Transcript Protein-Protein Interactions

Protein-Protein Interactions
Networks
“A comprehensive analysis of protein-protein interactions in Saccharomyces
cerevisiae”P.Utez et al, Nature 2000

“Functional organisation of the yeast proteome by systematic analysis of protein
complexes” G. Gavin et al, Nature 2002


“Global Mapping of the Yeast Genetic Interaction Network” Tong et al, Science 2004
“Global analysis of protein activities using proteome chips” Zhu, H. et al. Science
2001

“Conserved patterns of protein interaction in multiple species” R. Sharan et al,
PNAS 2005

Genomics

Genomics – “The
large scale study
of genomes and
their functions”

Why protein
network?
Why protein network?


Assemblies represent more than the sum of their
parts.
`complexity' may partly rely on the contextual
combination of the gene products.
19,500 genes
14,000 genes
24,000 genes
26,000 genes
50,000 genes
Yeast as a model

Why yeast genomics?
A model eukaryote organism …
Saccharomyces cerevisiae
The best-studied organism
 ~5,500 genes.

16(!) chromosomes.

13 Mb of DNA (humans have ~3,000 Mb).


We know (?) the function of >1/2 of the
yeast genes.
All the essential functions are conserved
from yeast to humans.
Example: cell cycle
Lee Hartwell, Nobel Prize 2001
4 methodologies for
high throughput research

Two hybrid systems

Analysis of protein complexes

Synthetic lethal

Protein Chips (?)
Two hybrid system

Aim:


Identify pairs of Physical interactions.
Solution:

Use the transcription mechanism of the
cell
The central dogma
DNA
TRANSCRIPTION
RNA
TRANSLATION
PROTEIN
3
Transcription factors
Movie – transcription (molecular model, real time) 7.2
Transcription – real time (viedo)
Reporter gene
Eukaryotic mRNA
Two hybrid system
 Isolate
double plasmids using
reporter or selection methods.
All against All
Focus on the baits
Baits are analyzed separately.
 192 baits vs. ~6000 pray yeast
strains.

A component of RNA polymerase I, III, identification of three new interacting proteins
Two hybrid system
Two hybrid system
“A comprehensive two-hybrid analysis to explore the yeast
protein interactome“ Ito T. et al, PNAS 2001.

Analysis of protein complexes

Aim:
Identification of complexes and their sub
units.

Solution: a two step method


Isolation of only relevant complexes
Identification of complex units.
Double Isolation
Identification of the members

Divide and conquer-
• Denaturate assembly
• Digest with protease
• Mass spectrometry
How does it work?


The deflection route of
ionized molecules is
used to determine the
molecule’s mass.
The output:
Analysis of protein complexes

Cross results of peptide mass with protein database.

Mass spectrometry can be implied again if the data is
not sufficient, this time for the peptides.
Analysis of protein complexes
•
Systematic(1): 1739 bait proteins.
•
•
232 complexes with 589 baits.
Systematic(2): 725 bait proteins.
•
3,617 interactions with 493 baits.
Analysis of protein complexes





About 25% false positive rate.
Covers 56/60%, 10/35% in Y2H, of known
complexes.
Only 7% of the interactions were seen by Y2H
assays.
But,
Can evaluate protein


Concentration.
Localization.
Post-translational modifications.
Synthetic lethality






First, few words on essentiality.
Create new strains, each strain with one
gene deleted (96% coverage)
Tag each strains with a unique sequence.
Grow all the strains.
Measure the amount of each seq.
Some 18.7% (1,105) are essential.
Synthetic lethality



High genetic redundancy hardens the discovery
of many gene functions (30%).
Only the double mutation is lethal, either of the
single mutations is viable.
Why?



Single biochemical pathway.
Two distinct pathways for one process.
…
The naïve approach

But how do you genomics it? …
All vs. All

~5100 non essential mutants.
• Main tricks:
1. Haploid strains
2. Resistant markers.
3. Extra marker for the
library haploid.
Synthetic lethality …
Making it genomics


Mass analysis: Crossing the query haploid with a
library (synthetic genetic array)
Tetrad analysis: Validation and finding synthetic sick
The genetic interaction map

8 genes against all produced a network of synthetic
lethal pairs.
Synthetic lethality …
Making it genomics




132 query genes vs. 4700
False negatives – 17-42%.
At least 4 times more dense than the PPI network.
Predicting ~100,000 interactions (?)
PPI Summery (2003)
PPI Summery
S. Cerevisiae (Yeast) C. Elegans (Worm)
• 4389 proteins
• 2718 proteins
• 14319 interactions • 3926 interactions
D. Melanogaster (Fly)
• 7038 proteins
• 20720 interactions
Sharan et al. PNAS 2005
We like Networks

Exploit graph theory methods.

Provide a general solution for data integration.
Network Structure and Function

Identify highly nonrandom network
structural patterns that reflect function:




Ideker et al: Finding co-regulated sub-graphs.
Lee at el: The repeated instances of each motif
are the result of evolutionary convergence.
Barabasi at el: Network motifs are associated
with specific cellular tasks.
…
Conserved patterns of PPI in
multiple species
Bacterial pathogen
(Helicobacter pylori)
~1500 interactions
~700 interacting genes
Baker’s yeast
(Saccharomyes cerevisiae)
~15000 interactions
~5000 interacting genes
Kelley et al. PNAS 2003
Goals






Separating true PPI from false positives.
Assign functional roles to interactions.
Predict interactions.
Organizing the data into models of cellular
signaling and regulatory machinery.
How?
Use approach based on evolutionary cross-species
comparisons.
Interaction graph (per species)



Vertices are the organism’s interacting proteins.
Edges are pair-wise interactions between proteins.
Edges are weighted using a logistic regression model:
 A: Number of times an interaction was observed.
 For Fly and worm observation In one experiment.
 B: Correlation coefficient of the gene expression.
 Shown to be correlated to interaction.
 C: Proteins’ small world clustering coefficient.
 Sum of the neighbors logHG probs.
How do we find Sub-network
conservation?


Interactions within each species should
approximate the desired structure:
 Pathway. Signal transduction.
 Cluster. Protein complex.
Many-to-many correspondence between the sets
of proteins.
Network alignment graph

Each node corresponds to k sequence-similar proteins.



Edge represents a conserved interaction.




BLAST E value < -7; considering the 10 best matches only.
Cannot be split into two parts with no sequence similarity between
them.
Match -> One pair of proteins directly interacts and all other
include proteins with distance <2 in the interaction maps.
Gap –> All protein pairs are of distance 2 in the interaction maps.
Match-Gap-> At least max{2, k −1} protein pairs directly interact.
A subgraph corresponds to a conserved sub-network.
A probabilistic model
(
)
q
e
(
)
=
S P
log q
e P
random
q(e) – interaction similarity
Searching for conserved subnetworks

Identifying high-scoring subgraphs of the network
alignment graph.




…This problem is computationally hard.
Exhaustively we find seeds - paths with 4 nodes.
Expand high scoring seeds. Greedily add/remove
nodes.
Filter subgraphs with a high degree of overlap
(>80%).
Statistical evaluation of subnetworks

Randomized data is produced:




Random shuffling of each of the interaction graphs.
Randomizing the sequence-similarity relationships.
Find the highest-scoring sub-networks of a given
size.
P-value is computed by the distribution of the top
scores.
The final product
← Protein sequence
similarity →
← Bacteria →
←
Yeast
→
3-way Comparison
S. cerevisiae
• 4389 proteins
• 14319 interactions
C. elegans
• 2718 proteins
• 3926 interactions
D. melanogaster
• 7038 proteins
• 20720 interactions
Sharan et al. PNAS 2005
Multiple Network Alignment
Network alignment
Preprocessing
Subnetwork search
Conserved paths
Interaction scores:
logistic regression on
#observations, expression
correlation, clustering coeff.
Filtering &
Visualizing
p-value<0.01,
80% overlap
Conserved
interactions
Protein
groups
Conserved clusters
Reduced false positives

Compared these conserved clusters to known
complexes in yeast 



Pure cluster - contain >2 annotated proteins and
>1/2 of these shared the same annotation.
94%(>83% in mono specie) pure clusters.
Did ‘‘sticky’’ proteins biased the clusters?
Of 39 proteins (> 50 neighbors), only 10 were
included in conserved clusters. And they were
annotated so.
Cross Validation: Function

Guilty by association.



Enrichment of GO annotation (p<0.01).
More then half of the annotated proteins had the annotation.
Species
#Correct
#Predictions
Success rate
(%)
Yeast
114
198
58
Worm
57
95
60
Fly
115
184
63
Outperforms sequence-based approach at 37-53%.
Cross Validation: Interaction


[1] Evidence that proteins with similar sequences interact within
other species.
[2] Co-occurrence of these proteins in the same conserved cluster.
Species
Specificity
(%)
77
P-value
Strategy
Yeast
Sensitivity
(%)
50
1e-25
]1[
Worm
43
82
1e-13
]1[
Fly
23
84
5e-5
]1[
Yeast
9
99
1e-6
]2[+]1[
Worm
10
100
6e-4
]2[+]1[
Fly
0.4
100
0.5
]2[+]1[
Wet Validation: Interaction


The tests were performed by using two-hybrid assays.
Of the 65 yeast predicted interactions:


5 were self inducing.
31 tested positive.
Conclusions

Associate proteins that are not necessarily each
other’s best sequence match.



177/679 conserved clusters.
31/129 conserved paths.
Inter module interaction is reinforced by interspecies observations.

40-52% >> 0.042% as a random PPI prediction.

Many PPI circuits are conserved over evolution.
!!!Thanks

Recoverin, a calcium-activated myristoyl switch.
GO – Gene Ontology

all : all ( 171472 )

GO:0008150 : biological_process ( 109503 )

GO:0007582 : physiological process ( 70981 )

GO:0008152 : metabolism ( 41395 )

GO:0009058 : biosynthesis ( 10256 )
GO:0009059 : macromolecule biosynthesis ( 6876 )
GO:0006412 : protein biosynthesis ( 4611 )

GO:0043170 : macromolecule metabolism ( 17198 )
GO:0009059 : macromolecule biosynthesis ( 6876 )
GO:0006412 : protein biosynthesis ( 4611 )
GO:0019538 : protein metabolism ( 12856 )
GO:0006412 : protein biosynthesis ( 4611 )

GO:0005575 : cellular_component ( 98453 )

GO:0003674 : molecular_function ( 108120 )
back
Interaction distribution
Expression data



Yeast - 794 conditions.
Fly - over 90 CC time points+170 profiles.
Worm - over 553 conditions.
back
Edge weight


where 0, . . . , 3 are the parameters of the
distribution.
Maximize the likelihood:
Positive: MIPS interactions.
 Negative: random or false positives in the cross
validation test.
Yeast - 1006 positive and negative examples.
Fly - 96 positive and negative examples.
Worm – 24 positive and 50 negative examples.




back
back
71 conserved regions: 183 significant clusters and 240 significant paths.
A probabilistic model





Ms - the sub-network model.
Mn - the null model.
Ouv - the set of available observations on u-v.
Puv- fraction of (u,v) in order preserving graphs family.
T/Fuv – True/False edge (u,v).
back
A probabilistic model


Each species’ interaction map was randomly
constructed.
Randomizing assumptions:
 Each interaction should be present independently
with high probability.
 The probability depends on their total number of
connections in the network.
Why Yeast?

“Comparative Genomics of the Eukaryotes” Rubin GM. et al. Science 2000
back
Analysis of protein complexes
1.
Isolation:
A straight forward method, using Affinity chromatography.
A target protein is attached to polymer beads that are
packed into a column. Cell proteins are washed through
the column.Proteins the interact with the target protein
adhere to the affinity matrix and are eluted later.
Analysis of protein complexes
2.
Isolation:
Co-immunoprecipitation. An antibody that recognizes the
target protein is used to isolate the protein.
Usually the there isn’t a highly specific antibody for the
target protein. A chimera protein is formed, using a the
target protein and an epitope tag.
The common tag is a enzyme glutathione S-transferase
(GST).
Analysis of protein complexes
2.
Isolation:
Isolation of complex using the Chimera
Glutathione coated beads
Cell extract
Glutathione solution
MIPS




Munich Information Center for Protein Sequences
(MIPS).
Hierarchy Structure.
Only manually annotated complexes from DIP.
Left with 486 proteins spanning 57 categories at
level 3.
back