Results

This page documents all of our results from building and testing validation, including images of gel confirmation, colonies, and sequencing results.

Results - iGEM

Overview

Our main goal was to build a dual-plasmid biosensor system that could detect nitric oxide (NO) and hydrogen peroxide (H₂O₂) and produce GFP as a reporter to temporarily replace IAA. We went through several rounds of cloning to get there, adjusting our process each time based on what went wrong before. Although we weren’t able to fully confirm the constructs by the deadline, we learned a lot from the build process and used modeling and literature to support the logic behind our design.

Experimental Results

We completed multiple Build cycles, each one focused on troubleshooting issues from the last:

Step Observations Notes
Round 1 – Digestion & Ligation Confirmed through gel electrophoresis
Gel 1
DNA yield was low after gel extraction, it is uncertain whether vectors are properly digested and ligated or not based on band size
Round 1 – Separate Transformation Colonies grew, but sequencing results showed empty backbones and recombination
Transformation result 2 Transformation result 3
Transformation result 1
Insert:vector ratio was likely too low due to lost DNA yield in the gel extraction step
Round 2 – Digestion & Ligation Switched to PCR clean up protocol after insert PCR to increase yield of insert

Incubated the digestion reaction overnight for increased effectivity, ran gel, and ligated
Gel 2
Gel results: digestion of pUC19 and pJUMP worked. For pUC19, the restriction enzymes weren’t cutting out a section of the plasmid, which is why the digested and undigested are similar in length. However, the undigested is still slightly further down because it is in circular form while the digested is linearized. Linear will move slightly slower in the gel than the circular DNA. For the pJUMP vector, the undigested form is heavier and intact while the digested moved farther down, which means it has been successfully digested by the restriction enzymes.
DNA recovery improved slightly through PCR cleanup as opposed to gel extraction
Round 2 – Transformation Standard transformation protocol

Colonies did not grow on the antibiotic plates
Transformation plate 1 Transformation plate 2 Transformation plate 3 Transformation plate 4
Possible reasons: incomplete ligation despite longer incubation time, ineffective PCR

The high background vectors and possibly incomplete digestion were our main issues, but we also realized that our strain choice (TOP10) may have been causing the recombination with the E. Coli genome.

Plasmid Maps

Plasmid Sequencing results from Plasmidsaurus during Round 1 of the Transformation

Puc19-PnorV refers to the plasmid containing the insert responsible for detecting nitric oxide and producing LuxR

pJUMP-PoxyS refers to the plasmid containing the insert responsible for detecting hydrogen peroxide and producing AHL, and then finally GFP.

Plasmid 1

puc19-PnorV from colony 5

Plasmid 2

Puc19-PnorV from colony 7

Both Ampicillin 5 and 7 vectors were empty, indicating that they consisted of an undigested pUC19 backbone without any DNA fragments from the insert. The ligation process was unsuccessful for ampicillin vectors.

Kanamycin vector 1 has recombined with some chromosomal DNA from E. Coli, resulting in an incorporation and expression of the GFP gene, which was not part of what we had intended to include. Kanamycin vector 2, similar to the ampicillin plasmids, was also an undigested vector.

For more confirmation steps, please visit our lab notebook, which includes a detailed list of all procedures we performed in this process, along with analysis of any validation steps.

Plasmid 3

pJUMP-PoxyS from colony 1

Plasmid 4

pJUMP-PoxyS from colony 7

Analysis

While we couldn’t confirm the plasmids experimentally, our modeling results and research back up the design:

  • Using Hill’s Equation, we modeled promoter activation for NO and H₂O₂ and found the expected response curves.
  • Literature supports that IAA reduces inflammation by blocking p65 nuclear translocation, lowering NF-κB activity, and neutralizing ROS.

Together, that gives us confidence that our system design should work once the build is complete.

Troubleshooting Takeaways

  • Low insert DNA concentration from PCR → redo with longer elongation step and more cycles
  • Low DNA concentration after gel extraction → switch to PCR clean up
  • No colonies for initial choice of chloramphenicol resistance vector → switch to pUC19 instead, it has a more compatible origin of replication too with pJUMP-1A and decrease antibiotic concentration
  • Low vector concentration after vector-only transformation and miniprep → inoculate E. Coli again with the vector and grow for a longer period of time in a liquid culture before miniprepping
  • Low vector concentration post-restriction → inoculate E. Coli again with the vector, grow in SOC media and improve DNA yield
  • No colonies after post-ligation vectors were transformed → Re-do transformation, allow the bacteria incubate on ice for 30 minutes before the heat shock and let the transformed cells recover in SOC media for one hour before plating on selective media.
  • Sequencing showed incomplete digestion and ligation → optimize PCR for longer elongation time and more cycles, let the restriction digest incubate overnight at room temperature, ligate with higher insert:vector concentrations.

Next Steps

For our next engineering cycle, we plan to:

  • Repeat digestion with fresh restriction enzymes and extended incubation.
  • Research maximal optimized protocols for each step in the cloning protocol.
  • The PCR protocol will be improved through a longer elongation step and more cycles to improve insert concentration after performing gel extraction.
  • Use PCR cleanup exclusively for inserts.
  • Increase insert DNA concentration and ligation ratio.
  • Transform into NEB Stable cells and test with GFP with fluorescence assays to validate the predicted AND gate logic.
  • Measure different concentrations of inducers to create a dose-response curve and compare with our Hill’s equation model
  • Conduct research into safe and optimal concentrations of IAA to use in our plasmid system such that the final product will be both safe to apply on the skin and effective in reducing inflammation.
  • Replace GFP with indole-3-acetic acid complex (iaaM and iaaH genes) and test inducer and output concentrations both in-vitro and in-vivo.

Overview

Our dry lab focused on two main projects: the first being the prediction of our biological circuit's behavior using mathematical modeling, and the second being the validation of our targets with an ML model. Both of these models provided us a strong proof-of-concept for our design and confirmed that our biosensor is biologically relevant and theoretically accurate. The mathematical model ensured that our system was sensitive and specific, while our machine learning model confirmed that our genes of interest do, in fact, contribute towards inflammatory skin.

Hill's Equation Modeling Results

We used Hill's Equation to make sure that our biosensor functions the way we intended it to. The equation models the dose-response curve of each of our three promotoers (PnorV, PoxyS and Plux). Our goal with this model was to confirm that our AND gate system would respond specifically to the presense of both NO and H₂O₂ at concentrations found in inflamed skin while remaining inactive in healthy tissue. This model helped to validate our circuit's logic.

The model simulates an AND gate which is represented by the 3D heat map below. Plux activation is only significant when NO and H₂O₂ concentrations are both past a specific treshold. Even when only one of the inducers is present, regardless of its concentration, the promoter's output remains minimal. This demonstrates the stingency of the gate, which is crucial to prevent our therapeutic compound from being unnecessarily produced in non-target environments.

Plux Two-Plasmid AND Gate Response

Figure 1: Plux promoter AND gate response, showing activation only when both NO and H₂O₂ are present.

This result provides a strong theoretical support for our design. It confirms that this circuit is specific enough to distinguish between inflamed and healthy states in skin, which minimizes the risk of off-target effects while ensuring that therapeautic action occurs solely in areas of inflammation.

Machine Learning Model Results

We built a logistic regression model in order to validate our choice of inflammatory markers. This model identified which genes are the most powerful predictors if inflammation in skill tissue. We analyzed a large dataset that compared normal and psoriatic skin, and the model analyzed over 13,000 genes to find the key drivers of inflammation. A psoriasis dataset was chosen because both acne and psoriasis inflammation are driven by similar pathways.

Our model's results were promising, and we achieved a testing accuracy of ~90% and an ROC-AUC score of 0.904. These values show that our model is both accurate and robust and can reliably distinguish between healthy and inflamed samples. After this, we used SHAP to interpret the model and then quantify each gene's contribution to predicting inflammation. This analysis helped to reveal that IL1B and NOS2 were among the top predictors.

Top Predictive Genes (by SHAP value):
#164. IL1B         SHAP: 0.0265
...
#311. NOS2         SHAP: 0.0120
...

The high ranking of both of these genes is critical to our proof-of-concept. Interleukin-1 beta is a pro-inflammatory cytokine that activates NF-κB signalling in acne lesions. iNOS is directly responsible for producing NO during acne-related inflammation. The emergence of IL1B and NOS2 in our rankings prove that they are key predictors of inflammation, validating the design of our NO-and-H₂O₂-sensitive-biosensor.

Overall Conclusion

These dry lab models provide a comprehensive and multi-faceted validation for our project. The Hill's Equation model proves that our circuit is logically sound and should function given the required specificity and sensitivity. Additionally, the machine learning model confirms that the biological signals that we chose to detect are key players in inflammatory skin disease. These models help to demonstrate that our design is both well-engineered and also targeted at a meaningful problem.