Saturday 09/14/13

Zenklusen

Intro

• use nascent mRNA to measure initiation frequency

live imaging

• short memory of how many transcripts initiated
• MS2 RNA labeling — look at intensity fluctuations
• initiations are uncorrelated. variable elongation rates coordinated with cell cycle (faster in G2/M)
• MS2 labeled RNAs are not translated. Good Tx reporter only.

Ms2 bind as dimer, add PP7 RFP system. (24 rtps. 12 rpts. still see single transcripts). RNAs still get degraded.

• proteins tend to aggregate. mean mRAN increases over time.
• mRNA moving too fast for tracking (measure degradation).

antisense RNA mediated transcriptional gene silencing of Pho84

• antisense transcription of gene clean back through the promoter.
• delete rrp6 nuclease, get exclusively antisense transcription
• mixed sense/antisense is it gradual or switch like. (=switch like, never co-expressed)
• antisense stays at site of nascent transcription, goes through promoter nucleosomes.
• infrequent initiation, 1.4 mRNAs/hour = mixed pop. 3.1 mRNAs/hour kep it off entirely. (knocking of modified histones)?

pombe house keeping gene

• pombe more time in G2
• less Poisson like expression.

other comments

• requirements for nascent expression: long gene (more transcripts on the gene at once).

Raj

intro

• PhD in Math, post-doc in bio
• what does it mean for a gene to be “on”
• fully induced genes are not always on.

does position in the nucleus affect whether gene is on or off?

• Hi-C
• why interesting? non-local cis-regulation. Beyond the 100 kb scale
• use intronic labeling
• look at bunch of introns along the gene, get ‘shape’ of gene
• 20 genes across the 60 Mb of chromosome 19.
• per chromosome transcriptional profile.
• most genes have no-pairwise correlations.
• 1 pair (14 Mb away) strongly anti-correlated. Inter-chromosomal interaction not anti-correlated.
• true in HeLa and primary fibroblasts.
• chromosome structure (on the multi-Mb scale) does not substantially affect gene expression.
• translocated fragment of chromosome 19 is up-regulated.

variation in cells and volume

• mRNA abundence for house keeping genes scales strongly with cell size.
• differences between quiescent, cycling, senescent cells volume driven.
• nuclear volume doesn’t correlate so much with cell size.
• is this why you put the degradation machinery on the membrane?
• degradation rate is the same in large and small cells.
• not sensing each gene — nascent transcription not up in siRNA
• ribosome count to polII count, to nuclear volume.
• fuse small cell to big cell, see if up-regulation.
• some genes don’t correlate with volume.

questions

• if this scaling is true for many genes, it is true for PolII (more PolII mRNA, more ribosomes)
• question for Johan, how does volume scaling actually work in the prokaryote? the concentration of the activators is the same, so the initation rate is the same, which means the number per unit time is the same, (i.e. not increasing with volume).

Elizabeth Read

• stochasticity in host-pathogen interactions
• phys chem. chem E

intro

• HIV progression: sharp peak in viral load, drops to low level after immune CD8+ lympohcytes kick up. Median time of 10 years of HIV positive before onset of symptomatic AIDS.
• dynamic equilibrium of asymptomatic phase: immune and viral turn-over.
• wide person-person variability
• rare individuals who can control HIV, viral load stays below threshold for transmission or progression (<.1%?)
• 50% of elite controllers have allele E, but having allele E only 2% prob of being elite controller.

model

• Resting and active T cells

compensatory mutations escape immune pressure

• escape immune, recover viral function,
• fitness landscape have to decrease before going up.
• in stochastic case it much harder to get double mutant (easier to go extinct in discrete model).

bistability

• deterministic system in this case is not bistable
• purely stochastic bistability, (e.g. from the existence of absorbing states see Artyomov or Samoilov)
• reduced network has a different steady state the full network.
• questions: mathematically bistable or just atime scale separation relative to lifetime? Early phase is bimodal.

summary

• stochastic networks can have qualitatively different behaviors even not in the small number limit.
• next steps: are there early predictors of viral escape?
• important role in the initial T-cell response for determining the sequence space the virus gets to explore.

Steve Abel

• signal transduction
• chem E U tenessee / Chakratoty at MIT (collaborate Groves).

Intro

• T cell signalling
• role of membrane in signal transduction
• Gillespie simulation on lattice
• slab geometry, vary thickness (effect of cytoplasm vs. membrane).

bistability

• driven by sequestration of enzyme. Substrate proteins sequester kinases.
• excess of available phosophotases will always reconvert
• effected by concentration and geometry.
• flattened, same concentration system, on average molecules are further apart, less sharp separation
• slow diffusion decreases effecive kinetic parameters, suppresses bistability (smaller fraction of bound pairs, less sequestration effect).
• rebinding suppresses bi-stability, (effective conc higher). Less bistability in 2D (more likely to rebind in 2D than 3D).

SOS allosteric control

• converts RAS-GDP to RAS-GTP at higher rate when bound by RAS-GTP
• SOS will allosterically switch between different catalytic rates. stochastically sample rates from distribution of rates.
• Allosteric regulation changes rate of sampling catalytic rates?
• expect effect at intermediate lifetimes of different kinetic rates
• good or threshold crossing.

Steven Altschuler

• “a scientist”

introduction

• look at complex signaling network, characterize its behavior (i.e. a switch, an oscillator etc. e.g. Ultrasensitivity in sginaling cascades)
• Modeling not such a great thing to do when the number of parameters gets large.
• we don’t even understand what the arrows mean in these signaling networks
• focus on the “reverse problem” which parts of the network are important for behavior you are interested in.
• approach: network gives some distribution of phenotype. Change network, shift distribution of phenotypes.
• use natural cell-cell variation to learn something about the network.
• Model system: neutrophil bacteria tracking. how does the front remain the front, keep attention?
• Fundamentally different cytoskeletal elements and chemical difference.
• expect mutual inhibition. need proportional balance? Stronger front need stronger back.

experimental setup

• can’t use traditional in neutrophil cell culture — too variable. Have to use fresh human neutrophils
• activate mobility with fMLP. cells polarize in an hour.
• measure actin for front, myosin for back, quantify intensity and localization.
• more or less flat relationship between back intensity vs. front intensity
• back polarity is robust — doens’t depend on front intensity

regulation of back buffering

• microtubules start to depolarize get an anticorrelated trend between front and back identity (also a broader distribution).
• affect localization of back/myosin signal but not the overall activity.
• compare effect of interating front intensity and having that promote back activity

question

• what are the consequences for attention tracking of the back buffering
• give a spatial signal is it different

Pieter Rein en Wolde

• cells are spatially heterogeneous at the molecule scale
• can spatial heterogeneity / clustering help cells compute?
• characteristic membrane clusters of 1-10 molecules (Raf signaling clustered, CheA E coli clustered).
• increasing cluster size decreases output in single step system (target size effect). (scales as root N).
• in two step system it increases the output of the system. Rebinding has local increased concentration.
• Need to have competing phosphatases.
• Non-monotic dependence on diffusion constant.
• partitioning linearizes response, reduces noise: larger fluctuations per partition, average out.
• particle exchange between clusters further reduces information transfer (allows molecules to be incorrectly trapped away from eachother).
• partitioning removes correlations but also isolates molecules. Can balance these effects to enhance information transfer. Partition as much as you can without making molecules isolated.
• provides an additional argument for scaffold based interactions.

Jeff

• M responds to CTMM
• alpha irreversibility of binding (0 for irriversbile)
• NL number of binding events, Ns number of other events.

Michal Komorowski

• sensitivity analysis and information theory
• information is related to a factor that changes a state of a system. Related to our sensitivity to a system
• Fisher information is a measure of sensitivity response. Fisher info = E[d log (X|Y)/dy]^2
• compare to mutual information
• Fisher information approach maximal information transmission (in bits) of system for optimal input, (whereas mutual information is for a particular given input).
• high number protein molecules, negligable experimental noise 7-9 bits.
• can decompose the variance from each reaction in the model (similar to our PLoS CB sensitivity analysis?)
• in an arbitrary system, degradation of an output contributes half othe outputs FanoFactor.

Mustafa Khammash

• control theory / engineering
• goal, manipulate the dynamic behavior of living cells using real time dynamic feedback control

System components:

• Model choice: random time change model (Markov model) or Moment dynamics (dE[m]/dt)
• GFP as readout with flow-cytometery
• Light as an actuator input. (build culture control feedback system in 3D printer)
• Model doesn’t have to be accurate description to e used for control applications.
• different setpoints (targeted levels of gene expression). Reached by giving light signal activation to induce transcription.
• open loop with no feedback steadily drifts away from setpoint. With feedback can correct.
• in vivo control? — cells are automonous don’t need to interact with optigenetics etc.

simple model: proporitional negative feedback

• theoretical differences between in vivo and in silico control (respond to population average or individual level). in silico system has only one input to give back, in vivo system doesn’t have population information, which is better controller? in vivo. In silico controller can’t use the variance, best thing to do is purely given by the mean.
• with two inputs (control transcription rate and degradation rate), can tune variance.

Johan Paulsson

History

• mRNA and protein birth death process. CV^2 1/>p> + 1/t1/t1_t2 ~ 1+b/

<

p> (negative binomial) similar results for mRNA with ON OFF gnees
* warning story on lynx and snowshoe hare (hare oscillates without the lynx).
* burst analysis can be misleading — might be bursty and look Poisson nd vice versa.

approach

• what the system can NOT do, based on what we know about it.
• technical challenge, without specifying system can not use many forms of approximation.

results

• variation being unexplained by the environment doesn’t mean explained by the system. So what you call as intrinsic noise experimentally might not be what you call intrinsic noise in the model.
• CV^2 vs mean can have lots of options. CV^2 – 1/mean = const = cov(x1,x2)/
• average life time is just average number of molecules / average steady state birth rate (or decay rate)
• apply to Sunny’s bacterial data.
• no model which has assumption of mRNA and protein being produced can explain the low lack of correlation between mRNA and protein levels.
• transcription noise alone can not explain the data
• model a experimental noise.

Discussion on major challenges moving forward

• Goal discuss major challenges moving forward in the field of stochastic biochemical reaction networks. First step clarify what are goals.
• what is the objective of measuring variability in cells in the first place?
  1. document its existence, here is a phenomenon that is apparently prevelant enough it should not be ignored.
  2. fundamental biological behavior is different because the system is stochastic. [Elizabeth examples of stochastic bistability
• what is molecular noise?
  1. Deviation from Poisson. simple stochastic model and can show this is not the case.
  2. same experiment, different result
• What is the role of modeling?
  1. idea generation
  2. form of clear communication.
  3. Jeff, Narenda eliminate biological models that seem very intuitive.
• extrinsic sources of noise (keeping track of volume [Arjun], cell cycle [Narendra], recent transcription history [Daniel] )
• Is the answer more live imaging (follow cell cycle, have individual cell history). Need less perpturbative live imaging?
• molecules or concentration

Hyun

• experimentalist physicist
• Synthetic multicellularity ‘cell circuits’
• simultaneous secretion and sensing: quorum sensing or autocrine signaling? (depends on relative receptor levels and affinities)

experiment setup

• test by hijacking yeast mating pathway. cell can bind alpha factor from self or others. will give more GFP. can tune secretion of alpha factor with dox
• also have sense only cells (marked with RFP)
• at low density mostly sense self (sense only cells get little signal).
• high density high secretion rate both strains respond the same.
• add positive feedback (alpha factor signaling directly triggers more alpha factor secretion.
• strong positive feedback low density OR weak feedback high density, all turn ON.
• positive feedback + signal degradation get bimodal population.
• self assembly of higher order structures based on adhesion proteins.

Bastian Drees

• Biophysical taxonomy of quorum sensing architectures

types of architectures

• Basler gram negative, freely diffusive signal
• B subtilis S aureus quorum sensing molecules mediated by transport proteins.
• Receptor location inside or outside. Active or passive transport. Modified or unmodified signaling peptides

Results

• modification determines complexity / number of species
• encoding characteristics: sensitivity and add Poisson production noise in molecule numbers and add extrinsic noise by hand “to match experimental estimates”
• classes of results: non functional (don’t important, measure internal). Band-pass: sense intracellular and produced intracellular
• sense extracellular and no import implies linear density sensing
• complex architectures allow ultra-sensitive response to cell density (including negative)
• total molecule concentration does not generally increase with cell density (though only with modifications can you get a system where they decrease)
• V fischeri, Ecoli band pass
• others have negatie senstivity, ultra-sensitivity, linear response.
• some observed concentration responses not predicted by model, must have additional feedback mechanisms.

Pankaj Mehta

• collaboration with Thomas Gregor: Tryo esler and Allsyon Sgro.
• theoretical physicist in defense of phenomenology
• complex network, near a Hopf bifurcation. ‘Universality’

Fitz-Hugh Nagumo model explains/predicts many aspects of cAMP signaling.

• adiabatic ramp, no spike, jump to level should get spike (experiment works)

noise is essential for collective behavior

• coupled Fitz-Hugh Nagumo
• noise gives large scale excursions for single cells. One cell firing triggers group.
• combined with fold change sensing.

Sunday

Jane Knodev

• few mm of DNA in yeast, somehow folded
• students do experiments, Jane does theory
• what are the rules for folding? (for later). Does folded state affect function?
• yeast involves very simple rules of folding — random walk polymers. Probability of contact as a function of distance scales between -1.5 and -1.
• polymer linker length = Lk~100nm N steps separation <r^ = N Lk^2. length scales with L^1/2, L^3/2 for volume contact

data

• lacO repeats + centromere marker.

organization

• telomeres attached to nuclear perifery at random
• all chromosomes thethered at centromere by SPB (16 chromosome). Tethering points at middle and ends, random walk chromatin in between.
• nucleolus (ribosome RNA) is volume excluded region.
• only free parameter is 200 nm (z-precision is 100nm). Can you do astigmatism?
• 10 nm fiber — “definetely no 30 nm fiber in a yeast).
• can’t have fixed telomere location.
• Effect of untethering — see expected tightening of distribution of inter-spot distance.
• why yeast are unique random walk? Human chromosomes too large to equilibrate.
• diffusion scaling (.5 instead of 1 for MSD um^2) or random polymer.

mating switching

• Program cut in the MAT locus (specifies mating locus)
• homologous recombination with silenced genes on either side of the mating locus
• 90% of the time use the new one (which is further away) 200 kb aay
• should be some tether 20 kb away. Tether needs to appear after the break.
• recombinant enhancer “targeting element — binds something near double stranded break
• measure distance between loci.
• can estimate the ‘stickiness’ of the data.
• can you estimate the time for the leash to shorten and how does this compete with the shorter distance.

Zimmer

• deterministic inference for stochastic models
• deterministic model can not independently estimate birth and degradation rate from observations at steady state. But with stochastic data can ID a qualitative difference
• can we use this for improved parameter estimation.

setup

• observed and unobserved species. Use likelihood function
• chemical master equation of states
• complete system of with of order 12,000 protein copy molecules becomes very computationally inhibitory
• Mean is approximated by the solution of the ODE on the last measurement. Assume variance is constant
• inaccuracy in estimating the variance only effects the uncertainty of parameter estimation, it does not bias the value.
• test with simulation studies. Additive Gaussian measurement noise, show uncertainty increases
• test partial observation also reasonable
• resolves identifiability problems (requires sufficient data).
• how robust is this to the structure of underlying model?

Georg Seelig

• ‘trained as a physist’. now a renaissance man

what is a miRNA?

• folds as hairpin, interacts with protein machinery as negatie regulatory of complementary RNA (6-7bp compliment)
• play a role in stress responses, providing robustness to gene expression.

setup

• RFP with intronic miRNA that targets its own UTR (incoherent feed-forward motif. Capable of biochemical adaptation and noise supression, See Belris et al MSB 2011 and Bosia et al BMC Sys Bio 2011).
• is this a good system for stable transgene expression?
• can we better understand these biological mechanisms
• mRNA does show perfect adaptation.
• is the adaptive behavior robust to competitive targets
• self targeting system has different initial responses but response level quickly converges to give the same output despite different strengths of input. —
• A nice way to get flat expression response without saturation