Transcript Noise in gene expression networks?

Bla
h ..
Shut up guys! I
can’t hear what
DNA it telling
me to do.
..
Bla
h ..
Bla
h ..
Rosenfeld et al.
Bla
h ..
Noise in gene expression
networks?
Ramu Anandakrishnan
March 14, 2006
1
My goal
1. Share with you what I’ve learnt
2. Get clarification and answers to
things I didn’t understand
3. Set the stage for my project
2
Agenda
1. Some basic background
2. Gene regulation at the single-cell level,
Rosenfeld et al.
3. Noise propagation in gene networks, Pedraza et
al. (time permitting)
4. Some final thoughts
3
Why is it important
Understanding what causes variations (noise) in gene expression
(protein production) can help prevent diseases .
• All biological organisms are essentially made up of
proteins and use proteins to function
• We are a finely tuned machines requiring the EXACT
right amount of the right type of protein, at the right
time
• Proteins are produced through the gene expression
process
• Slight variations or errors in the process can result in
disease or even death
4
Gene expression process
Gene expression in general is a very complex process with many
opportunities for variances from the optimal outcome
•
•
•
•
DNA to RNA Transcription
– Activator and Repressor proteins control affinity to Start site
– RNA polymerase attaches to Start site and locally cleaves DNA
– mRNA s formed by stringing together nucleoside triphosphates. The
reaction is catalyzed by pyrophosphatase
RNA processing (eukaryotic cells)
– 5’ cap added to mRNA to prevent enzymatic degradation
– Endonuclease cleaves 3’ end and poly(A) polymerase (a complex of
proteins) adds a poly(A) tail which specifies the correct number of A
residues to add (does not require a template)
– Noncoding “introns” are removed by splicing
Protein Synthesis
– Specific aminoacyl-tRNA synthetase attach a specific amino acid to
the transfer RNA (tRNA) which activates the tRNA
– Ribosomes, consisting of several different ribosomal RNA and more
than 50 proteins, binds to the mRNA and the tRNA to accelerate
protein synthesis
– Amino acids on the tRNA bind to form the polypeptide protein chain
and the tRNA is released
Protein Sorting and Secretion
–
–
Ribosome must carry mRNA to specific destinations
Protein moved to final destination, in some cases protected by transport
vesicles and Golgi complex ,
5
Controlled experiments
Biologists have devised ingenious ways of controlling and reporting the
gene expression process to get data that can be analyzed
•
Fluorescent Staining and Recombinant DNA
– Fuse gene for fluorescent protein (e.g. GFP) with gene for
the protein of interest creating a “chimeric” protein gene
– Insert chimeric gene DNA into the cell, which then produces
the chimeric protein
– The fluorescent dye lights up when the cell is illuminated
– Associated proteins can then be identified and counted
•
Transcription Control
– Inducer concentration
– Mutation to inactivate repressor production
– Operator site mutation to prevent repressor binding
– Promoter site mutation to prevent RNA polymerase from
initiating transcription
6
Research
Basic research is like shooting an arrow into
the air and, where it lands, painting a target.
-- Homer Burton Adkins (1892-1949, American
organic chemist)
I love fools' experiments; I am always making
them
-- Charles Darwin (1809-1882, British biologist)
7
Gene regulation at the single cell level, Rosenfeld et al.*
Temporal fluctuations due to extrinsic noise are large and long lasting,
so must be taken into account for realistic modeling
• Analyzes noise in the Gene Regulation Function:
Protein production rate = f (Transcription Factors)
• Variance separated into intrinsic and extrinsic noise
• Analyzes temporal aspect of noise
• λ-cascade strains of E. coli used to conduct
controlled experiments
• Experiments designed to permit measurement of
Rate of Production at an individual cell level instead
of steady state quantity or averages for multiple cells
* Gene regulation at the single-cell level, N Rosenfeld, JW Young, U Alon, PS
Swain, MB Elowitz
8
Experimental data
Conducted experiments designed to observe gene regulation in
individual cells over time
λ-cascade strain of E. coli
TetR+ background
represses YFP
Chromosomally integrated target
promoter (PR) controlling cyan
fluorescent protein
Induced by anhydrotetracycline
Yellow fluorescent
repressor fusion protein
9
aTc used to initially induce production of
cI-YFP, which then dilutes over time as
cells divide
Data Calibration
Individual cI-YFP molecules can not be directly detected due to cellular
autofluorescence, so it is indirectly estimated by fluorescence intensity
Number of copies of cI-YFP received by daughter cells follows a binomial distribution:
10
Theoretical model
The Hill function is often used to represent unknown regulation functions
Gene Regulation Function (Hill function)
Production Rate =
220
219.5
219
218.5
218
217.5
217
216.5
216
0
1
2
3
4
5
6
7
Matlab plot for PR
11
8
9
10
Variance
The Hill function does fit the mean values, however it’s parameters are
calculated from the data and the standard deviation is high (55%)
Standard deviation = 55%
12
Explanation of variance – intrinsic and extrinsic noise
Even after normalizing for cell cycle phase standard deviation is high (40%)
• Average CFP production rate for cells about to divide = 2x that of
newly divided cells
• Why?
– True for all proteins?
– Need to make two of everything so each of the daughter cells
will have enough to function?
– Cell grows twice as fast just before division?
• Production rates normalized to the average cell cycle phase. How?
• Normalized standard deviation = 40%
13
Explanation of variance – intrinsic and extrinsic noise
After separating out the effect of intrinsic noise (~20%), extrinsic noise
represents 35% of variance
• Intrinsic noise (~20%)
– Caused by randomness in biochemical reactions
– Identical copies of genes would show different production rates
– Measured by comparing expression of two cells with identical
levels of regulatory proteins
• Extrinsic noise (~35%)
– Resulting from fluctuations in levels of “cellular components such
as metabolites, ribosomes and polymerases”, but not in the
concentration of repressor
– Remaining variance after excluding intrinsic noise
– This must include variance due to cell cycle phase?
14
Temporal analysis of intrinsic and extrinsic noise
Extrinsic noise varies over cell cycle time scales
• Autocorrelation function
– Correlation between observations at time i
and i+m
– Rm = 1 indicates high correlated
• Intrinsic noise
– Decays rapidly, correlation time  < 10 min
• Extrinsic noise
– Correlation time  ~ 40 min, comparable to
cell cycle times of ~ 45 min
15
Conclusion
• Single-cell Gene Regulation Function can not be
expressed by a single-valued function because slow
extrinsic fluctuations give the cell a memory or
individuality lasting roughly one cell cycle
• This implies that any accurate cellular response in faster
time scales are likely to require a feedback loop
16
Statistics
On the Gaussian curve:
Experimentalists think that it is a mathematical theorem
while the mathematicians believe it to be an experimental
fact.
-- Jules Henri Poincare (1854-1912) [French mathematician]
17
Gene propagation in gene networks, Pedraza et al.*
Even though intrinsic noise may be low for each gene in the network,
transmitted noise may be high due to correlated sources of noise
• Studies propagation of noise from one gene to
another in a network
• Synthetic network of four genes used to conduct
controlled experiments
• Noise correlation model used to analytically calculate
noise transfer from one gene to the next
• Demonstrated that transmitted noise has a
logarithmic gain that depends on the interactions
between upstream and downstream genes
* Gene propagation in gene networks, JM Pedraza, A van Oudenaarden
18
Experimental Data
IPTG and ATC inducers are used to regulate the expression of Gene 1
and Gene 2.
Repressors
Inducers
19
Not part of cascade
Theoretical Model
The Langevin method can be used to analytically derive the correlation
function for intrinsic and global noise
20
Theoretical Model
The noise transfer gain, Hij is computed by fitting experimental data to
the correlation function derived using the Langevin method
21
Validation of the model
Parameters calculated by varying IPTG concentration, correctly
predicted the impact of varying ATC concentration
22
Conclusion
• Langevin method can be used to analytically predict
noise in gene networks
• Noise has a correlated global component due to
which fluctuation can be substantial despite low
intrinsic noise in all components
23
Noise
Noise \’noiz\ n 1: loud, confused or senseless shouting; 2 a:
sound that lacks agreeable musical quality or is noticeably
unpleasant; b: … ; c: an unwanted signal or disturbance
interfering with the operation of a device or system; d: …
Noise in the gene expression network:
Is this really noise or a limitation of our model?
Maybe it is part of the “design” to make sure that the
organism will die off over time and make way for the
new and improved next generation?
Does it matter or are we talking about semantics?
24
What happened to my hair?
1985
2006
Is it a result of accumulated errors from 20 years
of “noisy” gene expression?
25