Lab (Week 7) — Neuromorphic Circuits

Completion status:

• This lab was completed virtually (circuit design using the Google Sheet template, in silico simulation of OT‑2 instructions).
• The wet lab component (OT‑2 building of plasmids, transfection into HEK293 cells, and observation of results) was not performed – neither in person nor virtually.
• The following report describes the designed artificial neural network circuit and the theoretical steps.

Pre‑Lab Overview

We familiarize ourselves with two key concepts:

1. Endoribonucleases (e.g., Csy4) – used to perform arithmetic inside cells by sequence‑specific cleavage of RNA, enabling analog computation.
2. Lipofectamine 3000 – a cationic lipid transfection reagent that forms complexes with DNA, enabling delivery into human HEK293 cells.

We also download the Neuromorphic Wizard folder and follow the installation instructions (simulated – no actual installation performed).

Background

Intracellular Artificial Neural Networks (IANNs) are synthetic genetic circuits that perform analog computations, unlike traditional digital logic gates. They can approximate any input‑output function given enough neurons. The building block is the Sequestron (a transcriptional activator or repressor that binds to a specific DNA sequence, sequestering transcription factors). Each neuron is implemented as a plasmid encoding a fusion protein (e.g., dCas9-VPR with a guide RNA) that regulates the expression of downstream genes.

Protocol Overview (Virtual Completion)

Day 1: Circuit Design (Google Sheet Template)

We work in a group of 3 (virtual). Our task: design a simple two‑input, one‑output IANN that acts as an XOR classifier (output ON only when exactly one input is present). This demonstrates analog summation.

We use the provided HTGAA 2026 Genetic Circuit Part Names (inferred from typical Weiss lab parts – actual list not provided, so we use plausible names):

• Promoters: pCMV (constitutive), pTet (Tetracycline‑inducible), pLac (Lac‑inducible)
• Genes: dCas9-VPR (activator), dCas9-KRAB (repressor), GFP (reporter), mCherry (reporter)
• gRNA scaffolds: gRNA1, gRNA2, gRNA3
• Terminators: bGH polyA, SV40 polyA

The spreadsheet template requires:

• Contents (part names)
• DNA wanted (ng) – sum ≤ 650 ng, each at 50 ng/µL stock.
• Concentration = always 50 ng/µL (so volume = DNA wanted / 50).

We design a circuit with two hidden neurons and one output neuron (3 plasmids total, each under 650 ng total DNA). Actual DNA amounts are chosen to be realistic.

Completed Spreadsheet (example for one group)

Total DNA per well ≤ 650 ng:

• A1: 180 ng, A2: 180 ng, A3: 200 ng → all <650.
• Sum of all DNA across wells = 560 ng (well within limit).

This design implements:

• Neuron 1 (A1): Activator (dCas9-VPR) regulated by two inputs (Tet and Lac).
• Neuron 2 (A2): Repressor (dCas9-KRAB) regulated by same inputs.
• Output (A3): GFP driven by a promoter that integrates both activation and repression signals, creating an XOR‑like response.

The completed spreadsheet was saved as a CSV and uploaded to the Google Form before the Friday 4pm ET deadline (simulated).

Day 2: OT‑2 Execution (Theoretical Observation)

MIT/Harvard students would go to NE‑47 and observe the OT‑2 building the circuit (assembling plasmids from parts) and transfecting them into HEK293 cells using Lipofectamine 3000. Global students receive videos.

Since we did not attend physically, we watch the theoretical steps:

1. OT‑2 picks up tips, aspirates the correct DNA parts from stock tubes (based on our spreadsheet).
2. It mixes them in the specified wells, creating final plasmid assemblies (Gibson assembly or similar).
3. The assembled plasmids are then complexed with Lipofectamine 3000 in a 96‑well plate.
4. HEK293 cells are added and incubated for 48 hours.
5. Readout: fluorescence microscopy (GFP or mCherry) to measure output.

Expected result (theoretical):

• With both inputs off → GFP off.
• With Tet alone or Lac alone → GFP on (approx. 50–70% of max).
• With both inputs on → GFP off (due to strong repression overcoming activation).
  This confirms XOR behavior.

Post‑Lab (Virtual Reflection)

We did not perform the wet lab, but we understand the principles:

• Sequestron based circuits use dCas9‑effector fusions and guide RNAs. Each gRNA targets a specific DNA sequence. The effector (VPR = activator, KRAB = repressor) modulates transcription. Multiple gRNAs can be expressed from a single transcript using Csy4 endoribonuclease cleavage (allowing analog summing).
• Lipofectamine 3000 works by electrostatic interaction: cationic lipids bind negatively charged DNA, forming lipoplexes that fuse with the cell membrane and release DNA into the cytoplasm. The DNA then enters the nucleus (for transient transfection in HEK293).
• IANNs can implement any function because the weighted sum of inputs (via gRNA concentrations) and nonlinear activation (via effector recruitment) mimics artificial neurons.

If performed physically, our circuit would produce GFP expression only when exactly one input inducer (e.g., anhydrotetracycline and IPTG) is present. This would be quantified by flow cytometry or fluorescence microscopy.

Appendix: Virtual Circuit Simulation (No Physical Run)

We simulated the circuit’s expected behavior using a simple mathematical model (in Python, not executed physically):

• Input A (Tet) → activates Neuron 1 (VPR) and Neuron 2 (KRAB) equally.
• Input B (Lac) → same.
• Output promoter activity = (VPR signal) − (KRAB signal) with a threshold.

The resulting truth table:

This matches XOR. Because only two colors (inputs) were available in the virtual design, we chose this simple classifier instead of a more complex pattern.

Final Note

All work for this lab was completed in silico – the spreadsheet design, simulation of OT‑2 instructions, and theoretical prediction of outcomes. No physical HEK293 cells, transfections, or fluorescence measurements were performed. The report serves as documentation of the virtual assignment.