I got feedback on my recent DNA-brick posts (Cubic Micron DNA Structures and More Fun with DNA Bricks) from a DNA researcher (who wasn't comfortable being quoted by name) to the effect that velcro'd DNA bricks are showing promise, but he thinks I was significantly over-optimistic on the robotics.
The goal of the robotics was to deposit DNA bricks with nanometer precision onto a growing construct. What I proposed was having each brick attached to a bead, and the bead attached to a handle. The position of the brick relative to the handle would be detected, and then the handle would be transferred to a positioner that would push the brick into the construct at the right spot--essentially, working blind. In hindsight, I think I probably was over-optimistic; for example, it'd be hard to keep the temperature constant enough to avoid the handle changing size and moving the brick.
I think the specially modified AFM tip I alluded to in "More Fun with DNA Bricks" may provide a workable alternative. If you can put the brick at the tip of an AFM probe, then you can use the AFM to scan the sample, and when you get to the right area, just push the brick into place.
How to make a DNA-brick depositing AFM probe (in theory):
- Use a focused ion beam (FIB) to scoop out a brick-sized-plus-a-few-nm hollow in an AFM probe.
- Coat the probe with DNA that'll bind to the recognition strands on one face of the brick.
- Put the probe back in the FIB and burn the DNA out of half the hollow.
- Coat the tip with different DNA that'll bind to the recognition strands on an adjacent face of the brick. The first coat of DNA should prevent the second DNA from sticking, except in the cleared half of the hollow. (If the first DNA is too sparse, temporarily bind strands attached to bulky chemicals to it, to fill in the gaps.)
- Use the FIB again to burn all the DNA off the tip, except in the hollow.
Now, not only will the tip grab bricks from solution, but the bricks will have more or less known orientation.
The AFM tip will return rather different images depending on whether or not it has a brick bound to it. With a brick bound in the hollow, sharp points will look like the underside of the brick. Without a brick bound, sharp points will look like edge of the hollow (a low-walled corral, if the FIB work was precise enough).
So you take the tip, dip it in the solution of bricks, and image the roughened bottom of the container until the wall artifacts are replaced by lumpy points. (For extra credit, extract a clean image of the brick to make sure it's got the right shape and orientation. This is called "deconvolving.") Then you rinse the tip, and then image the construct to find out exactly where you want the brick. When the brick is imaging the hollow where it belongs, just press down the tip to overcome the mechanical energy barrier of the stiff velcro-caps (described in "More Fun with DNA Bricks") that kept the brick from binding to random spots on the workpiece while it was being scanned.
Dip-pen nanolithography (DPN) systems already have mechanisms for putting ink on tips, rinsing tips, and returning to a known spot on the workpiece.
Remember that the goal was to place a brick every ten seconds--which lets you build a cubic micron in a week (pretty impressive!) I haven't yet calculated how long it would take on average for a 10,000-base brick (in solution of reasonably achievable concentration, perhaps evaporated after brick fabrication) to blunder in and bind to recognition strands on a tip. If it takes a second, then we're fine. If it takes a minute, which would be my guess, then it should be possible to include a tip-changer and have six tips "soaking up" bricks at once; in fact, I think some AFMs already have automatic tip-changers to deal with damaged tips. If it takes an hour, there'd be some custom robotics required to handle 360 tips in rotation.
I also don't know how long it takes to replace the ink, to change a tip, or to re-register. The NanoInk DPN system(PDF) has multiple tips, and can scan its stage at 1/10 inch per second. Scanning a micron-sized image can take a long time, but registration should in theory require only a few horizontal and vertical scan lines.
I've only run this past one person so far, and it wasn't his field, so there will likely be some not-yet-solved problems with the idea. I invite everyone (including all technically knowledgeable skeptics) to comment.