Alternative Platform

Most therapeutics are molecules. You synthesise them, purify them, formulate them, ship them cold, and give a measured dose at measured intervals. That platform works, and it is the reason modern medicine exists.

It also fails in a specific way, in a specific place. Every step in that chain assumes infrastructure: a manufacturing facility, a cold chain, a distribution network, and a caregiver able to give the right dose on the right schedule. Childhood malnutrition is concentrated in exactly the settings where those assumptions hold least well.

So we chose a different platform. Instead of manufacturing the therapeutic molecule, we engineer an organism that manufactures it continuously, inside the patient, for as long as it is there. Our platform is E. coli Nissle 1917, a bacterium that is already licensed and sold as a probiotic in several countries, engineered to express fatty acid amide hydrolase from the gut commensal Faecalibacterium prausnitzii.

The last two rows are the ones that made the decision for us. Our target, oleoylethanolamide, is a satiety signal produced in the small intestine and acting locally. A drug would have to reach the gut lumen from outside. A gut organism is already there.

The appeal of a living therapeutic is obvious. The reason more teams do not build one is that the engineering is harder than it looks, and the difficulty is not in the enzyme.

The enzyme we want comes from an organism we cannot easily use. F. prausnitzii is one of the most abundant bacteria in a healthy human gut and a sensible evolutionary source for a gut-active enzyme. It is also a strict anaerobe that is notoriously difficult to culture, and it has no established genetic toolkit. Engineering it directly was never realistic for us, and would not be realistic for most teams.

Our answer was to decouple the two. Take the enzyme from the organism that evolved it, and put it in the organism we can actually engineer. That is the core move of this platform, and it is the reason the project exists in a form a student team can execute.

Getting a protein into a bacterium is not the same as getting it to its substrate. This is the problem that shaped our whole season. OEA sits in the gut lumen, outside the cell. An enzyme expressed in the cytoplasm of EcN is separated from its substrate by a membrane it cannot cross. Expression is necessary and nowhere near sufficient.

EcN is not a standard lab strain. It is used because it survives gastric transit and colonizes, which is exactly what a lab strain is engineered not to do. That same robustness means the protocols tuned for cloning strains do not transfer unchanged.

So the platform work split in two, in the same way Brno split theirs: making EcN programmable for this application, and making our workflow predictable enough to be repeated.

The localization toolbox

Rather than guess where the enzyme needed to be, we built a panel that tests the question directly. Every variant carries the same enzyme and the same N-terminal 6xHis tag, so they differ only in where the protein is directed.

Two of these are controls rather than candidates, and we think they are the most important entries in the table. The YebF-GFP construct exists so that secretion can be confirmed by fluorescence independently of whether the enzyme works. The catalytic mutant exists so that any activity we measure can be attributed to FAAH catalysis rather than to a host enzyme or to background chemistry. A localization panel without these two controls produces results that cannot be interpreted.

The cloning route

We assemble into the pEM106.2 backbone by in vivo assembly (IVA) cloning, replacing its YebF-hIL10 cassette with our insert. Backbone and insert are amplified separately with primers carrying 15–20 bp overlapping tails, the parental template is destroyed with DpnI, and the two linear fragments are co-transformed so the cell’s own recombination machinery joins them.

This choice matters for a team without much budget. IVA needs no restriction enzymes, no ligase, and no assembly master mix. The cost of a new construct is the cost of two primers. Over a season in which we built and rebuilt constructs repeatedly, that difference is the one that let us iterate at all.

The backbone contributes the rest of the platform: an inducible promoter driving the insert, chloramphenicol selection, and a secretion carrier already in place to be swapped or kept.

Status

On October 4 we transformed the verified YebF-ΔTm-FAAH construct into EcN as a full team run, and on October 6 we banked the resulting strain as glycerol stocks at −80 °C. That strain is the first version of the platform doing what it was designed to do.

The constructs are the visible output. The harder-won part is knowing how to run the workflow without losing weeks, and that knowledge is what we would most want to hand to the next team.

Selection: use chloramphenicol, not ampicillin

Our first three transformations went onto 100 µg/mL ampicillin. Colonies appeared. None of them established in liquid culture, and we spent two days assuming the bacteria were simply growing slowly before we worked out that they were satellite colonies: cells growing in the zone where secreted beta-lactamase from a genuine transformant has already degraded the ampicillin around it. They look exactly like real colonies on a plate.

Re-plating on 25 µg/mL chloramphenicol resolved it immediately and we never saw the problem again. The cost was a week.

Pick three colonies, not one

Of three sequenced colonies from one 6xHis-tagged clone, one carried a spurious extra base. Picking a single colony would have been a coin flip, and we would not have known.

In the same batch, two constructs turned out to be in each other’s tubes. Both were real and correct, and the error was invisible on a gel because the products were the same size. It was caught only by sequencing. That incident is why we now keep a formal clone tracker and label tubes before filling them.

When optimization stops, cut the band out

One reaction refused to give a clean single product no matter what we did to it. We ran annealing gradients from 55 to 67 °C and raised the cycle count, which improved it but never resolved it. What finally worked was physical: run the gel, excise the correct band, and gel extract it. Yield was low, 8 and 13 ng/µL, but enough.

The judgement call worth passing on is knowing when to stop optimizing a reaction and start purifying its product instead.

Let the enzymes do the ligation

Our longest single problem was a one-nucleotide insertion at a construct junction. We tried three times over eight weeks to remove it by amplifying the whole plasmid and letting E. coli circularize the blunt product during transformation. Every attempt returned either unchanged parent plasmid or an ambiguous base call.

The fix was to stop relying on the cell. Gel extract the product, phosphorylate the 5’ ends with T4 PNK, since PCR products carry no 5’ phosphate and cannot be ligated without it, circularize with T4 DNA ligase, and only then transform. Four of six clones came back correct on the first attempt.

One other thing mattered: DpnI digestion overnight rather than for thirty minutes. A short digest leaves enough parental plasmid to swamp the transformation, and the clones that come back are the ones you started with.

Anyone wanting to put a gut-active enzyme into a gut-resident chassis can start from here rather than from nothing.

• A verified chassis and backbone combination. pEM106.2 in EcN, with working selection conditions and a sequence-verified construct in hand.
• A localization panel design. Five routing strategies plus two controls, built so that variants differ only in localization and can be compared directly.
• A cheap assembly route. IVA cloning, where a new construct costs two primers.
• A troubleshooting record. The four problems above, each with the solution that worked, documented on our Notebook page with the gels and plates.
• A measurement strategy for the enzyme family, described on our Measurement page, designed around a plate reader rather than a mass spectrometer.

The pattern generalises past our enzyme. Many useful activities live in gut commensals that are anaerobic, fastidious, and genetically intractable. Moving the gene into an engineerable gut-resident chassis is a way of accessing those activities without having to make the source organism tractable first.

• We have not shown the enzyme works. We have sequence-verified constructs and confirmed expression by SDS-PAGE in BL21(DE3). We have not confirmed enzyme activity, OEA degradation, or any effect in an organism. The platform is built and not yet validated.
• The localization question is still open. It is the question the panel was designed to answer, and answering it needs the activity assays, which need the catalytic mutant control we have not yet built.
• Expression was tested in BL21(DE3), not in EcN. BL21 is the right strain for expression testing and the wrong strain for the application. Expression levels in EcN may differ and need separate measurement.
• The antibiotic resistance marker has to go. Chloramphenicol resistance on a plasmid in a gut-colonizing organism is acceptable for lab research and unacceptable for anything given to a person, because the gut is an environment where horizontal transfer happens. A translational version needs chromosomal integration and a non-antibiotic maintenance mechanism.
• A colonizing chassis needs biocontainment. The property that makes EcN attractive, that it persists, also means an engineered version would not be self-limiting if released. We discuss this fully on our Safety and Security page. We regard engineered biocontainment as a requirement for any step beyond the bench, not an enhancement.