Week 7 HW: Genetic Circuits Part II

PART 0

1. How do ERNs work + how this differs from proteases

They cleave mRNA transcripts before they can be translated into proteins, so the mRNA is not processed by the ribosomes and gets degraded quickly by cellular exonucleases. They work directly at the RNA level, unlike proteases which work on the final translated proteins. This means that they prevent new proteins from being produced, but they do not affect the existing pool of proteins.

ERNs recognize specific RNA sequences or structural motifs, while proteases recognize speciifc amino acid sequences (or structural features on folded proteins).

In summary: ERNs work at the pre-translational stage and detect specific RNA sequences or motifs sites that they cleave directly, while proteases act on the post-translational product guided by amino acid sequences or structural patterns.

2. Lipofectamine 3000 mechanism

Lipofectamine 3000 is a lipid-based transfection reagent that forms positively-charged lipid nanoparticles which bind to the negatively-charged plasmid DNA to create “lipoplexes”. These particles are then taken up by cells through endocytosis thanks to a similar mechanism, where the positively-charged lipoplex interacts with the negatively-charged cell membrane causing it to absorb it as an internal (endosomal) vesicle. They then “escape” the endosomal container by rupturing it (the lipids buffer the acidifying endosome), releasing the plasmid DNA into the cytoplasm before it can be degraded in lysosomes.

(Interestingly, I have been recently researching mRNA-LNP vaccines and they use a similar mechanism for “endosomal escape”!)

The DNA must then enter the nucleus, where it will be recognized by the host RNA polymerase II, which will transcribe it into mRNA (which will in turn be translated into the final protein once exported to the cytoplasm and processed by ribosomes).

3. Poly-transfection + why useful in neuromorphic circuits

Poly-transfection is the process of transfecting multiple plasmids into the same population of cells (so each cell receives and expressed multiple different genetic instructions). This is usually done by mixing all the target plasmids within the same lipofectamine reaction (the cell uptake of the plasmids can be uneven, so the final sample will be heterogeneous).

This is important for neuromorphic circuits because they require a variety of functional parts that could not fit into a single plasmid, due to size constraints and promoter interference (ie sensors, signal-processing nodes, output reporters). Some of these parts must be included within the same cell to operate as a singular functional/“computing” unit (like in a neural network).

PART 1 - IANNs

1. What advantages do IANNs have over traditional genetic circuits, whose input/output behaviors are Boolean functions?

Traditional Boolean genetic circuits produce discrete and binary outputs (“on” or “off”), and therefore cannot represent continuous values or perform weighted computations. IANNs can process continuous values as input and produce graded outputs, so they can be used to perform additional functions like weighted summation, thresholding and signal integration. They are also more flexible over time, as their weights can be tuned and upgraded over time through poly-transfection, which means that the same circuit can be adapted for different use cases. Lastly, they are more compact, as a single IANN can handle calculations that would require multiple Boolean circuits to compute.

(An analogy I liked is that traditional genetic circuits are like ON/OFF switches, while IANNs are more like dimmer switches with memory, which can represent a value between 0-100%)

Their ability to handle non-discrete values in computation makes them a much better fit for computations in biological environments (which are noisy in nature), since they can tolerate stochastic fluctuations (introducing noise won’t necessarily lead to a false negative or positive, since the final output is graded, instead of a “yes/no”).

2. Describe a useful application for an IANN; include a detailed description of input/output behavior, as well as any limitations an IANN might face to achieve your goal.

We could use IANNs to build a cell-state classifier, which could distinguish between states (ie healthy, stressed or apoptotic) based on expression levels of different biomarker proteins (the inputs). We would engineer the cells to have promoters that are sensitive to these specific proteins, and engineer a “hidden layer” that would compute weighted combinations that would act as signals for the output layer to express different fluorescent proteins to indicate each state (ie different colors for different states).

Example:

• Possible cell states: healthy, stressed, apoptotic
• Inputs: biomarker proteins (ie p53, caspase-3, which indicate apoptosis)
• Computation: 1. promoter detects biomarkers 2. hidden layer computes weighted combinations 3. output layer expresses different proteins based on the received values
• Output: signal proteins expressed (ie GFP for “healthy”, RFP for “stressed”, BFP fpr “apoptotic”)

The main difficulties would be the cell-to-cell variability after poly-transfection (explained in the relevant section above), which could introduce a lot of noise, and setting the correct weights (promoter strengths) to achieve classification accuracy.