CSBE Symposium Day 1: From Systematic to Synthetic Biology

I have been fortunate enough to be invited to go (as a CISBAN representative) to the two-day symposium on sytems and synthetic biology organized by CSBE in Edinburgh, UK for the 4-5 September.

Before I get to the nitty-gritty, here are my awards for....

...Most Fun Talk: Drew Endy, MIT. He gave his talk at the beginning of the first session, directly after Andrew Millar's introduction. The projector was still broken at this point, so he demonstrated some fantastic skills as a lecturer and did the entire thing on the blackboards. An interesting speaker well able to think on his feet!
...Best-Organized Slide Presentation: Angelika Amon, MIT. She is a fantastic speaker and had beautiful slides: no slide had more than one sentence on it, and she generally followed the formula 1) Ask a question 2) answer the question in one sentence 3) answer the question with pictures, generally with one slide per step. It was a beautiful thing. She also had a really interesting talk about aneuploidy, which definitely helped!

And no, I have no bias towards MIT - in fact I have no professional relationships to them at present - it just turned out that way!

Here are my notes from Day 1. Both me and my fellow CISBAN representative (hi Steve!) had a great time, and were extremely well-fed at lunchtime. Please note that these are my notes, and I may have not understood some things, and therefore made a mistake. Please let me know if there should be any corrections!

Notes CSBE Symposium Day 1: From Systematic to Synthetic Biology
September 4, 2007

Amdrew Millar, CSBE

He provided background into the development of CSBE and the main focus of their work. They have received £11 million in startup funding over 5 years, together with a £7 million grant from University of Edinburgh. This is primarily infrastructure funding, and they are currently working on getting grants for research work. In 2009 they will be moving into to the new building (Waddington Building) that is being built.

The research focus is not on a particular biological question, but instead on the process of systems modelling. It is very difficult for experimentalists to engage with SB and to transform their data into real models. Initially, there are three biological areas that will inform the larger systems modelling theme. All theoretical/informatics research will be integrated into the systems biology software infrastructure core (SBSI). Experimental projects will hang off this core as well, with the Kinetic Parameter Facility (KPF) included. The three projects are the RNA metabolism project (yeast), macrophage project (using human cell cultures), and the circadian clock project (arabidopsis). These projects are intentionally diverse. The wet lab projects differ in size/scale and in current levels of understanding.

Wet-lab biologists are generally neither rigourous or interested in providing kinetic parameters. However, the KPF will help resolve this problem. They are also working with new theoretical tools, e.g. ones that allow you to deal seamlessly with both discrete and stochastic models. In this, they're working with biological stochastic process algebra (PEPA adapted to create BioSPA). Network inference and network analysis are also important areas of research. To improve the interface between the experimental biologist and the netowrks, they are using and developing the Edinburgh Pathway Editor (EPE).

They also strongly feel that systems biology naturally leads to work in synthetic biology. For instance, to get a particular model tested, it may be necessary to create a synthetic system *just* containing the steps in the model to be tested. It is a biological test for a bioinformatics experiment.

Centres such as the CISBs have a particular role to play in playing the long-game, fostering community organizations and standards development and usage. They're planning extra collaboration with other centres within Europe.

Drew Endy, MIT:

"Synthetic Biology"

Drew is one of the organizers of the IGEN competition, and undergraduate competition in synthetic biology. In 1999, was looking and changing the genetic architecture of the phage he was working on, creating an autogene architecture where you get positive feedback. By doing this, he thought it would be a phage that grew faster than a wild type. However, the model wasn't right, as when the lab work was done, the growth wasn't as fast as the wild type. However, he had a problem publishing as his model didn't agree with his experiment. Eventually, got it published in PNAS. He thinks that perhaps natural systems haven't evolved to be optimized for modelling. *Really* Intelligent Design would have documentation! :-) And yet, we have no such insight yet.

So basically he works on trying to refactor natural biological systems to make them easier to model and manipulate. What should the theoretical "Dept of Biological Engineering" look like? The three lessons learned from engineering history are as follows: standards (to support reliable physical and functional composition, as a resulting product may end up with emebrgent properties), abstraction (borrowed primarily from computer science, allows you to implement much more powerful functions without having to bother with the nitty-gritty details: machine language = ATCG in this case), decoupling (separate complicated problems into simpler separate ones: in biology you could take as an example the automated construction of DNA).

Can you really make "biology" reliable? Alot of good process engineering would have to be done first. What other problems would be anticipated in making biology easier to engineer?
    + Noise: rest of the talk focuses on this.
        + Evolution

Combining synthetic biology and systems biology to try to combat noise. There has been, on average, one paper published per week addressing the concept of noise in biological data sets and in dealing with individual molecules in modelled reactions. Uses an example of a signalling pathway in yeast in a paper he contributed to, published in Nature: the experiment *actually* reports on the level of fluorescent protein levels, rather than in the values of the important proteins in the pathway - i.e. it's all indirect. What's interesting is not the variation in expression, but the fact that there's no observable phenotype for the vast majority of these. He appreciates the noise in biology, and the work gone into reporting it, but is not sure why its relevant. Is that noise really important to us?

Mentions a paper on the lambda (phage) vector from 1997 (published in 1998) by Adam Arkin, published in Genetics. How does phage lamda decide what to do? Two primary decisions are to either lyse the cell or to integrate its genetic material with that of the cell. In the Genetic Switch (book) he admits that there is no "perfect understanding" of what drives lambda to one over the other decision. The Arkin paper suggests that a model should be made of the lambda phage, and he isn't going to be content with a cartoon. Further, it will be discrete reaction events, not continuous. In cases of low multiplicity, you will almost always get lysis. If you infect with many cells, you almost always get chromosomal integration. There is also conditions where you can infect a genetically identical set of cells with 1 phage per cell, and you get a 50/50 split in the decision. Thus stochastic chemical kinetics is a relevant physical framework for describing the behaviour of the system. It's all driven by the noise = the stochastic chemical kinetics. Arkin makes explicit the fact that the molecular biology community did not find all the answers of the lambda phage, and also created a model, and also made a directly testable hypothesis. The lambda phage "rolls the dice": it doesn't know what to do when it gets in the cell. This means the hypothesis is running blind, and doesn't know the state of the cell. However, what we know of lambda biology points to it actually knowing the state of the cell.

Therefore the alternate model (trying to make a deterministic model) is as follows: if you knew what to look for, you could theoretically segregate the genetically identical population of cells into two sub-populations. In other words, it's not stochastic at all, but determined by the state of the cell at the time of infection. What, then, is the physical barrier that you look for? He tried all sorts of things, including looking at the genetic architecture of the phage more carefully. The lambda C1 repressor protein binds in an antiparallel fashion with the pro protein to R1, R2, and R3. If repressor binds first, it will shut off binding of cell, and you get genetic integration. If the reverse, you'll get lysis. The repressor binds with more cooperativity in the R1-R3 region than pro does. In small cells, the repressor will be more successful as it binds with more cooperativity, and in larger cells the pro will be more fit. So the 3 independent variables are: the abundances of repressor and pro, and the volume of the cell. What they find is that the binding energies of the proteins are matched to the volume of the e coli cell at various stages in its life cycle. In a small cell, 100 C1 = 400 pro, while in a big cell, 100 C1 can be balanced with 100 pro.

So, he took samples of independently collected fractions based on width (one sample had "1", the other was "1.6"). So could plot probability of lysis and lysogeny. Average volume of fractions plotted against the % of lysogeny. In smaller cells it is about 80%, and then declines linearly with cell size to about 25% at the high volume size. The lysis happens in the opposite way. This means at the intermediate fraction (intermediate cell size) you still get a 50/50 split. So the 50/50 split could be noise, but it might also be the distribution based on cell sizes. From this, determine the critical volume, which is pretty close to the middle of the cell division cycle in e coli, at about 1.4.

So, behaviour of such natural systems may not be stochastic (he means "noisy"), but actually deterministic. Next step is to make the appropriate mutants that would remove the assymetry in the R1-R3 region. Also, can you determine if it's absolute versus relative volume? Well, in exponentially growing cells you get about the same slope of the line as before, but still unclear as there is such a range of results (as you can tell, didn't quite hear the whole answer).

Mike Tyers, University of Edinburgh

"Size control: a systems-level problem"

Focusing on the dissection of a growth-dependent switch element in budding yeast by binding GFP to Sic1. The protein is degraded by the ubiquitin system and its recognition depends on phosphorylation at multiple sites. Elimination of Sic1 allows the onset of a B-type cyclin CDK activity. Elements that control the cell cycle were highly enriched in this system. In the system of Cdc4 and Sic1 (recognized by Cdc4 in a phosphorylation-dependent manner). A threshold of G1 cyclin-CDK activity is required for Sic1 elimination. Most individual Sic1 CDK sites are not required for degradation in vivo. 6 of 9 CDK phosphorylation sites appear necessary for efficient Sic1 recognition by Cdc4. A SPOTS peptide array defines the Cdc4 Phospho-Degron (CPD). However, a single optimal CPD site is sufficient for Cdc4 binding and degradation in vivo. Then he showed a video of precocious elimination of Sic1(CPD). Why multi-site phosphorylation? For ultrasensitivity. Might electrostatic repulsion lower Cdc4 affintiy for natural CPD sites? Re-engineering Cdc4 to reduce the phosphorylation threshold was the next step they did. A sharp transition in affinity of Sic1 for Cdc4 was discovered, using surface plasmon resonance analysis. They also performed NMR analysis of the Sic1-Cdc4 interaction.

We understand the equilibrium engagement of Sic1 with Cdc4. It is tuneable, evolvable and adaptable, and is ultra-sensitive. 30-40% of the proteome contains disordered regions.
L. Landau: There are two kinds of models in this world: those that are trivial and those that are wrong. (Paraphrase of a quote).

Joerg Stelling, ETH, Zurich

"Analysis and synthesis of biological networks"

Alternative quote: All models are wrong, but some are useful.

The challenges of biological networks include complexity and uncertainty. Approaches for creating mathematical models include graph theory (topology), structural analysis (stoichiometry), and dynamic analysis (biochemistry). Use the right level of description to catch the phenomena of interest. Don't model bulldozers with quarks. Synthetic biology is a new dimension of biology engineering. It is a case of forward-engineering. It promotes the creation of standardized interfaces to biology/wet lab work. Derived characteristics of synthetic circuit performance should meet the following criteria: robustness, tunability, feasibility, and stability. An example he used was a Synthetic Time-Delay Circuit. You begin with a simple electrical engineering circuit for a time-delay function. The biological analogy is as follows: biotin as a chemical signal (input), covalent protein modification (rectifier), protein accumulation (buffer), protein degradation (resistor), genetic switch (switch) and protein production (output).

Complexity is due to side reactions / coupling between activating input, internal components and inactivating input. They made an ODE model of this system. You can fine-tune the circuit by using non-linear dependencies of performance characteristics on parameters, inputs and component features (protein stability). This helps identify targets for fine-tuning of circuit functions. There were discrepancies between model and experiment. Qualitatively it matches, quantitatively it did not.

Are there any possible shortcuts?

Structural analysis does not need kinetic parameters, but just the structure and stoichiometry of the network. Feinberg talks about chemical reaction network theory. Gatermann discusses algebraic geometry.

Angelika Amon, MIT

"Systematic analysis of aneuploidy"

Studying the mechanisms that control chromosomal segregation, and specifically what prevents mis-segregation from occurring. What actually happens to cells that end up with an extra chromosome? She's discussing the effects of aneuploidy on on cell growth and division in yeast and mouse. Two take-home messages:
+ Aneuploidy causes a proliferance of damage (is bad for the cell!).
+ Additionally, there are a set of phenotypes/consequences to aneuploidy that is independent of which is their extra chromosome.

Yeast:

Comparative genome hybridization analyses confirm the karyotype, and can show you the stretch of the genome that is present twice. They have about 20 strains that carry extra chromosomes, and they studied a variety of properties of these strains:

+ Cell cycle properties of the aneuploid strains: many aneuploid yeast strains are delayed in G1.
Cells disomic for chr 13 are delayed in cell cycle entry. Budding and DNA replication are both delayed by about 15 minutes. A large number of the disome strains had their delays in the G1 phase. There appears to be a correlation with the amount of extra DNA present in the cells does contribute to the length of the G1 delay. Doesn't seem to be the only factor, but still an important one. All aneuploidy strains seem to have growth defects via problems with the G1-S transition, and it occurs upstream of the Cln/CDK pathways.

+ An aneuploidy signature: all such strains shared a common gene expression pattern.
This pattern is not seen when you just grow up the strains normally (the extra transcription for the disomal chromosome will mask any patterns of similarity between strains). So, you have to correct for the extra transcription. The pattern is seen in two clusters of genes, one up- and the other down-regulated (via a Phosphate-limiting experiment). in rRNA transcription and processing genes, which are all upregulated. Don't know what the significance of this cluster is yet. The downregulated genes seem to have something to do with amino acid metabolism.

+ Metabolic properties of aneuploid cells: Aneuploid strains stop dividing at a lower OD.
Something in the media is something that the aneuploidy cells need to grow more than the WT cells. They checked glucose amounts, and found that these strains take up more glucose and have a lower biomass per unit of glucose than the WT. All aneuploid strains show an amplification of two glucose transporters, HSD6 HSD7. They tried to knock out these genes, though, and it didn't make a difference, so still some work to do.

+ What is the extra glucose needed for?: Most genes on the additional chromosome are made into proteins (e.g. more than 90% when chr 2 is the extra one).
Cells devote about 60% of their chemical energy to making proteins. So, check to see if the extra proteins are actually produced, or just transcribed. 3 of 16 proteins analyzed show an increase in protein levels in accordance with RNA levels. 13 of 16 did not show a corresponding increase in their levels. Rather than thinking that they are NOT transcribed, it seems that feedback mechanisms kick-in that re-create the right stoichiometries in the cell for those proteins and they are quickly degraded.

+ Most aneuploids are sensitive to drugs that interfere with rna synthesis, protein synthesis, proteosome; this raises the possibility that these extra proteins create imbalances in the cell which the cells try to fix. The pheonotypes of aneuploids should not show up in cells that have large amounts of DNA, but have very little that are normally translated. Tested this next..

+ What are the consequences of foreign (non-translated) DNA on yeast cells?: No cell cycle delays or sensitivity to conditions interfering with protein synthesis, folding and turnover.

In Summary, a set of phenotypes that is independent of the identity of additional chromosomes.
Hypothesis: cellular homeostasis is disruped in aneuploids due to the RNAs and proteins synthesized from them. Also some of the phenotypes shared by aneuploids may represent the cell's effort to re-establis homeostasis.

+ A strategy to isolate mutations that allow cells to tolerate aneuploidy: select mutations that improve the growth rates.
they evolved strains disomic for chr 5. Choose strains that (via CGH) showed that both copies of chr 5 are intact. Doubling time shortened (3.9 hours rather than 5.1 hours in the original disomic chr 5 strain). They are still sensitive to cycloheximide. However, they are significantly less temperature sensitive.
    + SNP analysis showed 4 mutations in the evolved strains, though they haven't checked if it is those 4 mutations that have caused the changes in phenotype.
        + truncation of ubiquitin-specific protease
        + point mutation in RAD3
        + point mutation in SNT1
        + promoter mutation in the putative ribosome associated factor YJL213W.

+ Analysis of aneuploid mouse cells: analyzed trisomy 1, 13, and 16 in mouse embryonic fibroblasts.
First, transcript arrays confirm genotype (trisomy). Trisomy 16 cells exhibit proliferation defects; for example, such cells are bigger (is it a increase in growth, or a decrease in apoptosis etc mechanisms?).

+ Analysis of Human Ts21 Cells (down's syndrome: foreskin fibroblasts, work from another group in 1979)
These cells are also bigger than the "WT" human cells.

+ There was also an increase in glutamine uptake by their mouse trisomy fibroblasts. They also seem to produce more lactate than WT cells. This indicates a shift from oxidative phosphorylation to glycolysis. This shift is often seen in primary cancer cells. Perhaps, the aneuploid state itself somehow contributes to this metabolic change (still wild speculation ;) ).

+ Effects on immortalization: Immortalization is delayed/not occurring in Trisomy 16 cells.
Virtually every solid human tumor has its karyotype completely messed up. Having an extra copy of any of these 3 chrs inhibits immortalization. In WT fibroblasts, once immortalized they become neotetraploid. The Trisomy cell lines seemed much different - perhaps hexaploid. Interested in continuing to look at growth and immortalized properties of these trisomy cells. Immortalization in itself does not cause a switch from oxidative phosphorylation to glycolysis (aka does not cause an increase in lactate production).

Final Summary:
+ Aneuploidy causes a proliferation disadvantage
+ Loss of a tumor suppressor gene or gain of an oncogene comes with baggage: a whole extra chromosome.
+ During transformation (immortalization), such a disadvantage needs to be overcome.

Hans Lehrach, MPI of Molecular Genetics

"Vertebrate genomics"

Discussed the work required to finish the euchromatic region of the human chromosome.
Can we understand why certain genes are expressed? Try to understand gene regulation by systematically knocking-down TFs using RNAi. Project is progressing rapidly. Have selected 200 human TFs to work with that have endogenous expression in human cell lines. Looking at, among others, the effects on Chr21 due to its effects when trisomic (i.e Down's Syndrome).

Doing a pilot study of a monte-carlo simulation of large metabolic networks.
ConsensusPathDB.

Christina Smolke, California Institute of Technology

"Biomolecular engineering and Riboswitches"

Working on engineering a scaleable communication and control system, namely a sensor-actuator control system in biological networks. Such a system would be comprised of an actuator, an information transmission and a sensor (aptamer) element. Can have either an open or closed loop system. Closed loop systems are often seen in pathways that include feedback processes.

Synthetic riboswitch engineering: 3 methods in recent years:
+ trial and error integration of an aptamer into target places of a transcript
+ direct coupling of a regulatory element to an aptamer
+ screening randomized linker sequences for switching activity

They have been attempting to build a framework for building such a system. These include the creation of composition rules of ribozyme-based regulatory systems. Design strategies should support
+ portability and reliable functional composition,
+ integrated RNA component systems (e.g. instead of loop replacement - where one of two loops is replaced eliminating tertiary interactions - with "direct coupling", where the function of the regulatory element is maintained)
+ reliable functional switch composition
+ programmable ribozyme switch platforms

Design strategies for synthetic regulatory ribozymes (necessary for a universal ribozyme switch platform).

Have built an "on" and "off" switch to up- and down-regulate gene expression.
applications:
+ non-invasive concentration measurement
+ integrating RNA devices with survival genes
+ integrating RNA devices for programming cell behaviour (e.g. engineering T-cell proliferation)

David Baulcombe, University of Cambridge

"Small RNA silencing networks in plants"

+ RNA silencing as an antiviral defense
As the virus replicated and accumulating, this RNA-based system would be activating, slowing the accumulation of the virus. However, it isn't *quite* and virus defense system. Rather than just defense, it is really a virus regulatory system. Yes, it is used to defend the plant, but the virus "uses it" to prevent it from damaging its host system.

+ The basics
ssRNA -> (via RNA-dependent RNA Polymerases RdRP(RDR)) -> dsRNA -> (via a dicer) -> 21+24nt RNA -> argonaute (AGO)/slicer enzymes use the short RNA as a guide to the enzyme's target -> nuclease action on target RNA
There is negative feedback in RNA silencing. The cis-target RNA inhibits the creation of ssRNA.

+ Silencing spreads in two senses:
    + silencing can move beyond the originating cell
    + silencing can also spread along the gene that is the target: the *effective* silencing eventually involved the whole of the transcribed sequence, in both directions (3'<->5')

A primary siRNA is recruited by an AGO protein. The cleaved target then becomes the target for the RdRP. Once you have the secondary siRNAs, they can take over what was done by the primary siRNA. This is why the whole process can be maintained, and the primary siRNA is only needed transiently. This means there is an epigenetic process, i.e. completely independent of the DNA. There is also an amplification process (one initial siRNA -> many produced siRNAs).

Is siRNA a transcription mechanism? If so, there would be methylation of the target DNA. However, this is not a transcription mechanism, even though there *is* methylation. It is RNA virus-induced DNA methylation. Either directly or indirectly, there is some interaction between the viral RNA and the target DNA.

Did an experiment where one virus had GFP added and the other had a promoter sequence (the former gave posttranscriptional silencing and the latter gave transcriptional silencing). Studying the progeny shows a genetic imprint that persists through several generations, therefore RNA silencing can induce trans-generational effects.

+ screening for RNA silencing signal mutants
In theory the signal for silencing followed along the veins of the plant. Mutagenize these plants, and found some that lost the ability for silencing, and others had enhanced silencing ability. This means the amount of silencing is related to the negative feedback effect. If you knock-out the cis-targeting then you get increased silencing, or you could knock-out the trans-targeting effect which will reduce the silencing. (unclear to me which is the trans-targeting part of the pathway).

They did deep sequencing of Arabidopsis siRNA and miRNA. When you align the sequences of the sRNAs against the whole genome, the alignment is not random: there are certain areas of the genome that have a propensity for producing the endogenous sRNAs (== siRNAs and miRNAs). A minimum of 1% of the arabidopsis genome has the potential to generate siRNA. siRNAs: 1 siRNA = many siRNA, miRNAs: made from precursor molecules that can fold back on themselves, causing 1 precursor = 1 miRNA. In his opinion this difference is not profound, and is essentially moot.

tasiRNA: transacting small interfering RNAs.

We think of the following types of RNAs: initiators (foldback RNAs that form miRNAs, or transcribed on both strands; a perfect match to sequenced small RNA; sRNA loaded into AGO slicer complex), node RNAs (like the secondary siRNAs), and end point RNAs (perfect or imperfect match to siRNAs, no evidence for dsRNA, e.g. micro RNA or tasiRNA).

+ computational assembly of sRNA networks.
These networks are large and have non-random characteristics. Many nodes have a very low degree of connectivity (lower than expected with random networks), and there are a very few that are highly connected. They are now working on an empirical analysis to see if these networks do indeed exist.

+ if the networks exist, what could they be doing?
Influencing growth and development of the plant; influence epigenetic effects taking place during flowering, transition from juvenile to adult growth phases etc; heritable silencing by endogenous sRNA loci?
possibly: Altered expression of endogenous RNA -> novel rna directed DNA methylation and transcriptional silencing of target locus -> maintenance of impring through meiosis "heritable epimutation" -> natural selection -> meC to T transition by deamination, results in transformation if epimutation to mutation -> further rounds of natural selection
alternatively: because many sRNA loci are associated with genome repeats and transposons, the sRNAs from inverted repeats have the potential to affect mRNA and therefore the natural variation between species/strains.

+ plan to move into some work with chlamydomonas reinhardtii, which can be considered a model system for silencing. (green algae grown in liquid culture). It produces micro RNAs and siRNAs. Hope to use this organism to do a "truly" quantitative systems-level measurement experiment.