The Method

I take a small amount (10uL) of a slurry of diatomaceous earth (DE), add some (10uL, 50mM) calcium chloride and some (~1ug, in ~2uL) plasmid DNA and place it on a piece of parafilm stretched over a 3D-printed tube. I dry it in a filtered stream of air, then the tube adapts to the end of my cheap airsoft pistol, and I fire it at a leaf from a few cm distance.

The DNA I have codes for the ‘RUBY’ reporter, which expresses a red pigment. I extracted the DNA from bacteria carrying this plasmid which I got from Sebastian’s Biotech Bazaar. I used this miniprep kit to purify out the plasmid DNA.

Findings

I shot some tobacco leaves, some duckweed, a slice of carrot and some onion… In the leaves, around the spots shot with DNA (but not the controls) there are some, well, reddish bits that show up after a few days! And after lots of squinting at them I’ve convinced myself that at least some of these are cells expressing RUBY :)

My gallery is filled with blurry attempts to capture this, and none make it as obvious as I’d like. For example, there are also plenty of cases where, around an impact point where a clump of DE has hit, there’ll be some orange-ish crud welling up. Around most impacts (and all that I saw in the no-DNA controls) this stuff is pale white. Could this be RUBY too, expressed in some damaged cell(s) and being included in whatever is happening around the hole? Plus there are random red things just around the place - bits of fiber, random flakes - here’s a gallery of things I was more sceptical of:

The almost magenta dots in the carrot are apparently pretty much what RUBY looks like in carrot, that was a YOLO shot with some leftover stuff, I’ll do some follow-on ones with carrot callus that I have growing and hopefully get a clearer signal. But in general it was much harder looking for red/orange reporter in orange carrot tissue!

I didn’t see anything in the onion I shot. Nor did I see anything in onion that I tried my micro-needle idea on. But then, RUBY doesn’t work everywhere - for example this paper documents it being a poor marker in coconut (also a good ref for optimization strat once I can easily bombard and measure.)

For comparison: I’ve also been trying transformation with agrobacterium, which gives a much more macro result - here are some duckweed fronds expressing RUBY to various degrees:

Thoughts, What didn’t work, What Next

What didn’t work:

• Using graphite as the carrier (I wanted something more visible than the DE, so I sanded some graphite rod to get powder, but I didn’t ever see transformation using it)
• Using my tungsten micro-needles seemed to kill leaf cells, and when I tried them more gently on onion I didn’t see any transformation
• Bombarding duckweed was tricky (it tended to fly away in the blast) and I didn’t see any transformation of its tiny cells.

I think the red spots are enough to indicate that we’ve gone from 0 to 1. From here, we can optimise - playing with

• DNA, carrier and buffer amounts
• Nozzle diameter and geometry (I already switched to a converging-diverging nozzle which seems to give a much nicer shot). Ref for flow guiding barrel design (just found)
• Shot distance
• etc.

But I’m not happy with RUBY as a readout - too hard to see and measure. I’d love to use a flourescent reporter, but after months of trying I wasn’t able to get my hands on some. I have a construct with a fluorescent reporter ready to go for another project, which I’ll be ordering soon, for about $500… Once I get that I’m hoping we’ll be able to easily quantify how well a given shot does, and start optimizing everything. I wish individuals could order plasmids from addgene! Anyway.

Once you have some transformed cells, you can try to grow them out in tissue culture, optionally with something like herbicide resistance to select for the trasnformed ones. I’m hoping we can find a visual-selection-only flow since I don’t want to me making herbicide resistant plants or having other people work with them.

My dream is to work out a path for people who haven’t done this before, and don’t have the tools to do assembly and cloning, to still order a construct with their own protein/design, and get it into a plant, all as easily as possible. Wish me luck :)

Sequencing with Plasmidsaurus

I went with Plasmidsaurus’ ‘Standard Bacterial Genome Sequencing with Extraction’ service. I sent the sample (~15mg cells suspended in Zymo DNA Shield) off on Tuesday morning and by Wednesday evening the results were ready.

They give you lots of data, including annotating the genome for you with bakta. Great stuff! We can see right away that our guess was right - this is a Pseudomonas species (P. putida is the closest match) and are ready to dig in further to see what we can learn.

Pyoverdine Investigation

Pyoverdines vary but all come with three key parts: a dihydroxyquinoline core, a peptide chain, and a side chain. The peptide chain especially varies strain-to-strain - see here for some examples.

Step one was looking for the pyoverdine Biosynthetic Gene Cluster (BGC). A Deep Research run gave some promising genes to start the hunt with. Some hits let us narrow in on a key region between 4.71 and 4.76M bp, with a number of promising-looking genes. A tool called antiSMASH likewise identified a region (4,696,705–4,796,912), labeled as ‘NRP-metallophore’ - this is our BGC alright :)

By looking at the A-domain predictions for this region, we can start on a predicted peptide sequence: Glu-D-Tyr-Dab─Asp─Ala─Asp─D-OHOrn─X─OHOrn (where Glu-D-Tyr-Dab forms the conserved pyoverdine chromophore after cyclization). That X is an unknown, which we’ll have to work on later - my best guess was Gly, although a later lit search makes me think Ser (see later update section).

The next step is to figure out which side chain is present. (Some strains make some of both). The side-chain type is determined by which gene is present:

• pvdN (PLP-dependent aminotransferase) → succinamide side chain
• ptaA (periplasmic transaminase) → α-ketoglutarate side chain

We BLAST the genome against known reference sequences, finding an 83% match to PvdN (NP_251095.1) but only a 23% match for PtaA (AAY94407.1) -> fair guess that this strain produces succinamide side chain.

This info gives us enough info to reasonably guess at the structure of the pyoverdine precursor (ferribactin):

Here’s a notebook running over the steps, with some light annotation. (I’ve since been fiddling a bit more, LMK if you’re wanting to dive into this deeply)

Future Plans

It’s one thing to poke at a genome and predict a structure, it’s another entirely to verify it. I’m hoping to work with a local university to run LC-MS on this to answer a few remaining questions and nail down the final structure exactly. Will require dusting off and improving my organic chemistry and learning a lot more! Stay tuned for that in some future blog post hopefully :)

Also, it’s probably obvious but worth stating explicityl: I AM OUT OF MY DEPTH HERE AND EVERYTHING IN THIS POST SHOULD BE TAKEN WITH A GRAIN OF SALT, ESPECIALLY WHILE IT IS MARKED ‘DRAFT’ :)

For the curious, raw data is on Google Drive, I’m open to questions @johnowhitaker. This solveit dialog has the key code in a nice rendered form.

Update: A Potential Match

In A combinatorial approach to the structure elucidation of a pyoverdine siderophore produced by a Pseudomonas putida isolate and the use of pyoverdine as a taxonomic marker for typing P. putida subspecies BioMetals, 2013 I found the following: P. putida BTP1/90-40 Asp–Ala–Asp–AOHOrn–Ser–cOHOrn (citing Jacques et al. (1995)). This is pretty much spot on for ours, and would mean the mystery X is Ser (serine), not Gly as guessed. It also answers a mystery - I’d been searching fruitlessly for pvdF in my genome to figure out if OHOrn gets formylated, this would indicate it doesn’t in this case.

PS: Molecule viewer test

Here’s the predicted ferrobactin precursor molecule, copied from the output of this code:

    <canvas id="undefined" width="1000" height="750" style="width: 400px; height: 300px; padding: 0px; position: absolute; top: 0px; left: 0px; z-index: 0;"></canvas><canvas id="undefined" width="1000" height="750" style="width: 400px; height: 300px; padding: 0px; position: absolute; top: 0px; left: 0px; z-index: 0;"></canvas></div>

CNC Pipette

I’ve enjoyed hearing Jason Kelly’s take on why bio should abandon the lab bench and let automated labs (like the ones his company Ginkgo makes) do the work of moving liquids and things around, freeing up the smart minds to actually think about cool experiments rather than spending all their time on menial lab tasks. Sounds good, although I simultaneously think far more people should do a little lab stuff to open their minds to the realities of the real world, and worry how much might get missed if you only see what you set out to measure!

Anyway, with all that discussion fresh in my mind, one of the things I did this weekend was whip up an attachment for my 3D printer that holds my pipette and can suck up and dispense small amounts of liquid. End goal: agar art with colorful microbes, of course :D A Dynamixel servo that I had on hand pushes the plunger thingee on the pipette, with both it and the printer controlled by a raspberry pi or laptop (the latter after I ran the wrong G-Code and spilled colorful water on the former!).

The CAD and code are still WIP, contact me if you’re interested in this or wait until the next weekend I feel like tidying it up + making a more thorough video documenting the project. My favourite result so far comes from the first test, before I dialed in liquid dispensing:

I’ve since done some agar art with my pet blue fluorescent Pseudomonas sp. that is growing out nicely - will upload pics once I make a few more artworks.

Update: I made a short with it in action: https://www.youtube.com/shorts/He5VJz9E1lQ

Update 2: Here’s an animation made with fluorescent dye

Biolistics - ASGG tests

I haven’t managed to find a way to get genes from addgene. A friend has kindly stepped in to help, and I might also cold-email some local profs in case any want to collaborate. But that means than for now, I don’t really have a way to test the gene gun idea and won’t for a few more weeks at least. Still, I did muck about a bit, starting with: carriers and DNA adhesion.

In regular gene guns, DNA is precipitated onto gold nanoparticles, which are then washed in ethanol and then mixed with something called PVP, which let’s them then stick to plastic tubes or disks. The result is an even spread of DNA-coated gold balls. I’m planning to try a few things differently. One carrier I’m looking at is diatomaceous earth - sharpish fragments about the right size, high surface area. I’ve been putting little droplets of DE slurry on film (cling film or parafilm) and firing them without even waiting for the droplet to dry.

One test was thus, can we stick DNA to DE? Both are ~slightly negatively charged, so I tried one test with DE slurry + plasmid soln (my ~failed BLY_GREEN) w/ methylene blue to stain the DNA + water, and a second with the same but 50mM CaCL2 solution instead of water. The CaCl2 contributes +ve calcium ions which might (the theory goes) form a bridge. Indeed, it seemed like the DE that settled out in the CaCl2 version was bluer than the whiteish version with just water. (A better test would be to use a better DNA stain to check the water bit to see how much DNA is left). In another test, DE mixed with DNA + MB both with and without CaCl2 was added to lots more water and then spun down in a diy centrifuge - the pellet in both was blue, but the water-only pellet rapidly disperse on handling while the CaCl2 version was stable - my suspicion is that it causes clumping (pro gene gun prep involves sonicating to break up clumps). If that’s the case, then skipping the CaCl2 might be better - the DNA is still likely to coat the carrier if e.g. we let a drop evaporate and it is left with tons of surface area of silica DE vs a relatively small area of hydrophobic cling-film. More testing to do :)

I also explored penetration. My first attempt just had a wide attachment after the barrel, with film stretched across it. You had to get close to get much penetration, which also seemed to cause damage to the agar at least (using it as a transparent stand-in for plant matter). Testing it on onion it mostly penetrated only the 1st layer of cells. So, I modelled up a converging-diverging nozzle, with the film placed in the throat, and got much nicer results - less damage from air, more even penetration of the (presumably faster-moving) particles. In onion it sent some particles 5-10 cells deep on the first try :)

Biolistics - W wire tatoo approach

I did more tests with my mini tatoo gun concept, using a bundle of flame-sharpened tungsten wires (50-micron) to poke at an onion with some MB as a dye just to see what happens. The dye wipes away except where the wires (attached to an electric toothbrush) stabby-stabby stabbed the tissue.

The wire is perhaps too flexible, and rapidly gets bent without penetrating much beyond the first layer - perhaps I should have started from thicker wire. Still, with some fiddling it seemed to eventually yield a setup that got some poki-ness happening. We’ll see how it goes when I can try it with some actual DNA.

Tissue Culture

As for what we’ll actually transform, and how we’ll grow out plants from a few transformed cells, I’ve been exploring tissue culture a little. Most of my experiments so far ended in contamination - some from an early media test just heating the media up to 96C rather than microwaving or autoclaving (which left one or two live spores or something per tub, nothing if you’re doing a petri dish of bacteria for a few days, a dead pain if you’re wanting to grow a piece of plant matter for months) but most from the explants themselves; it turns out you need really aggressive sterilization with bleach etc to kill everything but the plant tissue (and in the process you usually do kill a fair bit of the plant tissue too). Might try some other options soon.

Anyway, now that I’ve switched to media sterilized in an Instapot and started being more aggressive with the explant sterilization, I feel like I’m slowly taming the beast. It’s all super slow though - aaaah my poor software brain is spoiled from same-second results :D

Glowing Gelatinous Blobs

I grow some of the Pseudomonad I isolated in ~500ml of ‘marmite broth’ in a flask, with parafilm cover and occasional stirring. Then I added 1g CaCl2 to it, and dripped it into some water with sodium alginate - something the molacular gastronomy folks call ‘reverse spherification’. The result is gooey blobs, which (thanks to all the pyoverdine the bacteria made) glow bright blue under UV light. I pictured cool fluorescent spheres I could sprinkle around the base of our plants, to provide them with a bacterial iron boost (but mostly for aesthetics). What I got was more of a congealed mess, but it was fun nonetheless.

Handling the containers (post boiling + alcohol steriliation) stained my hands a lovely fluorescent yellow-green, some cool modification of the blue-fluorescing molecule(s) this makes. Neat! Invisible under regular light, and now mostly gone after a day of occasional washes and a shower.

Misc Updates

• Ran some Thai Tea concentrate (low C) and Puerh (very high C for tea) through caffeine TLC process

• Tried friction welding 3D prints with a dremel, cool technique!

• Made a batch of beer from a kit - turned out extremely well, fermentation was stalled in my cold basement until I warmed it to 22C. Friends all agreed it was tasty, but none of us really like beer so I threw out most of it - sorry! Now the brew tank is destined for more exotic projects :D

OK that’s about enough typing for now, apoligies for a lower-effort post but I figure better to document some stuff now rather than try to remember it all in months time when I start getting more final results on things.

Background

There are various ways of getting DNA into organisms. For some bacteria, you can use heat shock - this is what I’ve done for my engineering of E. coli. For many cells, ‘electroporation’ is an option - basically, zap them and the DNA will get in by magic (but only into a small fraction of cells). (I think a piezo sparker from a BBQ lighter might do the trick TBH - future project right there). For plants, scientists sometimes use a bacterium called Agrobacterium tumefaciens that naturally infects plants and inserts its DNA into the plant genome. (Evolved genetic engineering! How cool!!). But it doesn’t work on everything, and also some places are touchy about you using a plant pathogen to do stuff, even if it’s found everywhere and also the lab versions have all the nasty bits removed. Anyway, so there are downsides to all of these methods.

An older technique that still gets used a lot, because it has a different set of pros and cons, is biolistics - shooting tiny particles coated in DNA into cells. This is done with a ‘gene gun’ that uses high-pressure gas to accelerate the particles to high speed, so they can penetrate cell walls and membranes. The particles are usually made of gold or tungsten, because they’re dense and inert. The DNA is stuck to these particles using calcium chloride and spermidine (salmon sperm). Machines cost tens of thousands of dollars, and are high-pressure and scary. They do a lot of damage to the tissue, and need fiddling to get the right settings. So people don’t use them much. Also you’re literally shooting gold, and everything involved is expensive. The disks (just pieces of kapton AFAICT) are expensive even.

The Idea

Anyway, as I said in the intro, why not leverage an airsoft gun to DIY one? So I handed papa Bezos $37 dollars for the ‘Elite ForceCombat Zone Enforcer Clear’ - a nice simple pistol (no blowback, so >100 shots per 12g CO2 cartridge). I 3D printed a little adapter that press-fits onto the front, with a way to press on a second part and twist to lock it on. Cad model here.

I don’t want to spend $$$ on facy metal powder and salmon sperm, so I’d like to experiment with different carriers. One idea I had is to use diatomaceous earth (DE) - a bunch of spiky silica shells about the right size, since I found a tub of it in the basement. Having a quick look around, seems like others have tried a few random things like clay with some success. I do have some spare plasmid but no good way to quantify how much that DNA sticks to a given carrier, so I might wait til later for that, or try some runs first in case it all magically works first time (unlikely).

Testing

Anyway, to visualize the spread I added a pinch of potassium permangate to some cling film, chucked it up (I’m smug everything fit perfectly first time) and fired it at some innocent MS media I had sitting nearby from some tissue culture work (TODO, blog posts about that). The result is a perfect ‘real life data viz’ example:

The idea is that in the center it might be too much and blast apart the cells, and at the edges too little, but in the sweet spot you might get transformations. And then the idea is you’ll select the transformed tissue and grow it out (e.g. with antibiotic that your plasmid confers resistance to, or with a visual marker like GFP).

I also tried adding some DE to Methylene blue dye (since it’s hard to see white on white) and fired it into a paper towel. It all looks promising to me - although I have no reference for what a ‘good’ gene gun shot looks like haha. It took 10 minutes to model and print the adapter though, so I have plenty of variables to play with. Distance to target, adding a diverging nozzle, different carriers, films (I used a 5uL drop of dye+DE on clingfilm for the test), etc etc. It’s promising that the liquid atomized nicely - others seem to use the solid particle carriers and dry them off. Dry would be good in some ways (‘bullets’ such as tubes with the pre-dried carriers are said to last months once prepped) but if e.g. I could add a few uL plasmid to some carrier and blast that right away it would save a lot of work.

Anyway, until I get plasmid and try this out in plant tissue, I should probably slow down on this. Anyone with an addgene account who feels like buying me a plasmid LMK ;) I’m also hoping to chat to someone I know who is far more knowledgeable about all this and actually has a gene gun, but that’ll be in a few weeks time.

PS: The Other Idea

For the curious, my other idea is plant micro-accupuncture. I got some 50-micron tungsten wire, and figured out how to sharpen it in a flame (inspired by this video on DIY gecko tape) - I figure I’ll make a bundle of these sharp micro-needles, lay down a layer of DNA over the plant tissue, and go to town. No idea if it’ll work, but if it does it’ll be a cool way to do localised transformations without all the damage of a gene gun.

Amazing how sharp you can get it, this is at 200X magnification. Reference is carrot cells (I’m working carrot callus culture as a nice model system for plant transformation).

Anyway, very much early days but I thought rather write these ideas down and get them out there, and then when I get results (could take a while) I can at least refer to part 1 and save myself some typing :)

There’s a joy to doing something like this. I know there are smart people who’ve done tons of research, but I also think the incentices in academia etc aren’t set up to tinker. You have a fancy gene gun, why work on a cheap one? Gold works, why swap? After all, you want to do your transform and write your paper and get your PHD. But for me, if we can lower the barrier to entry all the way, way more people can try stuff and maybe find new things. So, I have looked stuff up here and there, but I’m trying to also keep my outsider status as long as possible so I don’t get too stuck in “the way it’s always been done”. You,dear reader, will get to see if this approach works. There’s no losing though! Worst case, well, I got an airsoft gun, and may have already scattered some BBs around the basement with a big grin on my face :)

PS: This is an amusing area to try to use AI with - they’re all completely uninterested in assisting with ‘building a projectile weapon’ and will be vague to the point of uselessness on many related questions. More fun thinking for me :)

PPS: Relevant XKCD: https://xkcd.com/1217/

PPPS: I made a bundle of sharpened wires attached to an electric toothbrush, and it seemed to ~successfully tatoo MB into onion skin, which is promising. When they get all bent up you can re-sharpen in the flame. Penetration depth not excellent, stiffer thicker wire might have been a good idea.

Inspiration

I want to quantify an automated pipetting system using a precision scale, which required reading off the weight based on a photo of the scale. It seemed like the kind of simple thing that might work with a local VLM on my mac, and I wanted an excuse to play with LFM 2.5 1.6B that was released that day. So I snapped a few pics of my scale showing different weights, in different lighting and angles, and tried moondream v3 preview, zai-org/glm-4.6v-flash, and LFM 2.5 1.6B (code). I was a little suprised how bad they are! LFM was fast enough that I fiddled with the prompt a couple of times, and felt that if I kept the lighting constant and didn’t need fancy things like sign, it might mostly be OK with the occasional 8<=>0 mixup. Then I tried gemini-flash-3, which was near-perfect and cost $0.0005 per image. So the only sensible answer is to use that! It’s really tough to compete with a cheap, fast, API model like this with local models. Anyway, jagged frontier and all that. I figured I might as well make an eval inspired by this, and the obvious extension is to see if the models can guess the weight of the objects without seeing the scale reading too.

Collecting the data

This is the kind of thing that competent coding models really shine at. I made a data collection ‘app’ with this prompt to GPT-5.2-Codex (in Codex CLI):

**“I’d like to build a dataset for an AI eval. The way I’d like to gather the data is to run a Flask app on this Mac which we will create in this currently empty directory. It should, when I visit mac.local/whatever the port is for my phone, I should get a page that asks for camera permission and then it prompts me to take a picture of an object, take a picture of the object on the scale and then take a picture of the scale reading. So three photos for every object that’s going to be in the dataset. These three should all be saved with a consistent ID for that object or that data point followed by _object _scale _scale reading. The app should be kind of nice to use and give me options for retaking a photo right after I’ve taken it or moving on to the next photo, canceling an observation at any point. You know just basically a nice simple data gathering UI so that I can quite rapidly build up a fun little dataset to demonstrate this. Let me know if you have any questions otherwise feel free to start the implementation!“** (voice dictated)

It nailed it first try. The resulting app let me quickly snap some pics of 20 objects, with 3 pics each (object, object on scale, scale reading).

Testing Some Models

The data collection app saves triplets of images: the object, the object on the scale, and the scale reading. I used gemini flash to read the scale readings (all nicely in frame and right-way-up, so it was easy), and then fed the object images to various VLMs to see how well they could guess the weight of the objects in grams.

I used MAPE as a metric, since there’s a wide range of weights. The dataset is small, but I think it captures some real variation between the different models and how well they actually do on image tasks. Feel free to expand it from 20 imags to 200 :)

Code for the eval is here.

Thoughts + Takeaways

• VLMs have suprising gaps. Especially these small models - it can be easy to fall into a false sense of security since they appear to have so much deep general knowledge, but when you poke at vision tasks you quickly find that they have plenty of blind spots.
• SpecID (my multile-choice species ID eval) showed me that in these cases text-only performance can be suprisingly high! You’d think a 5-way multiple choice between obscure species names with no picture would mean ~20% accuracy, but no - more common species are more likely to have pictures taken, and larger models are smart enough to pick the more common (yet still relatively obscure) species names and do a lot better than that. Here I suspect something similar happens - rather than guessing the weight of this lemon, I think all that some of these models get from their kludged-on vision adapters is the concept of lemon-ness, maybe ‘small lemon’, and then from there they answer with their text knowledge of how much a lemon weighs on average.
• I have a friend with a related dataset for a similar but harder task, which might turn into a fine-tuning tutorial at some point, we’ll see
• Gemini 3 Flask is SO HARD TO BEAT. It almost bums me out. It’s so fast and cheap, really hard to think when you’d need local models! The big API models have their place, but cost and speed wise Flash covers a lot of the desirable part of the pareto frontier for tasks like this.
• AI is really got at one-shotting data collection apps now. Yes, you knew this already. Still, the fact that the fastest way to collect these 60 images was to spin up a bespoke web app is incredible :)

The Method

Here’s the approach I settled on after a few tests.

• Mix up reference concentrations: 1000mg/L, 500mg/L, 250mg/L, 100mg/L, 50mg/L and 25mg/L
• Spot on 7uL of the liquid to be tested near one end of a TLC plate. You can fit two comfortably on a plate - 3 is a squeeze, unless you spot much smaller quantities (you can spot, dry and re-spot to boost the signal, but it’s hard to be consistent).
• Place in a few mm of Ethyl Acetate (EA) in a container (after letting the EA sit and evaporate a bit) and let it run up most of the plate.
• Take out and dry flat.
• Photograph while illuminated by a 250nm UVC bulb. (Wear safety goggles, UVC is not good for skin or eyes, mine lives in a box and is handled with care)

This is already enough if all you care about is a rough measure of whether something is decaf, for example. But to make things a bit more quantitative, I loaded the image into solveit (an interactive jupyter-like environment) and grabbed the pixel values around each sample region, averaged and smoothed them, and plotted the results.

Plotting the delta between the peak and the baseline, we can get a measure that increases with concentration. Fitting a power law to the known concentrations gives us a way to estimate the concentration of the beverages. It’s pretty noisy at the low end though, and I was worried arbitrary choices like the size of the rect we draw around the blob might give different results. So, like a good scientist, I added error bars by re-running the analysis for a wide range of settings so we can estimate confidence intervals (“Monte Carlo sensitivity analysis” if you’re feeling fancy).

As expected, decaf coffee < hojicha < white tea < black tea << espresso*. The tea and yerba here are brewed my way, which is weak sauce. I tend to pour hot water (temp depending on tea) over the tea leaves or teabag, swish it around, and serve. Coming soon: results for brewing as the teas’ containers suggest.

More accuracy

The lowest two concentrations are hard to distinguish (indeed, I’m not sure if I got em switched to be honest) so all our teas are actually on the low end of a range this method can work for. There are two easy ways to fix this:

• Boil off 90% of the volume to get a stronger brew that will leave a clearer spot
• Extract + concentrate the caffeine with EA. Some will be left in the tea, but we can do this to our known concentration samples too and get a repeatable method that way too.

I’m going to try one or both of these soon - and will update this post when the results come in…

Update 1: shaking with EA then running that didn’t seem to give a useful signal

Stuff that mattered / didn’t work

• Acetone did move the caffeine, and with it I could actually see the spots under daylight while the plate was still wet with acetone. But it was smeary and hard to see - EA was much better.
• Spotting samples on the plates one at a time, then running them a few days later, gave smeary messes. Run them soon after the spots dry. Once you’ve run the plate, you can keep it and compare it to others - so e.g. as long as I stick with the same volume of liquid and the same solvent I should be able to run future samples without needing to re-do the calibration ones
• Running the plates in landscape mode lets you fit more samples on there, but makes it harder to get clean data out
• Illumination changes can be calibrated out, but ideally you want even lighting from the UVC and a steady, focused pic from a decent camera. Make sure it isn’t over-exposed and that the spots are clearly visible.

Re-Running on an energy drink with a known concentration

I did a run with “Melting Forest” MUSHROOM ENERGY, a curious beverage I spotted in our local New Seasons. Gratifyingly, the estimate (144.1mg +- 19.4mg) matches up perfectly with the stated dose on the can (150mg). Here’s the code. The video at the top walks through it.

Future plans + conclusions

Besides running this on a few more beverages and brew methods, I don’t really have grand plans for this. I hope this post + the accompanying video enable others who’ve wanted something like this to replicate the process, and who (like me) were frustrated by the lack of options for cheap testing. And more than that, I hope it inspires you to look for places in your own life where a bit of science could answer a question that you have :) Until next time,

J

PS: code from this first analysis, I’ll clean it up and share when I make the video (IF i make one, no promises)

PPS: coffee has other cool compounds that fluoresce under 360nm UV:

In the plate pictured in this post:

Top row: standards (last two possibly switched?) Bottom row: Hojicha teabag, black teabag, yerba mate teabag brewed extra weak, Bai Mudan white tea (leaves), decaf espresso, regular espresso. TODO: measure espresso volume of my home cup.

PPPS: I pulled another shot of espresso for my cousin, turns out the cup I’ve been using is more like 90ml, which gives a dose of 121mg per cup not the 67 quoted in this post! (But the one I made may have been 70ml, and the extra smears might change things, so this measurement has extra high uncertainty)

PPPS: Featured on Hackaday! https://hackaday.com/2025/12/31/measuring-caffeine-content-at-home/

PPPPS: I brew my tea weak! If I let an identical Hojicha teabag steep for 5 minutes rather than my usual quick pour and serve, the caffeine content more than doubles. Still, that in my insulated mug over the course of a morning is probably 1/10th the dose e.g. my grandparents-in-law get drinking their pot of drip coffee over the course of the day.

Take 1: Fusion

Inspired by this electric blue plasmid designed by Justin from the Thought Emporium, I figured I’d link together a yellow protein and a blue protein to get green. That’s how it works with paint, right? I dug around, found some candidate proteins, designed a construct, and had even started the order process for what I thought would be a very cool new protein:

I learned a ton about how the whole process works. It seems a lot of plasmid design involves copying strings of ATGACCGCGCTGACC… around, which I didn’t want to do, so I set about constructing it out of known pieces in a solveit notebook (an interactive coding environment). You can see the code here. Some bits I learned:

• Keeping track of start and stop codons
• Linking two proteins with a flexible linker
• Back-translating from an amino acid sequence to DNA that should express it
• The various other pieces of plasmid design - promoters, origins of replication, antibiotic resistance markers, terminators, Ribosome Binding Sites (RBS), etc.
• Cloning restriction sites, including checking for accidental ones in the parts I wanted to use and making substitutions to remove them

I’m pleased with the result, but as I waited for the DNA synthesis to start, a final LLM-assisted check caught a fatal flaw: aeBlue is a tetramer! That is, four monomers come together to make the final protein. Having yellow monomers hanging on could tangle up the process, leaving me with a non-functional mess. Oh no! Thankfully I managed to contact the kind folks at GenScript who were synthesizing the DNA for me, and they were able to cancel the order before it was made, and I frantically changed plan last minute and sent new DNA to them instead.

Seb’s work

More recently, chattingh to Sebastian, I mentioned this project and, lo and behold, he actually did just this way back in 2022!

He thought it hadn’t worked, but eventually after putting it in a more productive strain and growing it out for a while, he got some visible green color. And he used aeBlue as his blue piece - so maybe mine would have worked after all! Alas. Still, having the chance to chat to someone experienced who had already tried this gave me some vital info for the rest of the project.

My revised plasmid

I somewhat frantically fell back on a plan B when I was rushing to get a new sequence to GenScript. The new plan: make the yellow and the blue separately, each with their own RBS and terminator, on the same plasmid. This way, they can fold up independently, hopefully avoiding the tetramer problem. And if the balance is off, this has the benefit that I can switch to a different promoter or RBS for one or the other to try to tune the expression levels. Here’s the new design (and the notebook that I designed it in):

I switched to a weaker promoter than Seb’s designs, worrying about burdening the cells. This was probably a mistake given that it took him ages to see any expression - but then I didn’t know about his work then. I also gave up on worrying about cloning sites, and tried to keep it minimal. I did change the amino acid sequence of the fwYellow protein though, to add an easter egg I’ll reveal later ;)

Getting it in cells

I got the new plasmid in the mail, and an hour after it arrived the DNA was in some cells sitting replicating on a plate. The protocol I followed was essentially the same as that in my last post on glowing bacteria - with a few shortcuts and refiniements. I did have to whip up some sterile water to dilute the DNA - fortunately I have some syringe filters that did the job nicely. And now I have extra on hand in case I need it again in a hurry. The more you do, the more you learn and the moretricks you aquire! I’m feeling a lot more competent as I practice all the pieces of this puzzle.

Alas, my transformed colonies grew out with ~no visible sign of color (well, with some imagination I think they’re sliiiightly darker than normal). I presume they took in the plasmid, since they’re living quite happily on kanamycin plates, but I guess the expression levels are too low to see anything. The strain I used is an old one, MM294, which isn’t optimized for protein expression. Seb had to switch to a BL21 derivative to get his to show up. So, I hold out hope.

Next steps

I’m currrently ‘curing out’ some of Seb’s bacteria - streaking and re-streaking on plain plates in the hopes they ditch the current plasmid, leaving me plain BL21 ready to transform with my own plasmid. I actually tried a transformation a little impatiently this weekend, and was left with a lawn of cells - there were plenty that still had his kanamycin resistance, so I’ll need to be more patient curing it out :)

You can see in the pic above some pink colonies lurking among the white ones, after I streaked out the first ‘white looking’ dot. I tested some candidate colonies on plain vs kan plates and saw no growth for one of them, but I guess I got unlucky when I then picked the cells for the transformation step. Lesson learned: wait for individual colonies to grow out, and test carefully :) Once I get a successful transformation into these BL21 cells, I’ll grow some on plates and some in liquid culture and see if we can see any green at all. Either way, it’s already been a great learning experience.

I’ll also say, it’s wild that an amateur can, with the help of some friends and AI, whip up a custom plasmid design, get it synthesized, and have it expressed in bacteria in a matter of weeks for a few hundred dollars. What a world! But also, a little scary. Mostly exciting though - I see the majority of uses for tech like this being positive, creative projects and hope more and more people get the joy of working with the living world. See you in part 2, hopefully ft some green!

PS: even if the proteins misfold or I fail from here, I’ll let you in on my secret. The amino acids at the end of the fwYellow sequence: VIKAVDLETYRGSGREEN. So, technically, I made GREEN! Uncountable copies in my plain-looking colonies, spelling out a small victory even in the midst of a setback. (A project for the future is running some PCR+gels to verify this small victory if we don’t get a more obvious one!)

Until next time,

J

Update!

I went to sterilize and throw out the plate with that wasn’t showing color, after it had been sitting at room temp for a few weeks. Lo and behold, rings of color - looks like at least aeBlue is expressing!!!

Cool that this sequence of DNA from an anemone (Actina equina) coding for a blue protein works in this random bacteria, albeit badly :D

Update 2: cured out BL21, transformed my BLY_GREEN plasmid into it, still ~no color. Not going to worry about this more, on to new things.

The Details

I’m working with this kit from Carolina Biological Supply Company, which contains enough materials for four students, but realistically can stretch to ~8 transformations, with easy ways to extend it further.

The kit comes with 4ml AMP @10mg/mL. They say to add it all to 200ml agar, but for the 8-person kit they give the same qty and say to add it all to 400ml agar. I decided to keep back 2ml (1ml + 2x 0.5ml aliquots) to freeze for later.

The process for pouring the plates I went with is:

• Wipe down surfaces and gloves with IPA
• Melt the agar (bottle in boiling water with loose lid for a while, or microwave in short bursts. Make sure it is not cloudy and lumpy, and has all properly melted)
• Let it cool to 55C. Work carefully by a flame
• Pour 8 plates, and a few ml in a falcon tube for later stab of the starting e. coli
• Add the remaining 2ml AMP (to half the bottle of agar, i.e. 200ml -> 100mg/L AMP which IIUC is standard) and swirl to mix
• Pour 8 more plates
• Invert, label, bag em up in ziplocs to store in the fridge, keeping one LB plate out to streak the starter

And here’s my protocol, cutting a few minor corners:

• Use a sterile pipette to add 250uL CaCl2 soln (50 mM) to a 15ml tube.
• Place tube on ice for 10 minutes (along with the plasmid which I got out the fridge)
• Transfer a few isolated colonies (goal is 2-5mm diameter from one or more) to the CaCl2, spinning the loop to dislodge the mass and then pipetting up and down a few times to mix (I used same pipette as measuring CaCl2 since still clean). Should look cloudy white, no clumps
• Cells go right back on ice as soon as you’re done adding em and mixing.
• Add 10uL of plasmid DNA. They say 1 loopful, I couldn’t get a bubble to form, I did 10uL with one of the janky pipettes I have around (luckily I’d practiced this, it’s ~2mm up the tubes I have give or take).
• Incubate on ice for 5-10 minutes. Label the plate while waiting. I added 6 glass beads at this time too, for spreading the cells later.
• HEAT SHOCK: move from ice directly into 42C water for 90 seconds, then back to ice for >=1 minute. Gently agitate while in water bath.
• Using fresh sterile pipette, transfer 250uL LB to tube, tap to mix, and leave at room temp for 5+ mins to recover and start expressing antibiotic resistance
• pipette 100uL onto plate (I used same LB pipette) then spread by shaking the balls back and fourth
• Let it rest a few minutes to absorb then dump out the balls (and disinfect them with IPA in my case)
• Place plate upside down on 3D printer bed and ‘incubate at 30C for 24-36h or 36-60h at room temp’. (with 3D printer set to 35 plates are more like 28 but I don’t want to push too high so I left it at that).

Tips - Add the plasmid gently - Work clean, don’t panic - Make sure the agar is fully melted (otherwise you get goopy lumps, don’t ask me how I know) - You can up your chances by adding more cells / more DNA. My first run I didn’t get many cells on the loop, the second go I used the wire innoculating loop and got a bigger colony or three that had had 12 hours more growth compared to the first run.

There are some great resources for learning more about this, my top rec is probably the Thought Emporium YT channel.

Anyway, stay tuned for more bio fun. I might have to set up a separate lab notebook to keep protocols and updates from polluting the blog post feed, organisation TBD :)

Agar Plates and Yummy Broth

It can be fun to just make some agar plates and see what grows. I’ve been trying out just DIY-ing the broth. Recipe:

• 250ml distilled water
• 1/2 tsp marmite
• 1/4 tsp ‘better than bouillon’
• Add 4g agar agar if wanting solid media

Dissolve the marmite and bouillon in hot water, then add to a glass bottle, add agar if using, and shake. Put the lid on loose, and boil for a while (shake if agar isn’t dissolving well). I left it simmering for a while on the stove then popped it into the insulated pot. Goal is to sterilize it all thoroughly.

You then wait for it to cool to ~55C before pouring into petri dishes. I also tried using little sauce tubs from the store - so far I haven’t had any contamination with them, I’m carefully opening the bag with gloves that have been sterilized with alcohol, working near a flame on a clean, sterile surface, and pouring then capping without much chance for stuff to float in.

The first batch I made I used tap water, which here in Portland has some treatment that sticks around. Less grew on those than on the second batch where I used distilled water.

I did a test comparing some supplements a friend had that she hoped were sterile (they were not) with some probiotics I had that had one species alive and present (as the selling point) vs a control plate or two with nothing added. I also swabbed some of the flocs that I still have sitting around (they have algae growing there too now!) and got a much more diverse plate as a result, including a few pink colonies I’ve streaked out :) Probably a yeast like Rhodotorula or something else…

Under a scope they look like yeast (too big for bacteria) and stain with methylene blue. I might send them off for sequencing later.

DIY Optical Density Logger

The idea with this machine is to log how much light gets through a tube with some broth and bacteria in it. As the bacteria grow, they block more light, so the amount of light that gets through decreases. By measuring this over time, we can plot a growth curve. This comes in useful later when, for example, you’re wanting to know when your bacteria are in the right phase of growth for transformation. On one side of the tube I have a white LED with a 220 ohm resistor. On the other, there’s a TEPT4400 phototransistor with a 10k ohm pull-down resistor. The phototransistor’s output is connected to an analog input on the raspberry pi pico which reads the voltage and sends it over serial. I store a reading (averaged across a bunch of samples) every 10 seconds. I hooked it up to my raspberry pi 5 so I could set it logging and leave it for many hours. The tube slots in and a cap goes down over it, with the gaps taped up, to keep ambient light from interfearing too much. Here’s what the curve looks like when I pop in some E. coli (strain MM294) into some marmite broth:

(Using df['OD'] = -np.log10(df['volts'] / df['volts'].iloc[0])). Since they’re in a small sealed vial, it doesn’t take long for things to slow. Also, I probably need to tweak Rsense and the setup in general to get cleaner, less noisy readings and better range. Still, cool huh!

The bigger plan

I’ve been reading textbooks and having fun learning about palsmid design. Here is my current attempt at designing something fun - the resulting plasmids should show up in a week or so! As for the ‘larger plan’, I’m mostly just messing about and seeing where my interests take me. Here’s a rough attempt to map out the trajectory I’m picturing for now (I expect plenty more side quests will spawn along the way):

Base progression

• Make some plates (DONE)
• Culture some things, practice streaking, colony picking, sterile technique (DONE)
• Grow e. coli from carolina kit (DONE)
• Transform with glowy plasmid, from kit (WIP)
• Attempt a transform with my own plasmid
• Agar art with my own transformed bug
• Make new agar stabs of mine and sebs strains
• Do a mini-prep
• Run some gels (with a friend’s kit)
• Follow interests from there
• Teach - e.g. agar art kits, videos
• Attempt a more complex genetic engineering project.
• At some point: monitor turbidity in liquid culture for growth curves (This post!)
• At some point, make a centrifuge + heat block type thing

Pink side quest

• Swab flocs (DONE)
• Pick and streak pink colonies (DONE)
• Scrape a bunch up and extract pigment with acetone
• Run chromatography (inc. on TLC plates - new kit/skill)
• Agar art with them
• Try some liquid culture maybe?
• Test exposure to UV and how that affects pigment production.
• Write up - would be a cool exploration! Especially if can be consistent extracting + photographing

Read:

• Popsci (how life works (lame), song of the cell (better))
• Bioprocess engineering (summaries only for later chapters),
• Molecular biology of the cell (skim, research deeper for interesting bits). Ditto ‘Essential Cell Biology’, although I am enjoying giving some chapters more serious study.
• Watch: Thought emporium, everymanbio, some lab technique videos.

https://kevinhunter.opened.ca/ has great agar art info. TIL it started with flemming

What the floc is that?

Towards the end of my duckweed experiments, I noticed some things floating in some leftover stock solution of hydroponic nutrients I had made up. White, goopy blobs about a cm long. Was this some sort of weird precipitate? Some bacteria or yeast growing in there? A look under a microscope revealed a dense, tangled, squishy mess. A few tests ruled out mineral/chemical suspects, and after a bit of poking around my best answer was that these were probably biofilm flocs: mixed environmental microbes embedded in EPS (extracellular polysaccharide). That’s an answer, but how do we narrow it down to something more specific?

Enter Plasmidsaurus. They offer a number of services. In this case, I went for their “Microbiome 16S Amplification & Sequencing with DNA Extraction” service. Following the guidance in the sample prep docs, I suspended ~0.1g of the goo in a product called ‘Zymo DNA shield’ (after a quick wash + spin down in 0.9% saline) and shipped the result (in a tube in a tube in a bag in a bag just to be safe) off to their lab in the nearby town of Eugene.

The 16S gene is “highly conserved between different species of bacteria and archaea” and gets used to tell what somethig is without reading the whole genome - perfect for analysing environmental samples, gut bacteria etc. Plasmidsaurus will give you the raw reads, but they also do some processing to tell you some relative abundances of different species present (see above). In my case, it turns out these flocs are not a siongle species but instead a whole happy community of microbes! It’s fun to start researching and figuring out who might be doing what. Preliminary poking tells me this is not an unusual mix for a watery, nutrient-rich environment. Cool to see some Methylobacterium present - wonder how they’re getting food.

So, there you go. A much deeper answer to ‘what is that’, provided you’re willing to ship off a sample of goop and pay $60 + reagents and shipping. Is this as far as we can go though? No! Raw results here for the curous. To dive even deeper, I’m going to switch to exploring the example that inspired me to go down this rabbit hole: the case of the mysterious triton X infiltrator.

Sebastian’s Mystery Bug

Sebastian S. Cocioba🪄🌷 @ATinyGreenCell is an amateur biologist who is continually doing amazing science stuff. He noticed a contaminant growing in his surfactant solution - a rough environment to be a bacteria! We sent it off for sequencing, but this time being a pro the approach was a little fancier than the one I showed in the previous section. For one thing, he guessed that his contaminant was mostly one organism, a bacterium, and so went with the bacteria genome sequencing with extraction option on the friendlty dino site. This was a bit of an informed gamble - the kind of shortcut one can take with experience.

He’s been sharing the data and his explorations on X.

The main result from the run this time is a file with the reads: K9P4V9_1_CONTAM-1.fastq. In a solveit dialog I loaded up the file and started poking at it. With some AI help, here’s how you might explore this type of data:

from Bio import SeqIO
reads = list(SeqIO.parse(fn, "fastq"))
print(f"Total reads: {len(reads)}")
reads[0]

>>> SeqRecord(seq=Seq('CTGCATTGCGGGAATCGAGCTTTCGAGCGCAGCGAGAAGGTGATCTGCTGATTG...CGA'), id='427416c3-6c47-4f34-9889-f33a6ce52b84', name='427416c3-6c47-4f34-9889-f33a6ce52b84', description='427416c3-6c47-4f34-9889-f33a6ce52b84 runid=0a805c02-f153-4c61-8572-df4a5e0b5651 ch=1767 start_time=2025-10-21T16:20:17.800804+00:00 flow_cell_id=PBG22075 basecall_gpu=NVIDIA_A100_80GB_PCIe protocol_group_id=251021LV_BACT sample_id=D barcode=barcode60 barcode_alias=barcode60 parent_read_id=427416c3-6c47-4f34-9889-f33a6ce52b84 basecall_model_version_id=dna_r10.4.1_e8.2_400bps_sup@v4.3.0', dbxrefs=[])

from Bio.Blast import NCBIWWW, NCBIXML
result = NCBIWWW.qblast("blastn", "nt", chosen[0].seq)
blast_record = NCBIXML.read(result)

from Bio import Entrez
# Set your email (NCBI requires this)
Entrez.email = "your.email@example.com"  # Change this to your email
handle = Entrez.efetch(db="nucleotide", id="CP073697.1", rettype="fasta", retmode="text")
genome = handle.read()
handle.close()
with open("edaphobacter_flagellatus.fasta", "w") as f: f.write(genome)

import pandas as pd
df = pd.read_csv('~/Downloads/BVBRC_genome_feature.csv')
functions = ', '.join(df['Product'].unique())

Conclusions

Anyway, I think it’s super cool that we can ship off samples to get sequenced for tens of dollars, and that you can buy the devices to do this yourself for a few thousand dollars. I think it’s wonderful that open databases and tools mean I could BLAST some reads, download a similar genome, annotate it and poke around to find some interesting genes from my laptop in a hotel room during PyTorchConf. I love that AI gives me a starting point for digging into questions that I have. I love that delightful and generous experts like Sebastian share their knowledge and augment my learning with their own deep expertise. And I love that there are so many things I still don’t know, and so many things humanity still doesn’t know! There’s a fat pile of textbooks sitting on my desk, and lots of open questions I have about these bacteria, so I suspect this won’t be the last post from me on this topic :) Until next time, cheers, J

PS: Folding Proteins

I did a bit more messing around (vibe-research code here) and compared the wild-type genome to the reads of Seb’s bug. Found the genes common to both, and found 25 with differences ranging from single-base-pair or single-amino-acid up to pretty drastic changes. In this one, ‘Biopolymer transport protein ExbD/TolR’ it was cool to see a bunch of mutations but none in what I’m guessing is the ‘transmembrane region’ - i.e. the bit that is more hydrophobic, and important. I used alphafold 3’s free tier to fold the two variants and compare - although structures for floppy membrane proteins like this should be taken with a grain of salt.

Tweet: “I’m almost certainly making all sorts of mistakes, but I’m having lots of fun :) I found 876 annotated genes that occur in both genomes, of which 25 had changes ranging from a single base pair to big diffs like the prev tweet. Even a one-acid change like T->K (32) in the efflux transporter could be helping this bug live in the harsh environment of a flower designer’s surfactant solution? Anyway, fun stuff. So much to learn!” captures my mood :)