Integrated Genomic and Proteomic Analyses of a Systematically Perturbed Metabolic Network

1
Integrated Genomic and Proteomic Analyses of a
Systematically Perturbed Metabolic Network

• Trey Ideker et al. Science Vol. 292 4 May 2001
  p. 929-934

Alison Hottes BioNetworks Journal Club December
3, 2001
2
Main Theme in Paper Understanding a Large
Network

• Hypothesis generation and verification
• Microarray data
• Proteomics (mass spectrometry data)
• Visualizing network connectivity
• Protein-protein and protein-DNA interactions
• Self-organizing maps (SOMs)
• Hierarchical-clustering trees

3
Main Approach

• Propose an initial model for the subsystem under
  study
• Perturb the model and measure the results using
  mRNA microarrays and protein-expression
  measurements
• Integrate the new data with the original model
  for the subsystem under study as well as the
  global system.
• Generate new hypotheses and test.

4
Galactose Metabolism System
Figure 1 in (Ideker, 2001)

• Step 1 Define an initial model
• From previous work, a detailed skeleton of the
  network for yeast (S. cerevisiae) galactose
  utilization was already known

5
Step 2 Perturb the System

• Change the environment
• Grow with and without galactose
• Change the network
• Remove links through gene deletions

Figure 1 in (Ideker, 2001)
6
Analysis Focusing on GAL Genes
Figure 2 in (Ideker, 2001)
7
Analyzing the 997 Other Differentially Expressed
Genes

• Make 16 clusters using self-organizing maps
• Identify biological processes that
  correspondto each cluster

Figure 2
Figure 2 in (Ideker, 2001)
8
Comparing Protein and mRNA Levels

• Often assumed that protein levels are
  proportional to mRNA levels
• Between conditions, mRNA and protein levels
  corresponding to a single gene can change by
  different amounts
• Translational regulation
• Varying protein and mRNA stabilities

9
Finding Protein Ratios
Label sample 1 with normal tag
Label sample 2 with heavy tag
Mix samples, extract labeled protein, and digest
with trypsin
Use mass spectrometry to identify peptide and
relative amounts from each sample
Separate with multidimensional chromatography
10
Comparing Protein and mRNA Changes
Figure 3 in (Ideker, 2001)
Comparing wt gal with wt -gal
11
Comparing Protein and mRNA Changes

• Obtained ratios for 289 proteins
• Moderate correlation between protein and mRNA
  level (r 0.61)
• Of 30 proteins that had significant mRNA changes,
  15 did not have significant protein level changes
  may suggest posttranscriptional regulation
• Many ribosomal-protein genes had increased mRNA,
  but not protein, in galactose

12
A More Global View

• Obtain list of 2709 protein-protein interactions
  and 317 protein? DNA interactions
• Found 384 genes relevant to galactose
• Changed in some experimental perturbation in this
  study or
• Involved with 2 or more genes that were effected
  by the perturbations in the study

13
Effects of gal4? gal
Protein-Protein
Protein--gt DNA
Figure 4 in (Ideker, 2001)
14
Comparing Expression Profiles of Interacting
Genes

• Genes that interact physically tend to have more
  correlated (rgt0.4) expression profiles
• Possible causes
• A?B A causes/represses transcription of B
• (A,C)?B A and C cause/repress transcription of
  B
• C?(A,B) A and B are both regulated by a third
  protein C

15
Refining the Network

• Look for more genes directly regulated by Gal4p
• All known Gal4p binding sites were upstream of
  genes in clusters 1, 2, and 3
• Found 9 possible Gal4p binding sites upstream of
  other genes in clusters 1, 2, and 3
  (statistically more than found in other clusters)
  
• Hypothesize that Gal4p directly controls these
  genes (dotted line in model)

16
Additional Refinements

• Look for unexpected changes ( or -) in the mRNA
  data
• Gal7 and gal10 deletions reduce expression of
  other gal enzymes in presence of galactose

17
Forming Hypothesis

• Predict that build-up of Gal-1-P or one of its
  derivatives mediates the change in expression
  levels
• Test with gal1?gal10 ? mutant in galactose
  (prevents build-up of Gal-1-P)

18
Testing Hypotheses

• Use Euclidean distance to create
  hierarchical-clustering tree
• Included some double mutants and used for
  epistasis analysis

Figure 5 in (Ideker, 2001)

• gal1?gal10 ? mutant in galactose is most like
  gal1? in galactose ? supports hypothesis

19
Discussion of Method

• Puts a subsystem in the context of an entire
  network
• Proposed only a small number of network changes
• Used a variety of data types and analysis schemes
  
• Not an automatic method
• Displays data in a variety of formats to make
  network evaluation and iteration easier
• Human intervention is needed interpret the
  results
• More suited to refining an existing hypothesis
  rather than generating a new network

20
Possibilities for Wider Application

• Most applicable to well-studied organisms
• Needs a reasonable well-defined set of
  connections (protein-protein, protein-DNA) to
  start with
• Works best when large, diverse data sets are
  available
• Includes possible components of a more systematic
  system
• Such a system could determine its confidence in
  its results

21
We need YOU to volunteer to present a paper
next quarter!!!