complex networks are found throughout biology

Complex networks are found throughout biology

Can we define the basic building blocks of networks?

Generalize the notion of MOTIFS, widely used in sequence analysis, to the level of networks.

Sequence motif: a sequence that appears much more frequently than in randomized sequences.

‘Network motif’: a pattern that appear much more frequently than in randomized networks.

Database of direct transcription interactions in E. coli

X Y Z

Transcription factors

operons

X

Y

Z

Directed graph

424 nodes, 519 interactionsBased on selected data from RegulonDB + 100 interactions from literature search

Shen-Orr 2002, Thieffry Collado-Vides 1998

Algorithm that finds n-node network motifs

Find all n-node circuits in real graphN

T

Y

U

X

Z

W M

X

ZY

X

ZY

X

ZY

Examples of 3-node circuits

And in a set of randomized graphs with the same distribution

of incoming and outgoing arrows.

Assign P-value probability of occurring more at random than

in the real graphRandomization: Newman, 2000, Sneppen& Malsov 2002

The thirteen 3-node connected subgraphs

Two types of feed-forward loops are significant

X

Y

Z

X

Y

Z

Coherent Incoherent

Dynamics of the feed-forward loop system

Mangan, PNAS, JMB, 2003

X(t) = time varying input

dY / dt = F(X) – a Y

dZ / dt = F(X) F(Y) – b Z

F

Threshold

The feed-forward loop is a filter for transient signals allowing fast

shutdown

Mangan, PNAS, JMB, 2003

GFP Reporter plasmid system for promoter activity

Lutz, Bujard 1997Kalir, Leibler, Surette, Alon, Science 2001

Plasmid with promoter for gene Xcontrolling a reporter

Construct strains, each reporting for a different promoter

High throughput cloning:Promoter PCR, restriction, ligation, preps all in 96-well format.

Gene X intact on chromosome

Construct strains, each reporting for a different promoter

Grow strains under same conditions andmeasure reporter fluorescence or luminescence

Commercial fluorimeter/luminometershakes, temperature control, imjectors.Measure at the same time cell optical density.

Each well reportsfor a different promoter

0 100 200 300 400 500 600 700

104

105

106

minimal + aa LB minimal - aa arabinose 0.5mM

Activity of 96 promoters across 20 h growth in 20 conditionsG

FP

/OD

0 500 1000 1500 2000 2500 3000 3500 4000 4500

104

105

106

7 min resolution

*2

*1.1

*0.9

*0.5

GFP repeat 1

GF

P r

epea

t 2Day-day reproducibility of better than 10%

10%2-fold

Maximal response to a pulse of X is filtered by FFL

Simulation

X

ZX

Y

Z

Input pulse to X of duration T T

Response

Pulse duration T

Sign-sensitive filtering by arabinose feed-forward loop

Max response

crp

lacZ

crp

araC

araB

cAMP Pulse duration [min]Experiments: Mangan, Zaslaver, Alon, JMB (2003).

OFF pulse

Feedforward loop is a sign-sensitive filterFeedforward loop is a sign-sensitive filter

Vs.

=lacZYA =araBAD

Mangan Zaslaver, Alon JMB 2003

Shen-Orr, Milo, Mangan, Alon Nature genetics 2002

The single-input module can generate a temporal program of gene expression

Flagella operons are activated in temporal order

Kalir, Leibler, Surette, Alon Science 2001Laub, McAdams, Shapiro Science 2000

Temporal order matches position of proteinsalong the flagella motor

Turn ON (arg removed) Turn OFF (arg added)

Time [min] Time

Temporal order in Arginine biosynthesis system with minutes between genes

Zaslaver, Surette, Alon,

Nature Genetics 2004

Turn ON (arg removed) Turn OFF (arg added)

Time [min] Time

Glutamate

N-Ac-Glutamate

N-Ac-glutamyl-p

N-Ac-Ornithine

N-Ac-glutamyl-SA

Citrulline

Arginine

Temporal order matches enzyme position in the pathway

Klipp, Heinrich

199 4-node directed connected subgraphs199 4-node directed connected subgraphs

4-node motifs in E. coli network: overlapping regulation

X

Z

X

Y

Z1

Y

W Z2

Overlapping regulation Generalization of Feedforward loop

to two controlled genes

Hengge-Aronis

Mapping Logic gates using GFP reporter arrays (LacZ)

LacZIPTG

cAMP

ANDGATE?

Setty, Mayo, Surette, Alon, PNAS 2003

Can we draw complex networks in an understandable way?

E. coli and yeast transcriptional networks show the same motifs

X

Z

Y

W

X

Y

Z1 Z2

The same network motifs characterize transcriptional regulation in a eukaryote and a prokaryote

The same 4-node motifs

Also SIMS and DORs

X

Y

Z

Feed-forward loop: coli 40, yeast 74, over 10 STDs from random networks! Same two types out of eight, coherent and incoherent FFL

Milo et al 2002, Lee at al 2002

X1

Y1

Z1

AND

AND

X2

Y2

Z2

AND

AND

Z3

R. Losick et al. , PLOS 2004

Developmental transcription network made of feed-forward loops: B. subtilis sporulation

X1

Y1

Z1

AND

AND

X2

Y2

Z2

AND

AND

Z3

0 1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8

1 Z1 Z2 Z3

time

Z

Feed-forward loops drive temporal patternof pulses of expression

Food webs represent predatory interactions between species

Skipwith pond (25 nodes), primarily invertebratesLittle Rock Lake (92 nodes), pelagic and benthic species Bridge Brook Lake (25 nodes), pelagic lake species Chesapeake Bay (31 nodes), emphasizing larger fishesYthan Estuary (78 nodes) birds, fishes, invertebratesCoachella Valley (29 nodes), diverse desert taxa St. Martin Island (42 nodes), lizards Source: Williams & Martinez, 2000

Foodwebs have“consensus motifs”

• Consensus motifs – different from transcription networks

• Feedforward is not motif - omnivores are under-represented

Links between WWW Pages – a completely different set of motifs is found

• WebPages are nodes and Links are directed edges

• 3 node results

Full reverse engineering of electronic circuit from transistor to module level

Each node represents a transistor

Each node represents a transistor motif:Logic gate

Each node represents a gate motif:D-flipflop

Each node represents a gate-flipflop motif:counter

Itzkovitz 2004

Map of synaptic connections between C. elegans neurons

White, Brenner, 1986

FLP

AVD

AVA

Nose touch

Backward movement

Feedforward loops in C. elegans avoidance reflex circuit

ASH Sensory neurons

Interneurons

Thomas & Lockery 1999

Noxious chemicals, nose touch

Networks can be grouped to super-families based on the significance profile

Milo et al Science 2004

NormalizedZ-score

Summary –network motifs

Network motifs, significant patterns in networks

Three motifs in transcription networksand their functions

High-accuracy promoter activity measurements from living cells

Milo Science 2002, Shen-Orr Nat. Gen. 2002, Itzkovitz PRE 2003 ,algorithms at www.weizmann.ac.il/mcb/UriAlon

Acknowledgments

Alex SigalNaama GevaGalit LahavNitzan RosenfeldShiraz Kalir

Weizmann Institute, Israel

M. Elowitz, CaltechS. Leibler, RockefellerM.G. Surette, CalgaryH. Margalit, Jerusalem

Ron MiloAdi NatanShmoolik MangamShalev ItzkovitzNadav Kashtan

Alon ZaslaverErez DekelAvi MayoYaki Setty

Negative feedback loop with one transcription arm and one protein arm is a common

network motif across organisms

x

y

HSF1 HIF

iNOShsp70hsp90

NFB

IB

p53

Mdm2

Mammals

dnaKdnaJ

H Ime1

Ime2

smo

Ptc

Bacteria Yeast Fruit-Flies

Negative feedback in engineering uses fast control on

slow devices.

Heater

Thermostat

Powersupply

Temperature

slow

fastEngineers tune the feedback parameters to obtain rapid

and stable temperature control

Negative Feedback can show over-damping, damped or un-damped oscillations, depending on parameters

Negative Feedback can show over-damping, damped or un-damped oscillations, depending on parameters

Engineers usually prefer damped oscillations-fast response, not too much overshoot

Negative Feedback can show over-damping, damped or un-damped oscillations, depending on parameters

Engineers usually try to avoid parameters that give Undamped oscillations

real-time proteomics in single living cells for p53-mdm2 dynamics

pMT-I p53 CFP 0.2Kb 1.2Kb 0.7Kb

Hygro

pMdm2 Mdm2 YFP 3.5Kb 15Kb 0.7Kb

Neo

P53-CFP

Mdm2-YFP

p53-CFP and Mdm2-YFP fusion protein allow imaging of p53 and MDM2 protein in living cells G. Lahav

Undamped p53 pulses in individual cells following DNA damage

Lahav Nat. genetics 2004

‘Digital behavior’ in the p53-mdm2 feedback loop

Number of cells with multiple pulses increases with damage, but not the size/shape of pulse.

Analog design

Digital design

Dynamics of the p53-mdm2 loop in single living

cells, and the design principles

of biological feedback

Galit Lahav Uri Alon

Weizmann InstituteIsrael

overview• Introduction- negative feedback loop

motif

•System for real time proteomics of p53-mdm2 in living human cells

•Dynamics of p53-mdm2 at the single cell level

•Design principles of biological feedback

Negative feedback loop with one transcription arm and one protein arm is a common

network motif across organisms

x

y

HSF1 HIF

iNOShsp70hsp90

NFB

IB

p53

Mdm2

Mammals

dnaKdnaJ

H Ime1

Ime2

smo

Ptc

Bacteria Yeast Fruit-Flies

Negative feedback in engineering uses fast control on

slow devices.

Heater

Thermostat

Powersupply

Temperature

slow

fastEngineers tune the feedback parameters to obtain rapid

and stable temperature control

Negative Feedback can show over-damping, damped or un-damped oscillations, depending on parameters

Negative Feedback can show over-damping, damped or un-damped oscillations, depending on parameters

Engineers usually prefer damped oscillations-fast response, not too much overshoot

Negative Feedback can show over-damping, damped or un-damped oscillations, depending on parameters

Engineers usually try to avoid parameters that give Undamped oscillations

What about biological feedback control?

Kohn, Mol Biol Cell, 1999 Our model system- perhaps the best studied example in mammals, p53-mdm2 negative feedback loop

P53-CFP

Mdm2-YFP

Is the damage repairable?

Apoptosis

no

Cell cycle arrestG1/S G2/M

One cell death =

Protection of the whole organism

yes

DNA repair

Stress signals p53 MDM2

The p53 Network

Damped oscillations in p53-mdm2 and

NFB-IB negative feedback loops

Hoffman et al, Science, 2002

Measurements on populations!

Lev Bar-Or et al, PNAS, 2000

western

p53-mdm2

NFB-IB Damped oscillations

Electro Mobility Gel-Shift Assay

p53

MDM2

westernHrs after 0 1 2 3 4 5 6 7 8

p53 mdm2

System for real-time proteomics in single living cells for p53-mdm2 kinetics.

pMT-I p53 CFP 0.2Kb 1.2Kb 0.7Kb

Hygro

pMdm2 Mdm2 YFP 3.5Kb 15Kb 0.7Kb

Neo

P53-CFP

Mdm2-YFP

p53-CFP and Mdm2-YFP fusion protein allow imaging of p53 and MDM2 protein in living cells

p53-CFP shows damped oscillations in western following DNA damage

similar to endogenous p53.

p53-CFP

p53

0 ½1 2 3 4 5 6 7 8 9 Hours after

0

10

20

30

40

50

60

70

80

90

100

0 2 4 6 8 10

p53

p53-CFP

Mcf7 p53+/+

Mdm2-YFP kinetics are similar to endogenous

Mdm2.

Mdm2-YFPMdm2

0 ½1 2 3 4 5 6 7 8 9 Hours after Irradiation

0

20

40

60

80

100

120

0 2 4 6 8 10

Mdm2

Mdm2-YFP

Mcf7 p53+/+ Mdm2+/+

Mdm2-YFP appears to undergo the same decreaseas endogenous mdm2 at early times

p53-CFP and Mdm2-YFP expression after DNA damage in living human cells.

p53 pulses in the nucleus

Mdm2 pulses follow p53 pulses in the nucleus

Different behavior of individual cells following DNA damage

Different behavior of individual cells following DNA damage

Different behavior of individual cells following DNA damage

Different behavior of individual cells following DNA damage

Different behavior of individual cells following DNA damage

Two groups of behavior: one peak and two peaks, with small

variations inside each group

Peak Width=350±60min (SD)1st Peak Width=360±50min (SD)

Ym

ax1

Ym

ax2

Ymax2/Ymax1=1.1±0.2 (SD)

T(Ymax2)-T(Ymax1)=370±50min (SD)

2nd Peak Width=340±40min (SD)

Average of single cell data is similar to Western: damped oscillations are an artifact of mixing two behavior

classes

The irradiation level determines the distribution of

cells undergoing 0,1 or 2 oscillations

2.5Gy

Numbers of pulses0 1 2+

%of

cells

100

80

60

40

20

0

The irradiation level determines the distribution of

cells undergoing 0,1 or 2 oscillations

2.5Gy

10Gy

Numbers of pulses0 1 2+

%of

cells

100

80

60

40

20

0

The irradiation level determines the distribution of

cells undergoing 0,1 or 2 oscillations

2.5Gy

0.3Gy

Numbers of pulses0 1 2+

%of

cells

100

80

60

40

20

0

10Gy

The irradiation level determines the distribution of

cells undergoing 0,1 or 2 oscillations

5Gy

10Gy

2.5Gy

0.3Gy

0Gy

Numbers of pulses0 1 2+

%of

cells

100

80

60

40

20

0

Mean pulse widths do not depend on DNA damage level

0.3Gy 2.5Gy 5Gy 10Gy

First pulse width

2.5Gy 5Gy 10Gy

Second pulse width

Mean pulse height does not depend on DNA damage level

0.3Gy 2.5Gy 5Gy 10Gy

First pulse height

2.5Gy 5Gy 10Gy

Second pulse height

‘Digital behavior’ in the p53-mdm2 feedback loop

Number of cells with multiple pulses, not the size/shape of the pulse, increases with damage.

Analog design

Digital design

tStyaDDdt

tdS

tyαtydt

tdy

txβtydt

tdy

txtyαtxSSdt

tdx

βtxtyαtxSSdt

tdx

ss

yy

yy

x

20

0

00

20

10

*

**

Show limit cycle figure

Individual cells show large amplitude variation

But peak timing is more precise

What's next?

• What is the mechanism for the ‘digital behavior’ in the p53-Mdm2 loop?

• What about other negative feedback loops…do they also display ‘digital behavior’?

• Could different numbers of pulses convey different 'meanings' to the cell in terms of expression of downstream genes?

The goal: dynamic proteomics in individual living human cells

• Protein concentration and location at high temporal-resolution and accuracy

• In individual living human cells

• For most of the expressed proteins

Annotated library of cells each of whichExpressed a chromosomally YFP-tagged protein

+automated microscopy and image-analysis

Exon tagging a gene with a fluorescent protein

exon1 exon2 exon3 exon4intron1 intron2

intron3

DNA

transcription

exon1 exon2 exon3RNASA SD SA SD SA SD SA SD

YFP puro

splicing

mRNA exon1exon2 YFP exon3 exon4

Proteintranslation

YFPN C

SA SDYFP puro

clone

RACE stage

4c18 1d13 1d224d8 1e18 1f21

1 11111 2 22 2 22

Clone 1e18YFP

Clone 1e18

YFP

546

Added seqYFP AAAAAAAAAAExon n Exon n+1

472

Added seqAAAAAAAAAAExon n+1

A

C

B

Ribosomal protein L5

Tagged protein identified using RACE

A wide variety of localization patterns

0 min 10 min

20 min 30 min

Dynamic proteomics in individual cells

Summary

• Network motifs and their biological function

• Digital pulses of p53

• Dynamic proteomics in individual living cells

Acknowledgments

Galit Lahav

Alex SigalNaama GevaLior Ben-ArziNitzan RosenfeldShiraz KalirInbal Alaluf

Weizmann Institute, Israel

Zvi KamBenjamin GeigerMoshe OrenVarda Rotter

Stan Leibler, Rockefeller -coliMike Surette, Calgary -coliArnold Levine, IAS Princeton –p53Michael Elowitz, Caltech –p53

Ron MiloAdi NatanShmoolik MangamShalev ItzkovitzNadav KashtanIdo Zelman

Alon ZaslaverShai Shen-OrrErez DekelAvi MayoYaki Setty

The heat-shock single-input-module

k1

1

k2

2

k3

3

n

kn

n

dnaKJ groEL htpG lond

32

SINGLE INPUT MODULE NETWORK MOTIF

Heat shock promoters are activated after ethanol step, and then adapt

4% ethanolAdded at t=0 , Natan et al 2004

Heat shock promoters are turned ON together and OFF in a temporal

order

Natan et al 2004

Heat-shock promoters are turned OFF in temporal order

Normalized Promoter Activity

Time from ethanol shock [minutes]54 60 66 72 78 84 90 96

groEL

ftsJ

htpG

hslUV

htpX

Lon

clpB

clpP

dnaK 0

0.1

0.2

0.3

0.4

0.5

Temporal turn-off order ~3 minutes difference

Natan et al 2004

Turn-off order matches severity of repair

Same principle found in SOS DNA repair [Ronen, Shraiman, Alon PNAS 2002]