Structure, evolution and dynamics of transcriptional regulatory networks M. Madan Babu, PhD National Institutes of Health

Networks in Biology Nodes Links Interaction A B Network Proteins Physical Interaction Protein-Protein A B Protein Interaction Metabolites Enzymatic conversion Protein-Metabolite A B Metabolic Transcription factor Target genes Transcriptional Interaction Protein-DNA A B Transcriptional

Evolution of the transcriptional regulatory network Dynamic nature of the transcriptional regulatory network Evolution of local network structure Evolution of global network structure Dynamics of local network structure Dynamics of global network structure Structure of the transcriptional regulatory network Local network structure: network motifs Global network structure: scale-free structure Outline

Evolution of the transcriptional regulatory network Dynamic nature of the transcriptional regulatory network Evolution of local network structure Evolution of global network structure Dynamics of local network structure Dynamics of global network structure Structure of the transcriptional regulatory network Local network structure: network motifs Global network structure: scale-free structure Outline

Structure of the transcriptional regulatory network Scale free network (Global level) all transcriptional interactions in a cell Albert & Barabasi Madan Babu M, Luscombe N, Aravind L, Gerstein M & Teichmann SA Current Opinion in Structural Biology (2004) Motifs (Local level) patterns of Interconnections Uri Alon & Rick Young Basic unit (Components) transcriptional interaction Transcription factor Target gene

Properties of transcriptional networks Local level: Transcriptional networks are made up of motifs which perform information processing task Global level: Transcriptional networks are scale-free conferring robustness to the system

Transcriptional networks are made up of motifs Single input Motif - Co-ordinates expression - Enforces order in expression - Quicker response ArgR ArgD ArgEArgF Multiple input Motif - Integrates different signals - Quicker response TrpRTyrR AroM AroL Network Motif “Patterns of interconnections that recur at different parts and with specific information processing task” Feed Forward Motif - Responds to persistent signal - Filters noise Crp AraC AraBAD Function Shen-Orr et. al. Nature Genetics (2002) & Lee et. al. Science (2002)

N (k)  k  1 Scale-free structure Presence of few nodes with many links and many nodes with few links Transcriptional networks are scale-free Scale free structure provides robustness to the system Albert & Barabasi, Rev Mod Phys (2002)

Scale-free networks exhibit robustness Robustness – The ability of complex systems to maintain their function even when the structure of the system changes significantly Tolerant to random removal of nodes (mutations) Vulnerable to targeted attack of hubs (mutations) – Drug targets Hubs are crucial components in such networks Haiyuan Yu et. al. Trends in Genetics (2004)

Summary I - Structure Transcriptional networks are made up of motifs that have specific information processing task Transcriptional networks are scale-free which confers robustness to such systems, with hubs assuming importance Madan Babu M, Luscombe N et. al Current Opinion in Structural Biology (2004)

Evolution of the transcriptional regulatory network Dynamic nature of the transcriptional regulatory network Evolution of local network structure Evolution of global network structure Dynamics of local network structure Dynamics of global network structure Structure of the transcriptional regulatory network Local network structure: network motifs Global network structure: scale-free structure Outline

Evolution of the transcriptional regulatory network in yeast How did the regulatory network evolve? What are the underlying molecular mechanisms? Organismal evolution Growth of the regulatory network E. coli Yeast

? ? ? ? What are the mechanisms for creating new genes? Network Growth ? ? ? Transcription factor Creation of a new target gene

Mechanisms for the creation of new genes OR Recombination/Innovation Duplication & Divergence Duplication Divergence A A’ A’’

Dataset used in the analysis Transcriptional regulatory network in Yeast 477 proteins (109 TFs TGs) 901 interactions Guelzim et.al. Nat. Gen. (2002)

Duplication of the basic unit Basic unit Duplication of target gene transcription factor Duplication of Duplication of transcription factor and target gene

Divergence Scenario Duplication of TF Duplication of TG Duplication of TF+TG InheritanceLoss & gainInheritanceLoss & gain Loss & inheritance Gain Inheritance, loss and gain of interaction

Divergence Scenario I - Inheritance Duplication of TF Duplication of TG Duplication of TF+TG InheritanceLoss & gainInheritanceLoss & gain Loss & inheritance Gain Inheritance of interaction

Consequence of duplicating a target gene Basic unit Duplication of target gene same TF regulates homologous TGs (inheritance of interaction) Divergence of TG 20% of all interactions in Yeast (166 interactions) Eno1 Eno2 Gcr1

Consequence of duplicating a transcription factor Basic unit Duplication of transcription factor Same gene regulated by homologous TFs (inheritance of interaction) Divergence of TF Pdr1Pdr3 Flr1 22% of all interactions in Yeast (188 interactions)

Consequences of duplicating a TF and TG Basic unit Duplication of transcription factor and target gene Homologous TFs regulate homologous TGs (loss/inheritance of interaction) Divergence of TF and TG Mal11 Mal31 Mal13 Mal33 4% of all interactions in Yeast (31 interactions)

Contribution of duplication with inheritance to network growth 188 duplication of TF (21%) 166 duplication of TG (20%) 31 duplication of TG + TF (4%) 385 interactions in Yeast = 45% ~1/2 of the regulatory network has evolved by duplication followed by inheritance of interaction after the duplication event

Divergence Scenario II – Gain of new interaction Duplication of TF Duplication of TG Duplication of TF+TG InheritanceLoss & gainInheritanceLoss & gain Loss & inheritance Gain Gain of interaction

Duplication followed by innovation of new interaction Basic unit Duplication of transcription factor Duplication of target gene Divergence of TG + BS Different transcription factors regulate Homologous genes (loss/gain of interaction) Divergence of TF to recognise a different BS Homologous TFs regulating different genes (loss/gain of interaction)

365 duplication & gain (43%) Duplication with gain of new interactions “~ 1/2 of the regulatory network has evolved by duplication and gain of new interactions after the duplication event”

Network growth by Recombination Recombination Innovation

“~1/10 of the regulatory network has evolved by innovation” 101 innovation (12%) Innovation of new interactions

385 duplication & inheritance (45%) 365 duplication & gain (43%) 101 innovation (12%) Evolution of the transcriptional regulatory network ~90% of the network has evolved by duplication followed by Inheritance, loss and gain of interaction

How can such events affect local and global network structure? Questions - are the motifs and scale free structure products of duplication events ? - are the motifs and scale free structure selected for in evolution ?

Duplication of TG Duplication of TF Duplication of TF & TG Motifs Growth models Single Input Module Feed-Forward Motif Multiple input motif Duplication models and evolution of network motifs

Very rarely we find instances where duplication events have resulted in the formation of network motifs Evolution of local network structure Network motifs have evolved independently (convergent evolution) multiple times because they confer specific properties to the network

Duplication and evolution of scale free structure Growth by gene duplication and inheritance of interaction can explain evolution of scale-free structure Regulatory hubs should control more duplicate genes (4) (2) Total = 6 “Rich gets richer” (5) (2) Duplication & inheritance Total = 7 (6) (2) Duplication & inheritance Total = 8

Regulatory hubs do not regulate duplicate genes more often than any other normal transcription factor Scale free structure has been selected for in evolution and is not a product of duplication events Evolution of global network structure

Summary II - Evolution 385 duplication & inheritance (45%) 365 duplication & gain (43%) 101 innovation (12%) Gene duplication followed by inheritance of interaction and gain of new interactions have contributed to 90% of the network Teichmann SA & Madan Babu M Nature Genetics (2004) Network motifs and the scale free structure are not products of duplication events, but have been selected for in evolution

Evolution of the transcriptional regulatory network Dynamic nature of the transcriptional regulatory network Evolution of local network structure Evolution of global network structure Dynamics of local network structure Dynamics of global network structure Structure of the transcriptional regulatory network Local network structure: network motifs Global network structure: scale-free structure Outline

Are there differences in the sub-networks under different conditions? Cell cycle Sporulation Stress Static network Across all cellular conditions Dynamic nature of the regulatory network in yeast How are the networks used under different conditions?

Dataset - gene regulatory network in Yeast 3,962 genes (142 TFs +3,820 TGs) 7,074 Regulatory interactions Individual experiments TRANSFAC DB + Kepes dataset 288 genes genes 356 interactions interactions ChIp-chip experiments Snyder lab + Young lab 1560 genes gene 2124 interactions interaction

Integrating gene regulatory network with expression data 142 TFs 3,820 TGs 7,074 Interactions Transcription Factors 1 condition 2 conditions 3 conditions 4 conditions 5 conditions 142 TFs 1,808 TGs 4,066 Interactions Target Genes Gene expression data for 5 cellular conditions Cell-cycle Sporulation DNA damage Diauxic shift Stress

Back-tracking method to find active sub-networks Gene regulatory network Identify differentially regulated genes Find TFs that regulate the genesFind TFs that regulate these TFs Active sub-network

DNA damage Cell cycle Sporulation Diauxic shift Stress Active sub-networks: How different are they ? Multi-stage processes Binary Processes

Network Motifs Milo et.al (2002), Lee et.al (2002) Single Input Motif (SIM) – 23% Feed-forward Motif (FF) – 27% Multi-Input Motifs (MIM) – 50%

Sub-networks : Network motifs Network motifs are used preferentially in the different cellular conditions

- Do different proteins become hubs under different conditions? - Is it the same protein that acts as a regulatory hub? Cell cycleSporulationDiauxic shiftDNA damageStress Condition specific networks are scale-free

Regulatory hubs change with conditions Cluster TFs according to the number of target genes active in each condition Different TFs become key regulators in different conditions CCSPDSDDSRTF Swi6

Hubs regulate other hubs to initiate cellular events Suggests a structure which transfers weight between hubs to trigger cellular events

Network Parameters Connectivity Path length Clustering coefficient

Network Parameters - Connectivity Outgoing connections = 49.8 on average, each TF regulates ~50 genes Changes Incoming connections = 2.1 on average, each gene is regulated by ~2 TFs Remains constant

Network parameters : Connectivity Binary: Quick, large-scale turnover of genes Multi-stage: Controlled, ticking over of genes at different stages “Binary conditions”  greater connectivity “Multi-stage conditions”  lower connectivity

Number of intermediate TFs until final target Path length 1 intermediate TF = 1 Indication of how immediate a regulatory response is Average path length = 4.7 Network Parameters – Path length Starting TF Final target

“Binary conditions”  shorter path-length  “faster”, direct action “Multi-stage” conditions  longer path-length  “slower”, indirect action  intermediate TFs regulate different stages Binary Multi-stage Network parameters : Path length

Clustering coefficient = existing links/possible links = 1/6 = 0.17 Measure of inter-connectedness of the network Average coefficient = possible links 1 existing link 4 neighbours Network Parameters – Clustering coefficient Ratio of existing links to maximum number of links for neighboring nodes

“Binary conditions”  smaller coefficients  less TF-TF inter-regulation “Multi-stage conditions”  larger coefficients  more TF-TF inter-regulation Binary Multi-stage Network parameters : Clustering coeff

Sub-networks have evolved both their local structure and global structure to respond to cellular conditions efficiently multi-stage conditions fewer target genes longer path lengths more inter-regulation between TFs binary conditions more target genes shorter path lengths less inter-regulation between TFs

Summary III - Dynamics Sub-networks have evolved both their local structure and global structure to respond to cellular conditions efficiently Luscombe N, Madan Babu M et. al Nature (2004) Network motifs are preferentially used under the different cellular conditions and different proteins act as regulatory hubs in different cellular conditions

Conclusions “This has resulted in a network structure that can efficiently re-wire interactions to meet the biological demand placed by the process” “Transcriptional networks are made up of network motifs at the local level and have a scale-free structure at the global level” “Even though close to 90% of the regulatory network in yeast has evolved by duplication, network motifs and scale free structure are not products of duplication events – instead they have been selected for in evolution”

Implications First overview of the evolution and dynamics of the transcriptional regulatory network of a eukaryote Identification of key regulatory hubs under different conditions can serve as good drug targets Provides insights into engineering regulatory interactions Methods developed to reconstruct and compare active networks are generically applicable

Acknowledgements Sarah Teichmann MRC-LMB, Cambridge, U.K MRC - Laboratory of Molecular Biology Cambridge Commonwealth Trust Overseas Research Studentship Trinity College, Cambridge Nicholas Luscombe Haiyuan Yu Michael Snyder Mark Gerstein Yale University, NH, U.S.A L Aravind NCBI, NLM National Institutes of Health, USA