Uri Alon
- Linear relations between quantitative traits.
- tradeoffs between tasks (e.g. teeth for meat or plants, toes for walking or grasping)
- metric decrease in distance space results in linear arrangement. (for the right choice of coordinate scale, [may be log]).
- multi-trait tradeoffs fall in convex hull of simple shape with specialities at vertices.
- look at all bacterial promoters driving GFP see where they lie in traitspace. Two principle components capture 80% of variance. PCs ‘enriched for stress genes’ or ‘ribosomal genes’ (data not shown).
- Under selection will a species evolve along the line? (testing it. maybe).
Tom Shimizu, FOM Institute AMOLF, Response rescaling in bacterial chemotaxis
- Adaptation
- Notes to self — don’t use bottom eight of screen, obscured by heads.
- Using FRET to measure internal state of chemotaxis system.
- Response threshold scales linearly with background concentration (ratio of input to background). Called Weber’s law (old observation).
- Mechanistic model (Ising model) MWC eising model, almost linear
- for dynamically changing systems need a different scheme (Shoval, Sontag, Alon et al 2010) Fold change detection system.
- Generally responds only to shape / difference from background of waveform, not actual amplitude. Breaks down at low concentration.
- Response to temporal and spatial gradients over 500 fold range.
- Does this result from low molecular arrival rate or is it completely covered by the deviations from linear at the extremes?
Sriram Chandrasekaran, University of Illinois, Urbana-Champaign, Brain Transcriptional Regulatory Network Quantitatively Predicts Behavior-Specific Gene Expression
- Behavior changes gene expression in brain.
- Are the honey bee behaviors determined by gene expression (in the absence of enviormental stimuli) or is environmental stimuli affecting?
- Observe bee starting behavior at cryo-freeze it, then micro-array.
- multiple timescales of behaviors — minutes (foraging) weeks (hive interactions to foreigners)
- Aging, maturation etc cluster in gene expression space. Time scales also cluster.
- Use pairwise mutual information with a threshold to connect nodes. Remove weaker links using data processing inequality. How do you choose threshold? what are you using to predict what? profile of gene expression to predict behavior?
- top 5 TFs regulate 40% of genes, top 40 regulate 90%: Cre, ftz, broad, dorsal found to play behavioral roles.
Gregor Neuert, Massachusetts Institute of Technology, Systematic Identification of Signal-Activated Stochastic Gene Regulation
- yeast MAPK (aka Hog1). changes in mean or variance, or fraction of responders?
- Does it change expression rates or just the probability of expressing (and rate is always fixed).
- Hog1 localizes to nucleus on activation in all cells (~within 4 minutes). Only ~80% of cells subsequently turn on gene expression. Levels of expression are variable. GFP intensity, detected 30 min later.
- smFISH. Detect gene expression after 2 minutes. Quantify frequency of number of mRNA per cell.
- increase complexity just enough to match data without over fitting (test overfitting with test set data)
- Predict and test correlation between: maximum fraction of cells with nascent transcription and time at which total mRNA is max.
- See poster Rui Zhen Tan
- Emily Capra, Massachusetts Institute of Technology, Evolution of two-component signaling systems. Poster 29: duplication leads to cross talk which promotes divergence.
- Xianrui Cheng and Socolar (sea urchin) cell differentiation poster #34 Drosophila segment polarity using autonomous boolean models.
- Lily Chylek, Cornell University, Interaction libraries for rule-based modeling p53 interaction libraries, develop library of rules, annotate and visualize. (tie rules to experimental results). How many conflicts?
Pieter Rein ten Wolde, FOM Institute AMOLF, Single-cell ultrasensitivity and stochasticity give rise to bimodal response on the population level
- circadian rhythms, period 24 hr, entrainment, phase resetting by light.
- Temperature compensation. Stable.
- proposed higher organisms clocks stabilized by cell-cell interactions
- Mihalcescu and Leibler (2004): cynobacteria clock stable at single cell level: moments of cell division vary. amplitude of oscillations vary. period very stable.
- correlation time 166 +/- 100 days. No evidence of cell-cell interactions.
- Background: genes: kaiA kaiB, kaiC. BC are in an operon with oscillating expression. continuous overexpress C represses BC. transient express resets phase.
- Kondo lab: phosphorylation of C oscillates with or without light (but gene expression only oscillates in light). and even though concentration of C is constant.
- fixed A B C concentration + ATP phosphorylation of C oscillates by itself. Similarly can shut off oscillations of phosphorylation by kinase over expression and the gene synthesis cycle still has oscillations. why two clocks?
- 2 classes of models for oscillations: differential affinity and sequestration models or monomer exchange models
- Allosteric MWC model (all subunits switch together between active and inactive). Active state binds ATP phophorylation rates rise, increasing stability of inactive state, system converts to inactive state, causing phosphorylation rate to drop, and so cycle.
- Develop corresponding free energy model.
- A single hexamer oscillates. Population mean stays constant (oscillators not synched). KaiA stimulates phosphorylation of active C. [A]>[C]. B promotes dephosphorylaion. Inactive guys sequester A. laggers take away the resources from the front runners.
- Experiments show too much A does indeed stop the oscillations of C-p.
- robustness to stochastic production and degredation? fast degredation cross hopf-bifurcation and fail. High growth rate (high dilution) will kill this clock. Intuition: hexamers degraded/diluted before they get a chance to complete a cycle.
- Genetic synthesis coupling: get new C when you need it, make less when you shouldn’t have it.
- Coupled system is robust against variation in growth rate.
- How about transcription translation model only? This system needs high degredation to work. decay too slow to reset oscillation, slow decay requires slow production — low amplitude not robust. + Costly to have high production and high degredation.
- Combined clock is an order of magnitude more stable in physiological range
Fang Huang, University of New Mexico, Fast, simultaneous multiple-emitter fitting for single molecule super-resolution imaging
- 3 steps to super-resolution: segmentation (which pixels to fit on) localization (gaussian fit) and reassemble.
- High density data – single emitter fitting fails.
- first single emitter fitting model which estimates uncertainty: COM initial guess, then Newton Raphson to converge to best fit.
- extend to multi-emitter model: use filtering operation (difference of gaussian type)
- sequentially check 2, 3, 4 emitter fits. subtract expected distribution from single emitter fit. Use log-likelihood to choose best fit. Includes attempted fit on emitters outside of test region. Test on simulated data :).
- Apply to actin in HeLa cells.
- GPU implementation to speed up analysis time. 1000 fold speedup.
Lei Dai, Massachusetts Institute of Technology, Experimental Observation of Early Warning Signals for Approaching Population Collapse
- Predicting population collapse by fold bifurcation
- Laboratory population of yeast growing cooperatively on sucrose.
- examples (cod fishing)
- model hysteretic system: have to improve conditions much more than they were previously to annihilate the low state in order to jump back up.
- Most of cleaved sucrose diffuses out of cells, hence a cooperative effect.
- Experimentally determine critical density at which yeast population survive or die.
- CoV doubled (mean was dropping), standard deviation also increased independently.
- autocorrelation. It looks like the drop in optical density itself is far more predictive — your bifurcation occurs after very substantial drop: unchallenged population is much larger than the the critical point and the deteriorating population has to drop very dramatically before it reaches the bifurcation point.
- (hence perhaps) seeing bifurcation is more difficult at low sucrose.
Xiling Shen, Cornell University, Asymmetric Cell Fate Decisions in Colon Cancer Stem Cells
- ‘stem cell’ self-renewal. pluripotent. Requires asymmetric division.
- do ‘cancer stem cells’ exhibit asymmetric division? yes both asymmetic and symmetric inheritance of classic cancer markers during division.
- see symmetric self-renewal, assymetric division, and differentiated division.
- claim self-renewal requires active notch signaling
- asymmetric localization of NUMB inhibits receipt of Notch signal.
- asymmetric localization of miR-34a also can determine asymmetric division. miR-34a alone is a better cell fate indicator.
- NUMB directly inhibits notch and indirectly inhibits through HDM2 and TP53 to mir-34a to knockdown Notch. (translational and post-translation repression).
- more mir-34a more asymmetric division, less miR, more symmetric self-renewal (bigger tumors). can match FACs measure with monte carlo simulation.
- Notch inhibits mir-34, mir-34 inhibits notch, threshold bistable.
- loss of p53 can break mir-34a activation pathway, leading to cancer proliferation.
Oleg Igoshin, Rice University, Self-organization mechanisms in Myxococcus xanthus swarms
- (Oster’s former post-doc. On Myxo )
- predator bacteria, needs to cluster to feed effectively
- why ripple when on top of prey?
- side to side signaling (rippling) increases speed of expansion to cover prey.
- Drift less when on top of prey due to myxo motion — helps cell remain there longer.
- Part II: disperse or stay?
- size is the most critical variable (of nearest neighbor distance, residence time, size of total aggregate etc) determines disperal or not. Inferred from data (support vector and mutual info): existing models for dispersal don’t match data.
- Need new model, in process.
Robert Egbert , University of Washington, Tuning Gene Networks with Simple Sequence Repeats
- (Klavins Lab) bacteria synthetic biology
- “towards pattern formation: building a bistable switch”
- based on Gardner Collins toggle switch. Green (lac cells) sickly, grow poorly.
- tune with promoter, RBS, transcript stability, and protein degredation tags.
- How to choose a tuning option: 1) Coverage (dynamic range). 2) predictability. 3) scalability (tune multiple components simultaneously). 4) rapid? 5) evovable? (good under selection).
- 5 orders of magnitude expression variation on RBS, predictive models, RBS calculator (voigt et al) tune each gene independently.
- tune ‘spacer’ between Shine-Delgarno and start codon (repeats of As) best when small.
- ssR has high mutation rate (insertion and deletions due to slippage). Can build the library with PCR, slippage.
- how stable is the resulting construct in bugs? (good for 16 generation, except in mis-match repair mutant).
Kevin Wood, Harvard University, Using Maximum Entropy to Decode the Multi-drug Response in Bacteria
- Cluzel lab
- predict effects of N Drugs? (pairwise may be synergistic or antagonistic)
- if know the effects of drug pairs, can we predict?
- Assume underlying stochastic variable X determines growth. Can measure it’s moments but don’t know the distribution. Estimate p(X) using maximum entropy.
- Test on 3 interaction of protein synthesis inhibitors, two antagonistic, one synergistic. Works surprisingly well: suggests no “new” interactions.
- what fraction of the total N-body interaction is captured by pairwise approximation?
- Q: can you tell based on mode of action. How about effects of gene knock outs on targets?
- knocking out separate redundant branches probably yields non-pairwise predictable effects: synergy of knocking out all 3 branches is much greater than any pairwise effect.
Poster spotlights: Kang #50 actin modeling
Eric Batchelor, Harvard Medical School , Stimulus-dependent dynamics of p53 in single cells. How a single hub works. stable p53-GFP cell line. double strand breaks = pulses. low dose or high dose similar pulses. UV single amplitude modulated response.
Eugene Yurtsev, Massachusetts Institute of Technology, Microbial Cheating Limits the Evolution of Antibiotic Resistance. (Gore lab). sensitive cells can still survive as long as resistant cells are present (R cells break down antibiotic). R+ cells bear the cost of maintaining and replicating the R plasmid which is providing group resistance. Bacterial cheating limits behavior. Inhibitors can speed spread of resistance (increase fraction of R+ cells).
Diane Lidke, University of New Mexico, Measuring Membrane Receptor Interactions Using Single Quantum Dot Tracking
(presented by Keith)
- EGFR binding models: bind ligand then dimerize
- how about ligand free dimers? one ligand one no ligand and dimerize?
- Approach: single quantum dot tracking high resolution localization (nice alternative to FRET). Tag receptors and ligands.
- look at cdf of r^2 where (r is the interframe jump distance). provides measure of mobility. Kinase inhibition increases mobility.
- multi-color channel registration: take multicolor array of dots, scan in both channels before and after every tracking data set, use to correct shift.
- Find evidence for “domain associated dimers” too far a part to be dimer but constrained to within a spatial range. Question: do you transition from through “domain associated” en route from free to dimer?
- ligand associated dimmers are most stable. phosphorylation state of ligand associated dimer does not effect off rate. unbound dimers have much higher off rate.
Eric Deeds, The University of Kansas, Optimizing ring assembly: the strength of weak bonds
- ODE models of ring formation. Dimers favored too much get all dimers and no monomers => missing last component (single monomer).
- Larger the ring the more half-assembled states that don’t combine to complete a ring. Have to wait longer for for partial rings to split up and rejoin to full rings. (though final stable state is still full rings).
- Can go faster with one weak bond. (or appropriately chosen weak bond). One weak one is often optimal.
Can Guven, University of Maryland, Stability Analysis of a Model for Collective Migration of Dictyostelium Discoideum. Renamed talk: Pattern Formation due to Cell-cell Communciation: Signal relay during cell Migration
- Dicty cAMP response – stream and clump towards signal source. (produce same signal from their back as they move forward. Also amplifies signal.
- Mutants lacking secretion of cAMP ability don’t form pattern, just migrate /aggregate (less efficient?).
- wildtype form swirls (clumps chasing tail).
- Model: suggests faster degredation = longer streams and more swirls. Stronger communication even more swirls / bigger clumps. Quantify these instabilities.
Maciej Dobrzynski, Systems Biology Ireland, University College Dublin, Single-cell ultrasensitivity and stochasticity give rise to bimodal response on the population level
- Stochastic network response without bistability