1. Bioinformatics: Applications
ZOO 4903
Fall 2006, MW 10:30-11:45
Sutton Hall, Room 312
Jonathan Wren
Protein-Protein Interaction Networks

2. Lecture overview What we’ve talked about so far Overview
Proteins & their domains
Protein 3D structure
Overview
Proteins do not function in a vacuum
Methods of detecting protein-protein interactions (PPI)
Structure and types of networks
Behavior of networks
PPIs are equally or more important than structure

3. Cells are crowded places!
Hopper & Mayer, 1999, Prokaryotes. Am.Sci. 87:518

4. Importance of protein-protein interactions
Many cellular processes are regulated by multiprotein complexes
Distortions of protein interactions can cause diseases
Protein function can be predicted by knowing functions of interacting partners (“guilt by association”)
A comparison of sequence (GenBank) and protein-protein interaction data (DIP database)
Adapted from S. Fields, FEBS, 2005

5. Types of protein-protein interactions (PPI)
Non-obligate PPI
Obligate PPI
usually permanent
the protomers are not found as stable structures on their own in vivo
Stable
(many enzyme-inhibitor complexes)
dissociation constant Kd=[A][B] / [AB]
10-7 ÷ M
Transient
Weak
(electron transport complexes)
Kd mM-M
Non-obligate transient homodimer, Sperm lysin (interaction is broken and formed continuously)
Intermediate
(antibody-antigen, TCR-MHC-peptide, signal transduction PPI), Kd M-nM
Strong
(require a molecular trigger to shift the oligomeric equilibrium)
Kd nM-fM
Obligate heterodimer
Human cathepsin D
Non-obligate permanent heterodimer
Thrombin and rodniin inhibitor
Bovine G protein dissociates into G and G subunits upon GTP, but forms a stable trimer upon GDP

6. Multiple interactions: Guanine-nucleotide binding protein
Guanine nucleotide-binding proteins regulate a variety of processes, including sensual perception, protein synthesis, various transport processes, and cell growth and differentiation. They act as molecular switches and timers that cycle between inactive guanosine diphosphate (GDP)-bound and active guanosine triphosphate (GTP)-bound states. Recent structural studies show that the switch apparatus itself is a conserved fundamental module but that its regulators and effectors are quite diverse in their structures and modes of interaction
Adapted from Vetter & Wittinghofer, Science 2001

7. Multiple interactions: Guanine-nucleotide binding protein
A: small changes in the interaction site could be very disruptive to many proteins
Question: How conserved are the interactive vs non-interactive portions of this protein?
Adapted from Vetter & Wittinghofer, Science 2001

8. Protein evolution - gene duplication
Right after duplication
Over time
Pair of duplicated proteins
Shared interactions
Duplication – provides redundancy & room for divergence

9. Methods of identifying PPIs
Experimental
Protein-protein arrays
Y2H assay
TAP assay
Computational/Inferential
Interolog analysis
Co-localization, co-expression
Correlated mutations
Text-mining

10. Interologs Homolog Interolog Ortholog Paralog Common ancestors
Common 3D structure
Common active sites
Ortholog
Derived from Speciation
Paralog
Derived from Duplication
Interolog
Conserved Protein-Protein Interaction
If A and B interact in organism X, then if organism Y has a homolog of A (A’) and a homolog of B (B’) then A’ and B’ should interact too.
Requires list of known interacting partners
Thus, finding one PPI may yield dividends!

11. Protein Arrays H Zhu et al (2000) “Analysis of yeast protein kinases
Protein chips are disposable arrays of microwells in silicone elastomer sheets placed on top of microscope slides.
H Zhu et al (2000) “Analysis of yeast protein kinases
using protein chips” Nature Genetics 26:

12. The Two-Hybrid System Reporter Gene
Two hybrid proteins are generated with transcription factor domains
Both fusions are expressed in a yeast cell that carries a reporter gene whose expression is under the control of binding sites for the DNA-binding domain
Activation
Domain
The key to the two-hybrid screen is that in most eukaryotic transcription factors, the activating and binding domains are modular and can function in close proximity to each other without direct binding
Prey
Protein
Bait
Protein
Binding
Domain
Reporter Gene

13. The Two-Hybrid System Reporter mRNA Reporter mRNA Reporter mRNA
Interaction of bait and prey proteins localizes the activation domain to the reporter gene, thus activating transcription.
Since the reporter gene typically codes for a survival factor, yeast colonies will grow only when an interaction occurs.
Activation
Domain
Survival gene – such as to create lysine (remember Jurassic Park?)
Prey
Protein
Reporter mRNA
Bait
Protein
Reporter mRNA
Reporter mRNA
Binding
Domain
Reporter mRNA
Reporter mRNA
Reporter Gene

14. Genome-wide analysis by Y2H
Matrix approach: a matrix of prey clones is added to the matrix of bait clones. Diploids where X and Y interact are selected based on the expression of a reporter gene.
Library approach: one bait X is screened against an entire library. Positives are selected based on their ability to grow on specific substrates.
Uetz et al Nature 2000 – 957 putative interactions in Yeast
Rain et al Nature 2001 – 1,200 putative interactions in H. Pylori
Ho et al Nature 2002 – 3,617 putative interactions in Yeast (Mass Spec)
Adapted from B. Causier, Mass Spectroscopy Reviews, 2004

15. Advantages of Y2H
In vivo technique, good approximation of processes which occur in higher eukaryotes.
Transient interactions can be determined, can predict the affinity of an interaction.
Can be used to detect potential interactions of genes not yet observed to be translated into proteins (e.g. rarely expressed) or novel constructs (e.g. therapeutics)
Relatively fast and efficient.

16. Disadvantages of Y2H
Fusion of a protein into chimeras can change the structure of a target
Protein interactions can be different in yeast and the organisms where the genes came from
It is difficult to target extracellular proteins
It is hard to detect interactions between proteins active only in a complex
Proteins which can interact in two-hybrid experiments, may never interact in vivo

17. Tandem affinity purification method (TAP)
Target protein ORF is fused with the DNA sequences encoding TAP tag;
Tagged ORFs are expressed in yeast cells and form native complexes;
The complexes are purified by TAP method;
Components of each complex are found by gel electrophoresis or MS.

18. Tandem affinity purification method (TAP)
TAP tag consists of two IgG binding domains of Staphylococcus protein A and calmodulin binding peptide;
7123 interactions can be clustered into 547 complexes (Krogan et al, 2006)
O. Puig et al, Methods, 2001

19. Differences and similarities between Y2H and MS-TAP
TAP permits protein complexes to be isolated, but cannot detect weak/transient PPIs
Both methods generate a lot of false positives, only ~50% interactions are biologically significant
Y2H is in vivo technique
MS can detect large stable complexes and networks of interactions

20. Text Mining Searching Medline or PubMed for words or word combinations
Co-occurrence of terms is the simplest metric, yet lends to a higher FP rate
NLP methods are more specific (e.g., “X binds to Y”; “X interacts with Y”; “X associates with Y” etc.) yet are difficult to detect so it has a higher FN rate
Normally requires a list of known gene names or protein names for a given organism

21. Pre-BIND Used Support Vector Machine (SVM) to scan literature for PPIs
Precision, accuracy and recall of 92% for correctly classifying PPI abstracts
Estimated to capture 60% of all abstracted protein interactions for a given organism
Donaldson et al. BMC Bioinformatics :11

22. Drosophila interaction map
Interaction map plus subcell-localization (from GO)
From:
A Protein Interaction Map of Drosophila
Giot et al. Science 302, (2003)

23. Comparing large scale data of protein-protein interactions
All methods except for Y2H and synthetic lethality technique are biased toward abundant proteins.
PPI are biased toward certain cellular localizations.
Evolutionarily conserved proteins have much better coverage in Y2H than the proteins restricted to a certain organism.
Von Mering et al, Nature, 2002

24. Functional organization of yeast proteome: network of protein complexes
Essential gene products are more likely to interact with essential rather than nonessential proteins
Orthologous proteins interact with complexes enriched with orthologs
Gavin et al, Nature, 2002

25. PPI Databases online DIP MIPS (small scale)
MIPS (small scale)
BIND (PPI, Prot-DNA, Prot-SM)
(now owned by Unleashed)
OPHID (predicted interactions)
MINT - Molecular Interactions Database
IntAct (EBI)
InterDom (domain interactions)
STRING (EMBL)

26. Interaction databases
Types
Experiment (E)
Structure detail (S)
Predicted
Physical (P)
Functional (F)
Curated (C)
Homology modeling (H)
*International Molecular Exchange (IMEx) consortium
The IMEx consortium is a group of major public interaction data providers intending to share curation effort and exchange completed records on molecular interaction data, similar to successful global collaborations for protein and DNA sequences and for macromolecular structures.

27. Comparing the DBs High FP rate in high- throughput exp.
Disagreement between benchmark sets
Experimental PPI data is sparse relative to all PPIs, so dataset overlap is small and hard to confirm with multiple sources

28. PPI network properties
Nodes & connections

29. Characteristics of networks
n = nodes, k = connections or “edges”
K=2
K=2
K=3
K=1
In biology, n refers to genes/proteins (and/or metabolites) while k refers to interactions

30. Examples of networks: Proximity-based interactions

31. Examples of networks: Distant interactions

32. Elementary features: node (n) diversity and dynamics

33. Elementary features: edge (k) diversity and dynamics

34. Elementary features: Network Evolution

35. Network properties Network Structure Metrics Network Structure Types
Average path length
Degree distribution(connectivity)
Clustering coefficient
Network Structure Types
Regular
Random
Small-world
Scale-free

36. Structural metrics: Path length & network diameter

37. Structural Metrics: Degree distribution (connectivity)

38. Structural Metrics: Clustering coefficient

39. Network properties Network Metrics Network Structures
Average path length
Degree distribution(connectivity)
Clustering coefficient
Network Structures
Regular
Random
Small-world
Scale-free

40. Regular networks – fully connected

41. Regular networks – Lattice

42. Regular networks – Lattice: ring world

43. Random networks

44. Random Networks

45. Small-world networks

46. Exponential network degree distribution.

47. Scale-free networks
New nodes preferentially attach to highly connected ones
Coined by A.L. Barabasi in 1998

48. Different network models: Barabasi-Alberts.
Model of preferential attachment.
At each step, a new node is added to the graph.
The new node is attached to one of old nodes with probability proportional to the vertex degree.
ln(P(k))
Degree distribution – power law distribution.
Evolution – old proteins have more connections than new proteins
ln(k)
Barabasi & Albert, Science, 1999

49. Properties of scale-free networks.
Multiplying k by a constant, does not change the shape of the distribution – scale free distribution.
From T. Przytycka
Small diameter
Tolerance to errors and attacks
But: sub-networks can be scale-free while underlying degree distribution is not.

50. Difference between scale-free and random graph models.
Random networks are homogeneous, most nodes have the same number of links.
Scale-free networks have a number of highly connected verteces.
Adapted from Jeong et al, Nature, 2000

51. The Topology of PPI Networks
Small-world
Scale free
Recurring motifs (Barabasi et al. Nature Genetics 2003)

52. Highway connections – random. Airports – scale-free.
Reprinted from Linked: The New Science of Networks by Albert-Laszlo Barabasi

53. The Internet

54. Category: Internet Topology Description: Internet connectivity snapshot using Skitter data and Walrus visualization

55. Category: Natural Networks Description: Yeast protein network map, inside cell

56. Category: Natural Networks Description: Portion of food web in North Atlantic Ocean

57. Social networks relevant to spread of diseases, rumors, fads, etc.
Category: Social Networks Description: Individuals and their professions in the network of activities during the German Revolution of

58. Category: Economic Networks Description: World Trade network, 1992

59. Metabolic Networks
The metabolic networks of all organisms in all three domains of life (prokaryote, eukaryote, archaea) appear to be scale-free (43 examined)
The network diameter of all 43 metabolic networks is the same, irrespective of the number of proteins involved.
Does this seem counter-intuitive?

60. Implications – Attack Tolerance
Robust. For <3, removing nodes does not break network into islands.
Very resistant to random attacks, but attacks targeting key nodes are more dangerous.
Max Cluster Size
Path Length

61. Order, complexity and chaos
Principles of evolution

62. Evolutionary puzzle
The cell is constructed from unreliable parts and is subjected to mutations, yet it behaves in a robust and reliable manner.
Can natural selection alone explain this? Or might some additional principles also come into play, such as self-organization?
Kaufmann’s hypothesis: Natural selection acts on self-organizing systems, rather than creating them. Without an innate tendency toward order, most mutations would be fatal.

63. Boolean networks
Given a set of nodes, assign to each a set of rules that governs their state (1 or 0, on or off)
Ruleset (switch)
0 1
1 0
Ruleset (XOR)
00 0
01 1
10 1
11 0

64. Boolean networks
Given a set of nodes, assign to each a set of rules that governs their state (1 or 0, on or off)
Ruleset (switch)
0 1
1 0
Ruleset (XOR)
00 0
01 1
10 1
11 0

65. Boolean networks
Given a set of nodes, assign to each a set of rules that governs their state (1 or 0, on or off)
Ruleset (switch)
0 1
1 0
Ruleset (XOR)
00 0
01 1
10 1
11 0

66. Network behavior
Given this type of setup for network structure, what does the network behave like?
Does behavior change as n increases?
Does behavior change as k increases?
What do sparsely connected networks behave like?
What do highly connected networks behave like?
We can answer this with a simulation…
Run network dynamics program after this slide

67. Complexity
Ordered behavior is characteristic of genomic and metabolic networks: they quickly settle down into periodic patterns of activity that resist disturbance or cycle thru states.
Chaotic behavior is characteristic of many non-biological complex systems: sensitivity to initial conditions, long transients, and very large limit cycles (strange attractors).
BUT… Life exists and evolves in a region between order and chaos, termed “complexity” because it exhibits enough order to be stable & reproducible, but enough variation & instability to adapt to change
Think about the simulation in terms of evolution & the environment present in the cell with activities for each protein being governed by a set of rules. Whether early (few genes) or late in evolution, there is selection to control interactions.

68. Summary
PPI networks are one of the most active areas of bioinformatics research interest
PPI networks are constructed by empirical & computational methods
Networks have a structure that dictates their behavior
Biological networks are scale-free
Essential proteins have high connectivity
Life evolves on the edge between order & chaos

69. For next time
Supplementary reading S4