Transcriptional Regulatory Networks

1
Transcriptional Regulatory Networks
ANSCI 490 M Instructor Lei Liu, PhD November
12, 2002 Daniel H. Barnett Department of Cell
and Structural Biology University of Illinois at
Urbana-Champaign
2
Biological Networks Coming of age?
October 25, 2002, Friday Gains in Understanding
Human Cells By NICHOLAS WADE (NYT) 993 words
Late Edition - Final, Section A, Page 18, Column
4 ABSTRACT - Scientists at Whitehead Institute
in Cambridge, Mass, have made significant stride
toward understanding how a living cell's
operations are controlled by information in its
genome insight, which gives detailed view of
complex, computer-like biological circuitry,
should help researchers understand cellular
programming errors that underlie diseases study
was made possible by several recent advances in
technology, such as DNA decoding machines
findings are reported in journal Science.
3
Background Transcriptional Networks

• How to cells coordinately control routine and
  diverse processes such as cell cycle,
  development, and metabolism?
• How do cells coordinately control routine
  processes AND properly respond to environmental
  stimuli?
• If gene expression is ultimately modulated by
  transcriptional regulators, then
  what regulates the regulators?
• Transcriptional Regulatory Networks
• Previous work has focused on global measurement
  of mRNA expression as an output of regulatory
  networks
• - reverse engineering by Singular Value
  Decomposition (SVD) to form nodes possibly link
  to transcriptional regulators
• - use of prior knowledge of regulatory network
  composition or architecture
• Is there a more direct way to test the regulation
  of gene expression by transcription factors
  organize them in a meaningful way?

4
Background Transcriptional Networks
Wyrick and Young, 2002
5
Background Models and Techniques

• Saccharomyces cerevisiae
• - or the functional genomics workhorse
• 1st eukaryote to have entire genome sequenced
• 200 proteins which regulate transcription of
  6200 genes Yeast Proteome Database
• Tremendous amount known about mechanisms of
  action of transcriptional regulators (e.g. Gal4)
• Genome-wide Location Analysis makes it possible
  to couple DNA-protein interactions with gene
  expression analysis to monitor coordinated gene
  regulation at whole-genome level
• Ren, B., F. Robert, et al. (2000). "Genome-wide
  location and function of DNA binding proteins."
  Science 290(5500) 2306-9.
• Simon, I., J. Barnett, et al. (2001). "Serial
  regulation of transcriptional regulators in the
  yeast cell cycle." Cell 106(6) 697-708.

Wyrick and Young, 2002
6
Background Genome-wide Location Analysis
Wyrick and Young, 2002
7
Transcriptional regulatory networks in
Saccharomyces cerevisiae

• Science 2002 Oct 25298(5594)799-804
• Lee TI, Rinaldi NJ, Robert F, Odom DT, Bar-Joseph
  Z, Gerber GK, Hannett NM, Harbison CT, Thompson
  CM, Simon I, Zeitlinger J, Jennings EG, Murray
  HL, Gordon DB, Ren B, Wyrick JJ, Tagne JB,
  Volkert TL, Fraenkel E, Gifford DK, Young RA.
• web.wi.mit.edu/young.regulator_network -
  Supporting website
• Features of note
• Effectively coupled Genome-wide Location
  Analysis with genome-wide expression analysis in
  model eukaryote Saccharomyces cerevisiae
• Uncovered network motifs which underlie
  regulatory capacities in entire genome
• Developed an automated process which was
  successful in building large network structures
  de novo by combining genome-wide location
  analysis with genome-wide expression analysis
  data without prior knowledge of regulator
  functions
• By use of this process, connections of cellular
  networks were noted to permit coordination of
  functions within cell which had been eluded to,
  but difficult to prove
• Provides a template for developing similar
  models of transcriptional regulatory circuits
  which will be helpful in understanding complex
  systems and how they are regulated.

8
Genome-wide Location Analysis (Figure One)
Attempted to examine all 141 known transcription
factors in Yeast Proteome Database (-) 17
proteins without viable myc tags (-) 18 tagged
but not expressed proteins 106 viable tagged
strains for Genome-wide Location Analysis
9
Analysis Determination of Binding Sites

• Visual examination of scans distribution of
  scatter plots about 45 deg.
• Computed by SD of log ratio
• Rank of all chips by SD, from low to high
• Avg. of ranks of each chip comprising an
  experiment were used as a score for experiment
• Below 300 good 300-350 acceptable
  above 350 poor

- Background-subtracted intensity values - each
spot yields fluorescence intensity information in
two channels (immunoprecipitated DNA and genomic
DNA). - Background hybridization to slides
accounted for by subtraction of the median
intensity of a set of control blank spots. -
Different amounts of genomic and
immunoprecipitated DNA hybridized to the chip
corrected median IP-enriched DNA channel /
median genomic DNA channel gt applied to each
genomic DNA channel. - Determine log of
(IP-enriched channel genomic DNA channel) for
each intergenic region across the entire set of
hybridization experiments. - Systematic bias
accounted for by normalizing the log ratios for a
specific intergenic by subtracting the average
log ratio for that intergenic region. - A whole
chip error model (Hughes et al 2000 Cell) was
used to calculate confidence values (p-values)
for every spot and to combine data for the
replicates of each experiment to obtain a final
average ratio and confidence for each intergenic
region.
10
Analysis Effect of p-value Cutoff
Rather than use bound vs. unbound criteria,
confidence measures (p-values) used. - Inherent
noise from microarray data - DNA binding
proteins are in equilibrium between bound and
unbound states Which p-value to use?? More
stringent p-values reduce the number of
interactions observed, but decrease the
likelihood of false positive results. Generally
used a p-value threshold of 0.001 to analyze,
discuss and generate regulatory models -
minimizes false positive results, but allows an
increase in false negative results.
11
Analysis Confirmation of Predicted Binding

• Experimental Confirmation Quantifying False
  PositivesConventional, gene-specific ChIP
  experiments confirmed 89 of 95 binding
  interactions (involving 28 different regulators)
  that were identified by location analysis data at
  a threshold p-value of 0.001.
• This suggests that empirical rate of false
  positives is 6.
• Quantifying False NegativesThe 0.001 p-value
  threshold may result in an underestimate of
  regulator-DNA interactions.
• The determination of a true false negative rate
  was not feasible, but gene-specific PCR analysis
  with selected regulators was used to test the
  results predicted at each of the different
  p-value thresholds.
• Nrg1 and Stb1 genes with p-values closest to
  one of four thresholds (0.001, 0.005,0.01,0.05)
  and performed chromatin IP and gene-specific PCR
  - at least 2,300 additional genuine
  regulator-gene interactions exist among our
  results at all p-values above 0.001.
• Computational Estimation of False-Positive Rate
• Estimate a false positive rate by determining
  the number of spots below our p-value threshold
  of 0.001 in these control DNA vs. control DNA
  arrays - 1,000 groups of three arrays by
  randomly selecting six sets of measurements from
  the control DNA arrays, 3/ fluorophore.
• Results indicate that an average of 3.7 out of
  6500 intergenic regions significantly enriched
  using the 0.001 threshold actual experiments is
  38, so from this we estimate an avg. false
  positive rate of 10.
• Literature Confirmation of the Data
• Authors found that the location data generally
  agree with the published literature.
• No accurate estimate false-positive rate
  because the literature is incomplete.
• Some regulator-gene interactions in literature
  not observed, indicating that some interactions
  are not reported by the location data at a
  p-value threshold of 0.001.

12
Promoter-Regulator Interactions (Figure Two)
Nearly 4000 regulator-promoter interactions at plt
0.001. The promoter regions of 2343 of 6270
yeast genes (37) bound by one or more of the 106
transcriptional regulators.
Location Data
Chance
13
Promoter-Regulator Interactions (Figure Two) Cont.
The number of different promoter regions bound by
each regulator ranged from0 to 181 (0.001
p-value) avg. 38 promoter regions per
regulator
14
Network Motifs (Figure Three)
Simplest units of commonly used network
architecture (network motifs) - provide
specific regulatory capacities such as positive
and negative feedback loops. Motifs can be
assembled into network structures that help
explain how a complex gene expression program is
regulated. Six different regulatory network
motifs identified in yeast by G-WLA.
15
Network Motif Search Algorithms
The overall matrix D consists of binary entries
Dij, where a 1 indicates binding of regulator j
to intergenic region i with a p-value of less
than or equal to 0.001, a 0 indicates a p-value
greater than 0.001. The regulator matrix R is a
subset of D, containing only the rows
corresponding to the intergenic region assigned
to each regulator, in the same order as the
columns of regulators. - Autoregulatory motif
Find each non-zero entry on the diagonal of R. -
Feedforward loop For each master regulator
(column of R), find non-zero entries, which
correspond to regulators bound. For each master
regulator / secondary regulator pair, find all
rows in D bound by both regulators. -
Multi-component loop For each regulator (column
of R), find the regulators to which it binds. For
each of these, find the regulators it binds. If
any of these are the original regulator, you have
a multi-component loop of two. For all others,
find regulators to which they bind. If any of
these are the original, you have a
multi-component loop of three. Repeat to find
larger loops. - Single input module Find the
intergenic regions bound by only one regulator.
That is, take the subset of rows of D such that
the sum of each row is 1. Then for each regulator
(column), find non-zero entries. Each set
(greater than three intergenic regions) is a
SIM. - Multi-input module Find the intergenic
regions bound by more than one regulator. That
is, take the subset of rows of D such that the
sum of each row is greater than 1. Then, for each
row, find any other row bound by the same
regulators. The collection of rows bound by the
same regulators correspond to a MIM. Once a row
is assigned to a MIM, remove it from further
analysis. - Regulator cascade For each
regulator (column of R), use a recursive
algorithm to find chains of all lengths. That is,
for each regulator whose promoter is bound by the
regulator before it in the chain, find the
regulator promoters to which it binds. Repeat
until the chain ends. There are three possible
ways to end a chain a regulator that does not
bind to the promoter of any other regulator, a
regulator that binds to its own promoter, or one
that binds to the promoter of another regulator
earlier in the chain.
16
Network Super Structure Assembly
Regulatory motif refinement Algorithm was
developed to explore all the genome-wide location
data together with the expression data from over
500 expression experiments to identify groups of
genes that are both coordinately bound and
coordinately expressed. The algorithm begins by
defining a set of genes, G, that are bound by a
set of regulators S, using the 0.001 p-value
threshold. A large subset of genes in G are
similarly expressed over the entire set of
expression data, and use those genes to establish
a core expression profile. Genes are then dropped
from G if their expression profile is
significantly different from this core profile.
The remainder of the genome is scanned for genes
with expression profiles that are similar to the
core profile. Genes with a significant match in
expression profiles are then examined to see if
the set of regulators S are bound. At this step,
the probability of a gene being bound by the set
of regulators is used, rather than the individual
probabilities of that gene being bound by each of
the individual regulators. Because assaying the
combined probability of the set of regulators
being bound, and relying on similarity of
expression patterns, the p-value can be relaxed
for individual binding events and thus recapture
information that is lost due to the use of an
arbitrary p-value threshold. The process is
repeated until all combinations of genes bound by
regulators have been considered. The resulting
sets of regulators and genes are essentially
multi-input motifs refined for common expression
(MIM-CE).
17
Assembly of Motifs into Network Structures
Assembling network structure The refined motifs
were used to construct a network structure for
the yeast cell cycle using an automatic process
that requires no prior knowledge of the
regulators that control transcription during the
cell cycle. Cell Cycle - Extensive genome-wide
expression data and literature to explore
features of model - use to determine if a
principled computational approach can reproduce
substantial portions of the simple network that
was previously modeled using a more directed
approach (Simon et al, 2001 Cell) determine
whether the computational approach would
construct the regulatory logic of cell cycle from
the location and expression data without previous
knowledge of the regulators involved. 11
regulators identified by using MIM-CEs
significantly enriched in genes whose expression
oscillates through the cell cycle. To construct
the cell cycle network, a new set of MIM-CEs was
generated using only the eleven regulators and
the cell cycle expression data. This two-step
procedure is a general method for constructing
other regulatory networks. To produce a cell
cycle transcriptional regulatory network model,
the MIM-CEs were aligned around the cell cycle
based on the peak expression of the genes in the
group using an algorithm described previously
(Bar-Joseph et al., 2002). Since MIM-CEs contain
genes that are co-expressed, the expression data
was used to instruct the assembly of the network
to represent this temporal process.
18
Yeast Cell Cycle Model Transcriptional
Regulatory Network (Figure Four)
19
Network of Regulator-Regulator Relationships
(Figure Five)
20
Network of Regulator-Regulator Relationships
(Figure Five) Cont.
21
Network of Regulator-Regulator Relationships
(Figure Five) Cont.
22
Network of Regulator-Regulator Relationships
(Figure Five) Cont.
23
Coordination of Cellular Processes
Coordination of gene expression programs is
likely to be particularly important for
coordinating fundamental cellular processes. -
Regulators bind genes encoding regulators within
same category (e.g. cell cycle). Cell cycle
regulators bound to other cell cycle regulators
(Simon et al 2002), and this phenomenon was also
apparent among transcriptional regulators that
fall into the metabolism and environmental
response categories. - Multiple regulators
bind promoters for genes which regulate other
cell processes. Multiple transcriptional
regulators within each category bind to genes
encoding regulators that are responsible for
control of other cellular processes. These
observations are likely to explain, in part, how
cells coordinate transcriptional regulation of
the cell cycle with other cellular processes.
These connections are generally consistent with
previous experimental information regarding the
relationships between cellular processes. The
control of most, if not all, cellular processes
is characterized by networks of transcriptional
regulators that regulate other regulators. It is
also evident that the effects of transcriptional
regulator mutations on global gene expression as
measured by expression profiling the direct
targets of a single regulator.
24
Conclusions, revisited

• Effectively coupled Genome-wide Location
  Analysis with genome-wide expression analysis in
  model eukaryote Saccharomyces cerevisiae
• Uncovered network motifs which underlie
  regulatory capacities in entire genome
• Developed an automated process which was
  successful in building large network structures
  de novo by combining genome-wide location
  analysis with genome-wide expression analysis
  data without prior knowledge of regulator
  functions
• By use of this process, connections of cellular
  networks were noted to permit coordination of
  functions within cell which had been eluded to,
  but difficult to prove
• Provides a template for developing similar
  models of transcriptional regulatory circuits
  which will be helpful in understanding complex
  systems and how they are regulated.

25
References
1.     Lee, T. I., N. J. Rinaldi, et al. (2002).
"Transcriptional regulatory networks in
Saccharomyces cerevisiae." Science 298(5594)
799-804. 2.     Ren, B., F. Robert, et al.
(2000). "Genome-wide location and function of DNA
binding proteins." Science 290(5500)
2306-9. 3.     Ren, B., H. Cam, et al. (2002).
"E2F integrates cell cycle progression with DNA
repair, replication, and G(2)/M checkpoints."
Genes Dev 16(2) 245-56. 4.     Simon, I., J.
Barnett, et al. (2001). "Serial regulation of
transcriptional regulators in the yeast cell
cycle." Cell 106(6) 697-708. 5.     Wyrick, J.
J. and R. A. Young (2002). "Deciphering gene
expression regulatory networks." Curr Opin Genet
Dev 12(2) 130-6.
26
Background Basic Example of Transcription
Factor Association
Lee and Kraus, 2001
27
ChIP
- Chromatin immunoprecipitation assay (ChIP)
Lee and Kraus, 2001
28
Background Combining G-WL Analysis and
Traditional Expression Analysis in Physiological
Models
Wyrick and Young, 2002
29
Lee TI et al 2002 Figure Three