Building meaningful models of glycoproteins

354 VOLUME 14 NUMBER 5 MAY 2007 NATURE STRUCTURAL & MOLECULAR BIOLOGY To the Editor: The building of crystallographic models of proteins is guided by understanding of the primary structure of proteins and by well-established and rigorously applied stereochemical principles. However, a cursory survey of Protein Data Bank entries containing oligosaccharides suggests that of the order of one-third of entries contain significant errors in carbohydrate stereochemistry, nomenclature or even consistency with the electron density maps. Many of the stereochemical errors can be detected by reference to conformational studies of glycans1,2 and to publicly available resources (http://www. glycosciences.de/tools/). However, these errors also indicate that there is a wide discrepancy in the sophistication of building and validation tools available for protein and carbohydrate models. An example of the difficulties that can be encountered when building crystallographic models of glycoproteins is the recent model proposed by Szakonyi et al.3 of the EpsteinBarr virus major envelope glycoprotein, EBV gp350. EBV gp350 was expressed in Spodoptera frugiperda Sf9 cells, and Szakonyi et al.3 report that they observed electron density corresponding to the oligosaccharide chains of fourteen N-linked glycosylation sites. The crystallization of such a heavily glycosylated glycoprotein is a notable achievement. However, the proposed model contains not only systematic errors in carbohydrate stereochemistry, presumably resulting from inadequate parameter files, but also hitherto unreported motifs in the primary structures of the glycans. The previously undescribed glycosidic linkages and motifs that Szakonyi et al.3 propose include Man-(1→3)-GlcNAc and GlcNAc(1→3)-GlcNAc linkages (of indeterminate anomericity) within the trimannosyl core, hybrid-type glycans containing a terminal Man-(1→3)-GlcNAc linkage on the 3antennae, and β-galactosyl motifs capping oligomannose-type glycans. We suggest that, in the absence of supporting evidence, electron density at 3.5-Å resolution should not be used to support linkages incompatible Building meaningful models of glycoproteins