Glycomics

1
Glycomics
2
What is Glycomics?

• Glycomic analyses seek to understand how a
  collection of glycans relates to a particular
  biological event.
• Glycomes can far exceed proteomes and
  transcriptomes with respect to complexity
• some estimates have placed the vertebrate glycome
  at more than one million discrete structures.
• Many aspects of glycobiology can be understood
  only with a systems-level analysis
• glycomic changes during development and cancer
  progression
• many GBPs are oligomerized on cells and interact
  with multivalent arrays of glycans on opposing
  cells
• multiple discrete glycan epitopes work in concert
  to engage two cells or deliver a signal from one
  cell to the other

3
Virtual glycomes

• More than 250 glycosyltransferases have been
  found encoded in the human genome as well as many
  nucleotide sugar biosynthetic enzymes and Golgi
  transporters
• Expression patterns of many glycosyltransferases
  have been determined in human and mouse tissues
  using northern blots, quantitative PCR and
  transcriptomic analyses
• To construct virtual glycomes based on this
  information is of limited value because the
  combinatorial action of glycosyltransferases in
  many competing biosynthetic pathways renders the
  complete glycome very difficult to predict with
  any accuracy
• The glycome can change dramatically in response
  to a subtle change in the cellular system
• Glycan synthesis is not isolated in the cell
  (other enzymes may compete for common
  intermediates)
• variations in dietary monosaccharides

4
Tools to characterize the glycome

• Mass spectrometry
• Lectin and antibody arrays
• Cell and tissue analysis using lectins and
  antibodies
• Imaging the glycomes by metabolic and covalent
  labeling
• Glycan arrays
• Comparative glycomics

5
Mass spectrometry

• Glycoprotein- or glycolipid-enriched sample are
  prepared from cell lysates and subsequently
  analyzed by (tandem) mass spectrometry
• For glycoproteins, the N-glycans can be
  selectively released enzymatically or chemically,
  separated by HPLC (high-pressure liquid
  chromatography) methods, and then sequenced. The
  O-glycans are released chemically and sequenced
  as well.
• Glycolipids can often be directly sequenced
  without separation of the lipid component.
• Glycosaminoglycans are difficult to be analyzed
  because of their large size.
• Small fragments can be sequenced by mass
  spectrometry in conjunction with enzymatic
  digestion

6
Glycan profiling for biomarker discovery
7
Mass spectrometry pros and cons

• Advantages high-throughput, any given subtype
  can be profiled at once
• There is no method at present by which highly
  complex samples possessing many glycan subtypes
  can be analyzed in one mass spectrometry
  experiment
• Many MS experiments tend to partially or
  completely destroy the sample or miss potentially
  important modifications such as sulfation and
  O-acetylation.

8
Glycan sequencing using MS

• Database searching matching fragment ions in an
  MS/MS spectrum
• Glycofragment / GlycoSearchMS against SweetDB
• GlycosidIQ against GlycoSuite
• Cannot identify new glycan structures Number of
  known glycan structure is limited
• De novo sequencing
• Enumerating all possible structures STAT,
  StrOligo, OSCAR
• Only suitable for small glycans
• Can be used for MSn analysis
• Dynamic programming algorithm
• Linkage prediction from fragmentation ions

9
Glycan MS/MS spectra
10
De novo glycan sequencing problem

1. Monosaccharide sequence
2. Branching structure
3. Linkages

Glycan MS/MS spectra
Tang. H. et.al. ISMB 2005
11
Solution dynamic programming algorithm
Tang. H. et.al. ISMB 2005
12
Using cross-ring ions to score paths
13
Structural Characterizations of glycans

• The challenge isoforms with different linkages
  (and sometimes different branching structures)
• Occurrences of cross-ring fragment ions are not
  sufficient to distinguish glycan isoforms
• Solution scoring functions for distinguishing
  different linkages of isoformal glycans

14
Relative intensities of cross ring fragments are
different
Dextran (1-6)
Maltooligosaccharides (1-4)
15
Ranking ion type based on intensity
Dextran_glc9_Oligosaccharide5
Maltooligosaccharides_glc10_Oligosaccharide6
16
Two tails students t-tests based on ranks of
cross-ring fragment ions
Ion Type Probability 1,4 1,6
0,2A 0.448175
0,2X 0.001822 v v
0,3A 0.85771
0,3X 0.947922
0,4A 4.33E-25 v
0,4X 2.98E-25 v
1,4A 9.93E-07 v
1,4X 4.80E-04 v
1,5A 0.315954
1,5X 0.442356
2,4A 2.28E-27 v
2,4X 1.05E-08 v
2,5A 0.21049
2,5X 0.011821
3,5A 8.70E-10 v v
3,5X 0.270859

• 1,3A/1,3X are not considered since they do not
  exist in either fragmentation of 1,6 linkage or
  1,4 linkage
• The significantly different cross-ring ion types
  (in red) were used to distinguish 1,6/1,4
  linkages
• Checked ion types are used in later rank
  comparison for linkage discrimination

17
Rank based discriminate analysis
18
Mass spectrometry for glycoproteomics

• Mapping sites of attachment of glycans to the
  underlying protein scaffold (i.e., for
  glycoproteomic analysis)
• Slow reaction collision-induced dissociation
• Fragmentation at the glycosidic bonds mainly
• fast reaction photodissociation, electron
  transfer dissociation (ETD)
• Fragmentation at peptide bonds mainly
• High energy HCD (scan low m/z)
• Small oligosaccharide fragments

19
Site-specific protein glycosylation anslysis
using mass spectrometry

• LC/MS (ion trap) identification of co-eluted
  Cluster of peptide glycoforms (CPG), i.e.
  glycopeptides with various glycans attached to
  the same peptide backbone
• LC/MS/MS identification of glycopeptides based
  on their fragmentation pattern

20
Identification of CPG from MS1 problem
formulation
Peptide backbone mass 1000
Masses of glycoforms 200, 250, 300, 350, 400,
450
Masses of peptide glycoforms (expected to
observe) 1200, 1250, 1300, 1350, 1400, 1450
Masses observed (other ions missing ions)
1100, 1200, 1210, 1250, 1290, 1300, 1310, 1370,
1380, 1400, 1430, 1450, 1490
21
Identification of CPG from MS1 spectrum
convolution
Masses observed Y 1100, 1200, 1210, 1250,
1290, 1300, 1310, 1370, 1380, 1400, 1430, 1450,
1490
Masses of glycan forms X 200, 250, 300, 350,
400, 450
650, 700, 750, 800, 850, 900, 750, 800, 850,
900, 950,1000, 760, 810, 860, 910, 960,1010, 800,
850, 900, 950,1000, 1050, 840, 890, 940,
990,1040,1090, 850, 900, 950, 1000,1050,1100, 860,
910, 960, 1010,1060,1110, 920, 970, 1020,
1070,1120,1170, 930, 980, 1030, 1080,1130,1180,
950, 1000, 1050, 1100,1150,1200, 980, 1030, 1080,
1030,1180,1230, 1000, 1050, 1100,1150,1200,
1250, 1040, 1090, 1140,1190,1240, 1290
Peptide mass 1000, with highest multiplicity (5)
in Y?X.
22
Implementing spectrum convolution practical
issues

• Incorporating peak intensity into the scoring of
  spectrum convolution
• The same glycopeptide may carry different
  charges
• There may be more than one clusters of sister
  glycopeptides co-eluted in the same LC window

23
Identification of individual glycopeptides from
their fragmentary (MS/MS) spectra
m/z
Finding the largest subset of peaks, such that
the mass difference between any consecutive peaks
corresponds to the mass of a monosaccharide.
24
Challenges ofdirect glycoproteomic analysis

• Augmented ion complexity
• Number of ion species multiplied
• Instrument duty cycle
• Glycopeptides often correspond to abundant ions
  (than peptides).

25
Our approach

• We developed a new experiment protocol using
• Iterative (replicated) experiments (to overcome
  the limit of duty cycle)
• with time-segmented inclusion list (to replace
  intensity-based ion selection)

26
Motivation of dynamic analysis

• Intensity-based ion selection is inflexible.
• Many glycopeptides have lower-than-average
  intensity in real sample.
• Dynamic exclusion helps a little bit. But the
  duty cycle is an inevitable limit.

27
Motivation of Dynamic Analysis
28
Our approach

• We select ions by their glycomic feature, not by
  intensity.
• We call it Targeted Data Dependent Acquisition,
  or TDDA.

Pick labeled ions
29
Our approach

• The time-segmented inclusion list is available
  through Thermo LTQ-MS instrument.
• TDDA is implemented using the time-segmented
  inclusion list.

User interface of time-segmented inclusion list.
30
Selecting ions for inclusion list

• Group ions into clusters of putative
  micro-heterogeneities.
• Insert ions of big clusters into inclusion list.

31
Iterative experiments applying TDDA
Initial Experiment
Software Analysis
Experiment with Inclusion List
Inclusion List
No
Terminate?
Sample Preparation
The end
Yes
32
CID vs. ETD supplementary fragmentations
J. M. Hoga, Journal of Proteome Research 2005 4
(2), 628-632
33
Lectin and antibody arrays
34
Lectin array pros and cons

• Cons providing global information about the
  types of glycan epitopes that are present in the
  sample but does not give any detailed structural
  information, nor does the experiment provide
  information regarding which proteins the glycans
  are attached to.
• Pros high-throughput platform (allows for rapid
  comparison of many glycomes in search of global
  changes that might motivate further mass
  spectrometry studies)

35
Imaging the glycome
36
Specific expressions of glycans
Gagneux Varki 1999 Glycobiology 9747-755
37
Glycan arrays

• Profiling the GBPs, e.g. plant GBPs, viral
  antigens, GBPs in the innate and adaptive
  immune system
• DC-SIGN and DC-SIGNR C-type lectins
• expressed on dendritic cells and plays a key role
  in adhesion of T cells as well as in the
  recognition of pathogens such as HIV
• sharing 77 sequence identity, but with distinct
  ligand specificities
• Influenza Virus Hemagglutinin (HA)
• The HA glycoprotein mediates host-cell
  recognition
• Human viral HA preferentially recognizes glycans
  terminated by NeuAca2-6Gal, whereas avian HA
  preferentially recognizes glycans containing
  NeuAca2-3Gal
• The upper airway epithelial cells (target) in
  humans contain mainly NeuAca2-6Gal, whereas in
  birds both the airways and intestine contain
  mainly NeuAca2-3Gal linkages

38
Hemagglutinin (viral lectin)

• The influenza virus hemagglutinin was the first
  GBP isolated from a microorganism (1950)
• 3D structure determined in 1981 (Wiley)
• complex structure with sialyllactose.
• Mainly bind to terminal residues, some can bind
  to internal sequences found in linear or branched
  glycans
• The specificity of these interactions can be
  highly selective.
• For example, the human influenza viruses bind
  primarily to cells containing Siaa2-6Gal
  linkages, whereas other animal and bird influenza
  viruses preferentially bind to Siaa2-3Gal
  termini.
• Influenza C, in contrast, binds preferentially to
  glycoproteins containing terminal 9-O-acetylated
  sialic acids.
• Many other viruses (e.g., reovirus, rotavirus,
  Sendai, and polyomavirus) also appear to use
  sialic acids in specific linkages for infection.
• Other viruses display glycosaminoglycan-binding
  proteins that can bind to heparan sulfate
  proteoglycans, often with high specificity for
  certain sulfated sequences

39
Glycan topology determines human adaptation of
avian H5N1 virus hemagglutinin (HA)

• Transmission and virulence of influenza viruses
  is the binding of HA to sialylated glycans on the
  epithelial cell surface
• Transmission from birds to humans is believed to
  be closely associated with the ability of the HA
  to switch its preference from 2-3 sialylated
  glycans (2-3) to 2-6 sialylated glycans (2-6),
  which are extensively expressed in the human
  upper respiratory epithelia
• Glycan arrays for the glycan binding specificity
  of wild-type and mutant H1, H3 and H5 HAs show
  confounding results
• The relationship between the HA glycan binding
  specificity and transmission efficiency has been
  demonstrated on the highly pathogenic and
  virulent 1918 H1N1 viruses.
• Switching the receptor binding specificity of the
  highly transmissible and pathogenic human H1N1
  (A/South Carolina/1/18 SC18) virus from 2-6 to
  2-3 has produced a virus (AV18) not
  transmissible. Although A/New York/1/18 (NY18)
  H1N1 virus, which shows mixed 2-3/2-6 binding,
  does not transmit efficiently, the A/Texas/36/91
  (Tx91) H1N1 strainthat also binds to both 2-3
  and 2-6transmits efficiently.

40
Data mining of glycan array data

• HA binding array
• Extracted features

Chandrasekaran, et. al. Nature Biotechnology 26,
107 - 113 (2008)
41
(No Transcript)
42
Data mining of glycan array data

• Correlations between glycan features and the HA
  binding to these glycans are given as logistical
  regression classifiers
• Variations around the trisaccharide 2-3 motif
  primarily influence the differential 2-3 binding
  of H1, H3 and H5 HAs.
• The 2-6 classifier common to the human-adapted H1
  and H3 HAs is consistent with its gain in ability
  to bind long 2-6. Although the glycan binding of
  wild-type and mutant H5N1 HAs is not supported by
  the long 2-6 classifier, it is consistent with
  both 2-3 and short 2-6 classifiers.

Chandrasekaran, et. al. Nature Biotechnology 26,
107 - 113 (2008)
43
Combining with other evidences

• Predominant expression of 2-6 in the human upper
  respiratory epithelium, and the expression of
  long oligosaccharide branches with multiple
  lactosamine repeats on the apical side of the
  upper respiratory epithelia
• Structure analysis of HA and glycans suggested
  the existence of two subtype Has the cone-like
  topology is characteristic of 2-3 as well as
  short 2-6 glycans such as single lactosamine
  branches, and the umbrella-like topology is
  unique to 2-6 and is typically adopted by long
  glycans with multiple repeating
  lactosamine.units
• Both SC18 and Mos99 HAs show substantial and
  preferential binding to the apical side of the
  tracheal tissue, in comparison to the deep lung
  tissue.

Chandrasekaran, et. al. Nature Biotechnology 26,
107 - 113 (2008)
44
Cone-like (left) and umbrella-like (right)
topologies of 2-3 and 2-6 siaylated glycans
binding to influenza viral HAs
Chandrasekaran, et. al. Nature Biotechnology 26,
107 - 113 (2008)