Protein Structure Factory - The Berlin "Protein Structure

	1.Challenges to structural biology in the genome aera
	2.The "protein structure factory": an integrative approach
	3.Conclusions
	4.Acknowledgement
	5.References

1. CHALLENGES TO STRUCTURAL BIOLOGY IN THE GENOME AERA

In the genome aera, the challenge to structural biologists is defined as follows: To determine the three-dimensional structures of a representative set of proteins such that all further studies of protein function, e.g. in a medical-pharmacological context, may be carried out on a firm structural basis. This challenge cannot be met in the conventional way whereby a protein crystallographer or an NMR spectroscopist applies her or his sophisticated methods to the study of that single protein structure that seems the most interesting at the time. For sure, this approach has been tremendously successful over the last decade, filling the Protein Data Bank at an ever increasing speed with structures of ever increasing beauty, complexity and biological relevance1. However, in the light of the above challenge, an all-out approach to structure determination is needed in much the same way as it was and is very successfully applied to genome research.
This approach has become known as "structural genomics".

1.1. Structural genomics
The term "structural genomics" has been in use for quite some time, but has acquired a completely new meaning very recently. Traditionally, it represented an effort to characterize the (physical) structure of a complete genome by gene mapping and sequencing ². Now, it stands for initiatives inspired by the genome sequencing projects that aim at the determination of three-dimensional protein structures in a systematic way3-6. The approaches taken towards this goal fall into two broad categories:
(1) In the first, the emphasis is on determining the structures of a set of proteins or protein domains which would yield a complete representation of all protein (domain) folds present in the biosphere. This approach is based on the notion that the number of folding types (folds) for globular protein domains is not unlimited ^7-9. Very probably, it does not exceed the number of structure entries now present in the Protein Data Bank. One may therefore hope to cover the complete universe of three-dimensional protein structures within a few years, provided that it is possible to identify new folds from protein sequence. Computer-based methods for fold recognition are currently being developed in a number of laboratories ^10-12. In a small bacterial genome, fold assignment with high confidence is possible only for a small subset of coding sequences¹³. However, advances in biocomputing methodology are likely to improve the success rate in the near future¹⁴. A convenient route towards fast structure determination targets proteins from hyperthermophilic bacteria or archaea, because they can be easily purified from recombinant Escherichia coli cells and lend themselves especially to crystallization or NMR structure determination. A number of crystal structures of these proteins has already been determined ^15-17. The knowledge of a representative set of protein domain structures is hoped to enable the complete fold prediction for newly sequenced genomes by homology modelling. The availability of the predicted tertiary folds for most proteins in a genome would in itself be of enormous value for many fields of biological research. In addition, it may considerably facilitate the detailed structure determination by protein crystallography and NMR spectroscopy of those proteins for which this is deemed necessary.

(2) A second approach to structural genomics focusses on structure analysis methodology. Here, the main idea is to closely cooperate with and learn from the genome sequencing projects. The use of the wide variety of available coding sequences and efforts towards parallelisation and automation of structure analysis are unifying features of this approach. As before, bioinformatics will play an important role in this brand of structural genomics for the identification of relevant proteins or protein domains that are amenable to structure analysis. The RIKEN NMR structure determination project ¹⁸ exemplifies the technology-oriented structural genomics efforts by attempting to establish a facility for the broad-scale analysis of three-dimensional protein structures in solution. The Berlin "Protein Structure Factory" initiative belongs into the same category of structural genomics. However, by employing both X-ray diffraction and NMR methods it does not rely on one structure analysis technique exclusively. A main ingredient of the Protein Structure Factory is the close collaboration with the German Human Genome Project (DHGP).
Common to all structural genomics initiatives are efforts to identify and eliminate bottlenecks in the structure determination process. For example, it is generally agreed that the availability of bright synchrotron beamlines is a prerequisite for the successful use of diffraction methods ¹⁹. Membrane proteins, constituting up to 30% of the protein inventory of an organism and against which more than 50% of the currently used and tested drugs are targetted, represent the most persistent bottleneck for all analytical methods, because they are only water-soluble in the presence of detergents and difficult to overproduce in quantities that are required for biophysical studies.

top

2. THE "PROTEIN STRUCTURE FACTORY": AN INTEGRATIVE APPROACH

The term "Protein Structure Factory" was chosen to represent a common initiative of the DHGP and structural biologists from the Berlin area aimed at the broad-scale analysis of proteins. The Protein Structure Factory will be established to characterize proteins encoded by the genes or cDNAs available at the Berlin Resource Center of DHGP. At a later stage, it may analyze various sets of input proteins selected by criteria of potential structural novelty or medical or biotechnological usefulness. It represents an integrative approach to structure analysis combining the computer-based analysis of genes by bioinformatics techniques, automated gene expression and purification of gene products, generation of a biophysical fingerprint of the proteins and the determination of their three-dimensional structures either in solution by NMR spectroscopy or in the crystalline state by X-ray diffraction. Here we briefly describe the main features of the planned Protein Structure Factory.

2.1. Bioinformatics

Bioinformatics has two main tasks in the Protein Structure Factory: To predict what can be done and to propose what should be done. Predicting what can be done is equivalent to identifying proteins that will permit their three-dimensional structures to be determined by X-ray crystallography or NMR spectroscopy. These proteins will have some properties in common. They will be soluble in aqueous buffers up to a critical concentration, they will have a defined globular structure, and this structure will be stable for at least as long as it takes to grow and expose crystals or to measure the NMR spectra. Proteins that contain long stretches of hydrophobic or charged amino-acid residues, have extended sequence repeats or use a limited repertoire of amino acids over long polypeptide segments often do not display these properties. However, they may still contain single or multiple domains that permit structure analysis. In addition, bioinformatics will provide valuable information aiding the structure determination by predicting sites of post-translational modification and identifying proteins of known, similar tertiary structure. Structural prediction will be used to decide whether a given protein will be studied by NMR spectroscopy or by X-ray diffraction or, for the latter case, whether its structure analysis will require experimental phase determination or can be based on a homologous model.

To propose what should be done is the more challenging task. It is equivalent to finding proteins with interesting properties such as novel folds or a function in biochemical pathways that may be associated with disease. The more interesting a protein appears, the more effort will have to be invested in its structure analysis. Computational tools for functional sequence assignments are currently being developed ²¹. This work addresses questions concerning the subcellular localization of proteins, their membership in families defined by function^22-24 and their involvement in pathological states ^25,26.

2.2. Automated gene expression

The method of choice to produce recombinant proteins for structural and biophysical studies is the heterologous expression of their genes in E. coli. Proteins that cannot be synthesized in E. coli may alternatively be made in Saccharomyces cerevisiae or Pichia pastoris. For structure analysis by X-ray diffraction methods, the methionine residues of many proteins will have to be replaced by selenomethionine. Likewise, NMR structure determination will often require that the proteins be labelled with ¹³C and/or ¹⁵N which can be introduced through cell growth on media containing these isotopes in the form of ¹³C glucose or ¹⁵NH₄Cl.
Within the Protein Structure Factory, gene expression systems will be obtained either by the cloning of PCR products or by the direct construction of cDNA libraries in expression vectors (expression libraries) ²⁷. Both techniques will rely on the automated manipulation of clones in multi-well microtiter plates or on high-density
membrane filters. Methods for the detection of protein coding or novel clones with antibodies directed against protein tags or by oligonucleotide fingerprinting are available^27,28.

2.3. Purification of tagged proteins

The concept of the Protein Structure Factory requires the high-throughput production of highly pure proteins in about 50 mg quantities for structure analysis by NMR spectroscopy and X-ray crystallography. This is accomplished in two production units for the parallel fermentation and online purification of recombinant organisms (one for E. coli and one for S. cerevisiae or P. pastoris). E. coli is the organism of first choice, since it can be cultivated easily and offers a large number of readily available expression systems. Genes exhibiting low expression in E. coli or yielding proteins which are produced as inclusion bodies are expressed in yeast.
The recombinant organisms will be cultivated synchronously in a battery of fermenters (Fig. 1). The cells from the different fermenters are homogenised successively with a high pressure homogeniser. The solubilised proteins are separated from the biomass by microfiltration and the processed filtrate is then concentrated by ultrafiltration. The following purification of the recombinant proteins takes advantage of two tags of these proteins: a his6-tag and a strep-tag ^29,30 whose corresponding DNA sequences are fused to the 3'- and 5'-terminus of the protein-coding gene. This allows a highly efficient separation of the recombinant protein from host cell protein. In the first step, the recombinant proteins are successively bound to a Ni-NTA column and eluted with imidazole. A second affinity chromatography on a streptavidin matrix is applied for the final purification. This semi-automated production and online purification will require two days for proteins synthesized in E. coli or three days for proteins from yeast. This production unit is designed to provide several homogeneous proteins for structure analysis per day.

top

2.4. A biophysical protein fingerprint

The goal of this unit is to characterize the proteins, as they become available from expression and purification, by conventional spectroscopic and calorimetric techniques. It will mainly serve to confirm and to complement the information obtained from biocomputing for further structure determination. The proteins will be analysed with respect to their secondary structure and stability, in dependence on temperature and pH.
The following techniques will be employed:

Fourier-transform infrared spectroscopy (FTIR), to obtain secondary structure information by analysing the amide bands, circular dichroism spectroscopy (CD), to confirm the data obtained by FTIR, fluorescence spectroscopy, to investigate stability as a function of pH, differential scanning calorimetry (DSC), to measure thermal stability.
Automated routines for the data acquisition and evaluation procedures will be necessary to keep pace with the expected throughput of proteins. In part, these routines are already available, some have to be developed.
In summary, this unit will furnish biophysical parameters concerning secondary structure and conformational stability of proteins, independent of and preliminary to the determination of high-resolution structures. It will help to establish experimental conditions for protein crystallization and NMR studies. The biophysical data may also be useful in those cases where high-resolution protein structures cannot be obtained.

2.5. NMR spectroscopy

The role of NMR will be in the structure determination of protein domains and of their functional complexes, and in the investigation of ligand binding to help in the design of bioactive small molecules. For this purpose, it is necessary to automate the key steps in the NMR structure determination procedure. These include data acquisition, sequence-specific resonance assignments and structure calculation. Currently, it takes weeks to months for the spectral assignments, especially those of the NOESY spectrum, to be accomplished. In order to be able to determine the structures for all three steps, some concepts and algorithms for automating the procedures exist, and more need to be developed.
Automated data acquisition is probably the easiest task in this project. It includes the definition of a data set which is suited for automated interpretation. Most modern NMR spectrometers already provide features which allow one to automate the data acquisition itself. The critical step for being able to determine the structures of a large number of proteins is in the necessary automation of the assignment procedure. To date, a number of computer programs for this purpose are available ³¹, but, in any case, manual interference is required. Most of these software packages will require peak lists obtained from the multi-dimensional spectra, which usually contain false peaks generated from noise or artifacts. The logics of the program are not then capable of handling this problem. In the context of the protein structure factory, it is required to generate a new piece of software which works directly on the spectra and is already able to recognize peaks, noise and artifacts as such. On the basis of a data set comprising CBCA N NH, CBCA (CO) NNH, HCCH-COSY, HCCH-TOCSY, and amino-acid-sensitive experiments, it is expected that the program will generate a list of chemical shifts comprising those of all protons, carbons and nitrogens present in the protein that can be used to evaluate the three-dimensional NOESY spectra.
This peak list is then subjected to an automated structure calculation protocol proposed by M. Nilges ³², which essentially allows one to assign the NOESY spectra during the structure calculation. In this manner, it is expected that approximately three months of manual work can be saved per structure. It is expected that the NMR structures of proteins with up to 120 amino acids can be solved routinely, if their solubility is high enough, and that sufficient signal-to-noise can be obtained in the 2- and 3-dimensional spectra. The protein structure factory also provides means to exploit the structural information generated. In this context, NMR spectroscopy will be used to study ligand-protein interactions in screening campaigns to detect binding in a site-specific manner. This information will be used to optimize ligands.

2.6. Protein crystallization

At present, the crystallization of proteins is still the bottleneck in the structure determination by means of X-ray diffraction. There is no simple correlation between properties of proteins and the large number of parameters that have to be considered during crystallization. Consequently, the crystallization of proteins is mostly an empirical process that requires a broad screening of different crystallization conditions. In the Protein Structure Factory, it is planned to have available a large number of purified proteins or protein domains per year that are considered for crystallization. Since a manual optimization of crystallization conditions on the projected scale is not feasible, the development and the utilization of a crystallization robot is a key issue of the crystal structure determination within the Protein Structure Factory. The necessary innovations will rely on two well established groups with ample experience in protein crystallography and in the construction of robots.
It is planned to build a crystallization robot that is pipetting protein solutions and a buffer screen consisting of about 100 different conditions (pH, buffer, salt, polyethylene glycol, alcohols, salts) for "hanging drop" vapor diffusion experiments ³³: a drop consisting of protein and buffer is equilibrated against the buffer at about twice the concentration, so that the protein solution in the drop is brought to supersaturation and eventually to crystallization. This is set up in trays with 24 wells, and the trays are automatically stored at two temperatures, preferably 4°C and 18°C. The robot examines the trays by light scattering to monitor aggregation of protein and, if possible, nucleation, and in later stages the trays will be observed by microscopes with suitable software to automatically recognize crystalline material.

2.7. Acquisition of X-ray diffraction data using synchrotron radiation

The use of synchrotron radiation will be crucial to the Protein Structure Factory: high brilliance and tuneable wavelengths are prerequisites for fast data collection, the use of small crystals and multiwavelength anomalous diffraction (MAD) phasing¹⁹. An example for a diffraction image obtained from a small crystal at a synchrotron is shown in Fig. 2. With the opening of BESSY II, direct access to a third-generation XUV storage ring source with excellent conditions is available nearby. However, to shift the maximum of the emitted spectrum towards the X-ray range, a high-field multipole wiggler has to be installed as has been done at other medium energy storage rings (ALS³⁴, MAX II³⁵, ELETTRA).
Two beamlines are planned within the Protein Structure Factory: the central beamline is optimized for rapidly measuring high resolution MAD data sets. This MAD beamline will be equipped with a focussing premirror, a double crystal monochromator and a refocussing mirror to serve in the wavelength range from 0.7Å to 2.75 Å which covers the absorption edges of all commonly used heavy atoms³⁶. To make use of the expected short exposure times a state-of-the-art CCD detector with fast bus and high capacity storage system will be installed at the MAD station. This will be especially useful in cases when fine slicing down to 0.1° is employed.
The other beamline is designed as a constant-energy station with a selectable wavelength around 0.9 Å and will be used for the fast checking of crystal quality and further preliminary examinations. It will accept radiation from the the side portion of the wiggler fan and will be equipped with a premirror and a bent crystal monochromator to select the appropriate wavelength and to focus and deflect the X-ray beam. Both stations will be equipped with gaseous nitrogen cooling and both need highly automated beamline control, efficient software protocols and organization schemes to act as high-throughput system.

top

2.8. Crystal structure determination

The high-throughput determination of three-dimensional protein structures based on the X-ray diffraction data collected at the synchrotron beamlines (see above) will have to employ robust and efficient methods at four essential steps: Phasing, model building, refinement and quality control. In some cases it will be possible to use homologous protein or domain structures for molecular replacement phasing. As the Protein Data Bank grows and the techniques for detecting homology at the level of three-dimensional structure improve, the frequency with which such search models are available will increase substantially. Crystal structures can be solved easily if the structural similarity of a search model is high enough.
The analysis of protein structures with unpredicted fold requires experimental phase determination. Once dreaded because of the tedious trial-and-error searching for isomorphous derivatives, phasing has become a routine process with the advent of MAD methods ³⁷. All proteins produced in recombinant E. coli can be labelled with heavy-atom markers in the form of selenomethionine and thus subjected to MAD phasing. The power of MAD phasing may be appreciated from Fig. 3 comparing the experimental electron density (from MAD) with the final, refined density in a portion of the structure of a bovine adrenoxin, Adx (4-108)³⁸. Here, the two iron atoms of the protein were sufficient for MAD phasing to produce density that not only clearly reveals the protein atoms around the C-terminus of Adx (4-108) but even some of the water molecules bound in this region.
Currently, methods for semi-automated model building into electron-density maps³⁹ and structure refinement ⁴⁰ are being developed in a number of laboratories. These methods will be incorporated into the crystal structure determination process of the Protein Structure Factory. Finally, it will be necessary to stringently assess the quality of the determined structures⁴¹ before they are allowed to enter a database.

top

3. CONCLUSIONS

Genomics does not end when all base pairs of DNA have been sequenced. In contrast, it may be argued that the interesting part of the work - aimed at understanding whole organisms by starting from the molecules of life - is the one involving studies of structure and function of the gene products. Structural genomics approaches as the one described above and and large-scale, high-throughput functional studies, functional genomics ⁴², are starting to provide the tools to performing these analyses.

top

4. ACKNOWLEDGEMENT

We are grateful to Jürgen J. Müller (Max-Delbrück-Centrum) for providing figures 2 and 3. Supported by the Bundesministerium für Bildung und Forschung through the Leitprojekt Proteinstrukturfabrik.

top

5. REFERENCES

1. Abola, E.E., Sussman, J.L., Priluski, J. & Manning, N.O. (1997) Protein Data Bank archives of three-dimensional macromolecular structures. Methods Enzymol. 277, 556-571.

2. McKusick, V.A. (1997) Genomics: Structural and functional studies of genomes. Genomics 45, 244-249.

3. Terwilliger, T.C., Waldo, G., Peat, T.S., Newman, J.M., Chu, K. & Berendzen, J. (1998) Class-directed structure determination: Foundation for a protein structure initiative. Protein Sci. 7, 1851-1856.

4. Shapiro, L. & Lima, C.D. (1998) The Argonne Structural Genomics Workshop: Lamaze class for the birth of a new science. Structure 6, 265-267.

5. Gaasterland, T. (1998) Structural genomics taking shape. Trends Genet. 14, 135.

6. Koonin, E.V., Tatusov, R.L. & Galperin, M.Y. (1998) Beyond complete genomes: from sequence to structure and function. Current Opinion Struct. Biol. 8, 355-363.

7. Finkelstein, A.V. & Ptitsyn, O.B. (1987) Why do all globular proteins fit the limited set of folding patterns?
Prog. Biophys. Mol. Biol. 50, 171-190.

8. Chothia, C. (1992) One thousand protein families for the molecular biologist. Nature 357, 543-544.

9. Orengo, C.A., Jones, D.T. & Thornton, J.M. (1994) Protein superfamilies and domain superfolds. Nature 372, 631-634.

10. Bork, P. & Eisenberg, D. (1998) Sequences and topology. Deriving biological knowledge from genomic sequences. Current Opinion Struct. Biol. 8, 331-332.

11. Fischer, D. & Eisenberg, D. (1996) Protein fold recognition using sequence-derived predictions. Protein Sci. 5, 947-955.

12. Rice, D.W. & Eisenberg, D. (1997) A 3D-1D substitution matrix for protein fold recognition that includes predicted secondary structure of the sequence. J. Mol. Biol. 267, 1026-1038.

13. Fischer, D. & Eisenberg, D. (1997) Assigning folds to the proteins encoded by the genome of Mycoplasma genitalium. Proc. Natl. Acad. Sci. USA 94, 11929-11934.

14. Huynen, M., Doerks, T., Eisenhaber, F., Orengo, C., Sunyaev, S., Yuan, Y. & Bork, P. (1998)
Homology-based fold predictions for Mycoplasma genitalium proteins. J. Mol. Biol. 280, 323-326.

15. Kim, K.K., Hung, L.-W., Yokota, H., Kim, R. & Kim, S.-H. (1998) Crystal structures of eukaryotic translation initiation factor 5A from Methanococcus jannaschii at 1.8 * resolution. Proc. Natl. Acad. Sci. USA 95,
10419-10424.

16. Lim, J.-H., Yu, Y.G., Han, Y.S., Cho, S.-j., Ahn, B.-Y., Kim, S.-H. & Cho, Y. (1997) The crystal structure of an Fe-superoxide dismutase from the hyperthermophile Aquifex pyrophilus at 1.9 * resolution: Structural basis for thermostability. J. Mol. Biol. 270, 259-274.

17. Kim, K.K., Kim, R. & Kim, S.-H. (1998) Crystal structure of a small heat-shock protein. Nature 394, 595-599.

18. Saegusa, A. (1998) Japan's genome programme goes ahead, with protein analysis. Nature 392, 219.

19. Kim, S.-H. (1998) Shining light on structural genomics. Nature Struct. Biol. 5, 643-645.

21. Bork, P. & Koonin, E.V. (1998) Predicting functions from protein sequences &Mac220; where are the bottlenecks?
Nature Genetics 18, 313-318.

22. Schultz, J., Milpetz, F., Bork, P. & Ponting, C.P. (1998) SMART, a simple modular architecture research tool:
identification of signaling domains. Proc. Natl. Acad. Sci. USA 95, 5857-5864.

23. Bork, P., Dandekar, T., Eisenhaber, F. & Huynen, M. (1998) Characterization of targeting domains by sequence analysis: glycogen-binding domains in protein phosphatases. J. Mol. Med. 76, 77-79.

24. Yuan, Y., Schultz, J., Mlodzik, M. & Bork, P. (1997) Secreted Fringe-like signaling molecules may be glycosyltransferases. Cell 88, 9-11.

25. Museghian, A.R., Bassett, D.E., Jr., Boguski, M., Bork, P. & Koonin, E.V. (1997) Positionally cloned human disease genes: New motifs and evolutionary conservation. Proc. Natl. Acad. Sci. USA 94, 5831-5836.

26. Bork, P., Hofmann, K., Bucher, P., Neuwald, A., Altschul, S.F. & Koonin, E.V. (1997) A superfamily of conserved domains in DNA damage-reponsive cell cycle checkpoint proteins. FASEB J. 11, 68-76.

27. Maier, E., Maier-Ewert, S., Bancroft, D. & Lehrach, H. (1997) Automated array technologies for gene expression profiling. Drug Discovery Today 2, 315-324.

28. Maier, E., Maier-Ewert, S., Ahmadi, R., Curtis, J. & Lehrach, H. (1994) Application of robotic technology to automated sequence fingerprint analysis by oligonucleotide hybridisation. J. Biotech. 35, 191-203.

29. Hochuli, E., Bannwarth, W., Dobeli, H., Gentz, R. & Stüber, D. (1988) Genetic approach to facilitate purification of recombinant proteins with a novel metal chelate adsorbent. Bio/Technology 6, 1321-1325.

30. Schmidt, T.G.M. & Skerra, A. (1994). One-step affinity purification of bacterially produced proteins by means of the "Strep-tag" and immobilized recombinant core streptavidin. J. Chromatogr. A 676, 337-345

31. Oschkinat, H. & Croft, D. (1994). Automated assignment of multidimensional nuclear magnetic resonance spectra. H. Meth. Enzymol. 239, 308-318.

32. Nilges, M., Macias, M.C., OÕDonoghue, S.I. & Oschkinat, H. (1997). Automated NOESY interpretation with ambiguous distance restraints: the refined NMR solution structure of the pleckstrin homology domain from ?-spectrin. J. Mol. Biol. 269, 408-422.

33. Weber, P.C. (1997) Overview of protein crystallization methods. Methods Enzymol. 276, 13-22.

34. Earnest, T. (1995) Conceptual Design Report for ALS Beamline 5.0, Lawrence Berkeley Laboratory PN941209-2.

35. Svensson, L.A., StŒhl, K., Cerenius, Y., Oskarsson, *., Albertsson, J. & Liljas, A. (1997) A new beamline for crystallographic measurements at the MAX II synchrotron, Lund, Sweden, Annual Report 182.

36. Ogata, C.M. (1998) MAD phasing grows up. Nature Struct. Biol. 5, 638-640.

37. Hendrickson, W.A. & Ogata, C.M. (1997) Phase determination from multiwavelength anomalous diffraction measurements. Methods Enzymol. 276, 494-523.

38. Müller, A., Müller, J.J., Muller, Y.A., Uhlmann, H., Bernhardt, R. & Heinemann, U. (1998) New aspects of electron transfer revealed by the crystal structure of a truncated bovine adrenodoxin, Adx(4-108). Structure 6,
269-280.

39. Fortier, S., Chiverton, A., Glasgow, J. & Leherte, L. (1997) Critical-point analysis in protein electron-density map interpretation. Methods Enzymol. 277, 131-157.

40. Lamzin, V.S. & Wilson, K.S. (1997) Automated refinement for protein crystallography. Methods Enzymol. 277, 269-305.

41. Dodson, E.J., Davies, G.J., Lamzin, V.S., Murshudov, G.N. & Wilson, K.S. (1998) Validation tools: can they indicate the information content of macromolecular crystal structures? Structure 6, 685-690.

42. Brownstein, M.J., Trent, J.M. & Boguski, M.S. (1998) Functional genomics. Trends Biochem. Sci. 23 (Suppl.) 27-29.

top

Contact, questions, remarks?
Email: [email protected]

• [home] • [scientific concept] • [status] • [project structure] •
• [news] • [links] • [jobs ] • [internal pages] •

•The Berlin "Protein Structure Factory" initiative•

Project