Week 3 HW: Lab Automation

1. Published Paper Using Opentrons for a Novel Biological Application

Paper: Bryant, J. A., Kellinger, M., Longmire, C., Miller, R., & Wright, R. C. (2023). AssemblyTron: flexible automation of DNA assembly with Opentrons OT-2 lab robots. Synthetic Biology, 8(1), ysac032. https://doi.org/10.1093/synbio/ysac032

What they built

AssemblyTron is an open-source Python package that turns the ~$10k Opentrons OT-2 (with a thermocycler module) into a hands-free DNA-assembly workstation. It plugs into existing assembly-design tools (j5, Cello, Benchling) and executes the resulting build plans directly on the robot, covering three of the most common synbio assembly chemistries:

  1. PCR with optimal annealing-gradient calculation the software computes the best annealing temperature for each fragment from primer Tm and uses the OT-2 + thermocycler to run gradient PCRs across a range of fragment lengths.
  2. Golden Gate assembly Type IIS-enzyme one-pot assemblies of multiple fragments into a destination vector.
  3. Homology-dependent in vivo assembly (IVA) short-overlap fragments co-transformed into E. coli, with assembly happening inside the cell.

What they showed

The authors simultaneously built four different four-fragment chromoprotein reporter plasmids on the OT-2 and showed assembly fidelity comparable to a human doing the same work by hand (verified by sequencing). They also used the same platform for site-directed mutagenesis via homology-dependent IVA, again with manual-equivalent fidelity.

why it counts as “novel biological application”

This is a textbook example of automating the Build step of the Design-Build-Test-Learn (DBTL) loop, which has historically been the slowest and most error-prone manual step. Two things make it novel rather than incremental:

  • It’s the first open-source software package to drive Golden Gate and homology assembly on a low-cost robot, so the price floor for automated cloning drops from ~$100k (commercial systems like Tecan or Hamilton) to ~$10k. That changes who gets to do high-throughput synbio.
  • It directly accepts output from automated design tools (j5, Cello), so you can go from a Cello-designed genetic circuit to physical DNA without a human pipetting step in between. That closes a real gap in the DBTL loop.

Limitations the authors note

  • The OT-2 isn’t as fast or as well-error-handled as a commercial Hamilton STAR.
  • No integrated colony picking, transformation, or QC. The human has to come back in the loop after assembly.
  • Plate-format constraints (96-well bottleneck) limit how parallel things can really get.

2. My Final Project Automation Plan

Project context

My final project builds on Week 2: I want to express a panel of rhodopsin (RHO) variants in a cell-free system to characterize how single-residue substitutions in the chromophore-binding pocket shift the absorption spectrum. The screen compares each variant’s lambda_max under blue/green/red LED illumination. The end goal is a small library of spectrally-tuned opsins for optogenetics, but for HTGAA the deliverable is the screening pipeline itself.

Why this needs automation

A meaningful spectral-tuning library is 50-200 variants, each tested in triplicate, each under at least 3 illumination conditions. That’s 450-1,800 CFPS reactions. Manual pipetting is the wrong tool: error accumulates, reagents drift over an 8-hour day, and you can’t realistically do replicates. Automation is the only way the experiment is actually run, not just designed.

What I would automate

The workflow maps neatly onto the Example 2 cloud-lab pipeline in the assignment, but I would run it on the Opentrons OT-2 + Ginkgo Nebula cloud lab combination:

  1. Design phase (no automation, human + Benchling). Pick residues around the retinal-binding pocket (Lys296, Glu113, and surrounding residues from PDB 1U19), generate variants in silico, codon-optimize with Twist tool, order as a clonal-gene plate from Twist.
  2. Echo acoustic transfer. Echo 525 dispenses the variant plasmid DNA from a source plate into the destination 384-well plate at 50 nL per well, three replicate wells per variant. Acoustic transfer is ideal here because the volumes are small and there’s no cross-contamination.
  3. OT-2 stamps the CFPS master mix. A multichannel pipette on the OT-2 dispenses 18 uL of NMP-Ribose master mix (from Week 11) + lysate into every occupied well of the 384-well plate. This is the step I’d write the Python protocol for.
  4. OT-2 supplements with 11-cis-retinal. Add 1 uL of 100 uM 11-cis-retinal to every well (final 5 uM) so the rhodopsin holoprotein can reconstitute as it’s translated. Light-protected throughout.
  5. PlateLoc seals. Heat-seal the plate to prevent evaporation over the 20 h reaction.
  6. Inheco incubates at 30 C (not 37 C – rhodopsin folds better cooler) for 20 h in the dark.
  7. XPeel removes seal.
  8. PHERAstar reads absorbance spectrum (350-650 nm) for every well under three illumination pulses: blue (470 nm), green (530 nm), red (625 nm). The active rhodopsin shows a characteristic ~498 nm peak that shifts with mutation; bleaching kinetics under each LED give an orthogonal readout.
  9. Data lands in a Jupyter notebook on Ginkgo Nebula, fits each spectrum, extracts lambda_max and bleaching half-life, and outputs a ranked variant table.

Example pseudocode for step 3 (OT-2 protocol skeleton)

from opentrons import protocol_api

metadata = {
    "protocolName": "CFPS Master Mix Stamp - RHO variant screen",
    "author": "rcd, HTGAA 2026",
    "apiLevel": "2.15",
}

def run(protocol: protocol_api.ProtocolContext):
    # Labware
    plate = protocol.load_labware("corning_384_wellplate_112ul_flat", 1)
    mm_reservoir = protocol.load_labware("nest_12_reservoir_15ml", 2)
    retinal_tubes = protocol.load_labware("opentrons_24_tuberack_2ml", 3)
    tips_p20 = protocol.load_labware("opentrons_96_tiprack_20ul", 9)
    tips_p300 = protocol.load_labware("opentrons_96_tiprack_300ul", 10)

    # Pipettes
    p300 = protocol.load_instrument("p300_multi_gen2", "left", tip_racks=[tips_p300])
    p20 = protocol.load_instrument("p20_single_gen2", "right", tip_racks=[tips_p20])

    # Step 3: stamp 18 uL CFPS master mix into every well
    mm = mm_reservoir.wells_by_name()["A1"]
    p300.distribute(
        18,
        mm,
        plate.wells(),
        new_tip="once",
        disposal_volume=2,
    )

    # Step 4: add 1 uL of 11-cis-retinal to every well (light-protected)
    retinal = retinal_tubes.wells_by_name()["A1"]
    for well in plate.wells():
        p20.transfer(
            1,
            retinal,
            well,
            new_tip="always",
            mix_after=(3, 10),
        )

    # Cooling block keeps lysate viable; protocol then hands off to PlateLoc + Inheco
    protocol.comment("Ready for sealing and 20h incubation at 30 C, dark.")

Custom hardware I’d 3D-print

Two pieces I think would be useful enough to design and print:

  1. Light-blocking enclosure for the OT-2 deck during retinal addition. 11-cis-retinal photoisomerizes under ambient light, so the addition step needs to happen under dim red light or in darkness. A black-PLA shell that drops over the deck (with a port for the pipette to enter from above) would solve this.
  2. A 384-well-to-96-well adapter plate for moving samples between Echo-output (384) and downstream PHERAstar reads where 96-well is more convenient. The Opentrons 3D Printing Directory probably already has something close.

Why Ginkgo Nebula vs. local Opentrons

I’d use Ginkgo Nebula for the high-throughput screen because:

  • 50-200 variants in triplicate exceeds what I can realistically QC on a single OT-2.
  • The cloud lab already has the Echo, PlateLoc, and PHERAstar integrated. On the local OT-2 those steps would need manual handoffs.
  • Reproducibility: the protocol file is the experiment. Someone in Berlin or Shanghai can re-run my best variant verbatim.

I’d use a local OT-2 for the design-iteration phase (10-20 variants, debugging the master mix recipe, getting the retinal-addition step working) because the round-trip time on a cloud lab is too slow for that loop.

Risk and what could go wrong

  • Cell-free yield drops at scale. What works in 20 uL in a tube may not in 18 uL in a 384-well plate with a higher surface-to-volume ratio (faster evaporation, more O2 depletion). Mitigation: pilot on 96-well first, optimize seal + headspace.
  • 11-cis-retinal is photosensitive and expensive. Aliquot under red light, work fast, and consider all-trans-retinal + retinal-isomerase regeneration as a backup.
  • Variant DNA from Twist arrives at different concentrations. Normalize on the Echo or with an OT-2 normalization step before the screen.
  • Spectral readout on PHERAstar. A microplate reader is not a true spectrophotometer; for the cleanest spectra I’d want a SpectraMax or similar. Mitigation: use the PHERAstar for screening, then confirm top hits on a benchtop spectrometer.