Protein Science
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Protein Science (2006), 15:2718-2728. Published by Cold Spring Harbor Laboratory Press. Copyright © 2006 The Protein Society
This Article
Right arrow Abstract Freely available
Right arrow Full Text (PDF)
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Malawski, G. A.
Right arrow Articles by Müller-Tiemann, B.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Malawski, G. A.
Right arrow Articles by Müller-Tiemann, B.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us   Add to Digg   Add to Reddit   Add to Technorati  
What's this?

Identifying protein construct variants with increased crystallization propensity––A case study

Guido A. Malawski1,3, Roman C. Hillig1,3, Felipe Monteclaro2, Uwe Eberspaecher1, Arndt A.P. Schmitz1, Kerstin Crusius1, Martina Huber1, Ursula Egner1, Peter Donner1, and Beate Müller-Tiemann1

1 Schering AG, Research Center Europe, 13342 Berlin, Germany
2 Berlex Biosciences, Richmond, California 94804, USA

(RECEIVED August 15, 2006; FINAL REVISION September 29, 2006; ACCEPTED October 2, 2006)


   Abstract
TOP
 Abstract
Introduction
 Results
 Discussion
 Materials and methods
 Acknowledgments
 References
 
This study describes an efficient multiparallel automated workflow of cloning, expression, purification, and crystallization of a large set of construct variants for isolated protein domains aimed at structure determination by X-ray crystallography. This methodology is applied to MAPKAP kinase 2, a key enzyme in the inflammation pathway and thus an attractive drug target. The study reveals a distinct subset of truncation variants with improved crystallization properties. These constructs distinguish themselves by increased solubility and stability during a parallel automated multistep purification process including removal of the recombinant tag. High-throughput protein melting point analysis characterizes this subset of constructs as particularly thermostable. Both parallel purification screening and melting point determination clearly identify residue 364 as the optimal C terminus for the kinase domain. Moreover, all three constructs that ultimately crystallized feature this C terminus. At the N terminus, only three amino acids differentiate a noncrystallizing from a crystallizing construct. This study addresses the very common issues associated with difficult to crystallize proteins, those of solubility and stability, and the crucial importance of particular residues in the formation of crystal contacts. A methodology is suggested that includes biophysical measurements to efficiently identify and produce construct variants of isolated protein domains which exhibit higher crystallization propensity.

Keywords: protein isolation; protein characterization; chromatography; protein crystallization; construct design; expression screening; high throughput melting point analysis; high throughput protein production


   Introduction
TOP
 Abstract
 Introduction
Results
 Discussion
 Materials and methods
 Acknowledgments
 References
 
Structure-guided drug design has become of increasing importance to the pharmaceutical industry. Early structural information on lead-target interactions accelerates the chemical optimization process and improves compound quality as to patentability, selectivity, pharmacokinetics, and toxicology. However, obtaining structural information on novel drug targets in time to support the drug discovery process can represent a major hold-up. Attrition rates in protein crystallography remain high when taking into consideration the entire process from generating high-purity soluble protein to X-ray structure solution (Lesley et al. 2002; Service 2002; Heinemann et al. 2003; Rupp 2003). This is particularly true for isolated catalytic domains of human drug targets (Jenkins et al. 1995; Chambers 2002; Heinemann et al. 2003; Scheich et al. 2003). Recent developments in high-throughput methods for protein production have increased success rates in generating soluble protein amenable for crystallography and have accelerated the process from gene to crystal (Lanio et al. 2003; Scheich et al. 2003).

In this study, we present a fully integrated industrialized automated process of multiparallel cloning, protein expression, purification, characterization, and crystallization. This process is illustrated by a case study of the human MAPKAP kinase 2 (MK2). For this protein, initial efforts based on low-throughput protein production methods did not prove successful in yielding suitable crystals of the kinase domain for structure determination. Instead, such crystals were later obtained via a higher throughput approach on protein expression and purification directed at a large set of catalytic domain variants.

MK2 is a Ser/Thr kinase and is considered an attractive molecular target for therapeutic intervention in indications such as inflammation and cancer (for reviews, see Kotlyarov and Gaestel 2002; Gaestel 2006; and references therein). It was originally identified as one of the downstream substrates of p38 MAP kinase {alpha} and beta (Stokoe et al. 1992). Upon phosphorylation MK2 triggers a series of pathways after lipopolysaccharide (LPS) challenge, resulting in increased production of proinflammatory cytokines such as tumor necrosis factor (TNF)-{alpha}, interleukin-6, and interferon {gamma} (Winzen et al. 1999; Kotlyarov et al. 2002). MK2-deficient mice (MK2–/–) are viable and fertile and grow to normal size, but these animals exhibit a reduced LPS-mediated inflammatory response because of decreased TNF{alpha} serum levels (Stokoe et al. 1992; Kotlyarov et al. 1999).

Human MK2 consists of 400 amino acids coding for a proline-rich N-terminal domain (Engel et al. 1993) followed by the catalytic domain most closely related to the family of calmodulin-dependent protein kinases (Stokoe et al. 1993; Zu et al. 1994), while the C terminus contains an autoinhibitory region (Meng et al. 2002; Underwood et al. 2003) and a nuclear localization signal (NLS) as well as a nuclear export signal (NES) (Fig. 1A). A splice variant ending at residue 353 exists that lacks the NLS and NES (Roux and Blenis 2004). MK2 contains six potential phosphorylation sites of which Thr222 in the activation loop and Thr334 in the autoinhibitory domain are essential for MK2 activity (Ben-Levy et al. 1995; Roux and Blenis 2004). Crystal structures of the MK2 kinase domain at medium to low resolution (2.7–3.2 Å) (Meng et al. 2002; Kurumbail et al. 2003; Underwood et al. 2003; see Table 1) gave room for aiming at higher resolution to ideally support the drug discovery process with MK2 structures in complex with lead compounds.


View this table:
[in this window]
[in a new window]

  		Table 1. Information on constructs used for crystallization that was available at the start of the parallel cloning and purification approach

 


Figure 1
View larger version (43K):
[in this window]
[in a new window]

  		Figure 1. Domain structure and construct design of MK2 variants. (A) Domain structure of human MK2, with the positions of a nuclear localization signal (NLS), a nuclear export signal (NES), and phosphorylation sites (P) indicated. (B) Initial construct comprising only the core kinase domain. (C) Set of constructs (1 to 16) designed to screen for soluble protein with improved crystallization propensity; residue numbers for N- and C-terminal amino acids are indicated.

 
MK2 kinase domain constructs of varying C and N termini were cloned and expressed in Escherichia coli. These were screened for soluble protein by affinity purification via recombinant tags in a 96-well format applying automation. Protein samples were purified in parallel by an automated multistep chromatography protocol including thrombin cleavage, and analyzed by high-throughput melting point analysis. Finally, all constructs were purified in parallel by an automated multistep chromatography protocol, and the five best-behaved constructs were characterized by dynamic light scattering (DLS) and processed into crystallization. This study illustrates the paradigm shift in protein production for structural biology toward an industrialized process including multiparallel expression and purification backed by a strong analytical platform. Screening a large series of catalytic domain constructs holds the potential of identifying clones with improved product properties regarding expression levels, solubility, stability, and finally, the propensity to crystallize.


   Results
TOP
 Abstract
 Introduction
 Results
Discussion
 Materials and methods
 Acknowledgments
 References
 
The construct used in initial efforts at producing MK2 kinase domain for crystallization trials was expected to represent the kinase core domain based on alignments of MK2 with other kinases with known structures. This analysis suggested that the core kinase domain comprised residues 43–330 (Fig. 1B). This segment was cloned, expressed in E. coli, and purified to ≥95% purity following a two-step purification scheme; however, the final samples could never be concentrated to >1.4 mg/mL because of aggregation. Attempts to obtain well-diffracting crystals of this first MK2 construct failed. Conditions were identified under which protein microcrystals grew, and intensive fine-screening and streak seeding resulted in larger crystals with improved morphology (Fig. 2). However, these crystals were still relatively small and did not diffract, even when using synchrotron X-ray radiation.


Figure 2
View larger version (58K):
[in this window]
[in a new window]

  		Figure 2. Crystallization of initial MK2 construct (residues 43–330). (A) Initial lentil-shaped microcrystals. Diameter 5–10 µm. (B) Optimized crystals. Diameter ~50–70 µm.

 
After these experiments had been carried out, successful crystal structure determinations were reported for three other MK2 constructs (Table 1): (1) apo MK2, residues 47–400 (Meng et al. 2002); (2) MK2 (residues 41–364) bound to ADP and staurosporine (Underwood et al. 2003); and (3) MK2 (residues 45–371) bound to AMPPNP (Kurumbail et al. 2003). Interestingly, all three groups had crystallized their respective MK2 constructs under conditions very similar to the ones we had identified, with sodium potassium phosphate and ammonium sulfate, respectively, as precipitants (Table 1). However, in contrast to our construct 43–330, these constructs could be concentrated to higher final concentrations (5 and 10 mg/mL, respectively). Comparison of the construct borders strongly implied that our original construct featured a too-short C terminus.

Although the longer MK2 constructs yielded diffracting crystals, all crystal forms diffracted only to low or medium resolution (Table 1). We therefore decided to design a new set of longer MK2 constructs. Construct 41–364 was chosen as the starting point, as this had already successfully been used to cocrystallize MK2 with small molecule ligands (Underwood et al. 2003). Judging from the disordered C termini of the different molecules in the crystal structures of this construct (Table 1), the C-terminal 7–22 residues were flexible, and such flexibility was also implied by the structures of the other two successfully crystallized constructs. To a lesser degree, this was also true for the N termini. Therefore, we decided to truncate the N and C termini in a stepwise manner to arrive at constructs that were expected to feature less flexible termini and would therefore be more likely to result in better packed and better diffracting crystals. The final set of constructs is depicted in Figure 1C. The C terminus of the splice variant at residue 353 was not included into this set because the crystal structure of apo MK2 showed residues 346–363 to form one continuous {alpha} helix (Meng et al. 2002). By choosing residue 350 as C terminus (Fig. 1C), we cut into the beginning of this helix, but thought this justified as the terminal Trp349 binds into a hydrophobic pocket and this terminus also came close to the ordered region described by Kurumbail et al. (2003).

The constructs were then test-expressed in E. coli using an automated miniaturized multiparallel workflow. This routine resulted in a distinct band of the correct respective size for all of the 16 constructs in SDS-PAGE. Background was negligible and the yield was ~30 mg/L. A parallel test expression of the respective 6-His-tagged variants resulted in significantly lower expression rates compared with the GST-tagged variants and was therefore abandoned.

As test-expression resulted in reasonable yields for all 16 MK2 constructs, all constructs were moved forward into a parallelized and automated generic two-step purification scheme on the ÄKTAxpress (Fig. 3). One aim of automation of multistep protein purification is the selection of constructs as potential candidates for further scale-up for structure determination. This selection considers parameters such as yield, tag cleavage, solubility, homogeneity, and purity. With GST affinity chromatography (AC) followed by an on-column tag-cleavage by a protease as the initial step, and size exclusion chromatography (SEC) as the second step, this process is independent of specific protein parameters, such as the isoelectric point or hydrophobicity, while still addressing all parameters for further selection. For all 16 constructs, MK2 protein could be obtained with the expected retention volume on the SEC column. Based on three independent purification runs, all 16 constructs gave reasonable yields after the one-step automated miniaturized purification scheme, with the amount of purified protein varying strongly between the different constructs in the multistep generic purification (Fig. 4; Table 2). Based on the expression rates, the different constructs fall into two groups, the first with an expression rate up to 19 mg/L and the second with an expression rate greater than 29 mg/L culture volume. The second group, comprising constructs 4, 8, 12, 15, and 16, was moved forward toward large-scale purification on the ÄKTAxpress system and crystallization.


View this table:
[in this window]
[in a new window]

  		Table 2. Calculated yield per liter of culture volume of obtained protein purified with the automated parallel multistep generic purification scheme

 


Figure 3
View larger version (36K):
[in this window]
[in a new window]

  		Figure 3. Overview of the parallel cloning, expression, purification, and crystallization procedure applied to 16 MK2 kinase domain constructs.

 


Figure 4
View larger version (32K):
[in this window]
[in a new window]

  		Figure 4. SDS-PAGE analysis after generic purification scheme. Coomassie-stained SDS-PAGE gel of 2 µL of protein purified with the generic automated multistep purification scheme. Numbers indicate the MK2 constructs.

 
To evaluate additional technologies for construct selection and prioritization for structure determination, a high-throughput melting point analysis was performed. We adapted a method previously described for rapid identification of small molecule inhibitors (Lo et al. 2004). We reasoned that the melting points of a series of related constructs processed in parallel in the absence of ligands might be related to their folding states. A protein sample folded to a more thermostable conformation and/or containing a lower fraction of unfolded protein would then have a higher probability to pass the protein production process and be also more likely to crystallize.

Eluates from the generic purification scheme described above were subjected to this methodology. Figure 5A demonstrates the normalized fluorescence emission of the reporter dye over temperature of one experiment. The midpoint of the thermal transition (melting temperature, Tm50) was then determined for each measurement. The curves fall into two distinct groups, one with a mean ± SD of 38.8 ± 1.5°C (7 constructs), the other at 46.0 ± 1.4°C (4 constructs). This interpretation is statistically relevant (unpaired Student's t-test; p < 0.0001). The individual values are shown in Figure 5B, revealing that the group with a higher thermostability encompasses constructs 4, 8, 12, and 16. This set of constructs was identified each time the experiment was performed (n = 3). All of these constructs end with the same C terminus and include four out of the five constructs with the highest yield in the generic automated multistep purification scheme.


Figure 5
View larger version (28K):
[in this window]
[in a new window]

  		Figure 5. Thermal stability of MK2 variants. The eluates from the generic chromatographic purifications were supplemented with the fluorophore SYPRO Orange and subjected to a thermal stability analysis using a real-time PCR machine as detailed in the Materials and Methods section. One representative of three independent experiments is shown. Construct 1 could not be interpreted; constructs 11, 13, and 14 were not available for measurement. (A) Normalized fluorescence data, including the mean ± SD for both classes of constructs shown in red. (B) Melting temperatures of the individual constructs.

 
For those five MK2 constructs with the highest yields after the generic automated multistep purification, a more refined purification strategy with an additional hydrophobic interaction chromatography was established according to Underwood et al. (2003). This was successful for all five constructs and resulted in samples of ≥95% purity (Fig. 6), which were concentrated, depending on the available amount of protein, to 7.6 mg/mL for construct 4, 9.8 mg/mL for construct 8, 13.3 mg/mL for construct 12, 5.1 mg/mL for construct 15, and 12.6 mg/mL for construct 16.


Figure 6
View larger version (56K):
[in this window]
[in a new window]

  		Figure 6. SDS-PAGE analysis of final protein samples for crystallization. Coomassie-stained SDS-PAGE gel. (Lane 1) 20 µL of the pooled protein after SEC; (lane 2) 1 µL and (lane 3) 5 µL of the concentrated protein. Numbers below the gel indicate the constructs.

 
To monitor the homogeneity of the protein preparations, to measure the oligomeric state of the purified protein samples, and to detect protein aggregation, DLS measurements were performed. The observed monomodal polydispersity for the five constructs used in crystallization screens was between 26.1% and 32.7% for 4, 8, 12, and 16 and 7.6% for 15, indicating a relatively good crystallization probability for all five constructs. However, while the first four constructs showed no aggregates at all and the measured hydrodynamic radii indicated approximately a molecular mass of a dimer species, construct 15 formed higher molecular weight aggregates with a hydrodynamic radius of >10 nm (Wilson 2003).

All five MK2 constructs 4, 8, 12, 15, and 16 were subjected to a parallel crystallization screen in the presence of Mg2+/ADP. An overview of the screening results is shown in Figure 7. Conditions that yielded crystals are listed in Table 3. The crystals grew in some cases within 1–4 d, but many conditions needed up to 3 wk. While the pH values of the successful conditions varied over a wide range (pH 4.0–8.2), all crystals were found in conditions with high concentrations of 1.0–2.2 M salt as precipitant (ammonium sulfate, sodium potassium phosphate, and lithium chloride). This is very similar to the earlier crystallization conditions found with the initial construct (Table 1).


View this table:
[in this window]
[in a new window]

  		Table 3. Crystallization conditions versus MK2 constructs

 


Figure 7
View larger version (143K):
[in this window]
[in a new window]

  		Figure 7. Comparative analysis of the crystallization screen. This representation of the crystallization set-ups provides a quick overview of the results. Trends, such as conditions that give crystals with several different constructs, can easily be detected. Shown are the results for the five MK2 constructs tested in a crystallization screen consisting of four commercial screens (see Materials and Methods), revealing that constructs 4, 8, and 12 crystallized under several conditions, while 15 and 16 did not. (Dark blue) clear drop, (gray) weak precipitation, (light blue) granular precipitation, (brown) strong precipitation, (green) crystal seed-like objects, (teal blue) phase separation, (red) crystals.

 
Most crystals showed a bipyramidal morphology (Fig. 8A). Crystals of this morphology were further optimized only for construct 4 and diffraction data were collected (details will be published elsewhere). This confirmed that the bipyramidal crystal form belonged to space group F4132 with unit cell dimensions of of a = b = c = 254 Å and was thus identical to the form described independently for MK2:ADP (Underwood et al. 2003) and for MK2:AMPPNP (Kurumbail et al. 2003). Unfortunately, as with the published cases of this crystal form (Table 1), our crystals of construct 4 did not diffract to better than 2.8–3.0 Å (data not shown). In a smaller number of drops, crystal morphologies different to the bipyramidal one were observed (Fig. 8B); however, these were not further optimized.


Figure 8
View larger version (54K):
[in this window]
[in a new window]

  		Figure 8. Examples of crystal morphologies obtained in the parallel screening experiment. (A) Dominant bipyramidal morphology, here with construct 4, AS Screen-E8 (1.6 M ammonium sulfate, 0.1 M citric acid at pH 5.0). Crystal diameter ~200 µm. (B) Rod-shaped crystal morphology different from the bipyramidal morphology, here found with construct 12, pH Clear2-F5, rod length ~5–10 µm.

 
Interestingly, crystals were only observed for three of the five constructs that were screened (Table 3). Despite only minor changes in construct length, only constructs 4, 8, and 12 crystallized readily and within very few days, whereas constructs 15 and 16 did not crystallize at all. Surprisingly, the only difference between construct 12, which crystallized under nine different conditions, and construct 16, which did not crystallize at all, were three amino acids at the N terminus (longer in the case of construct 12). It is worth noting that without any prior selection, the high-throughput thermostability assay identified all three constructs that ultimately yielded crystals. In addition, it delivered only one "false positive," which stood out in terms of thermostability but did not crystallize in the subsequent screen. The same three constructs were also identified in the generic purification screen on the ÄKTAxpress, which yielded two further "false-positive constructs."


   Discussion
TOP
 Abstract
 Introduction
 Results
 Discussion
Materials and methods
 Acknowledgments
 References
 
This study introduces an efficient multiparallel automated workflow of cloning, expression, purification, and crystallization of a large set of construct variants for isolated domains of proteins aimed at structure determination by X-ray crystallography. The described methodology allows for an efficient gene-to-crystal process in settings where high attrition rates of X-ray structure determination projects cannot be tolerated. This is particularly the case in the pharmaceutical industry, where structure-aided drug design has evolved to become an integral part of the drug discovery and lead optimization process. The methodology described in this report has proven to be successful in identifying suitable protein variants of isolated domains for crystallography. It has now been taken even further and routinely involves screening of not only prokaryotic, but also a variety of eukaryotic, expression systems. As long as protein expression and crystallography remain rather empirical sciences, a screening approach is the methodology of choice. The experimental realization of all the process steps from cloning via expression and purification to crystallization integrates a variety of recent developments in the protein science field such as miniaturization, parallelization, and automation. This represents a truly industrialized process.

The case study of MK2 clearly demonstrates the power of this screening approach. The initial approach of cloning and purifying only one construct of the bioinformatically predicted kinase domain of MK2 failed as the protein was not stable enough in solution to allow concentration to higher than 1.4 mg/mL, and the crystals obtained did not diffract. This obstacle was overcome when novel construct boundaries were identified by the described domain truncation screening approach.

The method revealed a subset of five constructs that exhibited significantly higher yields after a multistep purification, including tag-cleavage on the ÄKTAxpress system, than the remaining eleven constructs. All five yielded large amounts of pure protein when subjected to an in-depth preparative purification scheme and could be concentrated to at least 5–10 mg/mL, and thus to concentrations that had never been achieved with the initial construct. Crystallization screening finally revealed that three of the five well-behaved constructs (4, 8, and 12) produced crystals, whereas constructs 15 and 16 did not.

The common feature discriminating those MK2 constructs that ultimately crystallized from the initial construct and the less well-expressed and purified constructs is the longest C terminus in the set tested here (ending at residue 364). Our initial attempt to arrive at constructs with improved crystallization characteristics by shortening the C terminus thus failed. The C terminus is not well-packed in the majority of the MK2 crystal structures in the Protein Data Bank (PDB) (Table 1) in which the amino acids beyond residues 342–345 are often disordered. Also, the C terminus of crystal structures of our construct 4 is no longer defined in the density beyond residues 346 (data will be published elsewhere). Such a disordered or less-ordered region in the crystal lattice is usually considered counterproductive for optimal crystallization (Gao et al. 2005). Our finding that this flexible terminus was indeed essential for the best constructs suggests that it probably increased solubility and stability in solution, which outweighed the disadvantage of a flexible region in the crystal lattice. This is very likely caused by covering of a hydrophobic surface patch. The crystal structure of apo MK2 (PDB entry 1kwp; see Table 1) revealed that residues 346–363 form an amphipathic helix. Residues Trp349, Val352, Met356, Leu360, and Met363 of this helix cover a hydrophobic surface patch formed by residues Leu342, Phe147, Ile255, Pro261, Pro189, Ile252, Val251, and Trp247. The hydrophobic section on the C-terminal helix in fact serves as a nuclear export signal, which is masked when bound to the kinase domain (Meng et al. 2002). In those constructs where all or some of the residues 346–364 are omitted, the hydrophobic surface area on the kinase domain would be partially or fully solvent accessible, resulting in aggregation problems as observed with our short initial construct. Such a predominantly hydrophobic interaction between the C-terminal helix and the rest of the kinase domain would also allow for small differences in the orientation of the helix when comparing different molecules in the crystal and may therefore cause the observed lack of density for this region in many crystal forms.

The failure of constructs 15 and 16 to crystallize despite behaving well in the generic and in-depth purification screen can retrospectively be explained when taking into account the thermostability and DLS data and by analyzing relevant crystal contacts: For construct 15, DLS revealed the presence of higher molecular aggregates and thus an inhomogeneous sample composition. In addition, the high-throughput protein melting point analysis showed a reduced thermostability of this sample. Following the theory that one prerequisite for crystallization propensity is the stability of the respective protein domain, it will be difficult for a protein less resistant to thermal unfolding to form ordered crystal contacts. It should be noted that this approach cannot generate one-to-one predictive data for crystallization propensity of constructs as thermostability is not the only parameter influencing crystallization behavior. It is, for example, well known that lysine or glutamate residues on the surface of protein molecules may be unfavorable for crystal contact formation because their large amino acid side chains have a high conformational entropy (Czepas et al. 2004; Derewenda 2004).

Construct 16 did not crystallize, although it featured the favorable long C terminus and also showed very promising DLS and thermostability data. Therefore, while both methods hold the potential for prioritizing sets of constructs or protein preparations, they do not allow for a strict correlation to the ability to crystallize. The failure of construct 16 can be explained via crystal packing analysis at the N terminus. The three constructs that crystallized did so under conditions identical or similar to the condition established before for our initial kinase domain construct and also to those reported in the literature (Meng et al. 2002; Kurumbail et al. 2003; Underwood et al. 2003). Judging from crystal morphology, the majority of the identified crystals over all constructs also belong to this same cubic crystal form of Kurumbail et al. (2003) and Underwood et al. (2003), which we analyzed by crystallography only for construct 4. In this crystal form, the N-terminal residues 44–51 reach over to a crystal neighbor and form a tight crystal packing contact (Fig. 9). Phe46 and Val48 of this N terminus insert into a hydrophobic pocket on the surface of the crystal neighbor that is formed by Val65, Ile60, and Tyr128 (Fig. 9B). In addition, the backbone oxygen and nitrogen atoms of Lys49 form two hydrogen bonds to the backbone of Leu127 of the crystal neighbor (not shown in Fig. 9B). The step from the well-crystallizing construct 12 to construct 16 (and 15), which failed to crystallize, is equivalent to the removal of residues 47–49. This deletion would abolish the described crystal contact completely. This explains why construct 16 could not crystallize in the cubic crystal form anymore, although it featured the favorable long C terminus. It is remarkable, however, that it did not crystallize in any other crystal packing arrangement either, at least in none of the 384 conditions tested in the work described here.


Figure 9
View larger version (79K):
[in this window]
[in a new window]

  		Figure 9. Crystal contact at the N terminus of construct 4. In the bipyramidal cubic crystal form of construct 4, the N-terminal residues 44–51 reach over to a crystal neighbor and form a tight crystal packing contact. (A) Three neighboring MK2 molecules (blue, yellow, and gray; the first two in ribbon, the latter in surface representation, in a view approximately along a threefold crystallographic axis). The hinge region (residues 139–141) of all three molecules is shown in green to illustrate the position of the ATP binding site, between the N- and C-terminal lobes. All three molecules extend their N termini (residues 44–49, in red) toward the N-terminal lobe of a crystal neighbor (highlighted by a box for the contact between the blue and gray molecules). (B) Stereo representation of a zoom into this box, with the N-terminal residues of the blue molecule in stick-representation (carbon atoms, red; nitrogen, blue; oxygen, magenta), showing in more detail the hydrophobic pocket into which the N terminus binds with Phe46 and Val48. This tight contact cannot be established in the subset of constructs that start only at residue 50.

 
The results from the high-throughput protein melting point analysis of MK2 indicate that less thermostable protein variants pose difficulties not only at the crystallization level but already during the multistep purification procedure. This became particularly obvious after removal of the GST tag, which is well known to improve the solubility of fusion proteins, thus shielding insolubility or tendency for aggregation of the recombinant fusion partner, here MK2. Indeed, the clear differentiation into two subsets of lower and higher thermostability only became visible when testing the "naked" kinase domains. Analysis of the uncleaved GST–MK2 constructs showed hardly any difference in thermostability (data not shown).

High-throughput protein melting point analysis can also be employed at the earliest possible stage of the workflow after small-scale screening, as each measurement requires only minimal amounts of protein. We demonstrated this by performing three independent and blinded runs of the following experiment. As detailed in the Materials and Methods section, all 16 MK2 constructs were re-expressed in a 2 mL volume in deep-well plates, and the eluates were analyzed. Each run identified the same set of constructs 4, 8, 12, and 16 (data not shown). Based on these results, high-throughput protein melting point analysis provides an attractive basis for early prioritization of promising protein constructs to be moved forward into large-scale chromatography purification and crystallization.

In conclusion, the data presented in this report illustrate the dramatic effect of variations of only few, in this case three, amino acids at the N terminus on the propensity of a protein construct to form crystals. Therefore, this study provides strong arguments for a multiparallel expression and purification paradigm, which is becoming increasingly established in the structural biology community. In the example presented here, this screening focused on domain truncation variants; however, it can also be applied to screening for surface mutations, with the aim of removing hydrophobic patches, solvent accessible Cys residues, post-translational modification sites such as phosphorylation or glycosylation sites, or entropically unfavorable residues such as clusters of Lys and Glu. High-throughput protein melting point analysis turned out to be a useful tool in prioritizing the constructs emerging from the parallel approach. It identified the four most thermostable constructs. All four later turned out to be among the five that behaved best in the purification schemes, and three of these four constructs yielded crystals in the crystallization screen. Thermostability, as tested here with the apo proteins, or in conjunction with screening for stabilizing ligands or buffers, holds the potential to be a valuable tool for identifying the most promising constructs and conditions for crystallization. Together with the application of modern robotics technologies in cloning, expression, and purification, as well as protein crystallization, the described set-up provides a sound basis for strong structural impact on the rapid development of innovative drug candidates.


   Materials and methods
TOP
 Abstract
 Introduction
 Results
 Discussion
 Materials and methods
Acknowledgments
 References
 
Cloning of initial MK2 construct (43–330)
All PCR reactions were performed using the Advantage HF PCR kit from Clontech (BD Biosciences). Restriction enzymes from New England Biolabs and T4 DNA ligase from Invitrogen were used for all cloning methods.

The human MK2 catalytic domain (aa43–330) cDNA was cloned using human Universal QUICK-Clone cDNA from Clontech (BD Biosciences). The MK2 DNA fragment was generated by PCR amplification using the following forward primer fusing a BamHI restriction site (in bold) to the 5' end of the MK2 N-terminal sequence starting at residue 43 (Formula '--CGCGGATCCTTCCCGCAGTTCCACGTCAAGTCC-3'). The forward primer was paired with the reverse primers, which terminate with the carboxyl end at residue 330 plus an additional EcoRI site (in bold) for cloning purposes (Formula '--GATGAATTCCTCACTTTGTTGATTGCATGATCCAAGG-3').

The resulting PCR product was subcloned into the pGEX-KT vector (Hakes and Dixon 1992; Kervinen et al. 2006) providing an N-terminal GST tag followed by a Gly linker sequence and a thrombin cleavage site. The MK2 DNA was cloned in frame using BamHI and EcoRI. The DNA construct was sequence-confirmed by double-stranded DNA sequencing.

Expression, purification, and crystallization of initial MK2 construct (43–330)
GST-MK2 (43–330) was expressed in E. coli (BL21). After lysing the cell pellet in buffer A (50 mM sodium phosphate at pH 8.0, 150 mM NaCl, 1 mM EDTA, 10 mM DTT, 1:50 Complete inhibitor) using a French press, the centrifuged (30,000g) supernatant was applied to and eluted from a Glu-Sepharose column using buffer B (15 mM glutathione in buffer A). After an additional Superdex 200 gel fitration in buffer C (25 mM Tris-HCl at pH 8.0, 300 mM NaCl, 5% glycerol, 0.1% octylglucoside, 1 mM EDTA, 10 mM DTT) the fractionated GST-MK2 pool was completely cleaved with thrombin (1:2000 w/w, 4°C overnight), further purified (MK2 > 95%) using an additional gel filtration step (Superdex 75, in buffer C), and finally concentrated to 1.4 mg/mL (highest concentration possible before aggregates occurred).

Concentrated samples were subjected to crystallization screens using Crystal Screen 1&2 and several grid screens (PEG, ammonium sulfate, sodium potassium phosphate) from Hampton Research. All experiments were carried out in 24-well Linbro plates using the hanging drop method, and drops were made from 1 µL of protein solution and 1 µL of reservoir solution. Lentil-shaped microcrystals from sulfate and phosphate grid screens were optimized by fine-screening and streak seeding. Optimized conditions consisted of reservoir solutions of either 2.4 M ammonium sulfate/100 mM MES at pH 6.0 or 1.8 M sodium potassium phosphate at pH 6.0. The crystals were tested for diffraction using synchrotron radiation (EMBL outstation at DESY, Hamburg, Germany).

Cloning of 16 MK2 constructs
The forward primers used fuse a TOPO cloning signal and a thrombin cleavage site (5'-CACCCTGGTTCCGCGTGGATCC-3') 5' of the MK2 N-terminal sequence starting at residues 41, 44, 47, and 50, respectively, while the reverse primers introduce two stop codons (5'-TCATTA-3') after residues 342, 345, 350, and 364, respectively. Each forward primer was paired with each reverse primer in a PCR reaction using an MK2 cDNA as the template and the Advantage-HF 2 PCR kit (Clontech). The resulting 16 separate amplicons were cloned into pENTR/D-TOPO (Invitrogen) and sequence verified.

To construct the expression vector pD-ECO1, the GST sequence of pGEX-KT (GE Healthcare) was PCR amplified while introducing NdeI and XhoI sites at the ends, which were used to ligate the amplicon into pET-22b (Novagen). Then, the Reading Frame Cassette A of the Gateway Vector Conversion System (Invitrogen) was introduced into the SmaI site. The expression vector pD-ECO3 was obtained by replacing the NdeI/NcoI fragment of pD-ECO1 including the GST-tag with the NdeI/NcoI fragment of pDEST17 (Invitrogen) coding for an N-terminal His-tag. Both expression vectors were resequenced in the altered regions. Finally, the set of MK2 plasmids were recombination cloned into pD-ECO1 and pD-ECO3 using Gateway LR Clonase II Enzyme Mix (Invitrogen).

Expression screening, generic purification, and preparative purification of 16 constructs
For expression tests, cultures in the E. coli strain Rosetta2 (DE3) (Novagen) were grown in deep-well plates and induced using autoinduction (Novagen). Bacterial pellets were harvested by centrifugation and lysed by a bead mill (Hummel and Kula 1989). Lysates were clarified with filter plates (Pall), and the recombinant protein was isolated on His or GST 4B MultiTrap plates (GE Healthcare) with a Tecan liquid handling work station.

For the automated parallel multistep generic purification scheme, cultures were grown as above, but in shake flasks. The cell pellet was resuspended in one tenth of culture volume with buffer D (50 mM Tris at pH 8.0, 120 mM, NaCl, 1 mM DTT, and 10% [v/v] glycerol) and stored at –20°C. The stored material (3.5 mL) was thawed and, after adding 3.5 mL of buffer D, lysed by one passage through a One-Shot Constant Cell Disruption System (Holly Farm Business Forum) at 2 kBar. A total of 3.5 mL of the lysate was centrifuged for 40 min at 4°C and 100,000g, and 8.5 mL of buffer D was added. All parallel column-based purifications were performed with the ÄKTAxpress System (GE Healthcare). For the automated parallel multistep generic scheme, two prepacked 1-mL GSTrap 4B columns were connected to each module of the ÄKTAxpress. After equilibration of the columns with at least 5 column volumes (CV) of buffer D, the clarified lysate was loaded onto the columns with 0.2 mL/min, columns were washed with 10 CV, followed by a 10 CV high salt wash with buffer D plus 300 mM NaCl. Thrombin (0.7 CV) (Hematologic Technologies Inc.) diluted in buffer D (11.6 µg/mL final) was loaded onto the column. After an incubation of 8 h, the cleaved protein was eluted with buffer D and transferred onto a Superdex 75 16/60 column equilibrated with buffer D. The MK2 protein containing fractions were pooled and loaded on a gel, and the protein yield was detected by densitometry with a BSA standard.

For the preparative parallel purification for crystallization, the cell pellets were resuspended in one-tenth of culture volume with buffer E (50 mM Tris at pH 8.0, 150 mM NaCl, and 10% [v/v] glycerol) plus 1 mM DTT and stored at –20°C. Twenty-five milliliters of the stored material was thawed, disrupted, and centrifuged, as described above. The supernatant was applied to a 1-mL GSTrap 4B column equilibrated with Buffer E plus 5 mM DTT. The column was washed with 20 CV equilibration buffer and 5 CV cleavage buffer (buffer E plus 1 mM DTT and 150 mM NaCl). The protein was incubated for 8 h with thrombin; the cleaved protein was eluted with cleavage buffer and passed through a HiPrep desalting column equilibrated with Buffer F (50 mM HEPES at pH 7.4, 10 mM DTT, and 0.8 M ammonium sulfate) and loaded onto a 1-mL PolypropylA column. The protein was eluted with a gradient (buffer F without ammonium sulfate), and the main peak loaded onto a Superdex75 equilibrated with buffer G (50 mM HEPES at pH 7.4, 200 mM NaCl, 5 mM MgCl2, 10 mM DTT). The eluted protein was concentrated in a Millipore Ultrafree concentrator (Amicon Ultra-4; 10,000 MWCO, Millipore).

The purification was performed at room temperature at an ÄKTAxpress system (GE Healthcare) and all columns used were GE Healthcare products.

Dynamic light scattering (DLS)
DLS was measured with the different constructs using a DynaPro MS800 instrument (Protein Solutions Inc.) equipped with an 825-nm laser at room temperature. Protein samples were centrifuged for 5 min at 13,000g to remove possible particulates that would interfere with DLS measurements. The samples were detected in buffer G at their final concentrations for crystallization. At least 20 measurements were averaged with acquisition times of 5 sec.

Thermostability assay
High-throughput protein melting point analysis was performed on a 7500 Fast RT-PCR machine (Applied Biosystems) adapted from Lo et al. (2004). In a total volume of 50 µL, the protein was combined with SYPRO Orange (Molecular Probes/Invitrogen). This compound is sold as a 5000x stock solution and was used at 1x final concentration. The time/temperature control of the PCR machine was programmed to perform 90 steps of 1 min each and raising the temperature at each step by 1°C, from 5°C to 95°C. The fluorescence emitted after excitation at 480 nm was recorded at each step.

For chromatographically purified samples, 1 µg of protein was used per well. For a small-scale screening experiment, samples were obtained as follows. The pD-ECO1 constructs were expressed in deep-well plates in a volume of 2 mL and the bacterial pellets lysed and clarified as above. The samples were then applied to Glutathione Sepharose 4B MicroSpin columns (GE Healthcare), washed extensively, and incubated overnight at 4°C with 100 µL of a 2.7 µg/mL thrombin solution. The columns were briefly centrifuged to collect the eluate containing only the MK2 fragments. Fifty microliters of these samples were then supplemented with the fluorophore and analyzed.

The fluorescence raw data were exported into Sigmaplot 8 (SPSS). Using the software's transformation language, the lowest local minimum and the highest local maximum as the start and end, respectively, of the unfolding curve were identified and these values used for normalization (cf. ordinate in Fig. 5A). This temperature range was then searched for the temperature value that corresponds to half-maximal fluorescence intensity.

Parallel crystallization screen of MK2 constructs 4, 8, 12, 15, and 16
Proteins were adjusted to concentrations between 5 and 10 mg/mL, ADP was added (stock solution of 100 mM ADP/MgCl2, final ADP and MgCl2 concentration of 5mM), and the samples were subjected to a crystallization screen comprising 396 conditions (screens "Classic," "Ammonium Sulfate," "pH Clear," and "pH Clear II," purchased from Nextal, in prefilled Greiner 96-well plates). All crystallization screens were set up with a Cartesian Synquad crystallization robot using the sitting drop method. The plates were stored at 20°C and imaged at days 0, 1, 4, 10, 15, 21, and 60. The photos were inspected for crystals.


   Footnotes
 
3 These authors contributed equally to this work. Back

Reprint requests to: Beate Müller-Tiemann, Schering AG, Enabling Technologies, Protein Chemistry Department, 13342 Berlin, Germany; e-mail: beate.muellertiemann{at}schering.de; fax: +49-30-46892850.

Article and publication are at http://www.proteinscience.org/cgi/doi/10.1110/ps.062491906.


   Acknowledgments
TOP
 Abstract
 Introduction
 Results
 Discussion
 Materials and methods
 Acknowledgments
References
 
We thank M. Mann, A. Einfeldt, I. Künkel, S. Schnabel, C. Moldenhauer, A. Sparmann, S. Grabow, S. Dämmig, M. Wilke, and N. Mörbt for technical help and Norbert Otto for helpful discussions.


   References
TOP
 Abstract
 Introduction
 Results
 Discussion
 Materials and methods
 Acknowledgments
 References
 
Ben-Levy, R., Leighton, I.A., Doza, Y.N., Attwood, P., Morrice, N., Marshall, C.J., and Cohen, P. 1995. Identification of novel phosphorylation sites required for activation of MAPKAP kinase-2. EMBO J. 14: 5920–5930.[Medline]

Chambers, S.P. 2002. High-throughput protein expression for the post-genomic era. Drug Discov. Today 7: 759–765.[CrossRef][Medline]

Czepas, J., Devedjiev, Y., Krowarsch, D., Derewenda, U., Otlewski, J., and Derewenda, Z.S. 2004. The impact of Lys -> Arg surface mutations on the crystallization of the globular domain of RhoGDI. Acta Crystallogr. D Biol. Crystallogr. 60: 275–280.[CrossRef][Medline]

Derewenda, Z.S. 2004. Rational protein crystallization by mutational surface engineering. Structure 12: 529–535.[Medline]

Engel, K., Plath, K., and Gaestel, M. 1993. The MAP kinase-activated protein kinase 2 contains a proline-rich SH3-binding domain. FEBS Lett. 336: 143–147.[CrossRef][Medline]

Gaestel, M. 2006. MAPKAP kinases—MKs—Two's company, three's a crowd. Nat. Rev. Mol. Cell Biol. 7: 120–130.[CrossRef][Medline]

Gao, X., Bain, K., Bonanno, J.B., Buchanan, M., Henderson, D., Lorimer, D., Marsh, C., Reynes, J.A., Sauder, J.M., and Schwinn, K., et al. 2005. High-throughput limited proteolysis/mass spectrometry for protein domain elucidation. J. Struct. Funct. Genomics 6: 129–134.[CrossRef][Medline]

Hakes, D.J. and Dixon, J.E. 1992. New vectors for high level expression of recombinant proteins in bacteria. Anal. Biochem. 202: 293–298.[CrossRef][Medline]

Heinemann, U., Bussow, K., Mueller, U., and Umbach, P. 2003. Facilities and methods for the high-throughput crystal structural analysis of human proteins. Acc. Chem. Res. 36: 157–163.[CrossRef][Medline]

Hummel, W. and Kula, M.R. 1989. Simple method for small-scale disruption of bacteria and yeasts. J. Microbiol. Methods. 9: 201–209.

Jenkins, T.M., Hickman, A.B., Dyda, F., Ghirlando, R., Davies, D.R., and Craigie, R. 1995. Catalytic domain of human immunodeficiency virus type 1 integrase: Identification of a soluble mutant by systematic replacement of hydrophobic residues. Proc. Natl. Acad. Sci. 92: 6057–6061.[Abstract/Free Full Text]

Kervinen, J., Ma, H., Bayoumy, S., Schubert, C., Milligan, C., Lewandowski, F., Moriarty, K., Desjarlais, R.L., Ramachandren, K., and Wang, H., et al. 2006. Effect of construct design on MAPKAP kinase-2 activity, thermodynamic stability and ligand-binding affinity. Arch. Biochem. Biophys. 449: 47–56.[CrossRef][Medline]

Kotlyarov, A. and Gaestel, M. 2002. Is MK2 (mitogen-activated protein kinase-activated protein kinase 2) the key for understanding post-transcriptional regulation of gene expression? Biochem. Soc. Trans. 30: 959–963.[CrossRef][Medline]

Kotlyarov, A., Neininger, A., Schubert, C., Eckert, R., Birchmeier, C., Volk, H.D., and Gaestel, M. 1999. MAPKAP kinase 2 is essential for LPS-induced TNF-{alpha} biosynthesis. Nat. Cell Biol. 1: 94–97.[CrossRef][Medline]

Kotlyarov, A., Yannoni, Y., Fritz, S., Laass, K., Telliez, J.B., Pitman, D., Lin, L.L., and Gaestel, M. 2002. Distinct cellular functions of MK2. Mol. Cell. Biol. 22: 4827–4835.[Abstract/Free Full Text]

Kurumbail, R.G., Pawlitz, J.L., Stegeman, R.A., Stallings, W.C., Shieh, H.-S., Mourey, R.J., Bolten, S.L., and Broadus, R.M. 2003. Crystalline structure of human Mapkap kinase-II. Patent application WO 03/076333.

Lanio, T., Jeltsch, A., and Pingoud, A. 2003. High-throughput purification of polyHis-tagged recombinant fusion proteins. Methods Mol. Biol. 205: 199–203.[Medline]

Lesley, S.A., Kuhn, P., Godzik, A., Deacon, A.M., Mathews, I., Kreusch, A., Spraggon, G., Klock, H.E., McMullan, D., and Shin, T., et al. 2002. Structural genomics of the Thermotoga maritima proteome implemented in a high-throughput structure determination pipeline. Proc. Natl. Acad. Sci. 99: 11664–11669.[Abstract/Free Full Text]

Lo, M.C., Aulabaugh, A., Jin, G., Cowling, R., Bard, J., Malamas, M., and Ellestad, G. 2004. Evaluation of fluorescence-based thermal shift assays for hit identification in drug discovery. Anal. Biochem. 332: 153–159.[CrossRef][Medline]

Meng, W., Swenson, L.L., Fitzgibbon, M.J., Hayakawa, K., Ter Haar, E., Behrens, A.E., Fulghum, J.R., and Lippke, J.A. 2002. Structure of mitogen-activated protein kinase-activated protein (MAPKAP) kinase 2 suggests a bifunctional switch that couples kinase activation with nuclear export. J. Biol. Chem. 277: 37401–37405.[Abstract/Free Full Text]

Roux, P.P. and Blenis, J. 2004. ERK and p38 MAPK-activated protein kinases: A family of protein kinases with diverse biological functions. Microbiol. Mol. Biol. Rev. 68: 320–344.[Abstract/Free Full Text]

Rupp, B. 2003. High-throughput crystallography at an affordable cost: The TB Structural Genomics Consortium Crystallization Facility. Acc. Chem. Res. 36: 173–181.[CrossRef][Medline]

Scheich, C., Sievert, V., and Bussow, K. 2003. An automated method for high-throughput protein purification applied to a comparison of His-tag and GST-tag affinity chromatography. BMC Biotechnol. 3: 12.[CrossRef][Medline]

Service, R.F. 2002. Structural genomics. Tapping DNA for structures produces a trickle. Science 298: 948–950.[Abstract/Free Full Text]

Stokoe, D., Campbell, D.G., Nakielny, S., Hidaka, H., Leevers, S.J., Marshall, C., and Cohen, P. 1992. MAPKAP kinase-2; a novel protein kinase activated by mitogen-activated protein kinase. EMBO J. 11: 3985–3994.[Medline]

Stokoe, D., Caudwell, B., Cohen, P.T., and Cohen, P. 1993. The substrate specificity and structure of mitogen-activated protein (MAP) kinase-activated protein kinase-2. Biochem. J. 296: 843–849.[Medline]

Underwood, K.W., Parris, K.D., Federico, E., Mosyak, L., Czerwinski, R.M., Shane, T., Taylor, M., Svenson, K., Liu, Y., and Hsiao, C.L., et al. 2003. Catalytically active MAP KAP kinase 2 structures in complex with staurosporine and ADP reveal differences with the autoinhibited enzyme. Structure 11: 627–636.[Medline]

Wilson, W.W. 2003. Light scattering as a diagnostic for protein crystal growth—A practical approach. J. Struct. Biol. 142: 56–65.[CrossRef][Medline]

Winzen, R., Kracht, M., Ritter, B., Wilhelm, A., Chen, C.Y., Shyu, A.B., Muller, M., Gaestel, M., Resch, K., and Holtmann, H. 1999. The p38 MAP kinase pathway signals for cytokine-induced mRNA stabilization via MAP kinase-activated protein kinase 2 and an AU-rich region-targeted mechanism. EMBO J. 18: 4969–4980.[CrossRef][Medline]

Zu, Y.L., Wu, F., Gilchrist, A., Ai, Y., Labadia, M.E., and Huang, C.K. 1994. The primary structure of a human MAP kinase activated protein kinase 2. Biochem. Biophys. Res. Commun. 200: 1118–1124.[CrossRef][Medline]


Add to CiteULike CiteULike   Add to Connotea Connotea   Add to Del.icio.us Del.icio.us   Add to Digg Digg   Add to Reddit Reddit   Add to Technorati Technorati    What's this?


This article has been cited by other articles:


Home page
 Biophys. JHome page
M. Chruszcz, A. Wlodawer, and W. Minor
Determination of Protein Structures--A Series of Fortunate Events
Biophys. J., July 1, 2008; 95(1): 1 - 9.
[Abstract] [Full Text] [PDF]

Home page
 Appl. Environ. Microbiol.Home page
E. Blas-Galindo, F. Cava, E. Lopez-Vinas, J. Mendieta, and J. Berenguer
Use of a Dominant rpsL Allele Conferring Streptomycin Dependence for Positive and Negative Selection in Thermus thermophilus
Appl. Envir. Microbiol., August 15, 2007; 73(16): 5138 - 5145.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Full Text (PDF)
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Malawski, G. A.
Right arrow Articles by Müller-Tiemann, B.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow