Week 1 HW: Principles & Practices

Project: Physarum-on-a-Chip Environmental Sensor

The tool I want to develop is a Physarum-on-a-Chip environmental sensor – a microfluidic device that confines the plasmodium of Physarum polycephalum (slime mold!!) within a controlled chemotactic gradient array, and reads out the organism’s foraging behavior as a chemical-environment signal.

Why Physarum

Physarum is a single multinucleated cell that solves problems no single cell “should” be able to solve. With no neurons and no central controller, it:

• Finds shortest paths through mazes between food sources (Nakagaki et al., Nature, 2000).
• Recapitulates the Tokyo rail network when offered food at the cities surrounding Tokyo (Tero et al., Science, 2010).
• Remembers where it has been. Even without a nervous system, it lays down a trail of extracellular slime and avoids re-exploring already-visited areas. Reid et al. (PNAS, 2012) showed this functions as a kind of externalized spatial memory – the organism offloads its memory into the environment.
• Anticipates periodic stimuli. Saigusa et al. (Phys. Rev. Lett., 2008) showed Physarum slows down in anticipation of regular cold pulses, then re-anticipates after the stimulus stops – a form of habituated learning without synapses.

This is bio-intelligence without a brain. The intracellular pathways are doing the work: oscillating cytoplasmic streaming, calcium waves, actomyosin contractions, and reaction-diffusion dynamics in the cell. The whole organism is a wet, living analog computer.

What I want to build

A device with these layers:

1. A microfluidic chip with PDMS channels patterned as a 2D array of “chambers” connected by narrow passages. Each chamber can be loaded with a chemoattractant (oat flake extract, glucose) or a chemorepellent (light, salt, quinine, or a target environmental contaminant – heavy metals, pesticide residue, microplastic extract).
2. A Physarum plasmodium introduced at a central inoculation chamber. It explores the array, makes routing decisions, and lays down its slime trail.
3. A camera + time-lapse readout that records the network topology over hours. Image analysis converts the plasmodium’s tube network into a graph – nodes, edges, weights.
4. A signal interpretation layer. The pattern of which chambers Physarum colonizes, which it avoids, and how fast it gets there encodes information about the chemical environment. A trained Physarum (one that has previously encountered a contaminant and “learned” to avoid it) gives a different network than a naive one.

Why I find this exciting

Three reasons:

1. The memory question. How does an organism without neurons remember a route? The extracellular slime hypothesis is elegant but probably not the whole story; intracellular calcium oscillations and tube-diameter hysteresis also encode state. Building a controlled platform lets me actually test which mechanism dominates in different conditions.
2. Bio-intelligence as an alternative paradigm. Most “intelligence” we build is digital and silicon. Physarum is a counter-example – distributed, analog, embodied, and runs on oatmeal. If the next wave of computing is going to be biological or neuromorphic, slime mold is a useful reference organism for what computation without a CPU even looks like.
3. The sensor application is genuinely useful. A Physarum-on-a-Chip in a riverbank or wastewater stream could integrate over many chemical signals at once and give a single read-out – “this water is unusual” – in a way that a stack of individual electrochemical sensors cannot. It’s an integrator, not just a detector.

Class Assignment: Governance & Ethics

Step 2: Governance / policy goals

Because this tool integrates a living organism into computational and sensing infrastructure, ethical development requires attention to four areas: lab safety, ecological responsibility, transparency / scientific honesty, and equitable access.

Goal A: Foster lab safety

• A1: Ensure safe handling of Physarum polycephalum, which is BSL-1 (non-pathogenic in healthy humans) but can still trigger allergic responses to its spores and is a mild contamination risk in shared lab spaces.
• A2: Standardize protocols for the microfluidic device fabrication (PDMS curing, plasma bonding, solvent handling) so the chip-making process is no more hazardous than the organism it contains.

Goal B: Protect the environment

• B1: Prevent ecological release of the cultured Physarum strain. P. polycephalum itself is cosmopolitan, but lab strains have been selected for fast growth on agar – a fitness profile that may differ from wild populations.
• B2: Prevent contamination of test water/soil samples after they have been incubated with the device. If a sensor is used in the field, the post-assay sample must be inactivated before disposal.
• B3: Ensure environmental sensor readouts are truthful and reproducible. A false-negative reading on a contaminated water source is a real harm; a false-positive triggers expensive intervention.

Goal C: Promote transparency and scientific integrity

• C1: Avoid overclaiming “intelligence” or “cognition” in slime mold. The science is genuinely fascinating, but the popular framing tends to drift into anthropomorphism that is bad both for public understanding and for the organism’s welfare framing.
• C2: Open data, open protocols. If a sensor’s output depends on a proprietary trained Physarum strain, the result isn’t reproducible.

Goal D: Promote equity and constructive use

• D1: Keep the technology low-cost. The whole point of a slime-mold sensor is that it runs on oats and tap water – this should be accessible to community labs, smallholder farmers, and schools.
• D2: Open educational use. Physarum is one of the best teaching organisms for distributed computation; the chip platform should be usable in undergraduate and high-school labs.

A note on a question that doesn’t fit cleanly in the four-bucket framework: does a slime mold have welfare interests? I think the honest answer is “probably not in any morally weighty sense, but the question deserves to be open.” For governance purposes I treat Physarum as a non-sentient living system that nonetheless deserves the same baseline respect as other model organisms.

Step 3: Three governance actions

Option 1: BSL-1+ handling protocol for engineered/selected microbial sensors (technical strategy + new rule)

Option 2: Open data + reproducibility standard for bio-sensor readouts (incentive + technical strategy)

Option 3: Language and framing guidelines for “bio-intelligence” (governance + norms)

Step 4: Scoring against the rubric

(1 = strongly does it, 2 = somewhat, 3 = does not, n/a = not applicable)

Step 5: Recommendation

I would prioritize a combination of all three, weighted toward Options 1 and 2, with Option 3 as a low-cost cultural overlay.

• Option 1 (BSL-1+ handling) is the highest-impact, lowest-cost safety measure for living-sensor deployments. It addresses the real but currently unregulated risk of releasing lab-selected microbial strains in the field. The compliance burden is a checklist, not new equipment.
• Option 2 (open-data standard) addresses the reproducibility crisis specific to living-sensor results – a real concern because Physarum behavior is sensitive to strain history, temperature, and food state, and “it worked in my lab” is not enough. Open data is also the precondition for equity: smallholder users need replicable protocols, not magic strains.
• Option 3 (framing guidelines) is the cheapest of the three and addresses a problem most safety/biosecurity frameworks miss entirely – that scientific overclaiming is itself a kind of harm, both to public understanding and to long-term research credibility.

Trade-offs:

• Adding a BSL-1+ tier risks regulatory creep – the same logic could be used to over-regulate other benign citizen-science activities. Mitigation: the tier triggers only on out-of-lab deployment, not on lab work.
• Open data standards favor well-funded labs that can produce clean, depositable datasets. Mitigation: provide deposit infrastructure (NSF-funded repository) and accept “rough” data formats for community-lab submissions.
• Framing guidelines can become language policing. Mitigation: the document is short, voluntary, and explicitly preserves the right to discuss genuine open questions about non-neural cognition.

Audience for this recommendation:

• For Option 1: the CDC/NIH BMBL committee (the formal home of BSL guidelines) and the EPA Office of Pesticide Programs (for the field-deployment permit hook).
• For Option 2: NSF Division of Environmental Biology and the Open Source Hardware Association.
• For Option 3: the iGEM Foundation, the Physarum unconventional-computing community (the small annual workshops), and journal editors at PNAS / eLife.

Reflection – ethical concerns this week

Three things stood out:

1. The “is this organism deserving of moral consideration” question is not zero, even for slime mold. I’m comfortable saying Physarum has no welfare interests in the morally weighty sense, but I notice that I’m comfortable with that partly because of how I was trained to think about single-celled organisms. As bio-intelligence research advances, the categorization is going to shift.
2. Overclaiming is a quiet ethical issue. Most biosafety frameworks ignore it because it’s not a physical risk. But scientific overclaim – “Physarum is intelligent!” – erodes public trust in the same way physical incidents do, just slower and harder to attribute.
3. The dual-use question for sensors. A Physarum-on-a-Chip that detects pesticide residues can also detect pharmaceutical metabolites in wastewater, which is one step from population-level surveillance. The same chip, deployed by the wrong actor, becomes a surveillance tool. The platform is dual-use even when the organism is benign.

Week 2 Lecture Prep

Prof. Jacobson’s questions

Q1. Polymerase error rate vs. human genome length.

DNA polymerase alone has a base-misincorporation rate of roughly 1 in 10^5 (1 error per 100,000 bases) from intrinsic nucleotide-selectivity alone. With built-in 3’ -> 5’ exonuclease proofreading, the error rate drops to about 1 in 10^7. Then post-replication mismatch repair (MMR) – MutS/MutL in bacteria, MSH/MLH homologs in eukaryotes – catches most of the rest, bringing the final error rate to about 1 in 10^{9 to 10}10 per base per replication.

The human genome is ~6 x 10^9 bp per diploid cell. If we used raw polymerase fidelity (10^{-5), every cell division would introduce ~60,000 errors. With proofreading only (10}-7), still ~600 errors. With proofreading + MMR (10^-9), it’s about 0.6 errors per genome duplication on average – so most divisions are error-free, with the occasional one slipping through.

Biology deals with the discrepancy by stacking three independent layers of error correction, each catching ~99-99.9% of errors the previous missed. Fidelity is multiplicative. On top of that, biology tolerates some residual error rate because (a) most of the genome is non-coding and tolerant to single-base changes, (b) diploidy means a hit on one copy is usually backed up by the other, and (c) the residual error rate is the substrate for evolution.

Analogy: it’s like a camera with three layers of stabilization – in-body sensor shift, in-lens optical, and software post-stabilization. Each fixes a different scale of shake. The combination yields a sharp image even from a moving handheld shot; none of the three alone would be enough.

Q2. How many ways to code an average human protein – and why most don’t work.

The genetic code is degenerate: 64 codons code for 20 amino acids + stop. Most amino acids have multiple codons (Leu, Arg, Ser have 6 each; Met and Trp have only 1).

For an average human protein (~375 amino acids), the number of synonymous DNA sequences is the product of codon counts over each residue. With an average of ~3 codons per residue, the number is approximately 3^{375 ~ 10}179 synonymous coding sequences – vastly larger than the number of atoms in the observable universe (~10^80).

Why most of those don’t express well in practice:

1. Codon usage bias. Each organism has preferred codons matched to its tRNA pool. Rare codons (e.g., AGG/AGA Arg in E. coli) cause ribosome stalling and truncated products.
2. mRNA secondary structure. Some codon choices fold the mRNA into hairpins that block ribosome scanning, especially near the 5’ UTR / start codon.
3. GC content. Extreme high or low GC affects mRNA stability and transcription.
4. Hidden regulatory elements. Synonymous changes can create or destroy splice sites, miRNA targets, internal Shine-Dalgarno-like sequences, or polyadenylation signals.
5. Restriction sites. Sequences containing BsaI/BsmBI/EcoRI break downstream cloning workflows.
6. Repeats and homopolymers. Long stretches of one base, or large direct repeats, are hard to synthesize and prone to recombination.
7. Translation kinetics matter for folding. Some proteins fold co-translationally; the speed of translation through certain regions matters. Optimizing every codon to “fastest” can paradoxically misfold the protein.

This is exactly why codon optimization tools (Twist, IDT, GenScript) exist – to navigate the 10^179 sequence space toward sequences that actually express in the chosen host.

Dr. Leproust’s questions

Q1. Most common oligo synthesis method: Phosphoramidite chemistry, developed by Caruthers and Beaucage (1981) and still the workhorse. A 4-step cycle adds one nucleotide at a time to a growing chain on a solid support (CPG bead or microarray chip): detritylation -> coupling -> capping -> oxidation. Repeat per base.

Q2. Why 200 nt is the practical limit for direct synthesis.

Two reasons, compounding:

(a) Coupling efficiency compounds geometrically. Even at 99.5% per cycle (very good), the yield of full-length product is 0.995^N. For N=200, that’s ~37%. At 99% efficiency, ~13%. At 98%, ~2%. Every length doubling cuts the full-length fraction sharply.

(b) Depurination is the hard wall. The mild acid used in detritylation (dichloroacetic or trichloroacetic acid) cleaves purine bases (A and G) from the sugar at a low but non-zero rate per cycle. Every cycle adds another exposure. By ~150-200 nt, depurination produces enough abasic sites that the full-length fraction collapses regardless of coupling chemistry. Agilent’s published work on 150mer libraries was a depurination-control breakthrough; getting much past 200 nt with conventional phosphoramidite remains hard.

A third practical reason: side products (n+1, n-1 deletions, GG dimers from dG re-coupling) accumulate, making purification harder for long oligos.

Q3. Why you can’t make a 2000 bp gene by direct oligo synthesis.

Combining Q2: at 99.5% per step, a 2000-nt direct synthesis would yield 0.995^1999 ~ 0.005%, essentially zero. Depurination would have destroyed most molecules long before. No production chemistry can synthesize a 2 kb oligo as a single molecule.

In practice, 2 kb genes are built by assembly: synthesize ~200 nt oligos that overlap each other, then stitch them via PCR-based methods (polymerase cycling assembly, Gibson, Golden Gate) into the full-length gene. Twist, IDT, and Genscript all use this hierarchical approach. Newer enzymatic synthesis approaches (Ansa, DNA Script) aim to break through the length barrier by avoiding the acid detritylation step.

Prof. Church’s question

Choice: Q1 – The 10 essential amino acids and the Lysine Contingency.

The 10 essential amino acids in animals (cannot be synthesized de novo, must come from diet):

1. Histidine (H)
2. Isoleucine (I)
3. Leucine (L)
4. Lysine (K)
5. Methionine (M)
6. Phenylalanine (F)
7. Threonine (T)
8. Tryptophan (W)
9. Valine (V)
10. Arginine (R) – essential in juveniles, conditionally essential in adults

(Mnemonic: PVT TIM HALL.)

What this implies about the Lysine Contingency (the Jurassic Park plot device where dinosaurs are engineered to require dietary lysine, so they die without humans feeding them):

The premise is scientifically incoherent on its own terms. Lysine is already essential in all animals – the engineered dinosaurs, like every other animal, would already be unable to synthesize lysine. They would already need to get it from their diet (meat, plants, anything containing protein). The “contingency” only works if you add a new dependency on something that doesn’t exist in their food chain: a non-natural amino acid, or a vitamin/metabolite the engineered organism can’t get from any natural source. What Crichton called a lysine contingency is actually a generic essential-amino-acid contingency, and lysine is the worst possible choice because it is abundantly available in any meat or legume the animals would naturally eat.

My view of this as bioconfinement: the principle is good – engineer a metabolic dependency that doesn’t exist in nature – but the dependency has to be chosen carefully. Real biotech implementations (e.g., E. coli strains dependent on non-canonical amino acids via expanded genetic code, or auxotrophic strains requiring synthetic ligands) work because the supplemented molecule is not found in nature, not just because it’s nominally “essential.” This actually connects to my Physarum project: any future engineered Physarum strain deployed in the field could be made dependent on a synthetic small molecule that doesn’t occur in soil or water, so that escape into the environment is self-limiting.

Citations: Standard biochemistry references (Lehninger, Berg’s Biochemistry) for the amino acid list. Crichton, Jurassic Park (1990), for the original framing. No AI prompts used.

Lab Preparation – Pipetting

• Completed in-person. I LUV Pipetting as a Biologist <3
• Tried to finish both certifications. Not sure if one went through.