Wednesday, September 28, 2011

Thinking Outside the Box or Just Plain Nuts?

Please take the title in the spirit it is intended: as a bit lighthearted. Seeing the object pictured and reading the accompanying blog post from one of Jonathan Eisen's graduate students.  It's an unusual solution to a common problem, and gave me a good chuckle.

The problem is normalizing the concentration of a large number of samples.  In this case, it is to support Nextera construction of Illumina libraries, which relies on in vitro transposition.  The post does a spectacular job of explaining that fascinating technology, which is a bonus.  Nextera offers a route to efficiently preparing a large number of barcoded libraries for pooling, but each of the input DNA preparations may be at a different concentration and outside the range for efficient transposition.  So, you really need to fix each input DNA sample to the same concentration.

His big complaint is that commercial liquid handling robots are closed systems and expensive.  I understand that.  I've never been directly involved with the robots, but we had lots at Millennium and Codon.  My favorite at Codon were the Echos, which use a very cool acoustic technology to spit droplets on a picoliter scale from a source plate to an upside-down destination plate with amazing speed and accuracy.  Yup, the destination plate is upside-down; at this scale surface tension overwhelms gravity and holds the liquids in place. Echo's are great for performing complex rearraying, such as is needed for a lot of interesting synthetic biology applications.  They also, if memory serves, are several hundred thousand dollars each, which is definitely real money.

It is shocking that no standard XML format for liquid handling control has emerged.  It would seem ideal to have a standard platform-independent format, coupled with platform-specific verifiers to make sure a given program will run on a given instrument (with a given configuration, for those that have configurable decks) and then compilers to generate the native code for an instrument.  Instead, it appears each instrument is a whole new programming learning curve.

So what is Russell's humorous solution?  Instead of trying to customize the liquids, he's going to customize the plates.  Using a 3D printer, which can custom-build complex shapes from plastic, a specialized plate would be built for each dilution experiment; the photo above is an early prototype.  Part of the attraction for him of this is that the 3D printer is a smaller, more general purpose instrument than a liquid handler.  I've thought they'd be cool for building educational models of proteins and organelles, but clearly they have many other uses.  A challenge is making sure the custom plates are sterile, so he's even considering only making a mold with the printer, and then casting PDMS on that mold.  That strikes me as a lot of steps for a custom part, and certainly something that won't scale well in a large operation.  But, of course, he's interested in progressing his thesis, not building a production line.  That's a key difference in aesthetic sensibilities; he likes using small, cheap general purpose instruments and I'm worried about cranking things out.

When we faced a similar situation at Infinity, needing to normalize a large number of PCR amplicons upstream of 454 sequencing, we took a different approach.  If I had an Echo, we would have normalized each one precisely.  But, given that we had just manual pipettors, I tried a different strategy.  We binned the concentrations into a small number (I think it around 6) bins, and that made normalizing the wells practical.  Writing out 6 maps for manual pipetting wasn't bad, and while it was a tedious bit of pipetting for the molecular biology craftsman doing the real work it got the job done.  In the results, it seemed to make some difference, but it was a single experiment and we didn't establish a track record for the approach.

What about cheap, open liquid handlers?  Legos immediately came into my mind, and indeed such a robot has been built.  DIY Bio folks and others are also exploring other solutions, ranging from the obvious (refitting inkjet printer heads) to the less so (refitting 3D printers!).

Is the custom plate solution crazy or crazy like a fox?  I vote for crazy (in a good way), but I wouldn't be surprised if a lot of people side with the foxes.


Ian Peikon said...

Another liquid handling robot built from LEGO that looks quite nice...though I don't think the build instructions are open...

Russell Neches said...

Well, I hope to win you over into the "crazy like a fox" camp in about two weeks. :-)

You are exactly correct that this solution is not intended for factory-scale sequencing operations. What I wanted was a solution that would completely and utterly remove sequencing library construction as an obstacle to a Ph.D.-scale project.

Any such solution needs to be easily manageable by one person with modest funding, limited time, and poor to mediocre bench skills (i.e., me).

I'm also trying to anticipate my situation after I graduate. I want to be as independent as possible. That rules out strategies that depend on very expensive, specialized equipment or the funding and authority to hire a skilled technician.

Anyway, 10,000 libraries would only require 26 384 well plates, or 104 96 well plates. Including DNA extractions, that should be manageable in about a week's worth of work (baring any mishaps). So, I wasn't being completely unserious!

Keith Robison said...


Thanks for responding - you make a good case for your approach. I won't be shocked if I end up suggesting it some time :-)

Russell Neches said...

I hope so! I'm going to publish all the software as Free Software, and all the instructions and notes Creative Commons.

I'm tired of going to talks and poster sessions on microbial ecology, and seeing three to ten samples in every study. I want to see some real ecology in microbial ecology, and that means a lot more samples. BGI, JGI and Broad could probably do that right now, but three players does not a community make.