On the frequency of protein glycosylation, as deduced from analysis of

the SWISS-PROT database

1

Rolf Apweiler

a;

*, Henning Hermjakob

a

, Nathan Sharon

b

a

EMBL Outstation Hinxton, European Bioinformatics Institute, Wellcome Trust Genome Campus, Hinxton, Cambridge CB10 1SD, UK

b

Department of Membrane Research and Biophysics, The Weizmann Institute of Science, Rehovot 76100, Israel

Received 15 February 1999; accepted 19 May 1999

Abstract

The SWISS-PROT protein sequence data bank contains at present nearly 75 000 entries, almost two thirds of which

include the potential N-glycosylation consensus sequence, or sequon, NXS/T (where X can be any amino acid but proline)

and thus may be glycoproteins. The number of proteins filed as glycoproteins is however considerably smaller, 7942, of which

749 have been characterized with respect to the total number of their carbohydrate units and sites of attachment of the latter

to the protein, as well as the nature of the carbohydrate-peptide linking group. Of these well characterized glycoproteins,

about 90% carry either N-linked carbohydrate units alone or both N- and O-linked ones, attached at 1297 N-glycosylation

sites (1.9 per glycoprotein molecule) and the rest are O-glycosylated only. Since the total number of sequons in the well

characterized glycoproteins is 1968, their rate of occupancy is 2/3. Assuming that the same number of N-linked units and rate

of sequon occupancy occur in all sequon containing proteins and that the proportion of solely O-glycosylated proteins (ca.

10%) will also be the same as among the well characterized ones, we conclude that the majority of sequon containing proteins

will be found to be glycosylated and that more than half of all proteins are glycoproteins. ß 1999 Elsevier Science B.V. All

rights reserved.

Keywords: Glycosylation; Glycoprotein; Database

Glycosylation is a common and highly diverse co-

and post-translational protein modi¢cation reaction.

Perhaps because almost all proteins of human serum

and of hen egg-white are glycosylated [1], as are

those of animal cell membranes [2], the sweeping

statement has been made that `most proteins are gly-

coproteins' [3,4]. The recent development of compu-

terized protein sequence data banks allows us to put

this statement to a quantitative test. Here, we present

the results of such an attempt, based on an analysis

of the SWISS-PROT database [5]. This data bank is

manually curated and strives to provide a high qual-

ity of annotation, with a minimal level of redun-

dancy and a high level of integration with other bio-

molecular databases. It is thus more reliable than the

supplementary computer-annotated TrEMBL data

bank that contains translations of all protein coding

sequences in the EMBL nucleotide sequence data-

base which are not yet in the SWISS-PROT and is

therefore much larger than the latter.

In almost all glycoproteins, the carbohydrate units

are attached to the protein backbone either by N- or

O-glycosidic bonds or by both types of linkage. The

0304-4165 / 99 / $ ^ see front matter ß 1999 Elsevier Science B.V. All rights reserved.

PII: S 0 3 0 4 - 4 1 6 5 ( 9 9 ) 0 0 1 6 5 - 8

* Corresponding author..

1

Dedicated to Prof. Akira Kobata and Prof. Harry Schachter

on the occasion of their 65th birthdays.

BBAGEN 24913 16-11-99

Biochimica et Biophysica Acta 1473 (1999) 4^8

www.elsevier.com/locate/bba

N-glycosidic bond is always to the amide of an as-

paragine that is part of the consensus sequence NXS/

T, or `sequon', where X can be any amino acid ex-

cept proline. The sequons are often referred to as

`potential glycosylation sites', since, for reasons that

are not understood, in not all of these, the aspara-

gine is glycosylated. No consensus sequences for

O-glycosylation seem to exist.

The SWISS-PROT database contained by the end

of 1998 (release 36, including updates to 01/11/98)

74 988 entries. Potential N-glycosylation sites were

identi¢ed 151 993 times in 48 636 sequences, an aver-

age of 3.1 per protein. In 26 352 protein sequences,

such sites are absent, showing that about one third of

the proteins cannot be N-glycosylated. Examination

of the TrEMBL entries leads to similar conclusions

(Table 1).

The number of proteins in SWISS-PROT that

have been ¢led as glycoproteins is relatively small,

namely 7942 (10.6% of the total). In this database,

a protein is labelled with the keyword `GLYCOPRO-

TEIN' only when it is beyond reasonable doubt

that it is really glycosylated, even if information on

the nature of its carbohydrate units and their linkage

to the protein is lacking, as is the case for most of

these glycoproteins. Detailed annotation on the gly-

cosylation of these substances is done by the follow-

ing format: `FT CARBOHYD 6 position s 6 po-

Table 1

Protein entries and sequons in SWISS-PROT and TrEMBL da-

tabases by the end of 1998
Database

SWISS-PROT TrEMBL

Number of entries

74 988

156 187

Entries containing NXS/T sequon

48 636

107 551

64.9%

68.9%

Number of sequons

151 933

394 483

Sequon/sequon containing entry

3.12

3.66

Fig. 1. Frequency of occurrence of sequons and carbohydrate units in the 749 well characterized glycoproteins listed in the SWISS-

PROT database by the end of 1998. Sequons per glycoprotein (A), and carbohydrate units in N-glycoproteins (B), O-glycoproteins

(C) and N-,O-glycoproteins (D).

BBAGEN 24913 16-11-99

R. Apweiler et al. / Biochimica et Biophysica Acta 1473 (1999) 4^8

5

sition s 6 description s '. For example, a glycopro-

tein with a N-acetylgalactosamine at position 34 will

be annotated as `FT CARBOHYD 34 34 N-ACE-

TYLGALACTOSAMINE'.

As a thorough biochemical characterization of

most of the glycoproteins is lacking, numerous 6 de-

scription s tags contain the strings `BY SIMILAR-

ITY', `PROBABLE' or `POTENTIAL'. These de-

note that no biochemical characterization of the

glycosylation site(s) is available. For the purpose of

Table 2

Potential and real glycosylation sites in the 749 well characterized glycoproteins listed in the SWISS-PROT database by the end of

1998

Glycoproteins with at

least one biochemically
characterized (`real')

glycosylation site

Glycoproteins with

at least one real
N-glycosylation site

and at least one real

O-glycosylation site

Glycoproteins with

at least one real
N-glycosylation site

and no real

O-glycosylation site

Glycoproteins with

at least one real
O-glycosylation site

and no real

N-glycosylation site

Sites

Entries

Sites

Entries

Sites

Entries

Sites

Entries

Potential N-glycosylation

sites (sequons)

2 066

697

289

80

1 679

582

98

35

Real glycosylation sites

1 965

749

556

80

1 041

582

368

87

Real N-glycosylation sites

1 279

662

238

80

1 041

582

0

0

Real O-glycosylation sites

686

167

318

80

0

0

368

87

Fig. 2. Amino acid residues per sequon in all well characterized glycoproteins (A) and per real glycosylation site in the N-, O- and

N-,O-glycoproteins of the same group (B, C and D, respectively).

BBAGEN 24913 16-11-99

R. Apweiler et al. / Biochimica et Biophysica Acta 1473 (1999) 4^8

6

the present study, we denote as `real' glycosylation

sites only those sites where the 6 description s tag

does not contain any of the above strings. We con-

sider the glycosylation annotation of a SWISS-

PROT entry to be based on biochemical character-

ization if it contains at least one `real' (or occupied)

glycosylation site.

It should be noted that at present, at least one

third of all proteins in SWISS-PROT lack any bio-

chemical characterization and 12 921 are hypothetical

proteins, not yet proven to exist.

Furthermore, the proportion of biochemically

characterized proteins is getting smaller due to the

increasing number of predicted proteins derived

mainly from the fully sequenced genomes of di¡erent

organisms.

Of the 7942 listed glycoproteins, 1295 (16%) are

devoid of N-glycosylation sites, while 6647 (84%)

are N- (or N-,O-)glycoproteins. The latter contain a

total of 33 550 potential N-glycosylation sites, on

average 4.6 per molecule, which is more than that

found for all proteins containing such sites. Only a

small fraction of these, namely 749, have been thor-

oughly characterized with respect to their glycosyla-

tion patterns and complete information is available

as to the number and types of carbohydrate units per

molecule (Table 2 and Fig. 1). The majority of these

glycoproteins (83%) is from animals, while the pro-

portion of those isolated from other sources is rela-

tively small, 12% from plants, including fungi, 3%

from viruses and 2% from microorganisms (bacteria

and protozoa) (Fig. 2).

Of the 749 well characterized glycoproteins, 662

are N- or N-,O-linked, containing in total 1968 se-

quons. This is an average of three potential N-glyco-

sylation sites per molecule of sequon containing pro-

tein, which is slightly lower than found for all sequon

containing proteins (Table 1). Of the sequons in the

well characterized glycoproteins, on average 1.9 per

molecule (i.e. about 2/3) are occupied. Typically, one

or two N-glycosidic carbohydrates are found in a

glycoprotein (Fig. 1B,D). They average 1.8 U per

N-linked glycoprotein and 3.0 U per N-,O-linked gly-

coprotein (Table 2), but their number may be as high

as 24 per molecule, as in gp120 of HIV [6]. The

number of O-linked units in the glycoproteins is

larger (Fig. 1C,D), averaging four per glycoprotein

(Table 2), with as many as 96 carbohydrate units

found in polysialoglycoproteins of Salmon ¢sh eggs

Fig. 3. Origin of the 749 well characterized glycoproteins in

SWISS-PROT.

Table 3

Spacing of sequons and real glycosylation sites in the 749 well characterized glycoproteins

Amino acid per
Sequon

Real N-site

Real O-site

Real N+O-site

a

Minimum

18.00

23.11

3.10

3.10

Maximum

1 669.00

3 412.00

2 813.00

3 412.00

Mean

159.04

249.92

269.90

231.66

Median

121

144

167

133

Range

1 651.00

3 388.89

2 809.90

3 408.90

S.D.

164.07

348.09

338.83

335.80

a

In N-,O-glycoproteins.

BBAGEN 24913 16-11-99

R. Apweiler et al. / Biochimica et Biophysica Acta 1473 (1999) 4^8

7

[7]. The range of spacing of the sequons and the real

glycosylation sites is extremely wide (Fig. 3 and Ta-

ble 3).

Assuming that the well characterized glycoproteins

are a representative sample of the glycoproteins

present in nature, some three quarters of all glyco-

proteins should be N-linked, about one tenth N-,O-

linked and about one eighth just O-linked. Based on

the rate of occupancy of the sequons in the well

characterized glycoproteins, the 48 636 sequon con-

taining proteins of SWISS-PROT should each carry

on average two N-linked carbohydrates. This is close

to the average number of such units per molecule of

well characterized glycoprotein. It is thus highly

likely that most of the sequon containing proteins

will prove to be glycosylated. To this estimate should

be added the 10% proteins expected to be O-glyco-

sylated. We conclude therefore that more than half

of all proteins in nature will eventually be found to

be glycoproteins.

References

[1] N. Sharon, Complex Carbohydrates, their Chemistry, Bio-

synthesis and Functions. Addison Wesley, Reading, MA,

1975, pp. 33^35.

[2] C.G. Gahmberg, M. Tolvanen, Why mammalian cell surface

proteins are glycoproteins, Trends Biochem. Sci. 21 (1996)

308^311.

[3] J. Montreuil, The history of glycoprotein research, a person-

al view, in: J. Montreuil, J.F.G. Vliegenthart and H.

Schachter (Eds.), Glycoproteins. Elsevier, Amsterdam,

1995, p. 1.

[4] N. Sharon and H. Lis, Glycoproteins: structure and func-

tion, in: H.-J. Gabius and S. Gabius (Eds.), Glycosciences-

Status and Perspectives. Chapman and Hall, Weinheim, Ger-

many, 1997, p. 133.

[5] A. Bairoch, R. Apweiler, The SWISS-PROT protein se-

quence data bank and its supplement TrEMBL in 1999,

Nucleic Acid Res. 27 (1999) 49^54.

[6] C.K. Leonard, M.W. Spellman, L. Riddle, R.J. Harris, J.N.

Thomas, T.J. Gregory, Assignment of intrachain disul¢de
bonds and characterization of potential glycosylation sites

of the type 1 recombinant human de¢ciency virus envelope

glycoprotein (gp120) expressed in Chinese hamster ovary

cells, J. Biol. Chem. 265 (1990) 10373^10382.

[7] J.K. Kitajima, Y. Inoue, S. Inoue, Polysialoglycoprotein of

Salmonidae ¢sh eggs. Complete structure of 200-kDa poly-

sialoglycoprotein from the unfertilized eggs of rainbow trout

(Salmo gairdneri), J. Biol. Chem. 261 (1986) 5262^5269.

BBAGEN 24913 16-11-99

R. Apweiler et al. / Biochimica et Biophysica Acta 1473 (1999) 4^8

8