Bioinformatics Advance Access originally published online on September 13, 2005
Bioinformatics 2005 21(21):3976-3982; doi:10.1093/bioinformatics/bti666
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Prediction of glycan structures from gene expression data based on glycosyltransferase reactions
Bioinformatics Center, Institute for Chemical Research, Kyoto University Uji, Kyoto 611-0011, Japan
*To whom correspondence should be addressed.
 |
Abstract |
|---|
 TOP Abstract 1 INTRODUCTION2 DATASET3 RESULTS4 DISCUSSIONREFERENCES |
|---|
Motivation: Glycan chains are synthesized by a combination of several kinds of glycosyltransferases (GTs). Thus, once we know the repertoire of GTs in the genome, in the transcriptome or in the proteome, it should in principle be possible to predict the repertoire of possible glycan structures in an organism or at a specific stage of the cell. Here, we show that a repertoire of glycan structures can be predicted from the set of GTs in the transcriptome. That is, using knowledge about glycan structure characteristics, we can predict glycan structures from incomplete or noisy data such as DNA microarray data.
Results: First, we constructed a reaction pattern library consisting of bond-formation patterns of GT reactions and investigated the co-occurrence frequencies of all reaction patterns in the glycan database. This was followed by the prediction of glycan structures using this library and a co-occurrence score. A penalty score was also implemented in the prediction method. Then we examined the performance of prediction by the leave-one-out cross validation method using individual reaction pattern profiles in the KEGG GLYCAN database as virtual expression profiles. The accuracy of prediction was 81%. Finally, we applied the prediction method to real expression data. Using expression profiles from the human carcinoma cell, glycan structures with sialic acid and sialyl Lewis X epitope were predicted, which corresponded well with experimental results.
Contact: kanehisa{at}kuicr.kyoto-u.ac.jp
Supplementary information: http://web.kuicr.kyoto-u.ac.jp/~kawano/suppl/bioinfo2005/
|
1 INTRODUCTION |
|---|
|
TOPAbstract1 INTRODUCTION2 DATASET3 RESULTS4 DISCUSSIONREFERENCES |
|---|
Glycans, which attach to some lipids and Asn/Ser/Thr residues of proteins, draw attention as the third type of biological chain next to DNA and proteins, since they play key roles in many biological processes such as fertilization (Vo et al., 2003), embryogenesis (Schachter et al., 2002), immunity (Rudd et al., 2001) and diseases (Birkle et al., 2003; Brockhausen et al., 1998; Hakomori, 2002). Half of the proteins in nature are glycosylated, based on estimates by the analysis of the Swiss-Prot database (Apweiler et al., 1999). It is proposed that the glycosylated proteins in the cell membrane and glycosylated lipids form a lipid raft, and that they are involved in signal transduction (Simons and Toomre, 2000). It is well known that some pathogenic bacteria and viruses infect their hosts via glycan–receptor interactions (Cossart and Sansonetti, 2004; Sacks and Kamhawi, 2001). Since glycans may have many functions such as localization signaling, protein stabilization, degradation signaling, signal transduction and immune reaction via glycan–glycan (Bucior and Burger, 2004) and/or glycan–protein interaction (Kogelberg et al., 2003; Weis and Drickamer, 1996), it is important to understand glycan functions for understanding life.
To understand glycan functions, determination of their structures (sequences) like DNA and proteins is required. In spite of the improvements in purification and analytical methods for glycans such as high performance liquid chromatography, capillary electrophoresis, mass spectrometry and nuclear magnetic resonance technology (von der Lieth et al., 2004), the determination of the glycan structure is still difficult. Glycans have more complicated structures compared to nucleotide and amino acid sequences. While nucleotide and amino acid chains are linear and consist of 4 and 20 elementary components, respectively, glycan chains are branched structures and consist of various monosaccharides. In addition, they are multivalent, and linkages have anomeric configurations (alpha and beta). These complexities make it difficult to determine glycan structure. Furthermore, the amplification method of glycan is not yet fully established, while DNA and proteins are easy to amplify using polymerase chain reactions and cloning-expression systems, respectively. This means that only a few samples are available for glycan structure analysis. Therefore, a reasonable prediction method for glycan structures is useful for glycomics research.
While the amino acid sequence of proteins is determined by the genetic code and the templates in the genome, the carbohydrate sequence of glycans is determined by the biosynthetic code, which is a specific set of biosynthetic reactions catalyzed by different types of glycosyltransferases (GTs). Each GT catalyzes formation of a glycosidic-bond between the glycan precursor as an accepter and the nucleotide-activated sugar as a donor (Varki et al., 1999). Thus, once we know the repertoire of GTs in the genome, in the transcriptome or in the proteome, it should, in principle, be possible to predict the repertoire of possible glycan structures in an organism or at a specific stage of the cell (von der Lieth et al., 2004). Here, we construct a reaction pattern library consisting of bond-formation patterns of GT reactions to link genome to glycome, and we predict glycan structures from gene expression profiles.
However, gene set on DNA chips is incomplete, and gene expression data is noisy. To obtain appropriate prediction result, we extract knowledge of glycan structure from the glycan database and apply it to our prediction method. In particular, a co-occurrence frequency of reaction patterns is calculated from the KEGG GLYCAN database, which is a comprehensive resource encapsulating the latest knowledge of glycans and a part of the KEGG resource containing genomic information and pathways (Kanehisa et al., 2004; Hashimoto et al., 2005a), and we use it together with the reaction pattern library. First, we evaluate the prediction method using virtual expression data generated from the KEGG GLYCAN database. Then, we apply our method to publicly available DNA microarray expression data and find characteristic glycan structures in a particular cell.
|
2 DATASET |
|---|
|
TOPAbstract1 INTRODUCTION2 DATASET3 RESULTS4 DISCUSSIONREFERENCES |
|---|
A relationship of our dataset is shown in Figure 1.
|
2.1 Glycosyltransferase reactions
In order to construct a GT reaction library, GT genes were obtained from the human genome in the KEGG GENES database (Kanehisa et al., 2004) based on their annotations. The reaction specificity of each GT was determined according to the published literature and was characterized by the following three features: (1) acceptor monosaccharide residue in the glycan chain, (2) the donor monosaccharide residue and (3) the linkage between them (Fig. 2). In the human genome, currently, 98 GT genes are annotated, and the reaction pattern library contains 42 reaction patterns (Supplemental data S1). The reaction pattern library consists of nine kinds of monosaccharides: glucose (Glc), galactose (Gal), mannose (Man), N-acetyl-glucosamine (GlcNAc), N-acetyl-galactosamine (GalNAc), fucose (LFuc), xylose (Xyl), glucuronic acid (GlcA) and N-acetyl-neuraminic acid (= sialic acid, Neu5Ac). The number of reaction patterns was less than that of GT genes because the human genome has paralogous genes encoding similar functional proteins. For example, the enzyme reactions of ST8SIA1 (Entrez GeneID: 6489), ST8SIA2 (Entrez GeneID: 8128), ST8SIA3 (Entrez GeneID: 51046), ST8SIA4 (Entrez GeneID: 7903) and ST8SIA5 (Entrez GeneID: 29906) are all the same under our definition of ‘Neu5Ac a2-8 Neu5Ac’ (see Supplemental data S1).
|
2.2 Glycan structures
The primary glycan structures were collected from the KEGG GLYCAN database (rel.32, http://www.genome.jp/kegg/glycan/), which contains 10 938 entries. To obtain glycan entries consisting of only carbohydrates, non-carbohydrate residues in the entries, such as Cer (ceramide), Asn, Ser/Thr, S (sulfate) and P (phosphate) were deleted and duplicated structures were merged. Furthermore, glycan entries including monosaccharides that are not present in the reaction library were removed. Finally, our dataset contained 4107 glycan entries.
2.3 Microarray expression data
DNA microarray expression profiles of human were obtained from the Consortium for Functional Glycomics (CFG) (http://www.functionalglycomics.org/static/consortium/organization/sciCores/coree.shtml). CFG has published statistically processed DNA microarray data with expression status, indicated ‘P’resent, ‘M’arginal and ‘A’bsent, from five experiments using human cell such as lung, leukemia and carcinoma cell lines (Supplemental data S2). GT genes were collected from the expression profiles with their annotations and accession numbers. Corresponding to 37 reaction patterns 80 GT genes were mounted on the arrays (Supplemental data S1). When a GT gene was found to be expressed (P) in the majority of the same experimental conditions, it was determined to be positively expressed.
|
3 RESULTS |
|---|
|
TOPAbstract1 INTRODUCTION2 DATASET3 RESULTS4 DISCUSSIONREFERENCES |
|---|
3.1 Co-occurrence score
Although glycan structures are diverse, many combinations of reaction patterns are conserved. For example, both ‘GlcNAc b1-4 GlcNAc’ and ‘Man b1-4 GlcNAc’ are components of the N-glycan core. To obtain an optimal prediction result from DNA microarray expression data, we introduced a co-occurrence score between the reaction patterns calculated from the current glycan database (KEGG GLYCAN). All glycan structures in the database were broken down into reaction pattern components consisting of two adjacent monosaccharides and their linkage, from which a ‘reaction pattern matrix’ was constructed. Here, the correlation coefficient [SP, Equation (1)], the Tanimoto coefficient [TC, Equation (2)] and the Cosine coefficient [SC, Equation (3)] between reaction patterns in the matrix were used as the co-occurrence score.
 | (1) |
 | (2) |
 | (3) |
is the average of the values in reaction pattern vector i. Next, we investigated the relationship between the reaction patterns using the hierarchical clustering method. The Ward clustering (Ward, 1963) dendrogram was obtained using the R version 1.9.0 statistics program (http://www.r-project.org/). The co-occurrence score of the reaction patterns appearing higher in the database (top 50 of 302 reaction patterns) was calculated, and the scores were converted into negative values for the clustering. The result of cluster analysis using the scores calculated with the Cosine coefficient is shown in Figure 3. The clustering can be divided into groups according to structural features of glycans. Cluster I corresponds to the reaction pattern in the N-glycan core, including ‘GlcNAc b1-4 GlcNAc’, ‘Man b1-4 GlcNAc’ and ‘LFuc a1-6 GlcNAc’ and their terminal/internal structures such as ‘Neu5Ac a2-6 Gal’ and ‘Gal b1-4 GlcNAc’. Cluster II consists of the reaction patterns in the O-glycan and glycolipid core such as ‘Gal b1-3 GalNAc’ (O-glycan core 1), ‘GlcNAc b1-6 GalNAc’ (O-glycan core 2) and ‘Gal b1-4 Glc’ (lactosyl ceramide) and their terminal structures such as ‘Neu5Ac a2-6 GlcNAc’ and ‘LFuc a1-4 GlcNAc’. Cluster III consists of proteoglycan chains, and clusters IV and V contain the components of polysaccharides including xyloglucans and galactomannans, respectively. Thus, reaction patterns conserved in core structures were assembled into the same group, indicating that the coefficient score accurately represents the co-occurrence of reaction patterns conserved in the glycan entries. In the case of using other coefficients, although the topology of the dendrograms was slightly different, reaction patterns in the same core sub-structures also fell into the same groups (see Supplemental data S3 and S4).
|
3.2 Prediction method and evaluation
To evaluate the usefulness of GT reaction library, we developed a prediction method and evaluated its performance. When an expression profile is given as a query, it is converted to an ‘expressed reaction pattern list’, Q = {q1, q2, ..., qn}. Then, the glycan structures that contain each query reaction pattern are searched for in the database. The score for each candidate glycan, SE, is calculated as follows:
 | (4) |
To test the accuracy of prediction, the reaction pattern list in each glycan structure from the database was used as a virtual expression profile, and the performance of prediction was evaluated with the leave-one-out cross validation method. The evaluation method is summarized in Figure 4. One glycan profile was removed from the ‘reaction pattern matrix’ (Fig. 4A), and the ‘co-occurrence score matrix’ (Fig. 4C) was calculated from the ‘training matrix’ (Fig. 4B) using the three coefficients. In the case that a co-occurrence score could not be calculated, 1 was used between the same reaction patterns, and 0 was used otherwise. The removed glycan profile was converted into a binary ‘virtual expression profile’ indicating that a pattern exists (1) or not (0) because real expression profiles are generally qualitative rather than quantitative and it is only clear that a gene is expressed (1) or not (0). Furthermore, the virtual expression profile was converted into the ‘expressed reaction pattern list’ (Fig. 4D). SE was calculated using the ‘co-occurrence score matrix’ (Fig. 4C) and the ‘expressed reaction pattern list’ (Fig. 4D), and all predicted glycan structures were sorted by score (Fig. 4E). The rank of the query glycan in the predicted glycans was detected. This operation was iterated for every glycan as the query profile. The accuracy of prediction was defined as the ratio of the query glycan predicted within the given ranks. The random data was generated by shuffling the prediction results and evaluating the rank.
|
Figure 5 shows the accuracy of the prediction. The prediction results using all co-occurrence scores clearly performed well compared with those using random data. In particular, the cosine coefficient had the best performance; 72% of all entries could be predicted among the top 10 of the prediction results. The fact that the distribution of values in the matrix was not normal and that most of the values was 0 might have resulted in the low accuracy of prediction using the correlation coefficient and the Tanimoto coefficient.
|
As described above, expression data are binary, so it is possible to determine whether one enzyme is expressed or not. To improve the accuracy of the prediction, we implemented a penalty score into the system.
 | (5) |
 |
Figure 6 shows the accuracy of the prediction with the penalty score. Implementation of the penalty score improved the accuracy of the prediction using the cosine coefficient as the co-occurrence score. For example, the ratio of the query glycan appearing among the top 10 of the prediction results was improved from 72 to 81%. The performance of the prediction using all other coefficients also improved by the implementation of the penalty score with 
4–5% improvement for every coefficient (Supplemental data S5).
|
To investigate the possible causes of false positive results, we looked at the individual results. An example of a falsely predicted glycan structure is shown in Figure 7. The virtual expression profile from G10095 [GenBank] (G number is entry ID of the KEGG GLYCAN database) as the query resulted in G04790 [GenBank] as the top score (5.229), while the query glycan was predicted as the 279th (3.513) among 1750 predicted glycans. Although glycans that have a large number of reaction patterns and repeated structures such as N-acetyl-lactosamine (Gal b1-4 GlcNAc) tend to give more false positive, a fundamental core structure was commonly observed. In the example of Figure 7, the hybrid type N-glycan core structure (GlcNAcn-Man3-GlcNAc2) was shared among the top-scoring predicted entries.
|
While the accuracy of the prediction within the top 10 predicted glycans was >80%, the accuracy of the prediction of the top score was
50%. In particular, glycans with high-scored repeat sequence tend to falsely predict. However, almost every glycan predicted as the top score was similar in structure to each other (e.g. Fig. 4E). Furthermore, various glycans are expressed in the same cell, regardless of the set of GTs expressed (Mortz et al., 1996). Thus, we claim that our prediction results were rather favorable.
3.3 Application to DNA microarray data
Finally, we applied our novel prediction method to real DNA microarray expression data. It is noted that GTs mounted on the array are limited. In the application of our prediction method to real expression profiles, the penalty score may be falsely assigned to practically expressed but unidentified GTs. In this study, however, we presumed that all GTs corresponding to reaction patterns on the microarray have been identified, since homologous enzymes may catalyze the same reaction and can be found easily from genome data using homology search. When a reaction pattern in a candidate glycan was not on the CFG microarray, we did not apply the penalty score to the prediction method. Here, we give examples of the predicted results from the human carcinoma U937 cell line (supplemental data S6). Top 10 of predicted glycans are G05467
[GenBank]
(score: 2.627), G05814
[GenBank]
(2.627), G04844
[GenBank]
(2.568), G11846
[GenBank]
(2.549), G04206
[GenBank]
(2.442), G04056
[GenBank]
(2.421), G00197
[GenBank]
(2.419), G10271
[GenBank]
(2.419), G10278
[GenBank]
(2.419), G00193
[GenBank]
(2.413) and G00194
[GenBank]
(2.413). Their structures are listed in supplemental data S7, and some representative examples are shown in Figure 8. Briefly, they can be divided into two groups, the hybrid type N-glycans and ganglioside. Although they have different core structures, they share the non-reducing end terminal structure, sialic acid (Neu5Ac) and fucose, and the internal N-acetyllactosamine structure (Gal b1-4 GlcNAc). Some predicted glycans contain sialyl Lewis X epitope [Neu5Ac a2-3 Gal b1 (LFuc a1-3) 4 GlcNAc]. It has been reported that carcinoma cells over-produce sialyl Lewis X epitope and terminal sialic acid (Kim and Varki, 1997), supporting our prediction results.
|
As indicated above, not all GT have been identified and mounted on the array. The glycan database used here contained 302 kinds of reaction patterns, but only 98 GT genes corresponding to 42 reaction patterns (14%) were identified from the human genome. Furthermore, only 80 GT genes corresponding to 37 reaction patterns (12%) were mounted on the CFG microarray. More identification of GTs and update of DNA chips are required for more improvement of our prediction method to apply to real expression data.
|
4 DISCUSSION |
|---|
|
TOPAbstract1 INTRODUCTION2 DATASET3 RESULTS4 DISCUSSIONREFERENCES |
|---|
We developed a new method for the prediction of glycan structures from expression profiles. First, we constructed a reaction pattern library of GTs. Based on the annotations of the genome 98 GT genes were collected from the human genome. Although it is believed that GT enzymes generally recognize two or three monosaccharide residues of an acceptor glycan (Nakayama et al., 1997; Narimatsu, 2004), only a few enzymes have their exact acceptor residue determined (Kakuda et al., 2004), and it has come to be known that some enzymes recognize different acceptor residues depending on their condition (Ramakrishnan et al., 2004). Regardless, at the least, the acceptor monosaccharide residue to which the donor monosaccharide is added at the non-reducing end has been elucidated in most enzymes. Thus, we defined the reaction pattern of GT genes as the combination of donor monosaccharide, acceptor monosaccharide and the linkage between them (Fig. 2).
In our reaction pattern rule, many possible reaction patterns can be determined. Although 558 reaction patterns can be determined theoretically from 9 monosaccharides included in the library, only 302 patterns (54.12%) appear in the database. Furthermore, while an enormous number of reaction pattern combinations are possible (2558-1), only 2178 reaction pattern combinations actually appear in the database. This suggests that the structure of glycans is of great variety, but that the actual combination of reaction patterns is limited. Thus, we used co-occurrence between each reaction pattern as a score for prediction when an expression profile is provided. We make note that in our virtual experiment on expression data, the prediction was made for a single glycan structure for simplicity. However, in real experiments the nature of the data is complex in that it is a mixture of multiple glycan structures and responsible gene sets. Therefore, our prediction method should be considered as an upper bound.
In this study, we have challenged for the first time the prediction of glycan structures from genomic data, including GT expression data. This method has the advantage that it applies to any expression data, not only to DNA microarray expression profiles but also to protein expression profiles. Since the determination of glycan structure is still difficult in spite of the recent improvements in analytical technology, it is important to predict glycan structures using reasonable data such as DNA microarray expression profiles and protein array profiles. For further improvement of prediction, introduction of a relative position score between two reaction patterns in the glycan structure may be effective. In addition, the scoring method employed here depends on the training dataset. There may be some data biases, i.e. the glycans that bind to antibodies or lectins with high affinity are easy to purify and to determine their structures. The preparation of the training dataset also admits of improvement. Furthermore, the final glycan product is determined not only by GTs, but also by sugar metabolic enzymes, which synthesize and degrade monosaccharides, transporters, which transfer nucleotide activated sugar from cytoplasm to endoplasmic reticulum and Golgi apparatus, and glycosidases, which trim and degrade the glycan (Yarema and Bertozzi, 2001). Moreover, the conditions of the reaction field, such as pH, temperature, ion concentration, substrate concentration and product concentration, affect the activity of such enzymes (Ramakrishnan et al., 2004). Increasing the accuracy of the prediction requires the incorporation of such information of these enzymes and the condition of the reaction field. Furthermore, combining this method with other data such as mass data and antibody binding profiles will improve the accuracy of prediction.
In addition, the composite structure map (Hashimoto et al., 2005b) is created by connecting glycan structures in KEGG GLYCAN like metabolic pathways, and it is possible to map DNA expression data. Combination of our method and the composite structure map not only improves accuracy of prediction but also predicts glycan structures that are absent in KEGG GLYCAN and reveals glycans biosynthetic pathways.
The glycan structure prediction has been improved by combining glycan structural data and knowledge of glycans. For example, our method combined DNA microarray expression data and co-occurrence of reaction pattern to improve the prediction method, and the availability of the known repertoire of glycan structures in a given cell line allowed the validation of the prediction method. Furthermore, other glycan data analyzed by chromatograph, nuclear magnetic resonance and mass spectrometry would also enable better glycan structure prediction. Combination of various type data will permit improvement of glycan structure characterization.
Recently, prognostic expectation and appropriate use of medications adapted for individuals is possible using genotype data (SNPs) and expression data (Efferth and Volm, 2005; Ferrando and Look, 2004). However, it is almost unknown what types of pathways mediate genotype/transcriptotype to phenotype. The prediction of glycan structures from expression profiles illustrates a part of the pathway from transcriptotype to phenotype, and enables higher precision treatment and medication.
|
Acknowledgments |
|---|
We thank Prof. Hiroshi Mamitsuka, Dr Yasushi Okuno, Dr Kiyoko F. Aoki-Kinoshita and Dr Yoshihiro Yamanishi for critical reading of the manuscript and helpful suggestions. We also thank Masami Hamajima, Tomomi Kamiya, Yuriko Matsuura, Kana Matsumoto, Atsuko Yano, Ami Tanaka and Fujitsu Kyushu System Engineering Ltd for development and maintenance of the KEGG GLYCAN database. Computational time was provided by the Supercomputer Laboratory, Institute for Chemical Research, Kyoto University. This work was supported by the grants from the Ministry of Education, Culture, Sports, Science and Technology of Japan, the Japan Society for the Promotion of Science and the Japan Science and Technology Corporation.
Conflict of Interest: none declared.
Received on June 7, 2005; revised on September 5, 2005; accepted on September 6, 2005
|
REFERENCES |
|---|
|
TOPAbstract1 INTRODUCTION2 DATASET3 RESULTS4 DISCUSSIONREFERENCES |
|---|
Apweiler, R., et al. (1999) On the frequency of protein glycosylation, as deduced from analysis of the SWISS-PROT database. Biochim. Biophys. Acta, 1473, 4–8[Medline].
Birkle, S., et al. (2003) Role of tumor-associated gangliosides in cancer progression. Biochimie, 85, 455–463[Medline].
Brockhausen, I., et al. (1998) Glycoproteins and their relationship to human disease. Acta Anat. (Basel)., 161, 36–78[CrossRef][Web of Science][Medline].
Bucior, I. and Burger, M.M. (2004) Carbohydrate–carbohydrate interactions in cell recognition. Curr. Opin. Struct. Biol., 14, 631–637[CrossRef][Web of Science][Medline].
Cossart, P. and Sansonetti, P.J. (2004) Bacterial invasion: the paradigms of enteroinvasive pathogens. Science, 304, 242–248
Efferth, T. and Volm, M. (2005) Pharmacogenetics for individualized cancer chemotherapy. Pharmacol. Ther., 107, 155–176[CrossRef][Web of Science][Medline].
Ferrando, A.A. and Look, A.T. (2004) DNA microarrays in the diagnosis and management of acute lymphoblastic leukemia. Int. J. Hematol., 80, 395–400[CrossRef][Web of Science][Medline].
Hakomori, S. (2002) Glycosylation defining cancer malignancy: new wine in an old bottle. Proc. Natl Acad. Sci. USA, 99, 10231–10233
Hashimoto, K., et al. (2005a) KEGG as a glycome informatics resource. Glycobiol., in press.
Hashimoto, K., et al. (2005b) A global representation of the carbohydrate structures: a tool for the analysis of glycan. Genome Informatics, 16, 214–222.
Kakuda, S., et al. (2004) Structural basis for acceptor substrate recognition of a human glucuronyltransferase, GlcAT-P, an enzyme critical in the biosynthesis of the carbohydrate epitope HNK-1. J. Biol. Chem., 279, 22693–22703
Kanehisa, M., et al. (2004) The KEGG resource for deciphering the genome. Nucleic Acids Res., 32, D277–D280
Kim, Y.J. and Varki, A. (1997) Perspectives on the significance of altered glycosylation of glycoproteins in cancer. Glycoconj. J., 14, 569–576[CrossRef][Web of Science][Medline].
Kogelberg, H., et al. (2003) New structural insights into carbohydrate–protein interactions from NMR spectroscopy. Curr. Opin. Struct. Biol., 13, 646–653[CrossRef][Web of Science][Medline].
Mortz, E., et al. (1996) Does matrix-assisted laser desorption/ionization mass spectrometry allow analysis of carbohydrate heterogeneity in glycoproteins? A study of natural human interferon-gamma. J. Mass Spectrom., 31, 1109–1118[CrossRef][Web of Science][Medline].
Nakayama, K., et al. (1997) Substrate specificity of alpha-1,6-mannosyltransferase that initiates N-linked mannose outer chain elongation in Saccharomyces cerevisiae. FEBS Lett., 412, 547–550[CrossRef][Web of Science][Medline].
Narimatsu, H. (2004) Construction of a human glycogene library and comprehensive functional analysis. Glycoconj. J., 21, 17–21[CrossRef][Web of Science][Medline].
Ramakrishnan, B., et al. (2004) Structure and catalytic cycle of beta-1,4-galactosyltransferase. Curr. Opin. Struct. Biol., 14, 593–600[CrossRef][Web of Science][Medline].
Rudd, P.M., et al. (2001) Glycosylation and the immune system. Science, 291, 2370–2376
Sacks, D. and Kamhawi, S. (2001) Molecular aspects of parasite-vector and vector-host interactions in leishmaniasis. Annu. Rev. Microbiol., 55, 453–483[CrossRef][Web of Science][Medline].
Schachter, H., et al. (2002) Functional post-translational proteomics approach to study the role of N-glycans in the development of Caenorhabditis elegans. Biochem. Soc. Symp., 69, 1–21.
Simons, K. and Toomre, D. (2000) Lipid rafts and signal transduction. Nat. Rev. Mol. Cell. Biol., 1, 31–39[CrossRef][Web of Science][Medline].
Varki, A., Cummings, R., Esko, J., Freeze, H., Hart, G., Marth, J. Essentials of Glycobiology, (1999) , NY Cold Spring Harbor Laboratory Press.
Vo, L.H., et al. (2003) Identification of the ZPC oligosaccharide ligand involved in sperm binding and the glycan structures of Xenopus laevis vitelline envelope glycoproteins. Biol. Reprod., 69, 1822–1830
von der Lieth, C.W., et al. (2004) Bioinformatics for glycomics: status, methods, requirements and perspectives. Brief. Bioinform., 5, 164–178
Ward, J.H. (1963) Hierarchical Grouping to optimize an objective function. J. Am. Stat. Assoc., 58, 236–244[CrossRef][Web of Science].
Weis, W.I. and Drickamer, K. (1996) Structural basis of lectin-carbohydrate recognition. Annu. Rev. Biochem., 65, 441–473[CrossRef][Web of Science][Medline].
Yarema, K.J. and Bertozzi, C.R. (2001) Characterizing glycosylation pathways. Genome Biol., 2, REVIEWS0004.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  R. Ranzinger, M. Frank, C.-W. von der Lieth, and S. Herget Glycome-DB.org: A portal for querying across the digital world of carbohydrate sequences Glycobiology, December 1, 2009; 19(12): 1563 - 1567. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
F. J Krambeck, S. V Bennun, S. Narang, S. Choi, K. J Yarema, and M. J Betenbaugh A mathematical model to derive N-glycan structures and cellular enzyme activities from mass spectrometric data Glycobiology, November 1, 2009; 19(11): 1163 - 1175. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D. D. Marathe, E. V. Chandrasekaran, J. T. Y. Lau, K. L. Matta, and S. Neelamegham Systems-level studies of glycosyltransferase gene expression and enzyme activity that are associated with the selectin binding function of human leukocytes FASEB J, December 1, 2008; 22(12): 4154 - 4167. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. D. J. van Dijk, D. Bosch, C. J. F. ter Braak, A. R. van der Krol, and R. C. H. J. van Ham Predicting sub-Golgi localization of type II membrane proteins Bioinformatics, August 15, 2008; 24(16): 1779 - 1786. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. Kanehisa, M. Araki, S. Goto, M. Hattori, M. Hirakawa, M. Itoh, T. Katayama, S. Kawashima, S. Okuda, T. Tokimatsu, et al. KEGG for linking genomes to life and the environment Nucleic Acids Res., January 11, 2008; 36(suppl_1): D480 - D484. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. Hashimoto, A. C. Yoshizawa, S. Okuda, K. Kuma, S. Goto, and M. Kanehisa The repertoire of desaturases and elongases reveals fatty acid variations in 56 eukaryotic genomes J. Lipid Res., January 1, 2008; 49(1): 183 - 191. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D. M. Aanensen, A. Mavroidi, S. D. Bentley, P. R. Reeves, and B. G. Spratt Predicted Functions and Linkage Specificities of the Products of the Streptococcus pneumoniae Capsular Biosynthetic Loci J. Bacteriol., November 1, 2007; 189(21): 7856 - 7876. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. M. Comelli, M. Sutton-Smith, Q. Yan, M. Amado, M. Panico, T. Gilmartin, T. Whisenant, C. M. Lanigan, S. R. Head, D. Goldberg, et al. Activation of Murine CD4+ and CD8+ T Lymphocytes Leads to Dramatic Remodeling of N-Linked Glycans J. Immunol., August 15, 2006; 177(4): 2431 - 2440. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. Raman, M. Venkataraman, S. Ramakrishnan, W. Lang, S. Raguram, and R. Sasisekharan Advancing glycomics: Implementation strategies at the Consortium for Functional Glycomics Glycobiology, May 1, 2006; 16(5): 82R - 90R. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||