path: root/transcripts/2009-09-15-paul-rothemund.mdwn

A relatively informal visit. We'll have a wonderful seminar here. Afterwards,
folks should feel free to hang around and talk with Paul, and at one point he
will head on over to MBB I think so he can look around and talk some more, and
go eat and drink that's up to him. Grant Wilson. Grant's been pestering me
forever to get Paul Rothemund here. Paul has been out with the semitech people
this morning. Greg. Paul Pastella. Semi-independent. So, he does DNA crystal
stuff. 

Beyond Watson and Crick, 


I came here for the semiconductor industry but then wanted to visit everyone
here. It's a mixed bag of left over other talks, so bare with me. I was talking
to semiconductor industry about things towards integrated DNA self-assembly,
artificial DNA self-assembly with nanoelectronics using this idea of
programming DNA self-assembling. I come loosely from this field of DNA
nanotechnology. I just want to give a real brief background of the elements of
that field and a little bit about what I am interested in. I also want to
entertain any discussion about anything here. Let's just have an open questions
and discussion, and at the end, more towards biology and biochemistry type
applications that you might see in the talk. Today might actually help single
molecule biophysics. I might be able to hand wave about that to stimulate
discussions.

There are is programming versus Programming. To (p)rogram means merely to
"specify", this is what a chemist means by programming binding interactions. To
a computer scientist, Programming means embedding a real computer algorithm, to
embed something that has arbitrary computation in a molecular situation.
Programs from molecules. Specify what the molecules do. There might be no
computation in a formal sense. The reason I am bringing this up is that lately
people in DNA computation have re-branded themselves as molecular programmers.
Their related friends in syunthetic biology. Synthetic biologists nicely let us
hang out with them. This is just what the effort is. We're trying to take
molecules, principles from biology, and turn them towards our own purposes that
might not be biological purposes. Synthetic biologists start with cells.
Molecular programmers start with molecules. Some of them are (p)rogrammers,
others are (P)rogrammers. The far most proponent of that has been Erik Winfree.
Ron Weiss has interests in big piece stuff. Tom Knight is a big P kind of
person. Drew to a certain extent. A lot of people are little (p) type. That has
to do with philosophy, not any sort of value.

The field of DNA nanotech has three strains. The little (p) of programming
structure. Ned Seeman started the field roughly thirty years ago. Over the
course of thirty years, they used DNA as lego, and build any structure we might
want. They laid out the structural motifs, making junctions, the things that
were thermodynamically and kinetically stable. Ned had origins in
crystallography, he tried and succeeded to make rationally designed 3D crystals
for DNA. One is this programming functions in Dynamics. Bernie Yurke and Andrew
Turberfield. Non-equilibrium DNA structures. DNA hybridization is not
two-state. We can make little DNA tweezers, little DNA walkers, none of these
dynamic things are as good as your protein motors. We're making them more
autonomous. In 10, 20 years, we might catch up with the structural guys and
have the dynamics and functions we expect. You're familiar with other functions
of nucleic acids that are not just structural. This is very active.

Finally, Len Adleman started experimental DNA computation in 1994. Putting
actual computation in molecules. I won't go into this today. The idea here is
to put computation in molecules, but not to compete with electronic computers.
Thermodynamics, kinetic error rates, cost, scaling would not allow us to
compete with electronic computers. These can control other molecular systems.
They can build shapes, structures. A little computation to determine what
structures. Or computation to figure out GRNs, to turn genes on and off. Do a
set of computation on biochemical inputs. Does this cell have a particular
disease state? These are small and compact. A few dozen gates. Never the less,
they are computations. If you put a little computation into molecules there may
be many many things you can do. There are strains of structure, function and
computation. This is the kind of place the field is going. We want to
recapitulate what the living systems are doing. Along the way, we'll use the
things we come up with, and put them into applications.

What was the original impact of Len? I view them as two different intellectual
efforts. For construction tasks, Erik visualized this thing as not computation,
but.. this is an aperiodic DNA crystal. These are little DNA monomers, grey and
white bodies are 12 nanometers by 4 nanometers, that has the beginning of this
fractal pattern. It's binary addition modulo two. It was about figuring out a
complex pattern. The tie between computation and pattern formation was
immediate, not just "explicitly doing computation". The argument is if Len had
never existed, I was doing DNA computing before Len, if Len hadn't ... Len's
implementation, which is being pursued in eastern Europe, the reason I bring it
up, is that we don't have lots of things. What leads you to believe, "we're
going to program with a big P, Len's big P isn't going to go anywhere, if we
bring on these fast time cells, what you're learning now will help later?". One
answer to that question is that independent of what we're learning now, I can
give a whole talk on it, if you pick this one subfield, this has all the
richness of computer programs, all the Kalgoromov complexity rolled into it, in
non-trivial ways. It's the only implementation of computation in DNA, as far as
I'm concerned, captures all of the interesting stuff in computer science, in a
way that makes sense, all plays out in whether or not you can build something.
If you just carve off one thing, whether or not it's practical is a different
issue. There are particular subsets of this particular domain, that could
capture all of the interesting things in this domain. 

But, making structures towards things we might do with nanoelectronics. The
first real successes with DNA nanostuff was periodic structures. Cubes. And
tetrahedrons. This is a DNA tube. It's made of similar DNA motifs. They are
made by 4 by 12 nm. They make this crystal. They have a curvature when they
bind, so they make a tube, it's not a model for a microtubule. It's a
crystalline tube. It's not too interesting of a periodic pattern. Purdue made
these where they have these little 5-connected objects. 17 plane symmetry.
Periodic pattern when symmetry breaks. We can make any periodic pattern you
want. We can make any periodic pattern in the plane with any symmetry. That's
great if you want to make a regular structure like computer memory, but what if
you have the greatest amount of arbitrary and unique things in a finite
structure. Arbitrary color pixels and make whatever pattern. A grad student
posed this question to me, so I came up with DNA origami. This is a process
where we take a long strand of DNA, 7k nt single stranded DNA, take a very long
ssDNA and fold it into a desired shape or pattern. The basic way this is done
is via DNA staples. In the real implementation they are 20 to 30 nt. Each
staple conceptually has a left half, and that left half binds that long black
scaffold strand. The right half doesn't have a continuous complementary. The
right and left sides are complementary in other areas. You have to design this.
It folds this into some shape or pattern. You have to respect the geometry of
DNA. The idea is that these colored staple strands fold this long strand into
this path where it makes a rectangle. The black hash marks make the hydrogen
bonds. The blue ones bind here, 8 bases here, here by 16, these things are more
complicated than what I described before, each one binds the scaffold together
in different places. It's shocking and somewhat confusing that this works at
all. This was a video by Shawn Douglas from William Shih's group at Harvard. We
can't make decent movies of stuff that happens with molecules. It gives you
some feeling of what must happen for these things to come together. These
staples are binding together, falling off, being invaded, etc. etc. The
molecuiles are wigling. You can heat it up to almost boiling. A double helical
character emerges. The whole thing is composed of things composed of double
helices. This is not a simulation. There are no kinetics. This is a movie.
Nothing is moving. This was made in a time-reversed way. 

Have you found issues with knot formation? That's been most rigorously explored
by Shih because things in 2D are easier than things in 3D. You have these
double helical axises held together by these crosslinks, where a staple strand
winds around one hnelices and then reverses directions. One of these crosslinks
enforces the other crosslink. It's kind of flike in homologous recombination,
but not quite. There are no unbaired bases. These are just drawn to show you.
There is a gap between helices, but not between helixes jump.

In each of these linkers, you have to worry about if a sequence in the linker
chain is an exact match in the seven thousand .. by design, each staple has an
exact match, it has domains that match exactly. Many things are wrong. The long
strand itself can have self-complementary to itself. The short strands can have
complementary to each other, complementarity to places they are not supposed to
bind. In DNA nanotech, we typically use computers to design these sequences. We
design the hell out of these sequences to avoid these problems. Here I used a
natural scaffold from the M13 virus. You're stuck with the sequence of the
virus. You're effectively stuck with the sequence of the staples because they
are somewhat induced by the virus. You have no choice. I just close my eyes and
get lucky. There are reasons you get lucky in this setting. In DNA
self-assembly, Burnie Yurkie on non-equilibrium DNA stuff, exact matches have a
mechanism- biochemists knew this for years- mechanisms by which they can
displace incorrect matches. If they mismatched by a length of 10 here, if
there's a guy who matches by 16, the other one can bind by four on the end, and
through a random rock can get there. The reason why this doesn't get trapped in
a billion different wrong ways, well, DNA hybridization is not two-state. The
10-mer binding domain, if you had to wait off, you don't have to wait centuries
to drop off, well, there's this milliseconds mechanism before step. Without
this mechanism, this would be dead probably. It's not just a difference in
equilibrium constant. It has to be able to "walk off" bad things that occur. If
you have something that matches inappropriately, and nobody will compete him
off, you will have some garbage stuck there. Now since you can synthesize that,
is it worthwhile to design a perfect scaffold for folding origamis? There have
been no pathologies so far. Shih and others are going up to 50 kb. The address
space gets more cramped. It might get worthwhile to design very carefully. The
answer is that I don't know when it becomes necessary. Did you ever actually
write some program that would ensure some tearing between the stitching in the
molecules? I didn't have choices about the sequences. The virus induces those
sequences. I had to close my eyes, disregard my best practices, and do it on a
small scale. But I did do the following: I checked this thing for self-binding,
it has a biologically important 20 nt hairpin. That hairpin is really
thermodynamically stable, so I only folded the molecule, except that really
important molecular hairpin. That's the pin that hangs off in everything. Shih
will fold through that section sometimes. Oour ability to test if something is
screwed up is not that great. Depending on what strands are bridging that
hairpin, and whether or not they are overlapping enough, that's where you
figure it out. Secondary structure can screw you up. Ned wrote algorithms very
early. Absolutely. Oversequences of this length, you can totally get rid of it.

I made a variety of structures early on. Once I had basic geometry under
control, I just wanted to fold in any way I wanted. So this was a scaffold
path. Start up, make a turn around, fold over, fold back down, and be back
here, and leave the little pin down here and out of the way. I wanted to do
this to show that I could make structures with mechanically stable holes. I
designed staples to do this. To give you a sense of scale, I had this 100 nm in
both directions, within 6 nm in one direction, 3 nm in the other. To compare
its size to a red blood cell, there's the smiley face guy. 70 of these can fit
across a red blood cell. It's slightly smaller than lithography. The line
widths are 5 to 10 times smaller than modern photolithography. Is there some
reason why it needs to be continguous? You think it would cause pins. You could
leave these structures as gaps, and the DNA helices stack across gaps. There
are staples across the gaps. In this design, there are DNA strands that go
across this gap and suture things with stability. It has problems with
stability elsewhere, but not there. Can you make a multi-layered structure if
you wanted to fold it? William Shih has been doing that.

Atomic force microscopy is where you have a micromachined sharp tip, you
measure the deflection with a laser beam bounced off of it and build up an
image one scanline at a time. You're not taking a picture, you're physically
dragging a tip, or tapping a tip, interacting strongly with a sample. This is
how we're going to visualize this structure. You design one, you send an email
to IDT, you get 250 strands back, you do it in a magnesium buffer so that they
form, then you're stuck with a hundred billions of these things in 100 micro
liters, you pour them over a surface. The DNAs are negatively charged. The
surface is negatively charged. You have magnesium in the buffer that serves to
stick the DNA to the surface. They stick strongly enough so that you can take
the AFM image and see a bunch of DNA origami like this. You can see that among
other things, they aggregate. Some number of them are broken, some of them are
fragmented. In this particular prep, 70% of them had the recognizable shape.
You can zoom in and see the fine structure. here's the fine structure. Here's
the fine structure of this thing. They take 4 to 8 minutes to image. Getting a
tip that is sufficiently sharp. In the past, you could sit there, it could take
two or three days, throwing away needles, changing the needle on your record
player until it sounds good. If you make a structure much more rigid, the
apparent yield rises. These triangles, which have, these are the schematics,
they have DNA helices in all directions. It doesn't have just a few helices..
it has greater than 95% yield. The yield of these, the gross morphology of each
staple, the gross morphology down to the 5 or 10 nm length scale, is dictated
merely by the purity of the scaffold. The better the purity of the scaffold,
when you break the structures, it's not hydrolyzed, you get the yielded
structures. But it's not staple-dependent? You can pick multiple sets of
staples to make the same structure. So there are some rectangles with no
apparent yield.. it's not measured in a fine way. We don't try to optimize the
yield. That 90% robustness is from different set of staples. It's shockingly
robust. The fine detail works out. 

A year after I did this, I got this picture in an email, Lulu Qian. Lulu is now
a postdoc at Winfree's lab. What a job application. So she sends me this thing
in an email. I was amazed that she did this on her own. She just wrote her own
code to do it, screw this guy, she designed this really great fold. She has
taiwan sitting right there. She has taiwan on the world's shortest leash. So
this goes to show that this can be reconpitulated by reading supplemental
materials. It's not rocket science. Layered structures. Shih has got 3D
structures of several times. Nature, 2009. Douglas et al. He has used one where
he uses DNA helices on a honeycomb lattice. On like a rectangular grid. I would
want to do a rectangular grid if I went to 3D. You can do hexagonal grid. These
are negative stained TM images. He has girders, interpenetrating structures,
all kind of stuff. These are three origami at once. Chains of origami. All of
these things have reasons. Everyone in the lab has different reasons for these
structures. The whole idea behind this stuff for biology is to make custom
instruments for biology. William Shih has a background in motor proteins. You
can ask hypothesis like, they take these steps, they walk along a protein
filament, you can ask, how does it know to release the backleg? The hypothesis
might be that the back leg is released when under tension. I want to test that
by varying the spaces of my track, maybe if I increase the step length, maybe
it walks faster. The probability of a walk might be less with shorter steps.
You can't engineer protein filaments. You can't change the step size on a
microtubule. This will be a substrate for motor proteins to work along. You can
do the DNA part and make it great, but the whole is meant to accept an active
tetramer. Getting that right will be the hardest part. At least we can think
about building custom instruments for biology. For little expense, you can
design a custom instrument. The interpenetrating barbel. Imagine machining that
thing. Andy's question about folding difficulties. To get these things to fold.
The 2D structures that I make, and 3D shells, fold quickly in 2 hours and
you're fine. To get these interpentrating bars, you have to let them fold for a
week. If you are clever you might do long staples on the inside, so they fold
first at higher temperatures, but nobody has been that detailed yet. Annealing
schedule. Staple length. You can get these things to fold. Several groups are
trying to get dense 3D structures. They made this structure or something. They
tried to make this structure, but they didn't let it fold in a week. The yield
of the interpenetrating bars after a week is 50%, for what they consider to be
"well formed". What's the definition? Well, it's at least better than 1% yield.
This is the point that Ned harps on, are these molecular and really all the
same? So that's fine. So you can assemble multiple pieces. They have little
extra DNA strands. They purify, they mix them together. The same for the
polymers on the top? There are many ways to string these together. They are
exploring that, a different one, if I have time for that. 

Hendrick Dietz, William Shih, 2009.

Are they flat? I don't know. If you make a 3D structure or a DNA tube, with a
total curvature, the mica might catalyze them open. I don't know if the smiley
faces are flat. I designed them with symmetries that would tend to take out,
would take out bowing in x or y, so according to the symmetries they might be
taken out.. so that same structure is completely compatible with trust. Shih
said that my structures are twisted. I went back two days later and realized I
screwed up. You have to nail the inner spacings between the cross-overs. It's
10 bp/turn. If you don't compensate for it, your twist will have this super
twist, not gentle at all. If you make a rectangle, it might be twisted from 90
to 180 degrees. They are twisted like pretzels in solutions. Everything I made
was really twisted. He actually, this just came out in nature, he had these
rectangles either left-hand twisted, right-hand twisted, or straight. We
reconpitulated this with our stuff, it really helps. Things were twisted, if
you know what to look for in the AFM. The long chains can also fragment.
Hendrick Dietz was the one who thought about this. Bending and twisting based
on this can lead to curved structures. Boomarang structures. They figure out
the model, they can nail it. Gears, they can make curved 3D stuff. This was
made before anyone was making this. This thing, if you look at the
reconstruction, it suffers from the same twist and deconstruction (the boxes).
Ebbe Andersen 3D box from 2D surfaces, Nature 2009. Rectilinear style. You can
make everything.

caDNano open source software

For the purposes of patterning, if you want to arrange stuff, every staple goes
to a unique position. The orange strand lands exactly there. In principle, you
have to put chemical functionalizations. For the paper on this, I didn't put
interesting functionalizations, just hairpins and bumps, which has been done in
DNA forever. With 6 nm resolution you can do North America, snowflakes, two
words of DNA with a continuous helix from one flake to another. There's a piece
missing from this D. It's the case that you're trying to figure out how many
functional groups you have. You can only visualize 94% of them. The AFM
actually knocks them off, they disappear while you're sitting there. If you're
trying to functionalize this thing. You need other techniques to try to nail
this to yield the functionalization of this technique. But if it all works out,
you want to put some active elements on some staples, for example, make a
circuit, and once it formed, you wash the thing away, and you have some
functioning nano circuit.

The design of a biochip of a self-assembling molecular scale memory device.
Robinson & Seeman. Protein Engineering 1 (4) 295-300. 1987. Nealey Lab, Ruiz,
have been using copolymers for 2D lithography and data storage. We get a little
amount of money from the semiconductor industry through the semiconductor
consortium. It has focused my thinking about what are the actual challenges to
nanoelectronics integration and how to surpass them. This chart was made by
people in this organiszation. Benchamrking Table for Nanopatterning
Technologies. Here's DNA nanotechnology. The things that are in this column
that might be a problem in yellow, and really definitely a problem in red. What
kind of devices and how to make them compatible with DNA? Higher order
structures? And foremost, registering them with underlying features on a
surface. Here's this problem. Here's this problem of DNA origami on solution,
and they go everywhere. We have FENA sponsored projects to address this
problem. Yield, defect rates, scanning, compatibility. CNT cross junctions.
And, in those projects, maybe some of these things, maybe the registration is
more yellow than red, and anyway. So.

We have a collaboration at caltech, and everyone has graduated. This involved
Siping Han, Robert Barish, Hareem Tariq, Marc Bockroth, Bill Goddard, Erik
Winfree. These students did everything. There's a history of DNA plus CNTs to
make devices. But there hasn't been any use of DNA self-assembly to assemble
anything other than a one-dimensional device. You can combine DNA with CNTs,
there's a nice paper by Keren et al. 2003. You might have a single CNT along
its length that might be functionalized different, like for attaching
electrodes, and the entire thing is gated by this carbon backing. It has a
symmetry breaking in some sort of mild way. You don't have anything that
approximates a two dimensional device. You have this same substrate-backing.
It's this whole giant device. We wanted to organize this thing in 2D, and have
2D devices. There are many ways to functionalize DNA to CNTs, you can sonicate
CNTs to DNA. Otherwise the CNTs won't go into aqueous solutions. Addressing the
compatibility of CNTs and DNA origami. You can combine some components to the
CNTs. As far as yield is concerned, the DNA is usually binding things later,
but those things can also bind to the CNTs. So, they will bind and be partially
accessible. So they stick DNA in hydrocellulose. Some stuff is bound, some
stuff is not. What the students did is took a note from Bernie and put
protecting groups on the strands. These little DNA sequences that are exposed
are very short. We haven't measured this rigorously. The protecting groups have
more available. This kind of tricks appears to have really helped us in
organizing these things. We took two different sequences orthogonally to each
other, made two kinds of CNTs with protecting groups on them. These protecting
groups only release if and only if these things start to bind with this toehold
with the appropriate DNA sequence. They wanted to make cross bars. If we used
random mixtures with metallilc and one semiconducting one you in principle get
transistors. One semiconducting, one metallic, they relied on random chance.
Maybe they don't make.. they got crosses. You can deposit them on silicon, wire
them up with electrodes. We got a few FETs, one of them was sort of stable.

Two dimensional organization of CNTs on DNA origami.

What drives the interactions of DNA and CNTs? It's not completely
characterized. The best way to separate CNTs. Metallic CNTs are semiconducting.
They have different chirality. You can close them and get chickenwire with
different chirality. You can exquisitely separate the CNTs with different
populations. It's fricking amazing. They had a paper in Nature in the last
couple of months, where they can do this massive screen over short DNA
sequences and selectively find 12 different types of CNTs. They proposed it to
be a DNA "beta sheet". They see the pyrimidines specifically binding between.
Once the tubes are put together, once you've directed the crossover, are there
tubes to each other, will the tubes remain stuck? There is this origami in
between them. They have no directions between them. This guy binds to this guy.
The DNA origami works. You could imagine aligning things on the CNTs such that
you have your toeholds and protecting groups. Breaking symmetry on CNTs is a
bitch. You can functionalize the ends more than the internals. You broke
symmetries, when you broke the CNTs, the whole tube is the whole thing. We're
not talking about bands. The whole tube has a particular chicken wire
structure. There are some CNTs that have heterogenerity between two structures.
Nobody knows how to grow CNTs to grow one chirality and switch. What I was
getting at was that, it's great to think towards this nanoelectronics era
driven by DNA, and then thinking that this is the world's most expensive tech
for this scale. But what if it was catalytic? As far as cost is concerned, if
we could organize things, we could paint a Boeing for pennies. In the grand
scheme in things, I haven't figured it out very precisely, you can paint large
areas with this stuff relatively cheaply. I think I am much more worried about
working with any yield whatsoever. That will stop it much longer than the cost.
The yields that we want are appropriate. They organized these crosses. What's a
metric for how well they actually did? We had CNTs that can bind either the
horizontal or vertical stripe. How many bind that and how well? 50% of the
tubes bind within 15 degrees. We didn't give them that long of a run to bind.
If we give them a longer binding stripe, this is an objective measure of how
well we actually did. There are some that are 90 degrees off from the desired
orientation. They might be binding only by one DNA, they are still floppy,
well, these are the distributions we get. Let's say we really make great
crosses. Even if you make great crosses, you have to find them. You have to use
ultramicroscopy to find them. You have deposition and aggregation. You try
increasing the yield, say you have not many crosses and not many DNA molecules
are binding to tubes, well, these CNTs are long. You get more aggregation, but
at the end of the day after you find them, you have to use lithography to go
down to these guys. A grad student is going blind trying to hook these things
up but it's not fun. What can you do about this? Under the FENA research. Grant
is gone? Looks like he ran off. Anyay, so this is a collaboration I have with
IBM Research Almaden. Ryan Kershner, Luisa Bozano, CHristine Micheel, Albert
Hung, Anne Fornoff, Charles Rettner, Marco Bersani, Jennifer Cha, Jane Frommer,
Greg Wallraff.

The largest length scale is the smallest length scale we can do with electron
beam lithography. The top-down length scale matches the bottom-up length scale.
So we're going to make locks that are sticky-patches for shape-patched origami.
Here's the idea. Ryan Kershner spearheaded one chemistry. Another two did
another chemistry. Everyone was really important. Someone did all the
lithography. The solution with the triangles, which are nice and rigid. Silicon
dioxide or a silicon substrate. So pattern a template layer over it. Pattern a
photoresist, use e-beam, transfer to the template layer using an oxygen plasma.
This is a dry etch, you impinge it on the template layer, you oxidize it and
change its chemical layer, you strip off the .. and in principle you get DNA
binding sites. Magnesium is what binds these things to a MICA surface. The
degree to which Mg will regulate this will depend on a few different factors.
For the conditions of the materials we had, we had 10 nM Mg to make it, but you
bind it in 100 nN Mg. If they haven't discoverred this, the whole thing
wouldn't have happened. We have a couple differentchemistries. Syline on a
straight dioxide surface. TMS/SiO2. We can do it on a diamond-like carbon
surface (DLC/DLC on Si). We've done this with two systems. This is DNA
triangles on diamond carbon. These have columns of up and down triangles. You
can score this very tediously and show that the diamond like carbon, you have 2
populations, and guys are usually 60% oriented in the correct direction. They
are oriented plus or minus 10 degrees. This is under buffer. Here, origami
triangles are being ripped off by AFM tip. If you do this under the buffer,
then you will necessarily rip them. If you dewett and get rid of the water, you
have to dry in a way that doesn't get them off, and some people are doing that.
This is the best data we could get. This data could be cleaned up. These are
20% off or something horrible. This is the random distribution, it would be +
or - 30 degrees. We tried to bind them into shape features. This is not the
route to getting th e organization of origami in features. Thsis is a movie
that shows the size dependence of the binding. This is all under AFM. These
things are binding and unbinding. For the small things, they have the highest
residence time. They are about 127 nm. The best alignment and binding in the
size-matched features. The important thing to know is that, if this binding
were not reversible, we would not get the results we get. One of the good
things about the noncovalent binding, the streptavidin would be stripped too
fast.. The triangles have a chance to settle down. So that's a key feature if
you're going to try this binding and alignment. So this is just some frames
from that movie. This shows the salt dependence. In too high binding, the DNA
itself actually overcharges, so it now, the DNA itself is nominally positively
charged and won't bind to the surface. If you're a biochemist and think of DNA
condensation, well, that's what this is, with a polycovalent cation. It doesn't
mediate the binding. If it does overcharge, the DNA falls apart. Leave one
strand on a triangle to be able to hybridize to another triangle. That's a
different project. There's a whole, so here we want individual triangles that
are not spaced, but then what if we have the triangles talk with substrands,
let them have molecular precision with their interactions. We want asymmetric
shapes. So that the origami goes down on the surface, all points north, a
triangle can go down with 6 different orientations. Place the shapes then add
the components. The components wouldn't cause the origami to aggregate in
solution. The student wouldn't have to do this crazy ultramicroscopy to figure
out where they are. Aggregation and finding the structures in one shot, using
this registration technology. In the next couple of years, high orientation of
the shapes, placing distinct shapes. I am seeking postdocs for this. IBM Albert
Hung, Jennifer Cha, now at UCSD. Gold triplets of nanoparticles on these
surfaces using DNA origami. They did however, they didn't want to do it on a
surface, the yield of the coupling was too low. They got some aggregation, but
they lucked out a lot, the triangle has this nice feature such that they
dimerize or trimerize, they bind and are oriented, there's a monomer, they got
lucky, but it's great stuff anyway. So anyway. There's lots of stuff to talk
about, so I'll end there.

Before the origami, did everyone build a FET out of CNTs? With randomly
spraying CNTs on a surface, you can find the crosses. The first one, where you
explicitly use molecular self-assembly to do it. Otherwise we don't achieve a
better yield than any of these other things. Are the behavior of these FETs,
their capacitance is very tiny? We didn't explore how fast they switch. You'd
be surprised. The number of devices that you can compare to the device that we
can make, is like 2. People abandoned messing around with these FET CNTs. So
everyone moved on because it was so hard. You can try to make 60 devices, we
got one. But we didn't do any better than people who did it before. We did it
in a different method, one in principle gives you 2D organization with
self-organization. When we combine with other stuff, do you see ways in getting
rid of building, getting some semiconductor straight out of it? Pulling a
semiconductor into a sandwich. I am not a big fan of CNTs. A lot of the things
we might go back and revisit, but organizing or functionalizing, pulling down
semiconductor quantum dots or silver wires, will probably be the next steps.
Even if you have 95% yield, how do you identify the things that work? How can
you purify the devices for those that work? In each stage where there is less
than 99.99% yield, what's the issue at that stage? The one point of
intervention is purity of the scaffold. There is a funnier purification in
things that are like impartial triangles don't bind to this site, so there's
some purification that happens there. At every step of the way you have to
figure out what you're going to do to maximize yield. I don't know how to say
it. You'll work out some system and say, this is the device, this is the
system, blah. When only exploring this, I don't care about yield. The way we
approach lithography now is that you can imagine a self-healing device. You can
build checksums like "hey look, I'm totally wrong", send a bunch of power, melt
me, re-anneal, and try again. Self-healing crystals that have those kinds of
properties, where the defects fall out and the thing can re-grow. The challenge
is figuring out an end-to-end process that works with the devices. A lot of the
concepts in this thing, when we actualyl build things, you're dry, the DNA is
dead, imagining a way where the self-assembly with later stuff, it could be
done, it's done in our bodies, but how you put that all together requires 50
years of cleverness. How do we adapt DNA nanotech to the wants of the
electronic industry? Well. Maybe. That's the thing. Is there a fundamental
incompatibility? We have error rates that are still too high. We have lots of
problems that we have to optimize and debug. I don't think so, if you look,
there are steps that are in the making of certain devices. There's this outfit
called Alien Technologies and they make RFID chips and whatever and what they
have are certain processes in the semicodncutor industry, RF subsets versus
whatever, are just incompatible. They have this big machine that makes this
surface, they make offline in another process, these little circuits. They pour
over a billion of them, they slot down in their holes, they have a defect rate
is low enough. It's a fluidic self-assembly process. That kind of existence
proof shows me that we could get there. I don't feel like it's fundamentally
incompatible. 

Tethering components to study molecular interactions. Shih is going in that
direction. The goal is 5 years from now to have 5 biophysical lab, and 5
physical sciences labs, to do some experiment they couldn't otherwise do
without this tech. Single molecule guys now, take some protein, or some protein
DNA complex, spray it, have a fluorphore or two, there is low signal, it looks
like noise, you get these blips. If you look at those on a grid, you could
forget everything except 1 or 2 microns on center. Orientation, position. Right
now, if you want to put a single molecule on a spot, you could spot it. You get
a Poisson number of molecules in each hole. That's great, you can assay him.
The bioscience guys have a big array, a polymerase in each hole. They can do
sequencing really fast. What if you want to study proteins in a group? By the
time you take Poisson statistics to the fourth pattern, the number of guys with
that, is very small. With DNA origami, you have 99.9% with those four guys, and
95% of the holes filled. Is anyone trying to do next gen sequencing. Tethering
single molecules on sheets?