the supersaturation gradient. When made of the appropriate polymer material, these counter-diffusion chips allow direct on-chip characterization of the crystals by X-ray diffraction, without any further (and potentially deleterious) sample handling [296, 297].

Another microfluidic technology designated ‘Microcapillary Protein Crystallization System’ enables nanolitre-volume screening of crystallization conditions and in situ X-ray diffraction studies [298]. The latest released method is based on controlled evaporation in the microfluidic device [299]. The many advantages of the microfluidic chips explain why the microfluidic technology has become a popular and affordable tool for various applications, such as condition screening, optimization, X-ray analysis and basic crystallogenesis research.

Laser technologies have also been miniaturized and new laser-based tools for crystal processing have recently been validated with HEW lysozyme crystals grown in semi-solid agarose gel and generalized for other crystals. Processing is carried out using a focused femtosecond laser, enabling the preparation of small well cut crystal fragments that are not damaged by the laser irradiation and are suitable for X-ray analysis [215]. Such protein microcrystals can be handled by micromethods [281] and can be used for X-ray studies by synchrotron microbeam technology [288].

Optimizing protein crystallization methods and crystals

Fabrication of protein crystals suitable for diffraction studies almost always requires optimization of the initial crystallization conditions. Seeding is probably the oldest optimization procedure, as already practiced in the 1980s [108], and has subsequently been constantly improved. Seeding techniques (either homogeneous or heterogeneous cross-seeding with seeds originating from a different protein) fall into two categories that employ either macroseeds [108] or microcrystals as seeds [300, 301]. In both cases, the solution to be seeded should be only slightly supersaturated so that controlled growth can occur. Several microseeding methods have been employed, such as streak-seeding developed by Enrico Stura in the 1990s [300], and recently automated [302], as well as a microseed matrix screening method [303], as also automated [304]. Most recent developments concern, for example, the adaptation of seeding methods to nanocrystallization [305] and the preparation of single microseeds by femtosecond laser ablation [306].

Most steps and variables in the crystallization process can be optimized [307, 308] (Table 7). For example, this concerns the choice of the crystallization method. Thus, changing from standard methods to the counter-diffusion technique improved the crystal of the core complex of a hydrophobic plant photosystem [309]. It also concerns the choice of the best temperature and pH when screening, as well as the size and volume of the crystallizing samples. Accordingly, temperature cycling [310] and pH optimization [311] strategies have been proposed that were shown to increase the possibility of obtaining crystals. Another optimization technology keeps the crystallization solution metastable during the growth process by controlled temperature variation of the crystallization solution [312]. Similarly, it was observed that ultrasound can optimize nucleation by decreasing the energy barrier for crystal formation [313]. Furthermore, Thermofluor-based high-throughput screening methods can be employed to optimize protein sample homogeneity, stability and solubility [275].

Optimization also concerns crystal quality, post-crystallization treatment for enhancing diffraction quality [314] and crystal size. For a long time, the production of crystals of a sufficient size and quality proved to be a bottleneck in structural investigations. Although techniques for screening crystals have improved dramatically, the methods for obtaining large crystals have progressed more slowly. Despite many structures were solved from small crystals with synchrotron radiation, it is far easier to solve and refine structures when robust data are recorded from larger crystals. In an effort to improve the size of crystals, a strategy for a small-scale batch method has been developed, which, in many cases, yields far larger crystals than attainable by vapour-diffusion [315]. Large crystals are required for neutron crystallography and, for that purpose, the crystal growth technique based on temperature variations is particularly appropriate [312]. It has been applied to grow high-quality large crystals of several proteins of interest, which, in the case of A. flavus urate oxidase, yielded neutron diffraction data in the range 1.9–2.5 Å [316].

Another paradigm change occurred with the advent of sophisticated X-ray optics, ultrasensitive detectors and microbeams at new-generation synchrotron sources [288]. Similar to microfluidic systems, this will revolutionize the practice of structural biology, with the consequence that large crystals are no longer a prerequisite in X-ray crystallography. Thus, crystals as small as 20 μm3, corresponding to not more than 2 × 108 unit cells, can yield usable diffraction data [288]. The same trend, although less extreme, occurred in neutron crystallography. In that case, a crystal of 0.15 mm3 of perdeuterated human aldose reductase yielded a structure at 2.2 Å resolution [317]. Importantly, from the viewpoint of structural biology, smaller crystals are potentially of enhanced quality (see ‘Fundamental crystallogenesis’).

An overall picture of crystallization strategies and their outputs for biology

Although much remains unclear, the ever deeper knowledge on crystallization has generated more rational strategies to produce protein crystals and to improve their diffraction quality. These strategies are diverse (Table 7) and have contributed to solving many bottlenecks in crystallization projects. They illustrate how the field of biocrystallogenesis has evolved in the last 50 years from mainly empirical methods to sophisticated trial-and-error strategies, as well as to idea- and basic science-driven methods that slowly infiltrate structural biology laboratories. Their number (and the accompanying crystallization methods) (Table 6) augmented progressively from 1971 until 2013, with a significant boost in the last decade.

Early strategies were based on understanding and modifying global structural properties of proteins in view of efficient crystallization; in other words, they considered the protein, as such, as a parameter affecting crystallization. Thus, simplified and more compact architectures obtained by proteolysis or genetic engineering, or stabilized by the addition of different types of structural chaperones, such as antibodies, ankyrins or macromolecular natural or designed ligands, showed enhanced crystallizability. This applies to all types of proteins, including membrane proteins, as well as RNAs. In that case, the chaperone can be a general RNA module [318, 319] or a protein [320-322]. Interestingly, this allowed the opportunity to crystallize biologically significant RNA:protein complexes [320, 322-324].

Sparse-matrix sampling combined with robotics (introduced in the 1990s) played an essential role in allowing quicker experiments and providing better reproducibility. Strategies for controlling the physical chemistry of crystallization were also of prime importance. They concern uncoupling nucleation and growth and procedures for optimizing crystallization. As an example, the screening space of crystallization in vapour-diffusion methods can be reduced by controlling water equilibration, protein solubility and drop preparation [325]. On the other hand, macromolecular engineering employed to modify physical properties of proteins that affect solubility or favour crystal packing allowed many difficult crystallization problems to be solved.

In case of difficulties in crystallizing an essential protein from a given organism, switching to another organism or, in a more systematic way, screening orthologues is one remedy [326]. This strategy has already been employed for proteins [50] and the ribosome [93], considering the relative ease of crystallizing macromolecules from extremophiles, and has been generalized in a screening procedure of orthologues [327]. Strategies to optimize crystallization can take advantage of the large panoply of available crystallization methods (Table 6). To date, this potential has only been partly explored, if not ignored, by practitioners of crystallization, especially hybrid methods combining, for example, crystallization in gel and laser pulses to induce nucleation or methods based on stirring or vibrations (although vibrations have likely induced many uncontrolled crystallizations in the past). Similarly, new devices allowing the growth of protein crystals in gradient magnetic fields [328] or assisting with protein crystallization electrochemically [329] await more thorough testing by practitioners. Note also an alternative approach of crystallization, orthogonal to current approaches, developed by Alex McPherson and colleagues, with the objective of doubling current success rates [330]. It is based on the hypothesis that many conventional small molecules, including new crystallants, might establish stabilizing, intermolecular and noncovalent cross-links in crystals.

Summarizing, it is rewarding to note that the crytallization toolbox of diagnostic tools, methods and strategies at the disposal of structural biologists (Tables 5-7) led to the structure determination of a variety of important proteins and macromolecular assemblies (Table 8). Note that many of these successes are based on recently developed crystallization strategies, such as the femtosecond laser technique [331], microfluidics [332], the crystallization of complexes with specific cross-links [333] or hybrid methods [334].

Table 8. Examples of emblematic crystallizations based on fundamental or practical advances in protein crystal growth, that led to structure determination. For references, see text and Data S10.
YearBiomacromolecular particleCrystallization strategy
1968S. cerevisiae and E. coli pure native tRNA speciesConventional method and/or first use of vapour-diffusion (organic solvents as crystallants)
1980S. cerevisiae AspRS:tRNAAsp complexAmmonium sulfate as crystallant (most salts disrupt protein:RNA complexes)
1980B. stearothermophilus ribosome (large subunit)Homologous crystallization (crystallizability of ribosome from thermophiles better than from mesophiles)
1982R. viridis photosynthetic reaction centre (first crystallzation of a membrane embedded assembly)Ammonium sulfate (crystallant), N,N-dimethyl dodecylamine N-oxide (detergent) and heptane-1,2,3-triol (additive)
1983Lysozyme in complex with a monoclonal anti-lysozyme antibodyCocrystallization with antibodies
1991Human ferritinEngineering crystal contacts by analogy with homologous rat ferritin
1994S. cerevisiae RNA polymerase IIEpitaxial growth on 2D crystals on positively-charged lipid layers
1994Beef mitochondrial cytochrome b-c1 complexCrystals grown in agarose gel
1998Group II intron domain 5–6 and hepatitis delta virus ribozyme RNA constructsCocrystallization with designed RNA motif
1999Complex between Eco tRNACys and Thermus aquaticus elongation factorHeterologous cocrystallization with partners from two organisms (access to targeted structure)
2001Human RhoGDI (cytosolic regulator of GTPases)Protein surface entropy reduction
2002T. daniellii thaumatinHybrid method combining microgravity and gel (data collected up to 1.2 Å resolution at room temperature)
2004Human aldose reductaseFine biochemistry; crystallization with cofactor and inhibitor (crystals diffracting at ultrahigh 0.66 Å resolution)
2004Maltose binding protein and two eukaryotic kinasesCocrystallization with an ankyrin repeat protein
2006Complex of tRNAGlu and MnmA (an enzyme that synthesizes 2-thioU at the wobble position of certain tRNAs)Femtosecond laser technique
2006R. viridis photosynthetic reaction centre (proof-of-concept experiment)Droplet-based microfluidic batch (at present ≥ 14 solved structures in PDB)
2006Human glutamate carboxypeptidase II (a large glycosylated)Fine biochemistry; heterologous overexpression
2011Complex of human gankyrin and C-terminal domain of S6 proteasomal proteinCrystallization of a specific photo cross-linked complex (via incorporation of a photophore by genetic code expansion)
2012Thermococcus thioreducens pyrophosphataseCounter-diffusion for neutron crystallography (Hughes RC, Coates C, Garcia-Ruiz J-M, Blakely M & Ng JD)
2013Human epidermal growth factor receptor (an apo cancer-associated mutant)Hybrid method: microgravity and counter-diffusion (in JAXA Crystallization Box)
2013Decameric bacterial SelA:tRNASec ring structure with heterologous tRNAChoice of the best bacterial orthologue (Aquifex aealicus) and heterologous cocrystallization
Protein crystals outside crystallography and structural biology

Although not the subject of the present review, it is important to note a few applications where knowledge of protein crystal growth was crucial. This is, for example, the case with respect to the design of protein-based biosensors [335]. It also concerns crystallization for protein purification [336] or safe protein storage for pharmacological formulations (e.g. the emblematic example of crystalline insulin) [337]. In this context, it is worth noting that nature uses this strategy under certain circumstances to protect macromolecules against degradation, as is the case of ribosome crystals found in hibernating animals [338]. More generally, protein crystals, spherulite-like aggregates and fibres have been found in vivo in many organisms [339, 340], including in the human body where they are often associated with severe pathologies, such as Alzheimer's and Parkinson's diseases [340]. Preventing or inhibiting their formation could therefore have therapeutic applications. This idea is being explored with respect to the rational inhibition of amyloid fibril formation [341]. On the other hand, it is known that antibodies can be raised against protein or small molecule crystals [342]. This opens the possibility of medical applications, such as for the diagnostic of crystal-based diseases (gout, Alzheimer's disease, etc.).

The status in 2013 and perspectives

  1. Top of page
  2. Abstract
  3. Introduction
  4. The time of physiology and chemistry (1840–1934)
  5. The birth of biocrystallogenesis as a science (1934–1990)
  6. Crystallogenesis in the era of technologies and structural genomics (1990–2013)
  7. The status in 2013 and perspectives
  8. Acknowledgements
  9. References
  10. Supporting Information

In 2013, biocrystallogenesis is a mature science based on strong interdisciplinarity between biology, physics, chemistry and associated technologies. Today, the physics and chemistry of protein crystallization are globally known, although some aspects remain elusive, such as an understanding of the growth of undesirable protein spherulites in crystallization trials, as apparently favoured by heterogeneous nucleation [343]. It is expected that the current studies on model proteins will contibute to finding methods preventing their formation [339, 342]. Predicting the likelihood of crystallization as well has made progress, although uncertainties still remain in crystallization experiments. Thus, despite a solid fundamental background, many protein crystals continue to be obtained by trial-and-error strategies. However, rational-based crystallization methods, such as counter-diffusion, as well as crystallization in gels or nanocrystallogenesis, are slowly being adopted by crystal growers, and a few others await more systematic testing, such as stirring methods.

On the other hand, in 2013, the PDB contains ~ 82 000 macromolecular structures solved by X-ray crystallography (but only 64 structures determined and/or refined using neutron diffraction data), representing a large panel of proteins originating from throughout the tree of life. This could mean that the bottleneck of crystallization is solved such that, in the future, protein crystallization will be straightforward. However, this is not true for three main reasons. First, the majority of these structures correspond to soluble proteins and there is a dramatic lack of membrane protein structures, which are predicted to represent approximately half of the proteome. Second, the presently solved RNA structures represent only 3% of the total and the crystallography of lipids is quasi-inexistent. An increasing awareness of the importance of RNA and lipids in biology requires a much better knowledge of their structures. Third, biologists are becoming more and more ambitious and want to know ever more intricate and larger macromolecular structures and assemblies; they especially want to comprehend the plasticity and dynamics of proteins and are even more ambitious regarding macromolecular machines. In addition, there will always be a need for structures solved at high and ultrahigh resolution.

Given this situation, one can anticipate further developments in the crystallogenesis of membrane proteins [109] and lipids [344], RNAs either free or in complex with proteins [345] and glycoproteins [346]. Improving crystallization methods and their application to ambitious biological problems will continue to be at the forefront of research (e.g. the gel method) [346-348]. This also concerns crystallization on solid nanotemplates [349] and other advanced nanocrystallogenesis methods [350]. Studying crystal polymorphs should also be pursued and could enable better access to the structural plasticity of macromolecules, and also erase possible artefacts resulting from packing effects [351].

From a more global perspective, concepts of macromolecular crowding and macromolecular confinement both in vitro and in vivo [352], should enter the field of biocrystallogenesis. Thus, one could question the actual physico-chemical properties of concentrated protein solutions in nanodrops and crystallizability (enhanced or inhibited) in crowded media. Being able to answer such questions could foster applications for more controlled protein crystallization and, importantly, shed light on in vivo protein crystallizations and their relation with pathologies. Moreover, in vivo-grown crystals could be usable for the emerging technology of free-electron laser-based serial femtosecond crystallography [339]. In conclusion, an exciting future is expected and it is anticipated that the interplay between science and technology will continue in the science of biocrystallogenesis [7].


  1. Top of page
  2. Abstract
  3. Introduction
  4. The time of physiology and chemistry (1840–1934)
  5. The birth of biocrystallogenesis as a science (1934–1990)
  6. Crystallogenesis in the era of technologies and structural genomics (1990–2013)
  7. The status in 2013 and perspectives
  8. Acknowledgements
  9. References
  10. Supporting Information

This text is based on lectures given at the FEBS Practical Courses on ‘Advanced methods in macromolecular crystallization’, held in Nove Hrady (Czech Republic) in 2004–2012. It is written to acknowledge FEBS with respect to its support for the field of biocrystallogenesis, which started in 1987 with a FEBS Lecture Course: namely ICCBM2, in Bischenberg (France). Warm thanks are extended to Ivana Kuta Smatanova, Pavlina Rezacova and Rolf Hilgenfeld, the organizers of the Nove Hrady Courses, to all my students and coworkers from Strasbourg, past and present, and to all my colleagues from the biocrystallogenesis and structural biology communities for the exchange of ideas and knowledge over the last 40 years. During all of this period, the support of CNRS and the University of Strasbourg was essential.


  1. Top of page
  2. Abstract
  3. Introduction
  4. The time of physiology and chemistry (1840–1934)
  5. The birth of biocrystallogenesis as a science (1934–1990)
  6. Crystallogenesis in the era of technologies and structural genomics (1990–2013)
  7. The status in 2013 and perspectives
  8. Acknowledgements
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. The time of physiology and chemistry (1840–1934)
  5. The birth of biocrystallogenesis as a science (1934–1990)
  6. Crystallogenesis in the era of technologies and structural genomics (1990–2013)
  7. The status in 2013 and perspectives
  8. Acknowledgements
  9. References
  10. Supporting Information

Data S1 to S10. Complete bibliography for Table 1–8 and Figs 1–4 : references [353–488] are supplemental.

Data S11. Contains additional bibliography and comments on ‘Books, historical accounts & reviews on crystal growth’, on ‘ICCBM proceedings’, and on ‘Specialized reviews & research articles’.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.