What you can learn from computer modelling

Speaker: Christine Jacobs-Wagner
Department: Casimir Research School
Subject: How to achieve cellular replication without fail: lessons from bacterial cells
Location: Delft
Date: October 13, 2017

Although Christine Jacobs-Wagner studied biochemistry in college, she would describe herself more as a cell biologist. According to her, one is not defined by education, but rather by the questions they pose themselves. That’s why her lab is very multidisciplinary, which made her seminar really interesting. I liked the way she didn’t just come to Delft to lecture us about what’s she’s done, but rather to get feedback from the scientists here, who might have a stronger modelling background than hers. This is how science should work, and it’s a great feeling to be a part of it.

She spoke about the importance of distributing plasmids evenly during cell division. If one of the daughter cells doesn’t inherit an essential plasmid for some reason, it will not survive. That’s why it is crucial to have a method in place to guarantee the survival of both daughter cells. The ParA and ParB proteins make up such a system. This was the focus of her research.

Plasmids should be distributed like these green dots

The ParA proteins bind to the nucleoid of a bacterium, and the ParB associates with plasmid DNA. When they interact, ParB induces a reaction which breaks the ParA-DNA bond. This is all well understood, but how would this have the desired effect of distributing all plasmids evenly? Well, it might be easier to observe what happens when there’s only one plasmid in the cell. It starts to oscillate along the long axis, but why?



Then she proceeded with an in silico model, a computer simulation. Unfortunately, the first model didn’t show the desired behaviour. The plasmid made a random walk, so the Par proteins didn’t do anything. At least, not until she adapted the model to make the DNA non-static. She could model ParA molecules as tiny springs, which oscillate around an equilibrium position. This was a very small change, but it had a great impact. It was all she needed to accurately model the oscillatory behaviour.

It was really cool to see such a small change make such a big difference. For further research and exploration of the ParA-B system, an analytical model would be useful, but that will also be a big challenge. I’m excited for the future!


How to achieve cellular replication without fail: Lessons from bacterial cells

Speaker:     Christine Jacobs-Wagner

Department: Molecular, Cellular and Developmental Biology, Yale University

Subject:       How to achieve cellular replication without fail: Lessons from bacterial cells

Location:    TU Delft (BN)

Date:           13-10-2017

Author: Antoine Rolland

This talk was given by Christine Jacobs-Wagner. She started her talk by saying that most of the work in her lab is based on different phases in the self-replication of the cell. The research is mostly done on bacterial cells, since these are for example simpler than eukaryotic cells. In this talk, she introduced a mechanism that made sure that some low copy number plasmids in bacterial cells are evenly distributed among the two daughter cells after replication.

This mechanism actually consists of a pretty simple biochemical cycle, where two molecules are the most important: parA and parB. Both of these factors are encoded by the plasmids themselves. ParB can bind to the plasmid and cover it almost completely and parA can, after it has bound an ATP,  bind to the chromosomal DNA, which is spread almost all around the bacterial cell. ParB has a high affinity for parA that is bound to the DNA, and in this way the plasmid, which is covered with parB, can be connected to a given loci of the chromosomal DNA where parA is bound. After a small time step, the parB unbinds and the parA loses its ATP, after which the cycle can be repeated. Experiments have shown that this cycle creates a near-perfect distribution of the plasmids  along the long axis of bacterial cells. In case all the plasmids are merged, so only one large plasmid is present, this plasmid oscillates along the long axis over time, following the gradient of the parA concentration. However, in a first model in which only this cycle was introduced, this behavior wasn’t observed. This was caused by the fact that it was assumed that the parA molecules, once they were bound to the chromosomal DNA, would be static. This is not the case, since loci oscillate from its equilibrium position in the cell, because of a certain spring force.

When this part was introduced to the model, exactly the same behavior was observed as in the in vitro and in vivo experiments. Also, changing some parameters, like the amount of parA or the cell shape or size, still produced a nice distribution of the plasmids. What I would be interested in, is which parameters would disturb this distribution, and how this is done.

model parA parB

One of the models of a bacterium cell which was used, with in green the parB covered plasmid and in red parA

What I conclude from this talk is that a simple system, like the biochemical cycle of parA and parB, can still create a very nice distribution of the plasmids before replication. Before this talk, I would think that some really complex processes with a lot of components would be needed to create such a distribution. Also, I think it is nice to see how the model first didn’t work at all, but after one small change, namely having the parA molecules experience a spring force, it made the model work perfectly.

Modeling of the ribosome and the RNA polymerase molecular motors

Speaker: Bahareh Shakiba
Department: Institute for Advanced Studies in Basic Sciences, Zanjan, Iran
Location: TU Delft
Date: July 12, 2017
Author: Teun Huijben

Bahareh has just finished her PhD and is now doing a Postdoc at the IASBS in Iran. In this talk she explained what she has studied during her PhD, in which she was interested in the two most important molecular motors present in the Central Dogma: the RNA polymerase and the ribosome. Both proteins can be seen as molecular motors since they use energy to move along a polymer (either DNA or mRNA). She was especially interested in the dynamics of these motors, so their pausing-behaviour and force-dependence.

The first part of the talk was about modeling the RNA polymerase (RNAP). Bahareh proposed a model to simulate the transcription steps using two components of the RNAP: the bridge helix and the trigger loop, which are both located very close the active site of RNAP (see Figure 1). The bridge helix promotes opening of the double stranded DNA helix and the trigger loop binds to the DNA and drags the DNA through the polymerase. The internal movement of components in the active site of RNAP during the transcription of a single nucleotide can be divided into multiple steps, which each having a energy barrier. Using the energy barriers calculated by other people, she computed the average duration of a pause during transcription. The average length of these pauses was the same as found in experiments.


ÈFigure 1: The bridge helix and the trigger loop are located near the active site of RNA polymerase. [M. Thomm, University of Regensburg]

However, I have some critical notes on this finding. First of all, she didn’t explain the model very well, so it was unclear to us in what way this model was new. Especially given the fact that many people do research on this subject and multiple models already exist. Besides that she didn’t mention the reaction rates she used in the simulations and where they came from.

Another important aspect is that she didn’t show the distribution of pauses. Normally it is very interesting to look at a histogram of pauses, to see how they are distributed, this can be Gaussian, exponential or Gamma, for example. Seeing the distribution will give a lot of information about the underlying processes and indicates if it is fair to simple take the average pausing time and compare this with experiments.

The second part of her presentation Bahareh talked about modeling ribosome dynamics. A ribosome is a protein complex that translated the mRNA to a protein. Bahareh studied how  mRNA-hairpins influence the processivity of the ribosome, since hairpins are formed in single stranded RNA and form roadblocks in front of the ribosome.

The active site of a ribosome has three free places to bind a tRNA molecule: the E- (exit), P- (polymer) and A-(active)-site (see Figure 2). The ribosome moves forward by first displacing the large subunit, followed by movement of the small subunit. If the mRNA in front of the ribosome has internally formed a hairpin, this can either result in necessary unwinding of the hairpin or a frameshift of the ribosome. The latter means that the ribosome temporarily detaches from the mRNA, resulting in errors in the produced protein.


Figure 2: The active site of the ribosome has three active sites, the E-, P- and A-site. A tRNA loaded with an amino acid can bind to the ribosome and transfer its amino acid to the growing peptide chain. A hairpin in front of the ribosome. [Shakiba 2016, arXiv: 1607.0719v1]

There are currently two ideas of how ribosomes solve the hairpins. One states that the moving ribosome applies a force on the hairpin forcing it to unwind. The other theory states that the ribosome itself actively uses helicase activity to make the mRNA single stranded. Both theories are simulated in the model. Then she compared the model with experimental data of ribosomes translating mRNAs having hairpins, while applying a force on the mRNA. This comparison revealed that the model with ribosomes having helicase activity did the best job in explaining the experimental data. Giving the indication that ribosomes indeed have their own helicase activity.

As already stated in this report, the talk of Bahareh was quite hard to follow, especially because she didn’t explain the models very well. Therefore, the first part of her talk didn’t really impress me, since I didn’t see what was new and surprising in her model. However, the second part was easier to understand and got a clear, well described message.

Diversity of immune repertoires

Speaker: Aleksandra Wolczak  
Department: Bionanoscience Department
Subject: Diversity of immune repertoires        
Location: Building 58, TU Delft        
Date: 09-12-2016 

Author: Nemo Andrea

Aleksandra Wolczak is a researcher at the Ecole Normale Supérieure (ENS) and the Centre National Recherche Scientifique (CNRS). She is known for applying statistical dynamics to cells in cases where traditional physics concept such as forces and energy are not a suitable approach. In her seminar, she covered a wide range of concepts concerning the immune system and experiments, with a particular focus on T cells.

The adaptive immune system consist of B and T cells. These cells should be able to detect and react to foreign pathogens. This is mediated through a great diversity of receptors on these cells’ membranes. It is estimated that there are around 10^9 receptors in the human body in healthy individuals. If all these different receptors were hardcoded in the DNA, it would have to be impossibly large. The way that this incredible variety is obtained is through alternative splicing of gene regions known as V, D and J. Various parts if these regions can be spliced together to create significantly more diverse set of receptors than if all the genes were simply transcribed. Additionally, random insertions and deletions in the regions where these V, D and J regions are separated allow for even greater diversity in the repertoire of receptors. So the staggering diversity in receptors comes not from the size of the DNA but from the combined effect of combinatorics and randomness.

[1] flow chart of the analysis pipeline of the model

They generated a probabilistic model for receptor generation, by creating a model that self learns and can calculate the probability that a certain receptor is generated. Furthermore, this model could also predict the specific mutations and splicing events that must have taken place (inference of the cause given an outcome). Additionally, they found that at the level of generation (the receptors generated in the immune system before any selection takes place) where very different between people. Their model predicted that you share as many receptors during generation with family members as complete strangers, with the exception of identical twins. The exception can be explained by the fact that identical twins share blood in the womb, thereby bringing their immune system together to some extent.

Another central question of research was what the optimal distribution of membrane receptors was. One can imagine that one might want to cover the widest range of receptors, while also producing more of the type that is compatible with the most common pathogen receptors. They modeled this with a model in which each receptors has a certain cross reactivity, which means each receptor can still bind to related receptors (albeit less strongly). Their model predicted that the optimal receptor repertoire was a peaked distribution with coverage following the antigen distribution. In practical terms this meant that the receptors were most strongly present at common antigen receptor types and the less common pathogens would be covered by cross reactivity. Such a setup still provided adequate antigen coverage, as can be seen from the image below.

[2] optimal receptor distribution (1D)

Another experiment conducted by her research group related to mutation and receptor effectiveness. Here, they mutated receptors and tested them against one single antigen. This way they could study the effect of mutations on receptor evolution. To asses the difference in affinity of the mutated receptor for the antigen, they used a new experimental approach called Tite-Seq[1]. This approach determines the affinity by means of something that can be seen as analogous to a titration curve (where antigen concentration is gradually increased and fluorescence is measured), rather than a traditional measurement which is usually done at a single concentration. This method would give a more accurate assessment of concentration, as a single concentration measure could easily be deceiving. These experiments showed that most mutations were detrimental, and only a small fraction of mutation actually improved affinity.

[3] Effects of mutations on receptor affinity and expression

I found this seminar to be very enlightening, as it covered such a wide range of experiments and disciplines. Seeing how theory of statistical dynamics, modelling, cell biology, and evolutionary dynamics were all covered in this short seminar, I think I will certainly read some more of her research group’s work. It was also fantastic to see more about the immune system, which is something we haven’t covered in any significant detail in the nanobiology course, showing that these seminars can be complementary to the course material. Added to that, I was intrigued by the probabilistic model they created, as it seems to be a form of a Bayesian Network, which is something I’m currently trying to code as a pastime project. Lastly, it surprised me that tite-seq was a new technology, since the arguments she made in favour of this new method seemed particularly convincing to me. Maybe there are some drawbacks of this method that I don’t see with my limited practical experience, but it may be prudent for the researchers working with binding affinity even in in the bionanoscience department to consider using this method.

[1] https://arxiv.org/pdf/1208.3925v1.pdf
[2] https://arxiv.org/pdf/1407.6888v1.pdf
[3] https://arxiv.org/abs/1601.02160

Collective Sensing by Communicating Cells

Speaker: Andrew Mugler
Department: Department of Physics and Astronomy, Purdue University
Subject: Collective Sensing by Communicating Cells
Location: Delft University of Technology

Date: 16 December 2016
Author: Romano van Genderen

This weeks talk at the department of Bionanoscience was given by prof Mugler of Purdue University. He started by showing us a video of a small cluster of epithelial cells called an organoid growing tentacle shaped protrusions that were growing in the same direction as a chemical gradient, showing that cells can sense gradients and use this information. He told us that he has developed a model capable of explaining this behavior. He said that his talk was split in two sections, one dealing with the fundamental limits to sensing and the other with how cells surpass these limits by communicating.

He started with one of the simplest models, namely E coli chemotaxis. These bacteria can move in two ways, either run forward or change direction by tumbling. The chance that a cell tumbles depends on the concentration gradient. But how accurately can a cell measure its local concentration? To answer this question he modeled the cell very simply, namely as a permeable membrane cube with edges a. So the mean number of molecules in the cell (μ) is equal to a^3*c. Because diffusion is a Poisson process the standard deviation is linearly dependent on the mean. If multiple measurements are done the standard deviation is equal to the mean divided by a refreshing factor T(a^2/D) in which T is the time between 2 measurements and D the diffusion constant. This leads to the Burg-Purcell limit:  μ/σ=(TDAc)^-0.5 In real E coli cells this relationship holds true, the cells are almost optimal measurement devices. Next he went on to a bigger cell, namely the amoeba. The amoeba has two detectors, one on the head and one on the tail. This cell also behaves in such a way to make an optimal measurement with a high signal-to-noise ratio.

Next he went on to show models involving cells talking to each other through autocrine signaling. He first showed some models for uniform cell gradients. This model had receptors, ligands and a messenger molecule. For this model he showed some scary looking rate equations with noise terms. He then solved them using a for me completely unknown method. This led to a function for the mean error. It had a term ½ in front for one cell. Two adjacent cells turned this term into 3/8, not ¼ because the cells are sometimes measuring some things twice. Then when he made the cells no longer adjacent the error became a function of the distance. The ideal distance according to this model was 8/3 cell radius. This is a balance between measuring a lot of different points and losing information due to communication. Comparing this to real life cell packing, the cells do in fact pack as predicted. This sparse packing is one of the two ways to minimize noise. They can also be directly attached through gap junctions.




Fig 1. The model used for long-range signaling. Ligand in blue, bound ligand in purple and communication model in green. The communication molecule concentration drops as a power law.

Finally he showed a setup for measuring a gradient. For this setup to work, the gradient must be relatively shallow. Then he placed a single epidermal cell in the gradient. This cell has no preferred growing direction. But an organoid of talking cells do have a bias. When the gap junctions are blocked, they have no bias. So there is juxtacrine signaling through the gap junctions. A new model was made. But the signaling is now done through two molecules, a local signaling molecule that does not diffuse and a global one that does. So the difference between the local and global molecule is a measure for the gradient. Using this model he showed that the optimal distance for a good signal is 4 cells. But like in the telephone tag game, the longer the signal is passed on from cell to cell, the more noise you get.

This was a very interesting talk, but some of the mathematics used was slightly too advanced for me to completely understand the talk. I did like that cells are way more optimal than I usually see them. They behave as “optimal noise reduction machines”. This is most likely something that is selected for a lot in evolution, because problems in signaling lead to swift death.