Analysis of the dispersity in carbohydrate loading of synthetic glycoproteins using MALDI-TOF mass spectrometry

Article abstract of DOI:10.1039/c0cc03420g

Statistical correlation of mass spectrum peak broadening with product dispersity in protein conjugation reactions allows more detailed characterization of putative therapeutic conjugates.

Full text of DOI:10.1039/c0cc03420g

View Article Online / Journal Homepage / Table of Contents for this issue
COMMUNICATION
www.rsc.org/chemcomm | ChemComm
Analysis of the dispersity in carbohydrate loading of synthetic
glycoproteins using MALDI-TOF mass spectrometryw
Mitul K. Patel, Balakumar Vijayakrishnan, Julia R. Koeppe, Justin M. Chalker,
Katie J. Doores and Benjamin G. Davis*
Received 23rd August 2010, Accepted 29th September 2010
DOI: 10.1039/c0cc03420g
Statistical correlation of mass spectrum peak broadening with
product dispersity in protein conjugation reactions allows more
detailed characterization of putative therapeutic conjugates.
is the variance, obtained from the FWHM.z The broadened
peak after modification is also approximately Gaussian and can
be represented as (2), where l2 is the variance of this peak.z
ꢀx2
1
2d2
The presentation of carbohydrates on proteins and cell
surfaces has far reaching biological significance.1 Carbohydrates
act as markers for critical signalling events and interact with
sugar-binding proteins using a high surface density of carbo-
hydrates to improve affinity.2 Synthetic glycoproteins that mimic
this surface coverage can interact with these proteins and are
useful for clinical applications3 such as vaccines,4 enzyme replace-
ment therapy,5 targeted drug delivery6 and gene therapy.7
Currently, most glycoprotein therapeutics are hetero-
geneous, formed from the indiscriminate conjugation of
glycans, for instance at primary amines on native lysine
residues and the N-terminus.8,9 Since lysine has a high natural
abundance, multiple modifications occur. A striking drawback
is the formation of a statistical mixture of products and the
challenge of measuring and controlling the product dispersity.
Conjugates can be characterized with matrix-assisted laser
desorption/ionization (MALDI) mass spectrometry (MS), but
for large molecular weights, the resolving power is insufficient
to distinguish each product.10 Hence, a broad peak is
observed, where the centre is the mean copy number of ligands
per protein. Information on the dispersity of the sample is
typically neglected; this is surprising since such proteins may
be used clinically. Fuller characterization would allow better
batch reproducibility11 and understanding of reactivity and
therapeutic structure–activity relationships.
pffiffiffiffiffiffiffiffiffiffi
e
ð1Þ
2pd2
2
ꢀðxꢀmÞ
2l2
e
1
pffiffiffiffiffiffiffiffiffiffi
f ðxÞ ¼
ð2Þ
2pl2
Indiscriminate modification produces a mixture of protein
products, each of which generates its own Gaussian ‘‘peaklet’’.
Inadequate resolution leads to peaklet coalescence and a
widened peak (Fig. 1). Therefore, the spectrum of the modified
protein can be represented as a sum of peaklets, weighted by a
function representing the product distribution. The theoretical
product distribution can be calculated by modelling the reaction
as a series of consecutive competitive second order events.15
This distribution and its analysis are sensitive to the reaction
conditions. For instance, at very low reagent concentration, a
Poisson distribution is likely most appropriate, whereas for a
reaction purely defined by statistics (where only the number of
reactive groups is considered to affect the rate constants), a
binomial distribution offers an exact analytical solution.
However, these probability functions are only valid over a
narrow range of reaction criteria. A Gaussian distribution is
more mathematically flexible and valid over a broader range of
conditions.w Assuming the variance of each peaklet stays
constant as d2 and ionization efficiency is unaltered by succes-
sive modifications, we can write (with a continuity correction):
At present, there are no simple methods to determine sample
dispersity. Top–down proteomics may reveal possible modifica-
tion sites, but product distribution is not readily apparent.12
Fourier transform-ion cyclotron resonance MS offers high
resolution and may be capable of directly measuring distri-
butions,13 but requires extensive computation14 and expensive
instrumentation. Broadened peaks of MALDI-MS can, in
principle, reveal information on the underlying distribution.
The parent peak of the unmodified protein can be approxi-
mated as a Gaussian peak which can be mathematically
expressed from its full width at half maximum (FWHM). The
peak shape is given by (1), where x is the measured mass and d2
2
3
nþ0:5
2
Z
2
n
ꢀðxꢀmnÞ
2d2
ꢀðxꢀmÞ
2s2
X
1
1
4
5
pffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffiffiffi
f ðxÞ ¼
e
e
dx ð3Þ
2ps2
2pd2
0
nꢀ0:5
where m is mass change per modification, n is the total number
of reactive sites per protein, m is the relative mean mass of
conjugates and s2 is the variance of distribution.
Eqn (2) and (3) represent different analyses of the same
function: that of the broadened peak of the synthetic
conjugate mixture. By equating the two formulae, s can be
Department of Chemistry, Chemistry Research Laboratory,
University of Oxford, Mansfield Road, Oxford, OX1 3TA, UK.
E-mail: Ben.Davis@chem.ox.ac.uk; Fax: +44 (0)1865 285002;
Tel: +44 (0)1865 285001
w Electronic supplementary information (ESI) available: Extended
discussion on the Gaussian approximation and peak tailing, experimental
details and MATLAB M-file scripts. See DOI: 10.1039/c0cc03420g
Fig. 1 Peaklet coalescence leads to peak broadening. By analysis of
peak widths, information on the product distribution can be obtained.
c
This journal is The Royal Society of Chemistry 2010
Chem. Commun., 2010, 46, 9119–9121 9119

View Article Online
determined. Thus, by simply measuring FWHM of modified
and unmodified proteins, the product distribution variance
can be calculated. Furthermore, since the area between ꢁ2s
equates to 95% of the total, m ꢁ 2s represents the range within
which 95% of adducts exist thus giving a direct, physically
intuitive measure of dispersity.
To demonstrate the versatility of the approach, we
synthesized various glycoconjugates and determined their
dispersity using MALDI with time of flight (TOF) detection.y
Mannose-bearing synthetic glycoproteins mimic the glycosyla-
tion on envelope protein gp120 and are potential HIV
vaccines.3,4 As a model system, we linked mannose residues
to immunogenic carrier proteins using thiophosgene conjuga-
tion chemistry, a strategy previously explored for synthetic
vaccine constructs.8,16 Such glycoconjugates are representative
of a type of vaccine design, which unlike many existing
vaccines, contain pure glycan components. Initially, we
conjugated monosaccharide 1 with bovine serum albumin
(BSA, 59 lysines + N-terminus), a common model protein
for vaccines (Fig. 2). MALDI peaks were used to determine
the extent of modification and sample dispersity for increasing
ratios of 1 : lysine. For instance, BSA modified with a mean of
15 copies of 1 has a dispersity (2s) of 8 (95% of proteins
incorporate 7–23 copies of 1)—a more informative description
of the actual product composition than simply the mean.
We then investigated coupling reactions of BSA with a
mannose disaccharide (Fig. 3). The greater steric demand of
2 required a larger excess of reagent to achieve similar levels of
modification as 1. Interestingly, glycoconjugates of 2 showed
more dispersity than the corresponding conjugates of 1
(Fig. 4). The dispersity is determined by the ratios of the
consecutive rate constants (the distribution constants).15 The
average rate constant of each successive modification is
expected to decrease due to the reducing number of available
lysines and increasing steric crowding from the protein struc-
ture. Thus, the rate of this decrease determines the dispersity
of the sample, with a sharper fall resulting in a less dispersed
Fig. 3 Top: reaction scheme for conjugation to BSA; middle: mass
spectrum of glycoconjugates with increasing equivalents of 2 per
lysine; bottom: mean extent of modification and dispersity versus
reagent excess.
Fig. 4 Dispersity of glycoconjugates of BSA with 1 or 2.
product distribution. Since 2 is bulkier than 1, a slower
reaction rate might have been expected. However, consecutive
rate constants are the critical determinant and may not be
affected to the same extent. If earlier rates in the sequence are
retarded to a greater extent than later ones, the progressive
decrease in reaction rates is much slower leading to a more
dispersed product distribution. The overall outcome revealed
that use of reagent 1 as compared to 2 has a greater effect on
earlier lysine couplings than with later, less solvent-exposed
residues which are already less reactive.
Applicability to other proteins was demonstrated by
conjugating 2 with the cross reacting material fragment of
diphtheria toxin (CRM-197), another common protein for
vaccine constructs (Fig. 5).17 Compared to the reaction of 2
with BSA, the extent of modification with CRM was reduced
due to the lower number of reactive sites (only 40 amines). The
dispersity of CRM conjugates is also lower compared with
BSA, since the statistical fall in sequential rate constants is
exaggerated when there are fewer reactive groups.
Having established the method on model systems, we
applied the analysis to a prospective HIV vaccine where a
mannose tetrasaccharide (D1 arm) was conjugated to a carrier
protein to mimic the glycan display on gp120. We have
recently evaluated such Qb glycoproteins for immunogenic
activity.18 Since the Qb monomer is a low molecular weight
protein (14 kDa, 8 amines), the individual peaklets are well
Fig. 2 Top: reaction scheme for conjugation to BSA; middle: mass
spectrum of glycoconjugates with increasing equivalents of 1 per
lysine; bottom: mean extent of modification and dispersity versus
reagent excess.
c
9120 Chem. Commun., 2010, 46, 9119–9121
This journal is The Royal Society of Chemistry 2010

View Article Online
in polymer science. Other applications of therapeutic relevance can
be envisaged. Indiscriminate in vivo protein glycation of lysines is
implicated in several diseases and measuring the dispersity of
glycated proteins would assist in disease diagnostics.21
We thank the International AIDS Vaccine Initiative (MKP,
BV, JRK, KJD), the Rhodes Trust and NSF (JMC) for
financial support. BGD is a Royal Society-Wolfson Research
Merit Award recipient and supported by an EPSRC LSI
Platform Grant. We also thank V. Hong and Prof. M. G.
Finn for their assistance in the preparation of the Qb
conjugates.
Notes and references
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffiffiffiffiffi
z FWHM ¼ variance ꢂ 2 2 ln 2
y MALDI-TOF detector settings, laser irradiance and matrix
preparation were identical for each set of comparisons to ensure
consistency.
Fig. 5 Top: reaction scheme for conjugation to CRM; middle: mass
spectrum of glycoconjugates with increasing equivalents of 2 per
lysine; bottom: mean extent of modification and dispersity versus
reagent excess.
z Other possible methods beyond MS that might aid analysis include
those based on differential electrophoresis such as isoelectric focussing.
1 R. A. Dwek, Chem. Rev., 1996, 96, 683.
2 N. Jayaraman, Chem. Soc. Rev., 2009, 38, 3463; J. J. Lundquist
and E. J. Toone, Chem. Rev., 2002, 102, 555.
3 K. J. Doores, D. P. Gamblin and B. G. Davis, Chem.–Eur. J., 2006,
12, 656.
4 R. Roy, Carbohydrate Based Vaccines, ACS Symposium Series,
2008, vol. 989; M. Hecht, P. Stallforth, D. V. Silva, A. Adibekian
and P. H. Seeberger, Curr. Opin. Chem. Biol., 2009, 13, 354;
R. Roy, Drug Discovery Today: Technol., 2004, 1, 327.
5 Y. Zhu, X. Li, A. McVie-Wylie, C. Jiang, B. L. Thurberg,
N. Raben, R. J. Mattaliano and S. H. Cheng, Biochem. J., 2005,
389, 619.
6 M. A. Robinson, S. T. Charlton, P. Garnier, X.-T. Wang,
S. S. Davis, A. C. Perkins, M. Frier, R. Duncan, T. J. Savage,
D. A. Wyatt, S. A. Watson and B. G. Davis, Proc. Natl. Acad. Sci.
U. S. A., 2004, 101, 14527; B. G. Davis and M. A. Robinson, Curr.
Opin. Drug Discovery Dev., 2002, 5, 279.
7 O. M. T. Pearce, K. D. Fisher, J. Humphries, L. W. Seymour,
A. Smith and B. G. Davis, Angew. Chem., Int. Ed., 2005, 44, 1057.
8 G. T. Hermanson, Bioconjugate Techniques, Academic Press,
2nd edn, 1996; M. Izumi, S. Okumura, H. Yuasa and
H. Hashimoto, J. Carbohydr. Chem., 2003, 22, 317.
9 For well defined glycoproteins see: B. G. Davis, Chem. Rev., 2002,
102, 579; D. P. Gamblin, E. M. Scanlan and B. G. Davis, Chem.
Rev., 2009, 109, 131; B. G. Davis, Pure Appl. Chem., 2009, 81, 285.
10 D. J. Harvey, Int. J. Mass Spectrom., 2003, 226, 1.
11 C. Jones, Carbohydrate Based Vaccines, ACS Symposium Series,
2008, vol. 989, ch. 2, pp. 21–35.
resolved by MALDI (Fig. 6) and provided direct validation of
the assumptions underpinning our dispersity analysis. Indivi-
dual peaklets have identical FWHM, and distribution profiles
match theoretical expectations indicating similar ionisation of
the different protein products.19 The product distribution is
directly evident, showing an average of 1.5 modifications and a
dispersity of 1.5. This allowed consistent levels of modification
between two constructs carrying different glycans, which was
vital for comparing biological efficacy. Most common carrier
proteins are larger than Qb and peaklet coalescence occurs, as
for the conjugation of the same D1 arm with BSA. Loading-
dispersity analysis reveals 10.8 ꢁ 8.8 modifications; similar
BSA conjugate with 1 has 10.6 ꢁ 7.2 modifications, consistent
with the trend of larger sugars producing higher dispersity.
Without this dispersity parameter, the two glycoconjugates
would be considered similarly modified. A dispersity analysis
is therefore vital to understand the true nature of the products.
The simple mathematical method we describe allows
dispersity in copy-number to be determined and expressed
intuitively, even for non-ideal cases exhibiting high-mass peak
tailing.20w Comparisons can be made between reagents:
bulkier sugars can increase dispersity, while proteins with
fewer reactive groups cause a decrease. Glycoconjugates can
be better characterized without additional experiments than is
routine,z allowing control of loading and distribution. Given
that very few synthetic vaccines in mainstream use are well
defined,4 our method enables better reproducibility and more
logical comparisons. The dispersity measure proposed here
may find use akin to the polydispersity index commonly used
12 B. A. Garcia, J. Am. Soc. Mass Spectrom., 2010, 21, 193.
13 A. G. Marshall, C. L. Hendrickson and G. S. Jackson, Mass
Spectrom. Rev., 1998, 17, 1; N. L. Kelleher, M. W. Senko,
M. M. Siegel and F. W. McLafferty, J. Am. Soc. Mass Spectrom.,
´ ´
1997, 8, 380; K. Nagy, K. Vekey, T. Imre, K. Lundanyi,
M. P. Barrow and P. J. Derrick, Anal. Chem., 2004, 76, 4998.
14 D. M. Horn, R. A. Zubarev and F. W. McLafferty, J. Am. Soc.
Mass Spectrom., 2000, 11, 320.
15 H. J. R. Maget, J. Polym. Sci., Part A: Polym. Chem., 1964, 2,
1281.
16 R. D. Astronomo, H. Lee, C. N. Scanlan, R. Pantophlet,
C. Huang, I. A. Wilson, O. Blixt, R. A. Dwek, C. Wong and
D. R. Burton, J. Virol., 2008, 82, 6359; D. H. Buss and
I. J. Goldstein, J. Chem. Soc. C, 1968, 1457.
17 G. Giannini, R. Rappuoli and G. Ratti, Nucleic Acids Res., 1984,
12, 4063.
18 K. J. Doores, Z. Fulton, V. Hong, M. K. Patel, C. N. Scanlan,
M. R. Wormald, M. G. Finn, D. R. Burton, I. A. Wilson and
B. G. Davis, Proc. Natl. Acad. Sci. U. S. A., 2010, 107, 17107.
19 F. K. Yeboah and V. A. Yaylayan, Nahrung, 2001, 45, 164.
20 E. Sachon, G. Clodic, T. Blasco and G. Bolbach, J. Am. Soc. Mass
Spectrom., 2007, 18, 1880.
Fig. 6 Glycoconjugates of the D1 arm with Qb (left) and BSA (right)
and MALDI mass spectra of the corresponding glycoproteins.
21 T. Niwa, Mass Spectrom. Rev., 2006, 25, 713.
c
This journal is The Royal Society of Chemistry 2010
Chem. Commun., 2010, 46, 9119–9121 9121