Russell's Blog

For my thesis project, one of the things I would like to do is sequence many different samples, perhaps on the order of several hundred or thousand. It's easy enough to build sequencing libraries these days, at least, with Illumina, anyway. Obviously, doing a couple of hundred lanes of Illumina sequencing would be ridiculous (not even Jonathan Eisen is that nice to his graduate students), and so I'll be using several barcoded samples pooled into each lane. The barcoding chemistry itself was fairly tedious, until people starting doing transposon-based library construction.

A transposon is a little piece of DNA that copies itself around inside the genome of an organism, via an enzyme called transposase. Here's what the genetic element looks like :

Transposase binds the element at the inverted repeats on either end, and coils it into a loop. Then it cuts the DNA at the inverted repeats, and the complex floats away. It leaves complementary overhanging ends in the chromosome, which are usually repaired by DNA polymerase and DNA ligase (DNA gets broken surprisingly frequently in the normal workaday life of a cell; that's why DNA repair mechanisms are so important). When it's complexed to DNA, transposase grabs the DNA like this :

The transposase we're using (Tn5) is a homodimer; the two subunits are in dark and light blue. The inverted repeats (red) are bound to the complex at the interfaces between the subunits. The pink loop is the DNA that gets cut and pasted.

The complex then floats around in the cell until the transposase recognizes an integration site somewhere else in the genome. It then cleaves the DNA and inserts the payload into the break. DNA ligase then comes along and fixes the backbones. You can see why this kind of transposon is also called a cut-and-paste transposon.

The reason these are interesting for library construction is that you can prepare a transposon complex where the loop of payload DNA is broken. When the transposon integrates, it pastes in a gap. If you add a lot of transposons that aren't too choosy about their binding sits, they will chop up your target DNA. Fragmentation is one of the steps needed for sequencing library construction. What's nice about transposons is that when you use them to chop up your target DNA, they leave the two halves of their payload stuck onto the ends.

If you stuck your sequencing adapters on there, the fragmentation process also includes adapter ligation. If you added barcodes along with the sequencing adapters, the reaction combines almost all of the library construction into a single digest. Epicentre whimsically named this process "tagmentation." Get it?

However, there's still a fly in this ointment. The distribution of transposon insertions is a function of the relative concentrations of charged transposon complexes to target DNA, and DNA extraction, even from seemingly identical samples, can have highly variable yields. So, it's very important to control the input concentrations and reaction volumes during the digest. This is fairly easy if you're only making a dozen or so libraries, but what if you want to make ten thousand of them?

Measuring DNA concentrations of lots of samples is relatively easy, and there are lots of ways of doing it. We have a plate reader that can do this by florescence on titer plates with 1534 wells, or we could (ab)use the qPCR machine to give us DNA concentrations on 384 well titer plates. There are other ways, too.

However you quantify the DNA concentrations, you have to dilute each sample to the desired concentration before you can start the tagmentation process. If you get the concentrations wrong, the library comes out funny.

A few dozen library constructions calls for hours of tedious work at the bench. I've gotten better at wetlab stuff since my first rotation, and the transposon-based library construction helps a lot, but staking my Ph.D. on reliably powering through lots of molecular biology would be a bad idea. Some people might not blink an eye at this, but as soon as I find myself repeating something four or five times, my computer science upbringing starts whispering there has got to be a better way in my ear. And lo, there is indeed a better way.

Hundreds or thousands of library constructions would call for a robotic liquid handling machine. I spent some time researching these things, and I'm not impressed. The hardware is nice, but programming the protocols involves wading into a morass of crumbling, poorly maintained closed source software, expensive vendor support contracts, and a lot of debugging and down-time. Oh, and they're terrifyingly expensive, and can be kind of dangerous.

Dispensing water into titer plates doesn't seem like a very challenging robotics application, so I thought about building my own robot. It would probably be about the same amount of work as ordering, programming and debugging one of the commercial robots, and it would be more fun.

But, robots are just such a mainframe-ish solution. If there is one thing my dad taught me, it's that a lot of little machines working in concert will beat the stuffing out of a single big machine. The trick is figuring out how to organize and coordinate lots of little machines. The key to this problem is to do lots and lots of little reactions in parallel; the coordination requires lots of precise dilutions simultaneously. Getting this part right would crack the whole thing wide open, allowing you to easily do more reactions than you probably even want.

So. I'm going to make my own custom microtiter plates, just for the dilution. This satisfies the coordination criteria, and allows me to treat a plate-load of reactions identically. If each well has the right volume for the dilution, I can just fill all the wells up to the top, pipette in the same volume of raw DNA with a multichannel pipetter, let the DNA mix a little, and all the wells will be at equal concentration. Then I pipette that into the tagmentation reaction, and I'm done. With a good multichannel pipetter, I can do 384 reactions about as easily as I could do one.

All that's necessary is a 3D printer, and the ability to procedurally generate CAD/CAM files from the measured DNA concentrations. As it happens, this is really easy, thanks to a little Python library called SolidPython :

These are the wells of a 96-well plate with randomly chosen volumes for reach well.

One of the things I'm worried about is contamination. 3D printers are not really designed for making sterile parts. So, what I've done here is design a mold, and I'm going to cast the plate itself in PDMS silicone elastomer. PDMS is easy to cast, and it has the nice property of being extremely durable once it's set. And, even better, when exposed to UV, the surface depolymerizes and turns into, essentially, ordinary glass. I can autoclave the heck out if it, blast it with UV, and indulge in all manner of molecular paranoia.

If I can figure out a way to reliably sterilize thermoplastic, I'll skip the business with the PDMS casting, and simply print microtiter plates directly, like this :

By the way, I used the dimensions of a Corning round-bottom 96 well microplate. You can download the model from my account on Thingiverse.

So, I ordered a personal 3D printer. It looks like the hottest Open Source personal 3D printer right now, and the only one with a build volume larger than a titer plate, is the Ultimaker. I'd have really liked to have gone with MakerBot Industries' Thing-o-Matic, but the build volume is just a scoche too small. Come on, guys! Just a few more millimeters? Please?

Unfortunately, the Ultimaker has a four to six week lead time, so I have to wait for a while before ours arrives. At the suggestion of Ian Holmes, I headed off to Noisebridge, a hackerspace in the San Francisco's Mission District where they have a couple of 3D printers available for people to use. The machines are Cupcake CNC's, MakerBot's first kit. The ones at Noisebridge are... well, let's just say they are well-loved. The one I used had to be re-calibrated before it would go. 3D printers are pretty straightforward machines when it comes down to it, so it only took me a couple of minutes of poking around at it to figure out how to make the right adjustments. Then, it worked like a charm!

As you can see, I was a bit conservative about the design, since I wasn't sure how good the print quality would be (especially after my cack-handed ministrations).

I'm experimenting with PDMS casting now, but I'm going try some tests to see how thoroughly I can clean thermoplastic with UV. I'd really like to just order up a nice 384 well plate, and get right to it!

Anyway, I need to thank (or perhaps blame) Aaron Darling for getting me interested in transposon-based library construction, and for pointing out their significance to me.