The crystal structure of human cytosolic serine hydroxymethyltransferase: a target … · 2017-02-14 · The crystal structure of human cytosolic serine hydroxymethyltransferase:

The crystal structure of human cytosolic serinehydroxymethyltransferase: a target for cancer chemotherapySuzanne B Renwick†, Keith Snell and Ulrich Baumann‡*

Background: Serine hydroxymethyltransferase (SHMT) is a ubiquitous enzymefound in all prokaryotes and eukaryotes. As an enzyme of the thymidylate synthasemetabolic cycle, SHMT catalyses the retro-aldol cleavage of serine to glycine, withthe resulting hydroxymethyl group being transferred to tetrahydrofolate to form5,10-methylene-tetrahydrofolate. The latter is the major source of one-carbon unitsin metabolism. Elevated SHMT activity has been shown to be coupled to theincreased demand for DNA synthesis in rapidly proliferating cells, particularlytumour cells. Consequently, the central role of SHMT in nucleotide biosynthesismakes it an attractive target for cancer chemotherapy.

Results: We have solved the crystal structure of human cytosolic SHMT bymultiple isomorphous replacement to 2.65 Å resolution. The monomer has afold typical for α class pyridoxal 5′-phosphate (PLP) dependent enzymes. Thetetramer association is best described as a ‘dimer of dimers’ where residuesfrom both subunits of one ‘tight’ dimer contribute to the active site.

Conclusions: The crystal structure shows the evolutionary relationshipbetween SHMT and other α class PLP-dependent enzymes, as the fold is highlyconserved. Many of the results of site-directed mutagenesis studies can easilybe rationalised or re-interpreted in light of the structure presented here. Forexample, His151 is not the catalytic base, contrary to the findings of others. Amechanism for the cleavage of serine to glycine and formaldehyde is proposed.

IntroductionSerine hydroxymethyltransferase (SHMT; EC 2.1.2.1) is aubiquitous enzyme that is central to one-carbon metabo-lism in most organisms, including prokaryotes, eukaryotesand archaebacteria. To date, the amino acid sequences ofSHMTs from 32 species have been deposited in theSWISSPROT data bank and exhibit sequence identitiesvarying between 32% and 93%. SHMT is a pyridoxal5′-phosphate (PLP) dependent enzyme and has beenassigned to the α class of PLP enzymes [1], but it exhibitsonly very low sequence identity with other members ofthis family (e.g. aspartate aminotransferase). The majorphysiological reaction catalysed by SHMT is depicted inFigure 1 and involves the retro-aldol cleavage of serine toglycine, with the one-carbon unit being transferred totetrahydropteroylglutamate. The resulting 5,10-methylene-tetrahydropteroylglutamate (5,10-methylene-tetrahydrofo-late) is the major one-carbon donor for the biosynthesis ofthymidylate, purines, methionine and choline. In additionto the physiological reaction, SHMT has also been shown tocatalyse transaminations, racemisations, and decarboxyla-tions, at least in vitro (for a review see [2]).

Of the trio of enzymes involved in the thymidylate syn-thase cycle, SHMT is the only enzyme yet to be studiedclinically as a target for cancer chemotherapy. The

5,10-methylene-tetrahydrofolate produced by SHMT par-ticipates directly in pyrimidine biosynthesis by donating amethyl group used for methylating 2′-deoxyuridinemonophosphate (dUMP) to 2′-deoxythymidine monophos-phate (dTMP), and indirectly in purine biosynthesis via itsconversion to 10-formyltetrahydrofolate. The glycine pro-duced in the SHMT reaction can also be utilised in purinebiosynthesis, providing two carbons and a nitrogen of thepurine ring.

With both reaction products involved in nucleotidebiosynthesis, it is not surprising that elevated SHMTactivity has been found in human leukaemic cells [3] andin a variety of solid tumour tissues of human and rodentorigin [4]. Furthermore, a positive correlation of SHMTactivity with tumour growth rate was found in Friendmouse leukemia and Pliss rat lymphosarcoma models [5].However, for serine to be preferentially channelledtowards nucleotide biosynthesis requires a reorientationof the enzymes involved in its synthesis and utilisation.Evidence for such a reorganisation can be seen in tumoursof hepatic origin where the increased activities ofenzymes involved in serine biosynthesis are accompaniedby the retention of SHMT and the simultaneous deletionof the two enzymes involved in serine catabolism —serine aminotransferase and serine dehydratase [4,6,7].

Address: Section of Structural Biology, Institute ofCancer Research, University of London, CotswoldRoad, Sutton, Surrey SM2 5NG, UK.

Present addresses: †Celltech plc, 216 Bath Road,Slough, Berkshire SL1 4EN, UK and ‡Departementfür Chemie und Biochemie, Universität Bern,Freiestrasse 3, CH-3012 Bern, Switzerland.

*Corresponding author.E-mail: [email protected]

Key words: cancer chemotherapy, pyridoxalphosphate enzymes, serine hydroxymethyltransferase,X-ray crystallography

Received: 26 May 1998Revisions requested: 25 June 1998Revisions received: 13 July 1998Accepted: 14 July 1998

Structure 15 September 1998, 6:1105–1116http://biomednet.com/elecref/0969212600601105

© Current Biology Publications ISSN 0969-2126

Research Article 1105

Consequently, the increased capacity for serine biosyn-thesis is coupled specifically to its utilisation fornucleotide biosynthesis as a part of the biochemical com-mitment to cellular replication in cancer cells. This obser-vation signifies the importance of SHMT as a therapeutictarget for cancer treatment. Indeed, inhibition of theenzyme in myeloma cells in culture results in dose-depen-dent inhibition of cellular growth [8].

The crystal structures of the other two enzymes of thethymidylate synthase cycle, dihydrofolate reductase [9,10]and thymidylate synthase [11,12], have already beendetermined. Both enzymes have proven to be clinicallysuccessful targets for anticancer chemotherapy [13,14].

Analogues of dihydrofolate, such as methotrexate, arepotent inhibitors of dihydrofolate reductase and havebeen used clinically as anticancer agents for over 50 years.Agents such as fluorouracil, and the newer antifolate drugTomudex, have proved to be effective inhibitors ofthymidylate synthase and are clinically established drugsin the treatment of cancer. Drugs targeting these enzymeshave also been used as antibacterial and antiprotozoalagents by exploiting the differences between the catalyticsites of isoenzymes from the host and target organism[15–17]. With such considerable effort put into designingdrugs against these enzymes, it is surprising that so littleattention has been paid to the inhibition of SHMT.Whereas inhibition of the other enzymes preferentially

1106 Structure 1998, Vol 6 No 9

Figure 1

Proposed reaction cycle catalysed by SHMT.Initially, the cofactor PLP is covalently boundas a Schiff base to the sidechain of Lys257.The retro-aldol cleavage is the reaction stepleading from intermediate II to III and involvesdeprotonation of the serine external aldimineby an unidentified base (B). The secondproduct of this step, formaldehyde, reacts withtetrahydrofolate to form 5,10-methylene-tetrahydrofolate. The absorption maxima ofsome intermediates are indicated in nm.

N CH3

OO3PO

NK257

H

H

NH2CH

CO O

N CH3

OO3PO

NK257

H

HNHO2C

OH

H

N CH3

OO3PO

NHK257

H

HNHO2C

OH

Internal aldimine422 nm

Geminal diamine343 nm

HO

I

N CH3

OO3PO

N

H

H

COOOB H

Serine external aldimine427 nm

II

BH

N CH3

OO3PO

N

H

H

COO

Quinoide aldimine495 nm

III

N CH3

OO3PO

N

H

H

COONH2

K257

External aldimine425 nm

N CH3

OO3PO

N

H

NH2

CO2

H

K257

N CH3

OO3PO

NK257

H

HNHO2CH

Retro-aldol cleavage

Serine

Glycine

THF

5,10-CH2THF

IV

K257-NH2

K257-NH2

Structure

disrupts pyrimidine biosynthesis, drug targeting of SHMTwould have the advantage of blocking DNA synthesis bythe concurrent action on both pyrimidine and purinebiosynthesis. This highlights the feasibility of using thethree-dimensional structure of this enzyme in the designof a clinically useful SHMT inhibitor.

In eukaryotes there are cytosolic and mitochondrial iso-forms of SHMT. Human cytosolic SHMT (cSHMT)exists as a homotetramer, with each monomer binding asingle PLP cofactor. In the absence of an X-ray crystalstructure, past studies have been performed by chemicalmodification and site-directed mutagenesis based onamino acid residues that are totally conserved across theknown SHMT protein sequences. Much of this work hasconcentrated on the dimeric enzyme from Escherichia coli[18–22] and on the tetrameric sheep liver enzyme [23,24].

A rational interpretation of these results requires aknowledge of the three-dimensional structure of theenzyme. The successful crystallisation of the E. colienzyme has been published [25], but as yet no structurehas been reported. We report here, for the first time, thecrystal structure of human recombinant cSHMT. Thestructure reveals a fold typical for α class PLP-depen-dent enzymes and a tetrameric architecture bestdescribed as a ‘dimer of dimers’. The active-site residuescan be readily correlated with the results of biochemicaland mutagenesis studies.

Results and discussionRefinementSome relevant refinement statistics are shown in Table 1.The final model contains residues 11–480, one PLP mol-ecule and 97 water molecules. Residues 1–10 and theattached His-tag are completely disordered, whereasresidues 275–280 have weak density. Cys110, which islocated at the surface, is apparently covalently modifiedbut the ligand could neither be modelled as dithiothreitolnor glutathione and its chemical nature remains unclear atthe present time. All other cysteine residues are in thereduced form. As expected with cytosolic proteins, thereare no pairwise arrangements of cysteines indicative of thepossible presence of a disulphide bond.

The compatibility of the primary sequence and three-dimensional fold was confirmed using the program ProsaII[26]. The z scores for Cβ–Cβ pair potentials, surface andcombined energies are –8.80, –7.47 and –11.62, respec-tively, and are well within the range expected for nativeprotein folds. All other conformations tested exhibitedmuch worse z scores.

The overall B factor is quite high which probably originatesfrom the high solvent content, no sigma cut-off and theinclusion of bulk-solvent corrected low-resolution terms.

Backbone conformationThe Ramachandran plot shows that 97% of all residuesfall into the core region as defined by Kleywegt andJones [27]. The two most prominent outliers are Glu133and Lys257. Both residues can be shifted to energeticallymore favourable regions by flipping the precedingpeptide bonds. In the case of Glu133 this flip wouldcreate too close a contact between the carbonyl oxygensof residues 128 and 132, whereas in the ‘Ramachandran-forbidden’ conformation this interaction is converted to ahydrogen bond between the carbonyl oxygen of residue128 and the amide nitrogen of residue 132. Similarly,Lys257 forms a hydrogen bond between the carbonyloxygen of residue 256 and the amide nitrogen of residue259. Difference electron densities of the ‘flipped’ (i.e.‘Ramachandran-allowed’) conformations seem to indi-cate that the original strained model is correct. In thecase of Lys257, a comparison with tyrosine phenol-lyase(TPL) indicates a similar, although not identical,strained conformation for this residue. Of course, theusual caveats hold for all these residues with respect tothe limited resolution of the crystal structure. Pro298 isin the cis-conformation; the electron density is unam-biguous here, so we believe that this is correct despitethe limited resolution.

Crystal packingThere is only one kind of crystal contact found in the struc-ture, which involves the ‘small’ domain (see below). Thecontacts are symmetrical and involve stretches aroundresidues 346, 423 and 467. The intertetramer contacts arecreated by the 62 screw operation and hence are along thec axis. As can be seen from Figure 2, the molecules packmost densely along the c axis and there are only intrate-tramer contacts in the a,b plane, which is consequently

Research Article Serine hydroxymethyltransferase Renwick, Snell and Baumann 1107

Table 1

Refinement statistics.

Resolution range (Å) 40.0–2.65Number of reflections 48,595Number of atoms (total) 3710Number of solvent molecules 97Observable/parameter ratio 3.0R/Rfree* 0.210/0.226Average B factorsall atoms (Å2) 61.1solvent (Å2) 53.2PLP (Å2) 53.5

Root mean square deviationbonds (Å) 0.006bond angles (°) 1.2dihedrals (°) 22.4impropers (°) 1.1B (bonded atoms) (Å2) 3.4

Residues in core Ramachandran (%) 97

*R = ∑ |Fo–Fc| / ∑ Fo. The Rfree was calculated using a test set of 2760reflections.

much more loosely packed. This observation explains theanisotropic diffraction properties of the crystals.

Monomer structureThe overall fold (Figure 3) of the monomer of humancSHMT is similar to that of other PLP-dependentenzymes of the α class, such as aspartate aminotransferase(AAT) [28]. A search for structural homologues using theprogram DALI [29] identified TPL [30] (apo form, PDBcode 1TPL; z score 20.9), dialkylglycine decarboxylase(DKB) [31,32] (PDB code 2DKB, z score 20.8), AAT[28,33] (PDB code 7AAT, z score 18.1) and ornithinedecarboxylase (ORD) [34] (PDB code 1ORD, z score16.5) as the most similar structures in the database. Thestructures of two other similar PLP-dependent enzymeshave been solved, namely ω-amino acid aminotransferase[35] and cystathionine β-lyase [36], the latter belonging tothe γ class. The coordinates of these two structures arecurrently unavailable. Clearly, all of these enzymes havethe same characteristic fold (Figure 3c), which will bedescribed briefly in the following section.

The monomer fold can be described in terms of threedomains (Figure 3): the N terminus, the ‘large’ domainand the ‘small’ domain. The first domain, the N termi-nus (residues 11–53), mediates intersubunit contacts andfolds into two α helices and one β strand. The second(large) N-terminal domain (residues 53–321) binds PLP.This domain is very similar to the corresponding cofac-tor-binding domains in AAT, ORD and TPL. It basicallyfolds into an αβα sandwich containing nine α heliceswrapped around a seven-stranded mixed β sheet. Thesheet topology is –5x,–1x,2x,1x,1x,1, in Richardson’snotation [37], with right-handed cross-over connections.The sheet is shielded from the solvent by five helices onone side and by three helices on the other. Finally, theC-terminal small domain (residues 322–480) folds into anαβ sandwich. The antiparallel β sheet has –1,–2x,1

topology. This sheet packs on one side against the largedomain and is shielded from the solvent by four heliceson the other side. The interdomain contacts are medi-ated by segments 143–146, 175–180, 204–207 and354–359. The long helix h13 (residues 322–345) has beenassigned to this small subunit, largely to maintain consis-tency with earlier assignments in AAT and TPL. Thishelix connects the large and the small domain and, incontrast to the corresponding helix in AAT, exhibits onlya slight bend.

Quaternary structureHuman cSHMT is a tetramer in solution and thisoligomeric architecture is also observed in the crystals(Figure 4). The tetramer is a crystallographic one, that is,the centre of the molecule occupies a special position inthe crystal lattice with 222 (D2) symmetry.

The four protomers interact with each other to signifi-cantly different degrees; upon tetramer formation asurface of 22,232 Å2 becomes buried. Formation of the‘tight’ dimers A–D and B–C buries 9249 Å2 each, whereasformation of the dimers A–C and B–D buries 39 Å2 eachand A–B and C–D each bury 1849 Å2 . These data showthat the tetramer is best described as a dimer of dimers.This situation is very similar to the one seen in DKB, butquite different from the oligomer association in TPL,where a more symmetrical interaction is found.

All mammalian SHMT proteins examined so far have beenfound to be tetramers. In contrast, the E. coli enzyme hasbeen described as both a tetramer and dimer, dependenton the expression level in E. coli. Under native conditions,however, the enzyme appears to be a dimer [38]. From thediscussion above it is easy to rationalise why SHMT canhave a dimeric or tetrameric quatenary structure: the func-tional entity is a dimer and tetramer formation does notrequire the rebuilding of a large number of contacts.


Figure 2

Stereoview showing crystal packing. Thesmall domain is coloured in red, the largedomain is in green, and the N terminus is inmagenta. The tetramer is a crystallographicone; each small domain mediates crystalcontacts to one other small domain of adifferent tetramer.

Structure

Site-directed mutagenesis studies on sheep liver SHMT[25] revealed that His134, corresponding to His135 in thehuman cytosolic enzyme, is involved in tetramer forma-tion. His135 in the human enzyme is positioned in thevery centre of the tetramer (Figure 5). The imidazolesidechain ring of a particular subunit stacks onto theequivalent ring of another subunit, that is, the interactions

are His135(A)–His135(B) and His135(C)–His135(D). Fur-thermore, within the same subunit the sidechain ofHis135(A) is hydrogen bonded to Glu168(A) which, inturn, forms a hydrogen bond to Arg137(B) of a neighbour-ing subunit. Hence, it is easy to understand why substitut-ing histidine for asparagine can interrupt the tetramerwhilst leaving the tight dimers intact. The dimeric E. coli


Figure 3

The overall fold of the SHMT monomer. (a) Stereoview ribbon diagram of themonomer. The N terminus is shown in purple,the small domain in red and the large domainin green. The cofactor PLP is shown in ball-and-stick representation. (The figure wasprepared using MOLSCRIPT [61] andRaster3D [62].) (b) Stereoview Cα plot of themonomer with residue numbering. Thedomains are coloured as in (a). (c) StereoviewCα overlay of SHMT (black), dialkylglycinedecarboxylase (DKB; red) and aspartateaminotransferase (AAT; green).

enzyme possesses a glycine residue at position 135, whichis entirely consistent with the interpretation above. It is,however, difficult to explain why the mutant enzymebinds PLP much more weakly than the wild-type enzyme,because interactions with PLP occur only within the tightdimer. Alternatively, one could assume that the dimerobserved in the mutant SHMT is not the catalyticallyactive tight A–D dimer but rather the ‘loose’ A–B dimer.Nevertheless, it would be difficult to imagine how a singlemutation could disrupt the tight dimer interface so com-pletely and yet the loose dimer (e.g. A–B) could still retainsome catalytic activity.

The N terminus is very important for the formation of thetight dimer. Site-directed mutagenesis studies on thesheep liver enzyme [39] have revealed that N-terminaltruncation destabilises the protein. The basis for this canbe seen from Figure 4: the two N-terminal helices act asclamps that are crucial for stabilisation of the tight dimer.

The active siteThe cofactor PLP is covalently bound to Lys257 in theactive site of the molecule (Figures 6 and 7). The proto-nated N1 nitrogen of the pyridine ring is hydrogenbonded to Oδ1 of Asp228 which, in turn, is hydrogen


Figure 4

Stereoview cartoon representation of theSHMT tetramer. Subunit A is coloured incyan, B in brown, C in green and D in purple.A–D and B–C form the ‘tight’ dimers, whichare related by a horizontal twofold axis. BoundPLP cofactor is shown in space-fillingrepresentation.

Figure 5

Stereoview of the tetramer centre showing thecentral role of His135. The carbon atoms ofthe A, B, C and D subunits are coloured inpurple, green, cyan and black, respectively.For the sake of clarity, only two of the fourhydrogen-bond networks involving His135and its symmetry-related mates are drawn(dashed lines).

bonded to the Nε1 of His151. The other carboxylateoxygen of Asp228 accepts a hydrogen bond from the main-chain amide nitrogen of Ala230. One side of the pyridinering of the cofactor is stacked onto the imidazole ring ofHis148. This interaction is also present in ORD where thecofactor stacks onto His223, while in AAT and DKB thishistidine residue is replaced by a tryptophan. On the otherside, the pyridine ring is held in place by Ala230, a residuethat is conserved in AAT and ORD, but not in TPL whereit is replaced by a threonine. The phosphate group of PLPis hydrogen bonded to various residues, including Tyr73′and Gly303′ from a neighbouring subunit. The O3′ of

PLP is hydrogen bonded to the sidechains of Ser203 andHis231, both residues being strictly conserved in allSHMT protein sequences. Not surprisingly, the two sub-units forming the active site are the same two that formthe tight dimer (A–D).

The similarity between the active sites of AAT, ORD andSHMT is revealed by their overlay shown in Figure 7. Inparticular, ORD and SHMT share the feature that thePLP-binding lysine residue is preceded by a histidine,making these two proteins members of a distinct subclasswithin the α class of PLP-dependent enzymes.


Figure 6

The active site of SHMT. (a) Stereoviewactive site cartoon depicting key residues. TheA and D subunits are shown as orange-brownand cyan traces, respectively. PLP carbonatoms are shown in green, carbon atoms inblack, oxygens in red, nitrogens in blue andphosphorus in yellow. Hydrogen bonds areshown as dotted lines. (b) Stereoview Fo–Fc‘omit’ electron-density map overlaid on thefinal model. The contour level is 3σ. (c)Stereoview showing a tentative model of theserine–PLP complex. The same colourscheme is used as for (a).

In previous investigations the role of some amino acidsconserved between SHMT sequences has been probed bysite-directed mutagenesis. These results can be readilyinterpreted with the structural data now available.

His256 has been implicated in reaction specificity [19].Besides the physiological reaction outlined in Figure 1,SHMT can catalyse L- and D-alanine transaminationand racemisations, but exhibits strict reaction speci-ficity with its natural substrates glycine and L-serine.The reaction pathway is stereoelectronically controlledby the requirement that the group to be released mustbe oriented orthogonal to the π-bond system of thepyridoxal ring. Therefore, it has been suggested thatHis256 binds to the γ-hydroxyl group of the serine andorients it with respect to the PLP-ring plane. Prelimi-nary modelling of the serine-aldimine was carried outusing the analogous structures of AAT and aligning theserine carboxylate group with the Arg402 guanidinogroup. In this simple model, which does not take intoaccount any possible conformational changes in theprotein, the distance between His256 and the serinesidechain is too large for any direct interaction to takeplace (Figure 6c).

Catalytic baseOne of the possible reaction mechanisms for the conversionof serine and tetrahydrofolate to glycine and methylene-tetrahydrofolate involves a retro-aldol cleavage [40,41](Figure 1). This mechanism requires that a catalytic baseabstracts the proton from the γ-hydroxyl group of serine; theresulting formaldehyde is transferred in a condensationreaction onto tetrahydrofolate, yielding methylene-tetrahy-drofolate. The catalytic base has not been identified so farin SHMT. Whereas in AAT the PLP-attachment lysineresidue has been implicated as the catalytic base [33,42],site-directed mutagenesis studies indicated that this is notthe case in SHMT [20] and ORD [34]. Mutagenesis studiesperformed on sheep liver SHMT were carried out to replacethe conserved residues His134, His147 and His150 (corre-sponding to His135, His148 and His151 in human cSHMT)with asparagine [24]. These studies revealed that His135 isinvolved in tetramer contacts (see above), His148 in PLPbinding and formation of the dimer interface, and His151has been implicated as the catalytic base. On the basis of ourstructure we can now readily account for these findings.

Mutation of His148 leads to the loss of PLP and partialdimer disruption. As mentioned above, His148 stacks onto


Figure 7

Steroview comparisons of the active sites fromdifferent PLP-dependent enzymes.(a) Overlay of the active sites of aspartateaminotransferase (AAT; cyan) and SHMT(green). (b) Overlay of ornithine decarboxylase(ORD; cyan) and SHMT (green).

Asp M228 Asp M228

His M231His M231

Lys M257Lys M257

His M151His M151

His M256His M256

His M148 His M148Ser M121Ser M121

His231His231Lys257Lys257

Ser203Ser203

His151 His151

Ser121Ser121

His148 His148

Structure

(a)

(b)

the pyridine ring of the PLP and thus is crucial in cofactorbinding. Because the phosphate moiety of PLP makescontact to the other subunit of the tight dimer, loss of PLPleads to a weakened dimer–dimer interaction.

His151 stabilises, via hydrogen bonding, the negativecharge on Asp228 which is crucial for stabilising the posi-tive charge at the N1 of the PLP cofactor. Protonation ofthe PLP pyridine ring is important for the electron-with-drawal properties of PLP. Hence, the loss of activity bymutation of His151 is readily explained, but His151 isclearly not the catalytic base.

As mentioned previously, in the case of AAT and cys-tathionine β-lyase [36] the lysine residue attaching to thecofactor has been implicated as the catalytic base, whereasin SHMT it was deduced that another residue must fulfilthis role. This is slightly surprising in terms of evolution-ary relationships between these enzymes and prompted usto find other possible candidates.

Besides Lys257 there are two other residues which couldact as the catalytic base, Glu75(D) and Tyr83(D)(Figure 6c), both from the partner subunit (D) of the tightdimer. Both residues are absolutely conserved amongstthe various SHMT sequences. The sidechain of Tyr83(D)is not well defined in the electron-density map. Glu75(D)is well defined and in a suitable position to abstract theproton from the γ-hydroxyl group of the external aldimine,as a crude modelling of the serine–PLP complex indicates(Figure 6c). Other residues that are within hydrogen-bonding distance include Ser53 and His231. His231 coor-dinates to the O3′ of the cofactor in a way similar toTyr225 in AAT.

Serine- and tetrahydrofolate-binding sitesIn the absence of bound substrates it is assumed that αclass PLP-dependent enzymes exist in the ‘open’ or ‘half-open’ conformations (i.e. the conformation observed inour crystals). Arg402 has been implicated as the residuethat binds to the carboxylate group of serine/glycine [21]and is well positioned to fulfil this role. We are not able toassign the tetrahydrofolate-binding site from the currentdata. Nevertheless, a possible hint might come from theconserved Arg263 residue, which is present in the activesite and might bind to the carboxylate groups of tetrahy-drofolate. Arg263 is conserved in all SHMT sequences,with the exception of those of the archaebacteria. Thesidechain of this residue is completely buried by theD subunit and there is no group that could neutralise itspositive charge. Such binding to tetrahydrofolate wouldrequire substantial conformational changes.

The catalytic mechanismOn the basis of mutagenesis data and our crystal structure,we suggest that the retro-aldol cleavage mechanism is the

most likely explanation of catalysis. Alternatively, a so-called thiohemiacteal mechanism has been discussed [41].This mechanism involves the nucleophilic attack of a cys-teine S-γ on the β carbon of the serine yielding a thio-hemiacteal. The same cysteine reacts with fluoroalanineintermediates to form an inactive adduct. Cys204 has beenidentified as the residue reacting with fluoroalanine orsimilar suicide inhibitors of SHMT. As outlined above, wedo not believe that this thiohemiacteal mechanism is oper-ating in the physiological reaction but nonetheless theinactivity of the modified protein has to be explained.

From our observation that all cysteine-modifying agentscause fracture damage to the crystals, it appears reasonableto assume that a conformational change involving the smallsubunit occurs. It is plausible that modification of Cys204leads to a conformational change involving the smallsubunit, similar to the one observed in AAT [33,42]. Thisconformation is inappropriate for the effective propagationof the reaction (e.g. it might be in the closed conformation).Modification of Cys204, which sits at the interface betweenthe large and small domains, would lock the enzyme, pre-venting any further participation in the reaction.

Biological implicationsSerine hydroxymethyltransferase (SHMT) is a centralenzyme in one-carbon metabolism, catalysing the trans-fer of the β-hydroxyl group of serine to tetrahydrofolateto create 5,10-methylene-tetrahydrofolate. The latter isthe major source of one-carbon units in cellular metabo-lism and is both directly involved in pyrimidine biosyn-thesis and indirectly involved in purine biosynthesis.Elevated levels of SHMT activity have been found inrapidly proliferating cells, particularly in tumours andvarious cancer cell lines. Of the three enzymes partici-pating in the thymidylate synthase metabolic cycle,SHMT is the only one yet to be exploited as an anti-cancer target. The rational development of a tight-binding inhibitor of the enzyme has been hampered byuncertainties over its catalytic mechanism and the com-plete lack of any three-dimensional protein structurefrom any species. Furthermore, although data fromchemical modification and site-directed mutagenesisstudies have been reported, the interpretation of thefindings has been constrained by the absence of anystructural model of the enzyme.

The crystal structure of human cytosolic SHMT, pre-sented here for the first time, reveals that the enzymebelongs to the α class of pyridoxal 5′-phosphate (PLP)dependent enzymes. The tertiary fold of these proteinsshows the monomer to comprise essentially twodomains where the larger N-terminal domain binds thePLP cofactor. In the case of SHMT, the active site isconstructed by residues from two different subunits ofthe tetrameric enzyme.


On the basis of the present structure, amino acids impli-cated in substrate binding and catalysis can be identifiedfor further site-directed mutagenesis studies. Suchresearch will aid the understanding of the catalyticmechanism of action of this enzyme. Furthermore, theprotein structure will also be of considerable value in therational design of enzyme inhibitors which may prove tobe useful in anticancer chemotherapy.

Materials and methodsOver-expression, purification and crystallisationDetails of the purification and crystallisation have been published else-where [43]. Briefly, a cDNA encoding human cSHMT was isolatedfrom the ZR75-1 breast cancer cell line cDNA library (P Byrne, per-sonal communication). SHMT cDNA was amplified by PCR and ligatedinto Nde I digested pET-14b (Novagen). Over-expression was carriedout in E. coli strain BL21(DE3)pLysS (Novagen). SHMT was then puri-fied in one step by metal chelation chromatography, a typical yield was~15 mg of soluble, pure cSHMT from 1 litre of bacterial culture.

Crystals were obtained using the commercial Crystal Screen kit(Hampton Research, Laguna Hills, CA, USA) [44] with HamptonCrystal Screen reagent 25 (1.0 M sodium acetate, 0.1 M imidazolepH 6.5). Hexagonal bipyramidal crystals were observed after 1–2 daysand grew to a typical size of 0.3 × 0.3 × 0.8 mm3 within one week. Thecrystals belong to space group P6222 with cell-dimensionsa = b = 155.0 Å, c = 235.5 Å and contain ~82% solvent which resultsin the surprisingly high Matthews parameter of 7.5 Å3/Da. Later, aslightly different crystal form was produced with cell dimensionsa = b = 154.1 Å, c = 235.4 Å. These small changes in the unit cellbetween the two crystal forms gave rise to significant intensity differ-ences in the order of 12–16%.

Data collection and structure solutionThe average life span of the cSHMT crystals on a rotating-anodesource was less than 10 min. Consequently, crystals were flash-cooledto 110K prior to data collection. Glycerol was introduced by serialtransfer of crystals into crystallisation buffer containing 5–30% (v/v) of

the cryoprotectant. Data were collected in-house using a RigakuRU200 rotating-anode generator operating at 5.4 kW with doublefocusing mirrors, and a Raxis IIc detector. Data were collected on eachcrystal in frames of 0.8° rotation and with an exposure time of 20 minper frame. The crystal-to-detector distance was 250 mm with a swingangle of 9°. In addition, data were collected at beam line BW7B at theEMBL outstation in Hamburg at the DORIS storage ring, DESY, usinga 30 cm MAR image plate at a wavelength of 0.9995 Å. All data wereprocessed using the HKL suite [45].

Some data collection statistics are given in Table 2. As the complete-ness in the shell between 2.63 and 2.60 Å was fairly low, we restrictedthe resolution range to 2.65 Å. The completeness of the merged dataset native 2 plus native 3 is 0.97 in the shell between 2.68 and 2.65 Åand the average I/σ(I) is 2.7 in this shell.

Preliminary trials to solve the structure by molecular replacement wereunsuccessful which was not too surprising considering the lowsequence homology between SHMT and other PLP-dependentenzymes. Hence, structure solution was effected by the method of iso-morphous replacement.

Heavy-atom derivatives were prepared by soaking crystals in motherliquor containing an appropriate amount of the heavy-atom reagent.Most mercury and platinum compounds destroyed the crystals veryquickly; one nonisomorphous mercury derivative was prepared byapplying a mixture of mercury(II) acetate and cysteamine [46]. Otherheavy-atom compounds caused severe changes in the a axis celldimension and these derivatives were useless beyond 6 Å resolution.Selenomethionine-substituted protein was prepared from E. coliB834(DE3)pLysS cells (Novagen) [47]. Eventually, a set of five deriva-tives were found. Derivative data were locally scaled to native datausing MAXSCALE [48]. The potassium osmate (KOS) derivative couldbe interpreted in terms of one major heavy-atom site using SHELX[49]. Further heavy-atom positions were determined by difference Pat-tersons and difference Fourier methods. Most crystallographic compu-tations were effected using CCP4 [50]. Phases were computed usingMLPHARE [51] and later SHARP [52]. The program SHARP, whichincorporates a full maximum-likelihood procedure, gave superiorphases and aided the discovery of minor sites. Phasing statistics are


Table 2

Data collection and phasing statistics.

Data set No. of reflections Resolution Rsym* Completeness <I>/<σ> Riso† No. of Phasing power‡ Rcullis

measured/ (Å) overall/ overall/ sites acentric/ (centric)§

unique last shell last shell centric

Native 1 158,836/28,884 3.0 0.058/0.247 0.857/0.446 21.1/2.71 – – – –KOS10 154,022/29,555 3.1 0.083/0.804 0.971/0.905 19.4/2.3 0.21 4 3.0/2.4 0.55KOS20 194,655/20,757 3.3 0.092/0.532 0.970/0.880 22.7/3.3 0.25 4 3.1/2.4 0.59PIP 78,169/19,647 3.5 0.105/0.403 10.5/1.4 0.29 3 3.3/2.9 0.53PIP05 107,150/26,066 3.3 0.094/0.327 0.937/0.853 8.8/2.0 0.25 2 2.9/2.6 0.55Semet 132,242/20,619 3.5 0.096/0.354 0.958/0.887 15.7/4.5 0.19 8 2.0/1.7 0.72Native 2 260,579/44,864 2.7 0.095/0603 0.999/0.100 18.6/2.5 – – – –Native 3 242,137/49,285 2.6 0.093/0.654 0.977/0.261 14.9/1.9 – – – –Native 2+3 500,988/49,366 2.6 0.102/0.659 0.978/0.259 22.2/1.9 – – – –

KOS10 and KOS20, 10 mM and 20 mM K2OsO4 overnight soak,respectively; PIP and PIP05, 1.0 mM and 0.5 mM di-µ-iodobis(ethylenediamine)diplatinum(II)nitrate overnight soak,respectively; Semet, selenomethionine-subsituted protein. Native 2 andnative 3 were collected at BW7A at the EMBL outstation in Hamburg.R factors were calculated according to Diederichs and Karplus [60]and gave the following results. For native 2, Rmeas = 0.104, pooledcoefficient of variation (PCV [60] ) = 0.163 and Imerge = 0.092. Fornative 3, Rmeas = 0.098, PCV = 0.166 and Imerge = 0.098. For native

2+3, Rmeas = 0.105, PCV = 0.174 and Imerge = 0.068. *Rsym = ΣΣ |I(hkl; i) –<I(hkl>/ΣΣ I(hkl; i), where I(hkl; i ) is the ith measurement ofthe reflection hkl; <I(hkl)> is the corresponding mean value. †Riso = Σ |Ideri(hkl) – Inati(hkl)| / ΣInati(hkl). ‡Phasing power = fH /<E>, where fH isthe heavy-atom amplitude and <E> the mean lack of closure. Thevalues given for the phasing power were calculated using SHARP;these figures are much higher than those computed using MLPHARE.§Rcullis = Σ | FPH ± (FP + Fh

calc) | / Σ (FPH ± – FP), where the summationis over centric reflections only.

given in Table 2. The mean two-dimensional figures of merit at 3.1 Åresolution were 0.54 and 0.65 for acentric and centric reflections,respectively. The electron-density map showed a clear solvent bound-ary and many secondary structure elements and was improved greatlyusing the solvent flipping procedure incorporated into the programSOLOMON [53]. Here the flip factor was optimised to a value of –0.6(solvent content 0.75) by monitoring the relative flatness of a solventpatch excluded from modifications. The improved phase set was usedto compute a map at 3.1 Å resolution which allowed tracing of thecomplete chain with the exception of residues 1–10 and 481–483which are disordered. Model building was performed using theprogram O [54] and was greatly aided by the program DEJAVU [55],which identified the fold in a very early stage of the model building. Theinitial model was subjected to slow-cooling torsion angle refinement[56] and restrained least-squares or restrained maximum-likelihoodrefinement using the programs X-PLOR [57] and REFMAC [58]. Theuse of individual restrained B factors was justified by an equal drop inR factor and Rfree of about 5% (2760 test set reflections) [59]. A bulk-solvent mask and an overall anisotropic B factor (b11 = b22 = –3.7 Å2,b33 = +30.6 Å2, b12 = 1.1 Å2) were applied. Model rebuilding waseffected using the program O and in the final cycles water moleculeswere put in at stereochemically reasonable positions. These water mol-ecules were only retained during refinement if the B factor remainedbelow 60 Å2.

Accession numbersThe coordinates of the cSHMT structure have been deposited in theBrookhaven Protein Data Bank with accession code 1BJ4.

AcknowledgementsWe gratefully acknowledge financial support from the Institute of CancerResearch. We thank S Neidle for sharing his X-ray and computing facilities.The award of beamtime at beamline BW7A at the EMBL outstation inHamburg is acknowledged. We are grateful to Paul Tucker and Ana Gonza-lez from the Hamburg EMBL outstation for help with the synchrotron datacollection. Thanks also to Chris Roome for his assistance before and duringthe data collections in Hamburg.

References1. Alexander, F.W., Sandmeier, E., Mehta, P.K. & Christen, P. (1994).

Evolutionary relationships among pyridoxal-5′-phosphate-dependentenzymes. Eur. J. Biochem. 219, 953-960.

2. Schirch, L. (1984). Folates in serine and glycine metabolism. InFolates and Pterins: Chemistry and Biochemistry of Folates. (Blakely,R.L. & Benkovic, S.J. eds.), pp. 399-412, Wiley-Interscience, NewYork, USA.

3. Thorndike, J., Pelliniemi, T.T. & Beck, W.S. (1979). Serinehydroxymethyltransferase activity and serine incorporation inleukocytes. Cancer Res. 39, 3435-3440.

4. Snell, K., Natsumeda, Y., Eble, J.N., Glover, J.L. & Weber, G. (1988).Enzymic imbalance in serine metabolism in human colon carcinomaand rat sarcoma. Br. J. Cancer. 57, 87-90.

5. Bukin, Y.V. & Draudin-Krylenko, V.A. (1980). Regulation of serinehydroxymethyltransferase activity in normal, tumour and host cells.Vest. Akad. Med. Nauk. SSSR 6, 67-69.

6. Snell, K. (1985). Enzymes of serine metabolism in normal andneoplastic rat tissues. Biochem. Biophys. Acta 843, 276-281.

7. Snell, K. & Weber, G. (1986). Enzymic imbalance in serine metabolismin rat hepatomas. Biochem. J. 233, 617-620.

8. Snell, K. & Riches, D. (1989). Effects of a triazine antifolate (NSC127755) on serine hydroxymethyltransferase in myeloma cells inculture. Cancer Lett. 44, 217-220.

9. Filman, D.J., Bolin, J.T., Matthews, D.A. & Kraut, J. (1982). Crystalstructures of Escherichia coli and Lactobacillus casei dihydrofolatereductase refined to 1.7 Å resolution. II. Environment of boundNADPH and implications for catalysis. J. Biol. Chem.257, 13663-13672.

10. Oefner, C., D’Arcy, A. & Winkler, F.K. (1988). Crystal structure ofhuman dihydrofolate reductase complexed with folate. Eur. J.Biochem. 174, 377-385.

11. Hardy, L.W., Finer-Moore, J.S., Montfort, W.R., Jones, M.O., Santi,D.V. & Stroud, R.M. (1987). Atomic structure of thymidylate synthase:target for rational drug design. Science 235, 448-455.

12. Schiffer, C.A., Clifton, I.J., Davisson, V.J., Santi, D.V. & Stroud, R.M.(1995). Crystal structure of human thymidylate synthase: a structuralmechanism for guiding substrates into the active site. Biochemistry34, 16279-16287.

13. Huennekens, F.M. (1994). The methotrexate story: a paradigm fordevelopment of cancer chemotherapeutic agents. Adv. Enz. Regul.34, 397-419.

14. Rustum, Y.M., et al., & Seeber, S. (1997). Thymidylate synthaseinhibitors in cancer therapy: direct and indirect inhibitors. J. Clin.Oncol. 15, 389-400.

15. Hartman, P.G. (1993). Molecular aspects and mechanism of action ofdihydrofolate reductase inhibitors. J. Chemother. 5, 369-376.

16. Then, R.L. (1993). History and future of antimicrobialdiaminopyrimidines. J. Chemother. 5, 361-368.

17. Costi, M.P. (1998). Thymidylate synthase inhibition: a structure-basedrationale for drug design. Med. Res. Rev. 18, 21-42.

18. Angelaccio, S., Pascarella, S., Fattori, E., Bossa, F., Strong, W. &Schirch, V. (1992). Serine hydroxymethyltransferase: origin ofsubstrate specificity. Biochemistry 31, 155-162.

19. Stover, P., Zamora, M., Shostak, K., Gautam-Basak, M. & Schirch, V.(1992). Escherichia coli serine hydroxymethyltransferase: the role ofhistidine 228 in determining reaction specificity. J. Biol. Chem.267, 17679-17687.

20. Schirch, D., et al., & Schirch, V. (1993). Function of the active-sitelysine in Escherichia coli serine hydroxymethyltransferase. J. Biol.Chem. 268, 23132-23138.

21. Delle Fratte, S., Iurescia, S., Angelaccio, S., Bossa, F. & Schirch, V.(1994). The function of arginine 363 as the substrate carboxyl-bindingsite in Escherichia coli serine hydroxymethyltransferase. Eur. J.Biochem. 225, 395-401.

22. Cai, K. & Schirch, V. (1996). Structural studies on foldingintermediates of serine hydroxymethyltransferase using fluorescenceresonance energy transfer. J. Biol. Chem. 271, 27311-27320.

23. Usha, R., Savithri, H.S. & Rao, N.A. (1992). Arginine residues involvedin binding of tetrahydrofolate to sheep liver serinehydroxymethyltransferase. J. Biol. Chem. 267, 9289-9293.

24. Jagath, J.R., Sharma, B., Rao, N.A. & Savithri, H.S. (1997). The role ofHis134, -147, and -150 residues in subunit assembly, cofactorbinding, and catalysis of sheep liver cytosolic serinehydroxymethyltransferase. J. Biol. Chem. 272, 24355-24362.

25. Stover, P., Kruschwitz, H., Schirch, V. & Wright, H.T. (1993).Diffraction grade crystals of Escherichia coli serinehydroxymethyltransferase. J. Mol. Biol. 230, 1094-1096.

26. Sippl, M.J. (1993). Recognition of errors in three-dimensionalstructures of proteins. Proteins 17, 355-362.

27. Kleywegt, G.J. & Jones, T.A. (1996). Phi/Psi-chology: Ramachandranrevisited. Structure 4, 1395-1400.

28. Kirsch, J.F., Eichele, G., Ford, G.C., Vincent, M.G. & Jansonius, J.N.(1984). Mechanism of the action of aspartate aminotransferase on thebasis of its spatial structure. J. Mol. Biol. 174, 497-525.

29. Holm, L. & Sander, C. (1993). Protein structure comparison byalignment of distance matrices. J. Mol. Biol. 223, 123-138.

30. Antson, A.A., et al., & Wilson, K.S. (1993). Three-dimensionalstructure of tyrosine phenol-lyase. Biochemistry 32, 4195-4206.

31. Toney, M.D., Hohenester, E., Cowan, S.W. & Jansonius, J.N. (1993).Dialkylglycine decarboxylase structure: bifunctional active site andalkali metal sites. Science 261, 756-759.

32. Toney, M.D., Hohenester, E., Keller, J.W. & Jansonius, J.N. (1995).Structural and mechanistic analysis of two refined crystal structures ofthe pyridoxal phosphate dependent enzyme dialkylglycinedecarboxylase. J. Mol. Biol. 245, 151-179.

33. McPhalen, C.A., Vincent, M.G. & Jansonius, J.N. (1992). X-raystructure refinement and comparison of three forms of mitochondrialaspartate aminotransferase. J. Mol. Biol. 225, 495-517.

34. Momay, C., Ernst, S., Ghosh, R. & Hackert, M.L. (1995).Crystallographic structure of a PLP-dependent ornithinedecarboxylase from Lactobacillus 30a. J. Mol. Biol. 252, 643-655.

35. Watanabe, N., et al., & Fukutani, H. (1989). Crystal structure ofω-amino acid: pyruvate aminotransferase with a newly developedWeissenberg camera and an imaging plate using synchrotronradiation. J. Biochem. 105, 1-3.

36. Clausen, T., Huber, R., Laber, B., Pohlenz, H.D. & Messerschmidt, A.(1996). Crystal structure of the pyridoxal-5′-phosphate dependentcystathionine β-lyase from Escherichia coli at 1.83 Å. J. Mol. Biol.262, 202-224.

37. Richardson, J.S. (1981). The taxonomy and anatomy of proteinstructures. Adv. Protein Chem. 34, 167-333.


38. Mansouri, A., Decter, J.B. & Silber, R. (1972). Studies on theregulation of one-carbon metabolism. J. Biol. Chem. 247, 348-352.

39. Jagath, J.R., Sharma, B., Bhaskar, B., Datta, A., Rao, N.A. & Savithri,H.S. (1997). Importance of the amino terminus in maintenance ofoligomeric structure of sheep liver cytosolic serinehydroxymethyltransferase. Eur. J. Biochem. 247, 372-379.

40. Jordan, P.M. & Akhtar, M. (1970). The mechanism of action of serinetranshydroxymethylase. Biochem. J. 116, 277-286.

41. Matthews, R.G. & Drummond, J.T. (1990). Providing one-carbon unitsfor biological methylations: mechanistic studies on serinehydroxymethyltransferase, methylenetetrahydrofolate reductase, andmethyltetrahydrofolate-homocysteine methyltransferase. Chem. Rev.90, 1275-1290.

42. Toney, M.D. & Kirsh, J.F. (1993). Lysine 258 in aspartateaminotransferase: enforcer of the Circe effect for amino acidsubstrates and general-base catalyst for the 1,3-prototropic shift.Biochemistry 32, 1471-1479.

43. Renwick, S.B., Skelly, J.V., Chave, K.J., Sanders, P.G., Snell, K. &Baumann, U. (1998). The purification, crystallization and preliminary X-ray analysis of human recombinant cytosolic serinehydroxymethyltransferase. Acta Cryst. D, in press.

44. Jancarik, J. & Kim, S.H. (1991). Sparse matrix sampling: a screeningmethod for crystallization of proteins. J. Appl. Cryst. 24, 409-411.

45. Otwinowski, Z. & Minor, W. (1997). Processing of X-ray diffractiondata collected in oscillation mode. Methods Enzymol. 276, 307-326.

46. Dirr, H., Reinemer, P. & Huber, R. (1994). Refined crystal structure ofporcine class Pi glutathione S-transferase (pGST P1-1) at 2.1 Åresolution. J. Mol. Biol. 243, 72-92.

47. Doublié, S. & Carter, C.W. Jr. (1992). Preparation of selenomethionylprotein crystals. In Crystallisation of Nucleic Acids and Proteins: APractical Approach. (Ducruix, A. & Giege, R. eds), pp. 311-317, IRLPress/Oxford University Press, Oxford, UK.

48. Rould, M.A. (1997). Screening for heavy-atom derivatives andobtaining accurate isomorphous differences. Methods Enzymol.276, 461-472.

49. Sheldrick, G.M. (1997). Patterson superposition and ab initio phasing.Methods Enzymol. 276, 628-641.

50. CCP4 Collaborative Computational Project, Number 4. (1994). TheCCP4 suite: programs for protein crystallography. Acta Cryst. D 50, 760-763.

51. Otwinowski, Z. (1991). Isomorphous Replacement and AnomalousScattering. In Proceedings of the Daresbury Study Weekend. pp. 80-85, SERC Daresbury Laboratory, Warrington, UK.

52. De La Fortelle, E. & Bricogne, G. (1997). Maximum-likelihood heavy-atom parameter refinement for multiple isomorphous replacement andmultiwavelength anomalous diffraction methods. Methods Enzymol.276, 472-494.

53. Abrahams, J.P. & Leslie, A.G.W. (1996). Methods used in thestructure determination of bovine mitochondrial F1 ATPase. ActaCryst. D 52, 30-42.

54. Jones, T.A., Zou, J.Y., Cowan, S.W. & Kjeldgaard, M. (1991). Improvedmethods for binding protein models in electron density maps and thelocation of errors in these models. Acta Cryst. A 47, 110-119.

55. Kleywegt, G.J. & Jones, T.A. (1997). Detecting folding motifs andsimilarities in protein structures. Methods Enzymol. 277, 525-546.

56. Rice, L.M. & Brünger, A.T. (1994). Torsion angle dynamics: reducedvariable conformational sampling enhances crystallographic structurerefinement. Proteins 19, 277-290.

57. Brünger, A.T. (1992). X-PLOR, a System for X-ray Crystallographyand NMR. Yale University Press, New Haven, CT, USA.

58. Murshudov, G., Vagin, A. & Dodson, E.J. (1997). Refinement ofmacromolecular structures by the maximum entropy method. ActaCryst D 53, 240-255.

59. Brünger, A.T. (1992). The free R value: a novel statistical quantity forassessing the accuracy of crystal structures. Nature 355, 472-474.

60. Diederichs, K. & Karplus, P.A. (1997). Improved R-factors fordiffraction analysis in macromolecular crystallography. Nat. Struct.Biol. 4, 269-275.

61. Kraulis, P.J. (1991). MOLSCRIPT: a program to produce both detailedand schematic plots of protein structures. J. Appl. Cryst. 24, 946-950.

62 Merrit. E.A. & Bacon, D.J. (1997). Raster3D photorealistic moleculargraphics. Methods Enzymol. 277, 505-524.


Documents

The crystal structure of human cytosolic serine hydroxymethyltransferase: a target … · 2017-02-14 · The crystal structure of human cytosolic serine hydroxymethyltransferase: