CISBIC Sub-project 1:

Stephen Muggleton, Brendan Wren, Victor LeskCISBIC, Imperial College London.www.imperial.ac.uk/cisbic

Modeling genotype-Modeling genotype-phenotype relations in phenotype relations in CampylobacterCampylobacter

Campylobacter jejuni : Unanswered questions

Are all strains equally pathogenic to humans?

Do strains from different sources share common features?

Why is C. jejuni a commensal in avians, but highly infectious in humans?

How does C. jejuni in some cases perturb the immune system to cause neurological sequelae?

Answers are likely to lie in glyco-surface structures

C. iejuni NCTC11168from gene sequence to glycostructure

Kelly J, 2006

Aim. Model variation in pathogen genome sequence of glycosylated surface structures & understand how this variation affects immune response.

Glycostructure questions

How do we get from sequence to full glycostructure?

How does glycan diversity affect disease potential and niche adaptation?

Use combination of traditional and systems approaches

1)Mutagenesis

2)Natural strain variation

3)Pathway modelling

PglF Cj1120

PglE Cj1121

PglD Cj1123

PglC Cj1124

PglA Cj1125

PglH Cj1129

PglJ Cj1127

PglI Cj1128

PglK Cj1130

NO

Basic sugars

UDP--D-GlcNAc

PglB Cj1126C L

PseB Cj1293

PseC Cj1294

PseH Cj1313

PseG Cj1312

PseI Cj1317

PseF Cj1311

PseA Cj1316

Strain collection

gDNA

Microarray

Quality Control

GeneSpring GACK analysis

Gene Calling

Bayesian analysis

Phylogenetic analysis

Statisticalanalysis Analysis

Values Ratio

Binary data

Evolution of virulence & pathogenesis

Phylogenomics pipeline

Natural strain variation and phylogenomics

•A livestock and non-livestock clade, but most clinical isolates fall in the non-livestock clade

•“Unknown” source of C. jejuni infection

•10 further sub-clades

•Water/wildlife sub-clade

Identification of sub-clade/niche specific genes

Identified glyco-specific genes for 1) Flagellin glyco (O)2) Capsule (C)E.g. Phosphoramidate

Genes Cj1321-6 present

Genes Cj1321-6 absent

Importance of Capsule

• Side branch sugars more widely conserved than core sugars?

• Phosphoramidate present in 86% of strains (232/270)?

Link to subproject 3

► Capsule prevents excessive cytokine production by dendritic cells

► Presence of phosphoramidate appears to reduce cytokine production

Machine learning approach

Biochemical Pathways

(KEGG,Biocyc)

Mutant data(MAS-NMR)

Glycan Structures(MAS-NMR)

Prolog Database

Machine learning

Background rules

Hypotheses

Examples of hypotheses

C. jejuni glycan synthesis pathway

Integrating multi-strain genetic data with machine learning of gene

functionsGenetic data for 75 serotyped

strains of C. jejuni

“If the product of Gene1 synthesises a sugar precursor which is transformed by the product of Gene2, then Gene1 tends to be present in strains with the same serotype as those expressing Gene2”.

Working assumption for relating gene function to co-presence

Integrating multi-strain genetic data with machine learning of gene

functions

Knowledgebase(KEGG, BioCyc)

Cj1432Cj1434Cj1438Cj1440Cj1442

Cj1432Cj1434Cj1438Cj1440Cj1442

Cj1432Cj1434Cj1438Cj1440Cj1442?

??

WT Cj1433-Cj1432-

Glycan mutant observations

Hypothesized transferase-encoding genes

Integrating multi-strain genetic data with machine learning of gene

functions

Knowledgebase(KEGG, BioCyc)

Cj1438

Cj1442 Cj1432Cj1434?

??

WT Cj1433-Cj1432-

Glycan mutant observations

Hypothesized transferase-encoding genes

Genetic data for serotyped strains

Learning from strain data (work in progress)

Learning curves with and without strain data

Gene functions provided to learning (%)

Gene functions identified by learning (%)

Acknowledgements

PI‘s and senior staff:Anne Dell, Stephen Muggleton, Jeremy Nicholson, Chris Rawlings,

Mike Sternberg, Brendan Wren

Postdocs:Richard Barton, Paul Hitchen, Emily

Kay, Victor Lesk, Alireza Tamaddoni-Nezhad