Russell's Blog

New. Improved. Stays crunchy in milk.

Spoon to Bench: A field DNA processing gadget review

Posted by Russell on September 30, 2011 at 1:24 a.m.
In my previous article, I outlined my plans for sequencing a very large number of metagenomes. Assuming that works, there also the problem of actually getting the samples in the first place. Aaron Darling likes to begin the story of metagenomics by saying, "It all begins with a spoon..."

So, how do you get the microbes from the spoon to the laboratory?

One of the things I learned from my experience in Kamchatka was just how tricky collecting samples in the field really is. From lining up permissions and paperwork, to dealing with cantankerous Customs officials, to avoiding getting mauled by bears, the trip from the spoon to the bench is fraught with difficulties. If you mess it up, you either don't get to do any science or you'll end up doing science on spoiled samples.

And then there is the DNA extraction. My lab mate Jenna published a paper last year where she created synthetic communities from cultured cells, and then examined how closely metagenomic sequencing reproduced that community. She found that the community representation was heavily skewed, but that the DNA extraction methodology was critically important. Because it was very difficult to know how well the extraction process was going to work on hot spring sediment, Albert Colman's group basically brought every DNA extraction kit they could lay hands on to Kamchatka. Also, they brought a whole lab with them; a 900-watt BioSpec bead beater (that almost killed our generator), a centrifuge, mini-fuge, a brace of pipetters, gloves, tips, tubes, tube racks, and a lab technician to run the show (see my Uzon Day Four post to see a little of that; also, most of the of heavy crates in the photos).

Albert, Bo and Sarah really did an excellent job pulling all of this together, but it was hard. Watching them (and helping them where I could) got me to think very carefully about how I want to conduct my field research. One thing is for sure; as much as I respect our BioSpec bead beater, I am not going to carry it into the field. Period. In fact, if I can possibly manage it, I am going to restrict my supplies and equipment to what I can carry in a daypack.

I'm still working on how I will do water sampling, but I think I might have found a solution to sediment sampling at the ASM meeting in New Orleans. Zymo Research just came out with a line of field DNA extraction kits that are intended specifically for field collection. The idea is pretty straight-forward; they combined a DNA stabilization buffer with a cell lysis buffer, and made a portable, battery-operated bead beater to go with it.

It's super cool, but I hemmed and hawed for a few months after ASM. I was a little suspicious of my own judgement; the system includes a cool gadget, and so of course I wanted it. I spent a month reading protocols and tinkering around before I finally decided that if the system works the way Zymo claims, it's just about the best thing for my purposes. What clinched it was re-reading Jenna's paper, which clearly shows the importance of thorough cell disruption.

So, I finally decided that I had to give it a try, and that's what this article is about. If you like, you can think of it as a parody of the tedious gadget reviews on Gizmodo and Engadget, with maybe a dollop or two of Anandtech's penchant for brain-liquefying detail.

I guess this wouldn't be proper gadget review unless I started with a meticulous series of photos documenting the unboxing. So, uh, here are the boxes.

The big one contains the sample processor, and the two smaller ones contain 50 DNA extraction mini-preps each. I'm going to leave the mini-prep kits sealed for now, since I'm going to use them for my field work. Zymo provides two DNA extraction mini-kits with the sample processor, so I'm going to use those to test out the system.

Underneath the documentation (directions are for suckers) and the mini-kits, there is the sample processor, a charging station, a 12 volt lithium ion battery pack, and an international power adapter. They also provide some little disks, which I think are for using with conical tubes (they recommend using skirted tubes, since conical tubes can shatter), and a couple of pairs of earplugs.

The earplugs turned out to be... prescient.

The sample processor itself is an modified Craftsman Hammerhead Auto Hammer. Upside? I can buy extra batteries from Sears! Downside? Seeing the $71.99 pricetag from Sears really makes Zymo's $900 pricetag hurt. Our super-powerful bench-top BioSpec bead beater is only about twice that.

When I asked, Zymo said that they've actually modified some of the internals of the Crafstman tool, but this might have just been to discourage me from traipsing off to the hardware store to buy some PVC pipe fittings and a hacksaw. Experience tells me, though, that I could easily fritter away $800 worth of time replicating their engineering. OK, $700. It's a really nice international power adapter.

I was a little disappointed to note that the Craftsman part is made in China. Not that I have anything against things being made in China, but I was under the impression that Craftsman was an American brand. It's a little like discovering that a jar of authentic-seeming salsa is made in New Jersey, or something. I'm sure they make perfectly good salsa in New Jersey. Nevertheless, I have a deep-seated belief that salsa should be made in a Southwestern state by grandmothers who each know five hundred thousand unique salsa recipes, and Craftsman tools should be made in Pennsylvania or West Virginia by guys who wear blue overalls and carry their lunches in pails.

OK, so maybe I do have something against everything being manufactured in China. While using the sample processor in the lab, it suddenly made a very loud click that I hadn't heard before. When I looked carefully, I noticed that there was a piece of metal debris caught in the motor vent. It seems to be made out of aluminum (it's not ferromagnetic). My guess is that this is debris from the manufacturing process, not a broken part of the device. I shook out two other smaller pieces, but lost them before I could photograph them. It looks like the three pieces are part of a square. Most likely this is the remains of an improperly handled punch-out, like a metal version of a paper chad. As you can see, it got kicked around inside the motor housing until it was ejected into the vent. I think Craftsman (or their subcontractor) should get the blame for this, rather than Zymo.

Here is the soil/fecal mini-kit. Each prep uses three sets of spin columns. The bead bashing tubes, as they are labeled, are in the upper right, along with two tubes of lysis/stabilization buffer and a tube of elution buffer.

The protocol says to add the sample first, and then add 750ml of lysis/stabilization buffer, and then bead-beat. But... then you would have to bring a p1000 and tips along with you. No thanks. The sample tubes and the beads had better be chemically stable, or they'd wreck everything. So, I aliquated the buffer into the bead tubes before leaving the lab, and left the p1000 behind. Zymo includes some very fancy spin columns with this kit; they have their own caps, and little nubs on the flowthrough channels that you need to snap off before you use the columns. I've not encountered anything quite like these.

The final step of the kit includes these green-capped columns that are pre-filled with buffer. I wasn't expecting any liquid to be in them, and so of course I spilled the first one on my foot. Don't do that.

So, I took a little miniature field expedition to the exotic environs of the Putah Creek Riparian Reserve to try this out. It didn't take long to find a place that promised to have plenty of microbes.

Here's a soil sample before processing.

I processed some of these samples for 45 seconds (the directions recommend a minimum of 30 seconds). Usually it seems to work fine, but occasionally the tube explodes and splatters mud and buffer all over the inside of the lysis chamber.

The exploding tube problem appears be caused by grit preventing the threads from closing correctly. In other words, it was my fault. Be extra careful to get the dirt actually inside the tube. Here's what it's supposed to look like.

After processing, the samples are noticeably warm. If you are going to process for much longer than 45 seconds, I suggest you stop and let the sample cool for a few minutes before continuing.

Here are the yields I measured for the mini-kit preps (minus the tube that exploded), eluted into 100 μL of buffer.

Source Yield
Potted plant 80.6 μg/mL
River muck 0.669 μg/mL
River muck 1.13 μg/mL
River muck 0.595 μg/mL
I messed up the extraction protocol a little bit (and I used too much elution buffer at the end), but still got enough DNA to work with. Not too shabby for a first try.

I decided I had to throw these samples and DNA away because I don't actually have permission to use samples collected on UC Davis's campus. That's also why I'm not showing a gel.

How to sequence 10,000 metagenomes with a 3D printer

Posted by Russell on September 19, 2011 at 1:15 a.m.
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.