Session 5: Nanofabrication via Structural DNA

00:00:07.15 My name is William Shih.
00:00:08.23 I'm an associate professor of Biological Chemistry
00:00:11.15 and Molecular Pharmacology
00:00:12.23 at Harvard Medical School,
00:00:14.11 Dana-Farber Cancer Institutes,
00:00:16.06 and the Wyss Institute for Biologically Inspired Engineering.
00:00:19.22 It's my pleasure to share with you today
00:00:21.25 some recent technical advances
00:00:23.13 in the field of structural DNA nanotechnology,
00:00:26.29 from my laboratory and those of my colleagues.
00:00:30.04 We're all familiar with the biological role of DNA
00:00:33.02 as an information repository,
00:00:35.22 principally for coding for protein sequence
00:00:38.06 and for regulation of protein expression.
00:00:41.08 And I'm not going to speak at all about that today.
00:00:44.08 Instead, I'll be talking about using DNA itself
00:00:47.12 as a building material
00:00:49.02 and harnessing that in order to us
00:00:51.04 to construct nanoscale objects.
00:00:53.17 For example, shown here is an electron micrograph
00:00:55.22 of actin filaments
00:00:57.19 that are about 7 nanometers (nm) in diameter.
00:01:01.04 And below, we can see this peculiar
00:01:03.10 Pac-Man-shaped object.
00:01:05.06 This is a structure that was built entirely from DNA
00:01:08.24 and we designed it for the purpose
00:01:10.27 of taking bite-size chunks
00:01:12.15 out of the actin filament.
00:01:14.10 Well, this project hasn't yet been successfully,
00:01:17.06 however hopefully this image drives home
00:01:19.05 the complementarity between the dimensions
00:01:21.16 of our designed DNA nanostructures
00:01:24.08 and biological macromolecular complexes.
00:01:28.01 So I've been working in this field
00:01:29.11 for over a decade now,
00:01:31.10 and I'm continually surprised
00:01:33.05 by advances in the field.
00:01:34.25 I have these preconceived notions of the limitations of DNA,
00:01:37.21 and they're always shattered
00:01:38.25 by the latest new discovery.
00:01:40.19 And today I'm going to be sharing with you,
00:01:42.10 in the first two sections, respectively,
00:01:44.25 two recent developments
00:01:46.16 that enable us to design and assemble
00:01:48.20 DNA nanostructures of the size and complexity
00:01:51.09 of this object shown here,
00:01:52.18 about 30 nm in diameter.
00:01:55.14 One of the most important goals for DNA nanotechnology
00:01:58.18 is to self-assemble objects
00:02:02.06 of ever-increasing complexity over time.
00:02:05.06 For example, is it possible that,
00:02:07.29 within the next decade or two,
00:02:09.14 that we can self-assemble objects
00:02:11.12 that are let's say a thousand times as complicated,
00:02:14.12 that have a thousand times as many unique components,
00:02:17.05 as the object shown here?
00:02:20.02 A second question is:
00:02:21.12 what are these things good for?
00:02:23.02 And in the third part of this lecture,
00:02:24.09 I'll be discussing some applications in my laboratory
00:02:27.11 as tools for molecular biophysics
00:02:29.22 and tools for future therapeutics
00:02:32.02 where we think these objects might prove useful.
00:02:35.00 We're inspired from natural systems,
00:02:37.02 we know that they can carry out many amazing behaviors:
00:02:40.18 they can build, they can adapt,
00:02:42.15 they can heal, they can reproduce.
00:02:45.05 And these are capabilities that human technology
00:02:47.04 struggles to reproduce on any kind of length scale,
00:02:50.00 but what's especially remarkable
00:02:52.00 is the ability of life to carry these out
00:02:54.08 on the molecular scale.
00:02:55.23 So, for example, here on the left
00:02:57.16 is a picture of a ribosome,
00:03:00.03 and this is of course a machine that's about 25 nm in diameter
00:03:04.05 that takes information encoded in messenger RNAs
00:03:07.10 and then translates that into
00:03:08.21 a specific sequence of amino acids,
00:03:10.26 to produce a polypeptide.
00:03:13.18 Quite amazing machine.
00:03:16.06 On the right, we have the T4 bacteriophage,
00:03:19.06 that's a bit larger in length scale.
00:03:21.02 We can see that the viral capsid itself -- head capsid --
00:03:23.26 is about 100 nm in diameter,
00:03:25.17 so it looks like a little nanoscale hypodermic syringe
00:03:27.19 that docks onto the surface of its bacterial hosts
00:03:30.26 and then injects its DNA cargo into the cell
00:03:33.25 against an osmotic pressure gradient.
00:03:37.05 Of course what makes this all possible
00:03:39.00 is that living systems have invented molecular manufacturing
00:03:43.08 and they've come up with a very robust and clever way of doing this.
00:03:46.08 They, of course, synthesize these biopolymers
00:03:49.07 -- they can be polypeptide chains, polynucleotide chains --
00:03:52.11 that then self-assemble into the desired structure.
00:03:56.25 Now, if somebody asked you about ten years ago,
00:04:00.08 "Would it be possible to generate an object
00:04:03.02 of this complexity using some kind of human-based technology?"
00:04:06.00 most people would have been skeptical.
00:04:08.17 And yet I'm going to show you a new technology,
00:04:10.27 DNA nanotechnology, developed in the last...
00:04:14.04 especially with advances from the last ten years,
00:04:16.24 that now make it possible for us to self-assemble,
00:04:19.12 in a programmable way,
00:04:21.12 structures of the same kind of complexity
00:04:23.12 as you see here.
00:04:24.26 Not yet of the functional complexity,
00:04:27.01 but nevertheless we think this is an encouraging first step
00:04:30.02 because along the pathway to this kind of functionality,
00:04:33.16 we first need to master structural complexity.
00:04:39.21 We're going to be using DNA as our building material,
00:04:43.04 and we know that DNA,
00:04:45.22 very similar to proteins and other macromolecules from life,
00:04:48.26 are very complicated molecules.
00:04:51.07 There's many different atoms,
00:04:52.24 but it turns out the key point for DNA nanotechnology
00:04:55.26 is that the robust base-pairing properties of DNA
00:04:58.20 allow us to abstract away those chemical details,
00:05:01.26 which is going to make the act of
00:05:04.00 designing the nanostructures much simpler.
00:05:06.23 And in fact there's only three characteristics of the DNA
00:05:09.19 that we need to remember
00:05:10.28 for the purpose of DNA nanoconstruction.
00:05:13.13 One, is that it's a ladder with antiparallel strands.
00:05:17.24 Secondly, there's a right-handed twist for B-DNA,
00:05:21.04 and we need to know that the twist
00:05:22.23 is around 10.5 basepairs per turn.
00:05:25.15 It turns out we can switch that around a little bit.
00:05:28.29 And finally we need to know that
00:05:31.11 A pairs with T,
00:05:32.16 C pairs with G,
00:05:34.08 and anytime you deviate from that pairing,
00:05:36.20 you're going to destabilize the structure.
00:05:39.25 And it's because the propensity of DNA
00:05:42.17 to form this very regular structure,
00:05:44.21 enforced very strictly according
00:05:46.08 to this Watson-Crick base pairing,
00:05:48.03 that gives us its power in being able to
00:05:50.19 generate these large structures with very little design work.
00:05:54.10 The father of the field of DNA nanotechnology
00:05:56.23 is Ned Seeman at NYU.
00:05:58.28 He invented this field about thirty years ago.
00:06:02.10 His training is as a crystallographer,
00:06:04.07 and the way he came up with the idea is as follows:
00:06:06.16 he was sitting in the campus pub one day,
00:06:08.20 just drinking his beer, and suddenly
00:06:10.24 what popped into his head was this woodcut
00:06:12.29 from M.C. Escher.
00:06:15.00 He had been collaborating with some of his friends
00:06:17.14 on DNA Holliday junctions,
00:06:19.28 and he had this eureka moment:
00:06:21.11 why not replace the flying fish
00:06:23.23 with DNA Holliday junctions?
00:06:25.24 The notion was that if you could rationally design
00:06:28.18 a porous crystal out of DNA,
00:06:30.24 and then he could take the target protein he was interested in,
00:06:33.02 and then dock that into each unit cell
00:06:35.02 into a stereotyped orientation,
00:06:37.13 then he would be able to impose that crystalline order
00:06:40.01 on the target protein,
00:06:41.19 and therefore make the X-ray crystallography problem easier
00:06:45.02 for these large macromolecules
00:06:46.20 that are otherwise difficult to crystallize.
00:06:50.20 He's been working on this problem for over thirty years,
00:06:53.00 it's an important goal,
00:06:54.01 and he's made some interesting progress,
00:06:55.25 I'll have more to say about that in the third segment.
00:06:59.26 But in the meantime, he's had some interesting
00:07:01.25 landmark successes.
00:07:03.25 So the first really noteworthy advance that he reported
00:07:07.13 came out in 1992 or so in Nature,
00:07:10.14 so this is a DNA cube
00:07:12.20 where each edge of the wireframe cube
00:07:14.20 is two turns of a double helix.
00:07:16.27 Each face is a circular strand of DNA
00:07:20.12 and the entire object has dimensions
00:07:22.09 of about 10 nm.
00:07:25.02 So the first time I think people saw this they thought,
00:07:27.15 "Wow, this is really cool,
00:07:29.21 but that doesn't look like biology to me."
00:07:32.22 And what I hope convince you today
00:07:34.17 is that this is in fact an extremely powerful technology.
00:07:38.12 Yes, it's fun,
00:07:39.18 but it's actually potentially very useful as well,
00:07:42.01 for many different applications.
00:07:48.06 Of course if we're building from DNA strands
00:07:50.19 and we're just making double helices
00:07:52.03 then that's boring.
00:07:53.15 The power of DNA nanotechnology
00:07:55.08 is that we can build with branched junctions.
00:07:57.27 With the previous example, the cube,
00:07:59.13 each one of those vertices
00:08:00.25 is a three-branch junction.
00:08:02.23 But it turns out the most powerful motif
00:08:04.21 so far in structural DNA nanotechnology
00:08:07.13 has been a four-branch junction;
00:08:09.09 a Holliday junction.
00:08:11.03 So on the upper left here we have a schematic
00:08:14.07 using simple letter notation of the strands,
00:08:17.22 so you can see the cyan strand starts
00:08:19.22 5'-CCGG, goes to it's 3' and,
00:08:23.05 and if you look closely at this you can see that
00:08:24.23 there's four different sequences
00:08:27.06 and they have the proper sequence complementarity
00:08:29.17 in order to generate a Holliday junction
00:08:31.20 that actually is immobile.
00:08:33.14 It can't branch migrate due to its sequence.
00:08:36.23 And we know from structural studies
00:08:38.15 that this object likes to stack into two double helices
00:08:42.12 that are connected at a joint,
00:08:44.18 and it turns out this is really the building block
00:08:46.10 that's been the most fruitful for DNA nanotechnology.
00:08:49.29 So the idea is as follows:
00:08:51.23 if you only have one Holliday junction,
00:08:53.28 now you have two helices that can wobble around.
00:08:56.13 In order to fix those two helices
00:08:58.01 to make a rigid building block,
00:08:59.22 what we do is we simply introduce
00:09:01.15 a second Holliday junction downstream,
00:09:04.00 and now when we fix those two double helices
00:09:06.04 with two Holliday junctions
00:09:07.17 we have that rigid building block that we want,
00:09:09.23 now with four sticky ends.
00:09:12.10 The next step is to build two versions of this building block.
00:09:15.24 In this example, we have a red one
00:09:17.24 and we have a blue one.
00:09:21.02 And we designed the sticky ends with the following
00:09:22.06 complementarity in this example.
00:09:24.17 So let's say we make the sticky ends
00:09:26.05 on the upper right-hand side of the red block,
00:09:29.02 and we make that compatible with
00:09:31.19 the lower left-hand side of the blue block.
00:09:35.00 And so on and so forth,
00:09:36.10 in order to create this kind of checkerboard fashion,
00:09:38.24 hopefully you can see that we would be able to self-assemble
00:09:41.15 these two bricks into an infinite two-dimensional lattice,
00:09:44.12 as shown below.
00:09:46.25 I don't have the experimental images for this,
00:09:48.29 but suffice it to say that this method actually worked.
00:09:52.24 Quite amazing - you can design a two-dimensional crystal.
00:09:56.29 The step after this would be to say,
00:09:58.21 "Well, instead of two bricks, two tiles,
00:10:02.02 what if I had ten tiles,
00:10:03.23 or what if I had 100 tiles?
00:10:05.05 Can I now make non-periodic structures
00:10:07.11 that are highly complex,
00:10:09.07 just with self-assembling tiles?"
00:10:10.26 And unfortunately, nobody has really
00:10:11.25 demonstrated this method,
00:10:14.12 extending this particular method to 100's of tiles,
00:10:17.09 although what I'll show you shortly is one method,
00:10:20.14 DNA Origami, that can achieve this kind of complexity,
00:10:23.03 and in the second segment,
00:10:25.06 something called single-stranded bricks,
00:10:26.17 that can do something very similar
00:10:27.29 to what I just described.
00:10:30.23 The method of DNA Origami
00:10:33.03 is a particular flavor of structural DNA nanotechnology.
00:10:36.05 It was developed by Paul Rothemund at Caltech,
00:10:38.18 he published this in 2006,
00:10:40.28 and the basic idea is as follows:
00:10:43.22 so imagine you have a long single strand of DNA,
00:10:46.16 the 7000 base genome of the M13 bacteriophage,
00:10:50.05 that's the gray strand in this animation.
00:10:52.21 We know what the sequence is,
00:10:54.01 and based on that known sequence,
00:10:56.01 we chemically synthesize 100's of short oligonucleotides
00:10:59.11 that are 20-60 bases long
00:11:01.17 that are programmed by Watson-Crick complementarity
00:11:04.16 to pinch that long strand
00:11:06.06 into a parallel array of helices,
00:11:08.17 after heating everything up to about 65°C
00:11:11.24 and then cooling down to room temperature
00:11:13.28 over the course of an hour.
00:11:16.11 At the end of the assembly,
00:11:18.05 you end up with this parallel array
00:11:20.09 of double helices
00:11:22.06 where adjacent double helices are held together
00:11:24.01 by these Holliday junction crossovers
00:11:25.28 that I described to you a couple slides ago.
00:11:27.27 So this is a half-crossover
00:11:29.11 and then here we have a full DNA crossover.
00:11:34.10 Importantly, what you just saw was an animation,
00:11:36.18 not a simulation.
00:11:37.24 In fact, we have a very poor understand
00:11:39.15 of the order of events of folding these objects,
00:11:42.19 we just know that if we program them
00:11:44.18 in a way where all of the scaffold ends up basepaired
00:11:47.24 to staple strands,
00:11:49.12 then we have an extremely high probability
00:11:50.27 of forming the desired structure.
00:11:53.04 So it's a very active area of research for us to
00:11:55.01 try to understand better the mechanism of folding,
00:11:58.03 and we hope that will actually help us to design more
00:12:01.01 complex structures in the future.
00:12:03.16 So Paul Rothemund used this method in 2006
00:12:05.29 to make structures such as this
00:12:07.24 'disc with three holes'
00:12:09.16 that has dimensions of about 100 nm x 100 nm x 2 nm,
00:12:13.17 this is an atomic force micrograph.
00:12:15.17 The example in the upper left-hand corner represents, in size,
00:12:18.14 just part of the upper lip of the object.
00:12:21.04 So this is quite large by macromolecular standards.
00:12:23.29 It's like we have two ribosomes worth of molecular Silly Putty
00:12:27.07 that we can mash into any desired
00:12:29.16 two-dimensional cookie cutter shape.
00:12:32.02 One of the very interesting things that he pioneered
00:12:34.12 was that he developed a way to
00:12:36.22 make this DNA origami
00:12:38.11 where he made each one of the staple strands
00:12:40.17 in two different flavors.
00:12:42.11 So one flavor
00:12:44.01 just made the structure as you saw.
00:12:46.17 The second flavor had the identical sequence,
00:12:49.01 but had a surface feature,
00:12:50.19 a dumbbell that's sticking out of one of its ends.
00:12:53.24 And so what that means is any time he used the original flavor
00:12:57.02 and he added it to the folding mix,
00:12:58.27 then you'd get a plain vanilla DNA Origami surface
00:13:02.05 at that location.
00:13:04.13 But then if he replaced that sequence
00:13:05.27 with the longer sequence, the one with the feature,
00:13:08.07 now you get that same shape,
00:13:09.25 but a bump over that feature.
00:13:12.22 And in that way,
00:13:13.27 he conceived that this rectangular DNA Origami
00:13:16.21 could be treated as a molecular breadboard,
00:13:18.28 where let's say it has 200 different positions,
00:13:21.17 we can decide at each position
00:13:23.06 whether we want to create a bump or have no bump.
00:13:26.18 In effect we have something that's like a bitmap,
00:13:28.28 and we can create new patterns
00:13:30.13 simply by repipetting different patterns
00:13:32.18 of the no-bump and plus-bump strands
00:13:35.01 for each one of the locations.
00:13:36.16 So for example, here we can see that he's designing
00:13:39.03 something that will say 'DNA'
00:13:41.01 and have a little picture of DNA.
00:13:42.20 These structures actually become very sticky at the ends
00:13:44.09 because they have lots of blunt ends,
00:13:46.08 and then they'll make a continuous ribbon
00:13:48.14 that says 'DNA'.
00:13:50.19 You can see that he made a map of the Americas.
00:13:53.14 He's a very humble guy,
00:13:55.01 so he apologized to the rest of the world
00:13:56.13 for stopping at the Americas,
00:13:57.25 but DNA is a little bit expensive,
00:14:00.14 so he stopped at the moment.
00:14:02.26 Maybe by now he's made the rest of the world.
00:14:06.20 And you can program them to link up in specific ways,
00:14:10.29 and in that way you can self-assemble
00:14:12.25 two-dimensional crystalline objects.
00:14:16.02 So what about getting to three dimensions, as I alluded?
00:14:18.28 Well, we can get our initial inspiration
00:14:21.07 from macroscale paper Origami,
00:14:23.14 where we're quite familiar
00:14:25.17 that if we fold flat paper in many ways
00:14:28.11 we can get quite intricate three-dimensional shapes.
00:14:30.29 So this is the famous crane.
00:14:34.00 And if you're really diabolical, like Robert Lang,
00:14:38.11 you might note that if you can fold these papers
00:14:42.02 in especially intricate ways,
00:14:44.13 then you can make incredibly complicated objects,
00:14:48.18 that we can see some examples of here.
00:14:51.18 Now, nothing I'm going to show you with DNA
00:14:53.15 is as complicated as this,
00:14:55.05 but again, as I mentioned,
00:14:56.13 one of our goals is to scale to ever-increasing complexity,
00:14:59.08 so we hope that someday we actually can
00:15:01.20 self-assemble DNA into objects of this kind of complexity.
00:15:07.00 So that group in Denmark that I just mentioned,
00:15:09.10 of Andersen, Jørgen Kjems, Kurt Gothelf,
00:15:12.23 they were able to design that M13
00:15:15.08 to fold into six different sheets,
00:15:18.13 and then they programmed those six sheets to fold up
00:15:20.18 into a three-dimensional box
00:15:22.00 with a hollow inside.
00:15:23.24 They designed a lid that can open in response
00:15:25.20 to some kind of molecular key.
00:15:27.26 So this was the first example of
00:15:30.02 a three-dimensional hollow DNA Origami.
00:15:35.18 So where my group wanted to contribute
00:15:37.14 was to make solid three-dimensional Origami structures,
00:15:41.02 and the idea is as follows:
00:15:44.00 so first of all, we know that we can
00:15:46.10 curl up DNA due to the helicity of the DNA helices,
00:15:49.16 and I'm going to go through a little thought experiment
00:15:51.17 just to give you a flavor of what this is about.
00:15:53.25 So here we have, on the far left,
00:15:55.25 three double helices that are arranged
00:15:57.13 into a little DNA Origami.
00:15:59.06 You can see, if you look closely,
00:16:00.25 they're connected by those Holliday junction crossovers
00:16:03.00 to keep the helices parallel.
00:16:06.14 And in this arrangement it's making a flat sheet of three helices.
00:16:10.12 So now imagine what would happen
00:16:12.11 if we moved these crossovers on the top
00:16:15.21 two base pairs to the left.
00:16:18.16 Then that's going to move that double helix
00:16:20.12 behind the plane of the page.
00:16:22.15 And likewise, if we move those two crossovers
00:16:24.14 two base pairs to the right,
00:16:26.09 that's going to move that double helix
00:16:27.29 in front of the plane of the page.
00:16:30.00 And the take-home message here is that
00:16:31.12 simply by shifting around the position of those crossovers
00:16:34.23 with respect to each other,
00:16:36.18 we can achieve curvature of these DNA Origami sheets
00:16:40.05 along the axis of the double helices.
00:16:42.21 So that's the first key.
00:16:45.00 So now let's extend that and build an actual
00:16:46.24 solid 3-dimensional structure.
00:16:49.00 So here we have another representation of a DNA Origami
00:16:51.03 where each one of these cylinders
00:16:52.19 represents one of those double helices,
00:16:54.15 so it's similar to the example in the upper left,
00:16:56.16 but now just rotated into this orientation.
00:17:00.00 So this would represent the pattern of the scaffold
00:17:02.12 running through those helices,
00:17:04.05 but for the purpose of this explanation,
00:17:06.09 I'm going to leave that invisible.
00:17:08.15 It's there, but I'm just not going to talk about it,
00:17:10.18 that or the staple strands.
00:17:12.15 And so what we're going to do is we're going to
00:17:14.08 shift around the position of those crossovers
00:17:16.24 so now these helices no longer prefer to be planar,
00:17:19.25 but instead prefer to curl up
00:17:22.25 into some kind of specific geometry.
00:17:25.14 And in this example what we're doing is
00:17:27.08 we're trying to curl up the structure into a corrugated S shape.
00:17:32.04 Furthermore, anywhere where we have the orange
00:17:34.29 that touches the white sheet that touches the blue sheet,
00:17:37.13 we're routing those staple strands through those interfaces.
00:17:40.15 So for example, we might have a staple strand
00:17:42.09 that starts 7 base pairs on this helix,
00:17:44.22 and then goes 7 base pairs here,
00:17:46.12 7 base pairs, 7 base pairs, 7, 7 base pairs.
00:17:50.04 And in that way, if the structure forms the way we intend it to,
00:17:53.03 it should be highly crosslinked
00:17:55.17 by these staple strands that are traversing the different helices.
00:17:59.26 So it looks good on paper, okay,
00:18:03.03 what happens in the test tube when we tried it?
00:18:05.19 And perhaps we can say,
00:18:07.18 "Of course, when we threw all the strands together
00:18:09.16 and tried to fold the object, then it didn't work."
00:18:12.24 We got a pile of molecular spaghetti
00:18:14.24 that we could see under the electron micrograph.
00:18:18.12 But we didn't want to give up,
00:18:20.04 and eventually Hendrik Dietz in the group
00:18:22.08 came up with a key insight,
00:18:24.01 which is it's not that these 3-dimesional objects
00:18:26.13 now are unstable thermodynamically,
00:18:28.28 simply they're more difficult to achieve kinetically.
00:18:32.23 And so what we found is that we could only get appreciable yields
00:18:34.28 of these objects
00:18:36.27 if we folded them instead of for an hour,
00:18:39.04 from 65°C to room temperature,
00:18:40.27 if we folded them for more like a week,
00:18:43.15 then we could start to get appreciable yields
00:18:45.16 of the objects ranging from 10-50% yield.
00:18:50.13 So we can see here one of the objects
00:18:52.16 that was built by Shawn Douglas.
00:18:54.10 Instead of 3 layers, it was with 10 layers.
00:18:57.26 And then we have the electron micrographs below.
00:19:00.29 We can see that we get a close resemblance
00:19:03.02 between what we observe in the electron micrograph
00:19:05.07 and the projection orientations
00:19:07.28 of our designed structure.
00:19:12.05 This is work that we published back in 2009,
00:19:14.22 in the meantime, our group and others have been hard at work
00:19:17.11 trying to improve the method.
00:19:18.26 So the important thing here was that
00:19:20.06 we could get something to fold at all,
00:19:22.08 and now we're trying to get better yields,
00:19:24.11 improve the folding times.
00:19:26.01 So there have been a couple of important discoveries since then.
00:19:28.21 One has come from Hendrik Dietz's lab in Munich,
00:19:31.20 where they've discovered that these structures
00:19:33.16 tend to have a favored temperature
00:19:35.20 at which they fold faster than the other temperatures.
00:19:37.27 So instead of spending the same amount of time
00:19:39.28 at 65°C down to room temperature,
00:19:43.10 for example this structure maybe folds faster at 50°C.
00:19:48.11 So what they found is if they do most of their folding at 50°C
00:19:51.13 then they can get it fold
00:19:53.01 maybe an order of magnitude faster,
00:19:55.25 which makes a lot of improvements
00:19:58.28 for our lives as scientists designing them,
00:20:00.19 they also suffer less thermal damage with a slower folding ramp.
00:20:04.28 We've also learned some details about
00:20:06.26 how to design the strands, the crossovers,
00:20:09.12 the breakpoints,
00:20:11.01 that I don't have time to go into in this presentation,
00:20:13.24 but I encourage you to look at some of our publications
00:20:16.00 if you want to see the latest discoveries
00:20:18.01 in how to make this process work better.
00:20:22.23 So now I'm going to go through a panel,
00:20:24.21 a gallery of different objects built using this method
00:20:27.17 by our laboratory to give you a flavor of the generality of the method. S
00:20:31.20 o the example on the top is what I just showed you,
00:20:33.28 we call it the "Monolith", it was built by Shawn Douglas.
00:20:37.08 You might say that it looks a little bit like a nanoscale crystal,
00:20:40.23 honeycomb array crystal,
00:20:43.06 but it's important to keep in mind that every element
00:20:45.23 of the object is associated with a unique sequence
00:20:48.25 and therefore is independently addressable.
00:20:51.24 This is quite different from most nanoparticles
00:20:53.22 that we see in synthetic nanotechnology today.
00:20:58.27 The example on the bottom was built by Franziska Graf,
00:21:01.15 we call it the "Genie Bottle".
00:21:02.29 We called it that because one version,
00:21:04.29 not shown here,
00:21:06.15 we only folded part of the M13 scaffold
00:21:08.15 and the rest of it was coming out of the lip
00:21:10.00 of the object kind of like wisps of smoke.
00:21:13.13 These are all 20 nm scale bars.
00:21:18.10 So here again, 20 nm scale bars,
00:21:20.20 on the left is an object built by Shawn Douglas,
00:21:22.26 we call it the "Square Nut".
00:21:24.28 It has a 7 nm hole in the middle,
00:21:27.23 it has a front end and a back end,
00:21:30.16 and if we make the sticky ends on the front end compatible
00:21:32.14 with the sticky ends on the back end,
00:21:34.07 then we can self-assemble filaments
00:21:36.24 that are somewhat reminiscent of
00:21:39.05 actin filaments and microtubules,
00:21:40.29 although in this case they don't yet demonstrate any dynamics.
00:21:43.23 They're just equilibrium formation of these long polymers.
00:21:48.07 On the right is an object built by Tim Liedl,
00:21:50.12 we call it the "Railed Bridge".
00:21:52.22 Again, every cylinder is one double helix,
00:21:54.22 and we can see as we go through cross-sections of the object
00:21:57.08 we have a different arrangement of double helices,
00:21:59.22 and we can understand from this example
00:22:02.20 that it is kind of analogous to sculpture.
00:22:06.02 That you could imagine the sculptor begins
00:22:08.01 with a solid block of marble,
00:22:10.02 in our case these parallel arrays of double helices,
00:22:12.28 and in design space we're chipping away
00:22:14.20 at that solid block to achieve whatever 3-dimensional structure
00:22:17.25 we actually want in relief.
00:22:20.15 Once we have our final design,
00:22:22.01 then what we're doing is we're compiling that 3-dimensional structure
00:22:26.07 into a series of DNA strands
00:22:28.12 that are going to self-assemble with the M13 scaffold
00:22:30.27 into that object.
00:22:36.00 Here's an object built by Björn Högberg
00:22:38.05 when he was in the group,
00:22:39.11 we called it the "Slotted Cross",
00:22:41.03 I'll have more to say about this object in the next slide.
00:22:43.20 This is another crossed object we called
00:22:45.24 the "Stacked Cross", built by Hendrik Dietz.
00:22:48.24 Again, these are all 20 nm scale bars.
00:22:50.19 This one looks a little bit like stacked molecular celery.
00:22:54.20 We even designed a little molecular cavity on the top
00:22:57.17 where we initially imagined we could host protein guests
00:23:00.00 on the inside of that cavity.
00:23:02.21 So let's take a closer look at that "Slotted Cross"
00:23:05.08 from Björn Högberg.
00:23:07.05 So here what he's done is he's generated an animation
00:23:10.25 where he's stylized the routing of the scaffold strand
00:23:13.25 through the structure.
00:23:15.18 It's designed as an "H-domain" and an "O-domain",
00:23:20.14 and the middle of the H-domain is designed
00:23:22.13 to pass through the middle of the O-domain
00:23:24.14 and it's all folding from just one long M13 scaffold.
00:23:28.24 I was quite amazed that this folded at all,
00:23:32.01 but the yields are not so great at the moment,
00:23:34.09 just a few percent.
00:23:38.25 So now what I'm doing is I'm zooming in
00:23:40.28 on what we call the strand diagram
00:23:42.22 that describes the blueprint of the object.
00:23:45.04 It's like we take all the helices
00:23:46.22 and then we splay them out
00:23:48.05 onto a 2-dimensional surface.
00:23:50.13 And in this case the blue represents that M13 scaffold strand
00:23:54.19 and those colored strands represent the staple strands.
00:23:57.17 And this part of the object is the upper left-hand corner
00:24:01.04 of the H-domain.
00:24:03.03 And so if you look closely you can see that the staple strands,
00:24:06.00 what they're doing is they're binding to part of the scaffold strand
00:24:08.12 and then they're crossing over to a different part of the scaffold strand
00:24:11.15 to pull those components together
00:24:13.23 to make the 3-dimesional shape.
00:24:16.19 We can zoom out,
00:24:18.07 and then here you can get an appreciation
00:24:20.26 that it is kind of like a blueprint.
00:24:22.19 You can make out which part is the H-domain
00:24:25.03 , which part is the O-domain,
00:24:27.02 and if you look closely you can actually see
00:24:29.15 where the H-domain and O-domain are being connected
00:24:31.16 by that long scaffold strand.
00:24:38.08 All of the examples that I've shown you so far
00:24:40.07 have been built using this honeycomb lattice paradigm
00:24:45.10 where we're using these corrugated sheets.
00:24:47.29 It turns out that it more naturally fits
00:24:50.07 the preferred twist of DNA
00:24:53.01 at 10.5 basepairs per turn,
00:24:55.13 but it turns out we can also self-assemble these
00:24:57.26 DNA sheets in a square lattice format.
00:25:03.10 The only proviso is that now we have to
00:25:05.29 underwind the DNA to 10.67 basepairs per turn,
00:25:10.05 which is slightly disfavored.
00:25:12.06 And, quite interesting,
00:25:13.22 what happens is the structure will still form,
00:25:15.27 but it then compensates
00:25:17.22 by having a global supertwist
00:25:20.11 in the right-handed direction.
00:25:21.26 So it's quite analogous to how plasmid DNA, for example,
00:25:25.15 will have a right-handed supertwist
00:25:27.11 when it's underwound, as we find in most cells.
00:25:34.13 One very important development in the field
00:25:37.24 is software with a graphical user interface
00:25:39.26 to make it accessible to people who are outside the field,
00:25:42.13 but also just to make the process faster,
00:25:45.09 more robust and convenient,
00:25:47.01 for experienced practitioners.
00:25:49.12 So for this really powerful software suite
00:25:52.06 called "cadnano",
00:25:54.05 we owe our thanks to Shawn Douglas,
00:25:55.22 he developed this software when he was a graduate student in my group,
00:25:58.18 now he's an assistant professor at UCSF,
00:26:01.24 at the time of this filming.
00:26:03.14 So I encourage you to check out the software,
00:26:06.01 he's continually improving it, at cadnano.org.
00:26:09.10 And what we can do, now again with the graphical user interface,
00:26:12.21 within an hour or so,
00:26:14.03 we can design different shapes
00:26:15.21 and then compile that into the sequence of DNA strands
00:26:18.20 that can self-assemble into that object.
00:26:25.07 What if you wanted to build larger structures?
00:26:28.08 Well, the most obvious idea is to
00:26:30.13 just get more parts.
00:26:32.00 So you can remember as a kid,
00:26:34.15 the first time you got a Lego set it was enthralling,
00:26:37.19 but then about two hours later,
00:26:39.07 you now were hungry for additional Lego pieces.
00:26:42.10 So that's the big drive for our field:
00:26:44.07 can we get more Lego pieces into the structure?
00:26:46.21 But in the meantime we can do other things
00:26:48.19 that will allow us to get a little bit bigger.
00:26:50.16 So one example is just to build with wireframes,
00:26:52.21 that have high strength-to-weight.
00:26:54.24 So in this example what we've done
00:26:56.10 is we've added staple strands that fold that M13 scaffold
00:26:59.18 into this wireframe structure.
00:27:03.09 Each one of these struts in this example
00:27:04.28 is a 6-helix DNA nanotube,
00:27:07.12 and then we designed sticky ends such that they're compatible,
00:27:10.21 and we can get this structure,
00:27:12.27 this Z-shaped structure,
00:27:14.17 to fold into a double triangle,
00:27:17.09 with now 10 of these 6-helix bundle termini,
00:27:20.17 each with a unique set of sticky ends.
00:27:23.15 In this example, what we did is we programmed
00:27:25.13 3 separate double triangles to form
00:27:27.11 in 3 separate test tubes,
00:27:29.15 and we programmed it to form this network on the bottom.
00:27:32.22 This is a Schlegel diagram,
00:27:34.08 and for those of you who might recognize this,
00:27:36.18 you might see that this is actually a Schlegel diagram
00:27:38.23 for a wireframe icosahedron.
00:27:41.25 This object has an overall diameter of about 100 nm,
00:27:45.18 each one of the struts has a length of about 45 nm.
00:27:49.25 And here on the lower left-hand corner
00:27:51.19 we can see an animation, macroscale animation, reenactment,
00:27:55.05 of the self-assembly of these double triangles
00:27:57.13 into a wireframe icosahedron.
00:28:07.24 What we find is that this process works in the test tube as well -
00:28:09.27 no hands required.
00:28:12.01 So again, what we do is we fold each of the double triangles
00:28:14.12 in three separate test tubes,
00:28:16.02 we then mix them together to form the desired wireframe object.
00:28:21.01 So let's take a look using electron microscopy.
00:28:23.18 So here we see with a 1 micron (um) scale bar,
00:28:26.01 we see a bunch of objects
00:28:28.06 that seem to have about the right size, about 100 nm in diameter.
00:28:31.07 There's aggregates as well,
00:28:32.19 so the self-assembly's not perfect,
00:28:35.14 but we're glass-half-full kind of folk,
00:28:38.03 we're encouraged by something that works
00:28:40.05 even partially.
00:28:42.04 So now we've zoomed in, you have a 500 nm scale bar,
00:28:45.00 and we can tell that there's some kind of wireframe action going on.
00:28:48.11 Zoom in some more,
00:28:49.23 now we have a 200 nm scale bar,
00:28:51.28 and it's starting to look like the wireframe structure
00:28:55.00 that we imagined.
00:28:56.27 Of course you have some mis-assemblies as well.
00:28:59.26 And then now if we go to the highest effective magnification
00:29:01.23 for this negative stain method,
00:29:03.17 100 nm scale bar,
00:29:05.11 we can see the objects, in fact,
00:29:06.29 they look like they have lots of these triangular faces,
00:29:09.18 they look like they have 5-fold vertices.
00:29:14.14 And we're able to make an object that now is
00:29:17.09 something like five times the mass of a ribosome,
00:29:20.17 it has overall dimensions the size of a medium size virus.
00:29:24.16 And this is all just powered
00:29:25.18 by Watson-Crick base pairing:
00:29:27.05 A pairs with T,
00:29:28.12 C pairs with G.
00:29:30.12 It's remarkable that we can push it this far,
00:29:32.25 ~but we're greedy and we dream about
00:29:34.13 being able to extend this to objects
00:29:36.17 that are a thousand times more complex
00:29:39.09 or even more than that some day.
00:29:43.26 Another kind of wireframe structure
00:29:45.25 from macroscale engineering
00:29:47.15 that inspired us are these floating compression sculptures
00:29:50.06 from the artist Ken Snelson.
00:29:53.10 And the idea here for these sculptures
00:29:55.11 is that you have these beams,
00:29:57.09 that are bearing compression,
00:29:58.14 that aren't touching each other directly,
00:30:00.09 but instead they're connected by cables
00:30:02.08 that are bearing tension.
00:30:03.20 And if you wire this up in the correct way,
00:30:06.15 then it's a balance between the
00:30:08.00 tension of the cables
00:30:09.20 and the compression of the beams,
00:30:11.00 and you end up with an object that has
00:30:12.03 high strength-to-weight
00:30:13.17 and has other interesting features.
00:30:15.09 For example, if those cables have some elasticity,
00:30:17.21 then if you put a global force on the object,
00:30:19.20 then it will deform,
00:30:21.17 and every individual strut will rearrange in 3-dimensional space.
00:30:26.05 When you now relieve that stress,
00:30:27.26 then it'll bounce back to the original shape.
00:30:30.05 So we wanted to see if we could implement this
00:30:32.07 using DNA Origami.
00:30:34.24 This is work that was led by Tim Liedl and Björn Högberg
00:30:37.22 when they were in the group,
00:30:38.16 in collaboration with Don Ingber.
00:30:40.26 So what they did was
00:30:42.09 to design the staple strands to fold this M13 scaffold
00:30:46.17 into 3 different struts,
00:30:49.11 each of the struts in this case is 13 helices.
00:30:52.05 It's actually grabbing 3 separate segments
00:30:54.11 of that scaffold in order to make each one
00:30:56.21 of those 13-helix struts.
00:30:59.28 So we again add everything together,
00:31:01.06 heat it up, cool it down,
00:31:03.00 and remarkably enough you can form structures
00:31:05.16 like this in the test tube.
00:31:07.27 In fact, we started to play games about
00:31:09.21 looking at how much stress we could put the objects under
00:31:13.05 and have them still fold.
00:31:14.19 So what happens is that you have these single-stranded
00:31:16.16 DNA elements that are acting like entropic coils
00:31:19.08 - they're exerting tension.
00:31:21.06 And if we simply design those cables to be shorter,
00:31:25.02 have fewer number of bases,
00:31:26.28 then it's going to exert a larger force
00:31:30.01 over the same design distance between the two compressed elements.
00:31:34.18 And what we found by continually shortening these cables
00:31:38.15 is that we could self-assemble the structures
00:31:40.18 up to about 14 piconewtons of force,
00:31:43.09 that was the calculated force for the shortest cable
00:31:45.20 that were able to self-assemble the objects.
00:31:48.25 In other words, we're able to self-assemble these DNA objects
00:31:52.02 against twice the force that can be generated by
00:31:55.14 powerful cytoskeletal motors such as kinesin or myosin.
00:31:59.09 This is all powered by just DNA base pairing.
00:32:03.17 We believe that these kind of structures may prove
00:32:07.26 useful for applications in tissue engineering
00:32:10.01 and regenerative medicine.
00:32:11.23 So of course cell biologists have noted for a while now
00:32:15.11 that cells, especially going through development,
00:32:17.29 can communicate with their outside environment,
00:32:20.28 with each other, using mechanics.
00:32:22.26 So they might pull on the extracellular matrix
00:32:25.00 and that extracellular matrix may pull back,
00:32:28.10 and you might have,
00:32:29.29 by introducing deformations into the extracellular matrix
00:32:33.03 or within the cytoskeleton of the cell,
00:32:35.25 you can trigger biochemical events.
00:32:37.29 So we envision a day where we can
00:32:39.11 use these kind of DNA nanostructures
00:32:41.15 that can deform in response to some kind of mechanical stress
00:32:45.10 and then translate into a biochemical event,
00:32:47.05 it could be release of a growth factor,
00:32:49.27 or maybe it could involve catalysis of some kind of chemical reaction.
00:32:54.14 So we believe that this could be useful for regenerative medicine.
00:33:00.02 So the last thing that I'd like to show you for this section
00:33:03.04 is work from Hendrik Dietz,
00:33:05.13 Shawn Douglas assisted on this work.
00:33:07.18 Everything that I've shown you thus far has involved double helices
00:33:10.18 that are straight.
00:33:12.14 And Hendrik wanted to ask the question,
00:33:14.07 "Could you build structures, curved structures,
00:33:16.24 where the helices now are following an arc,
00:33:18.28 instead of going straight?"
00:33:21.02 And the basic strategy for implementing this is as follows:
00:33:24.04 so here we have, again every cylinder represents a double helix,
00:33:27.21 these planes that are slicing through the double helices
00:33:31.06 represent the positions at which those crossovers are occurring.
00:33:34.14 So it turns out in this example
00:33:36.01 they're only occurring every 7 basepairs.
00:33:40.21 And he asked the question, well,
00:33:42.09 what would happen if he replaced
00:33:44.09 the double helical segments on the top, so the orange segments,
00:33:48.21 with shorter double helices
00:33:50.14 that only have 6 base pairs between planes.
00:33:53.23 And what if he replaced the helices on the bottom,
00:33:56.08 the blue ones,
00:33:57.25 instead of 7 basepair segments,
00:33:59.21 he had 8 basepair segments.
00:34:01.25 So mechanically, now on the top,
00:34:04.06 those elements are going to be under tension
00:34:07.09 because you have less material in the same amount of space,
00:34:09.15 they're going to be stretched out.
00:34:12.08 The helices are the bottom are going to be under compression,
00:34:14.29 because we now just stuffed more material
00:34:17.09 into the same amount of space.
00:34:19.24 And the system is under stress
00:34:21.07 and so it's going to relax, of course, by bending.
00:34:24.26 So this is the way to relieve that tension on the top
00:34:27.10 and compression on the bottom.
00:34:29.08 Does this actually work
00:34:30.06 when we attempt this in the test tube?
00:34:32.08 And the answer is yes.
00:34:34.08 So Hendrik implemented this,
00:34:36.02 using an 18 helix DNA bundle
00:34:39.13 that's illustrated on the upper left-hand side.
00:34:43.20 And so what he did was he had a stereotyped straight region,
00:34:46.18 these white regions,
00:34:48.09 and then he had an experimental region
00:34:50.19 that's indicated here in red.
00:34:52.24 So that's where he's going to be introducing
00:34:54.13 those longer and shorter elements
00:34:56.10 to induce the bending of the structure.
00:34:59.12 You can see for the control you get this nice rigid straight object.
00:35:04.18 So what happens when he introduces
00:35:06.17 some small number of shorter strands
00:35:08.21 in the double helices on top,
00:35:10.12 and then longer ones on the bottom.
00:35:11.27 He could get a reliably predicted
00:35:13.21 30 degree arc at that position.
00:35:16.27 If he has roughly twice the number of perturbations,
00:35:19.20 then you can get to a 60 degree angle.
00:35:24.14 Kept on going, you get 90 degree,
00:35:26.00 you can get a 120 degree angle,
00:35:27.19 that's quite remarkable.
00:35:30.03 This is now getting down to a 10 nm radius of curvature.
00:35:32.29 But then he kept on going,
00:35:35.07 and he found he could go all the way
00:35:36.21 to 180 degrees in this example.
00:35:39.29 So this is something that has a 6 nm radius of curvature
00:35:42.20 , it's comparable to the tightness of wrapping of
00:35:45.15 DNA double helices around histones in a nucleosome.
00:35:49.12 So in that case that's powered by protein-DNA interactions,
00:35:51.27 in our case this is powered by
00:35:54.04 DNA base pairing interactions.
00:35:56.22 So here what we have is an animation prepared by Shawn
00:35:59.28 that explains the bending principle.
00:36:02.20 So again, what we're doing is
00:36:03.28 we're introducing more basepairs,
00:36:08.00 or long double helices on the left,
00:36:10.05 and then shorter ones on the right.
00:36:14.05 And you can see a little graph on the lower left-hand side
00:36:16.15 that tells us how many basepairs per turn
00:36:18.13 that we have
00:36:20.05 for each of these different elements.
00:36:21.24 And at the most extreme example we're actually
00:36:23.17 getting 15 basepairs per turn on the left,
00:36:26.21 which is severely underwound,
00:36:29.02 and only six basepairs per turn on the right,
00:36:31.11 which is severely overwound.
00:36:33.19 And I was quite flabbergasted
00:36:35.27 that it should be possible for us to
00:36:37.26 torture DNA to this extent.
00:36:40.10 Now in fact once you get to those extremes,
00:36:41.27 our folding yields do start to go down,
00:36:44.05 so we can see that we're at the edge
00:36:46.02 of what we can do to DNA,
00:36:47.29 but still it's quite remarkable that DNA is so robust,
00:36:52.14 that the 10.5 basepairs per turn
00:36:54.26 is simply what it prefers to do,
00:36:56.11 but if you put enough stress on it,
00:36:58.01 you can make it do things that deviate
00:36:59.29 from that ideal by quite a bit.
00:37:06.20 So Hendrik and Shawn
00:37:08.14 now used the method to make a variety of different structures.
00:37:11.09 So on the upper left-hand corner
00:37:13.10 we have a 6-helix DNA bundle
00:37:15.03 that's folded into a series of
00:37:18.18 180 degree arcs of increasing radii of curvature,
00:37:21.27 so you make a spiral.
00:37:24.22 On the lower left-hand corner we have an object
00:37:27.02 that's programmed to self-assemble
00:37:29.07 into a beach ball, out of 6-helix bundles.
00:37:34.01 You can see objects that are making concave triangles,
00:37:37.19 this is designed by Shawn Douglas.
00:37:39.22 And then here we have those
00:37:41.19 120 degree arcs that are repurposed,
00:37:44.13 so we made sticky ends on the two ends
00:37:46.15 of this little boomerang to be complementary,
00:37:49.11 so that you can have three identical versions of them
00:37:51.27 will come together to make a larger triangle.
00:37:56.07 So in conclusion,
00:37:57.10 hopefully I've persuaded you that DNA Origami
00:38:00.09 is a highly versatile method
00:38:01.28 for building both 2-dimensional
00:38:03.19 and 3-dimensional structures of
00:38:06.06 quite remarkable complexity,
00:38:07.25 about twice the mass of a ribosome.
00:38:10.22 Where we're moving to next is to try to build structures
00:38:13.13 that are more complicated.
00:38:14.26 You might wonder what's preventing us from building
00:38:16.20 something 1000 times larger already, today.
00:38:19.12 And the main problem is that
00:38:20.21 we have errors in the self-assembly.
00:38:24.00 For example, for one of these objects,
00:38:26.09 we might have a yield, in the best cases,
00:38:29.06 75% of so,
00:38:30.22 which might sound pretty good.
00:38:32.11 But now if you wanted to build an object that's 1000 times bigger,
00:38:36.00 then you might argue that the probability,
00:38:38.02 if you just mixed these things together,
00:38:39.21 1000 of them together,
00:38:40.27 the probability that none of the 1000
00:38:43.13 would have any defect,
00:38:44.28 would be 0.75^1000,
00:38:47.08 which if you do the math, that's basically zero.
00:38:50.00 So there's a lot of activity in the field
00:38:52.12 trying to improve the fidelity of this self-assembly,
00:38:55.26 other methods like hierarchical self-assembly,
00:38:57.28 error correction,
00:38:59.08 that'll allow us to scale amount complexity
00:39:01.16 and build really very complicated objects of the future.

00:00:06.28 Welcome back -
00:00:08.04 this is the second part of the lecture
00:00:09.29 on structural DNA nanotechnology.
00:00:12.16 In the previous lecture, we discussed a method,
00:00:14.25 scaffolded DNA Origami,
00:00:16.13 that's proven powerful enough to self-assemble DNA strands
00:00:20.03 into objects that are about twice the mass of a ribosome,
00:00:23.00 about 5 megaDaltons in size,
00:00:25.02 involving a long single-stranded scaffold
00:00:26.28 that's folded by many short staple strands
00:00:29.12 into the desired object.
00:00:31.28 For the second segment,
00:00:33.08 I'm going to discuss a new method
00:00:35.17 that was just reported in the last year.
00:00:37.29 This is work that's primarily been led by
00:00:40.21 my colleague Peng Yin at Harvard
00:00:42.21 and the Wyss Institute
00:00:44.07 that he calls DNA Single-Stranded Bricks.
00:00:48.10 And it turns out that this method seems to be
00:00:50.24 roughly comparable in its power
00:00:53.06 of self-assembling structures
00:00:54.22 of this kind of complexity.
00:00:56.21 My group assisted in collaboration
00:00:58.22 at the endpoint of moving this into 3-dimensions.
00:01:01.15 I think it's a really interesting method,
00:01:03.01 that's why I'd like to discuss it in this
00:01:04.27 iBio seminar.
00:01:08.20 The scaffolded DNA Origami method
00:01:10.28 is analogous to some toys
00:01:12.24 that you might have played with.
00:01:14.10 So this is a DNA snake
00:01:16.00 that you can fold into 3-dimensional structures.
00:01:19.02 Here's another related idea
00:01:20.24 of a snake-like polymer
00:01:22.09 that we can fold into objects.
00:01:24.13 It's also familiar to biologists
00:01:26.17 who think about polypeptide chains
00:01:28.19 that are individual chains that fold up
00:01:30.16 into some kind of 3-dimensional configuration.
00:01:33.11 And in this way you can achieve
00:01:35.04 almost any shape by having a long polymer,
00:01:37.17 folding that up into that shape.
00:01:40.26 However, from a human point of view,
00:01:43.05 there might be a simpler way if that would actually work,
00:01:46.19 and that's using the example of Lego bricks.
00:01:49.08 So if you can imagine that you have a set of bricks
00:01:52.15 that have a stereotyped shape,
00:01:55.23 if we just have a lot of them
00:01:57.09 and we can connect them at different angles,
00:01:59.14 then now you could argue that gives us
00:02:01.08 even more design flexibility
00:02:03.07 in building these large 3-dimensional shapes.
00:02:05.10 You no longer have to worry about
00:02:06.21 the connectivity of the chain going through the object.
00:02:11.29 And what Peng Yin's group has demonstrated
00:02:14.03 is that, in fact, we can do this with DNA.
00:02:17.19 Now, when DNA Origami came out,
00:02:20.04 it was a shock to everybody that,
00:02:22.23 wow, we can build these very complex structures,
00:02:25.08 and then it was immediately assumed
00:02:27.02 that the key to success for the method
00:02:28.27 was the fact that you did have this very long strand
00:02:32.15 that keeps all the short strands into order.
00:02:35.04 It was a master template
00:02:36.25 and if you didn't have that
00:02:38.00 maybe everything would descend into chaos.
00:02:40.11 And what Peng Yin's group demonstrated is,
00:02:42.10 in fact, that's not so,
00:02:43.22 although we still think that long strand might be helping,
00:02:46.15 but now we know that it's certainly not a necessary component
00:02:49.07 for a successful strategy to build structures of this size.
00:02:53.21 So the first part of this story
00:02:54.24 was developed in Peng Yin's lab,
00:02:56.25 work led by Bryan Wei and Mingjie Dai in his lab.
00:03:02.17 And the idea is as follows:
00:03:04.05 so this is, if you remember from the first lecture,
00:03:06.12 we had that idea of a double-crossover tile
00:03:08.22 and we said,
00:03:09.25 "Oh, if we could only have double-crossover tiles
00:03:12.17 of many different sequences
00:03:14.03 and they would actually behave themselves,
00:03:15.22 then we could now make a complex tapestry
00:03:18.03 where every single one of the elements
00:03:19.14 has a unique sequence."
00:03:21.16 And what Bryan and his colleagues demonstrated
00:03:24.09 is that they could do that,
00:03:25.18 but with a slightly different motif.
00:03:27.25 The idea is as follows:
00:03:29.07 so you have each one of your bricks or tiles
00:03:31.12 has the same stereotyped architecture
00:03:33.06 where it has four different domains.
00:03:35.13 In each one of the domains
00:03:36.24 is one turn of a helix,
00:03:40.14 about let's say 10 bases long.
00:03:43.13 And it's a flexible polymer,
00:03:45.05 but in the final design tapestry
00:03:47.03 that it's supposed to self-assemble into,
00:03:49.06 each one of those tiles is supposed to adopt
00:03:51.16 a very fixed orientation
00:03:53.12 where it's like a horseshoe:
00:03:55.07 one half of the horseshoe is part of one double helix,
00:03:57.22 and then the other part of the horseshoe
00:03:59.09 is part of a second double helix.
00:04:01.23 And they assemble with each other
00:04:03.03 using the following rule:
00:04:04.23 that if you have, in your solution,
00:04:07.11 you might have Domain 1 of one of your bricks or tiles
00:04:11.04 is going to be compatible with Domain 3 of another tile,
00:04:14.09 and then Domain 2 of one of your tiles
00:04:16.07 is going to be compatible with Domain 4
00:04:18.21 of another one of the tiles.
00:04:20.24 Each one of these tiles in this assembly, in this tapestry,
00:04:23.18 has a unique sequence.
00:04:25.05 It has four unique nearest neighbors.
00:04:28.02 So if you can design all of these strands
00:04:30.14 with the same overall length and structure,
00:04:32.12 but each one with a different sequence,
00:04:34.04 and with the sequence complementarity rules that I just described,
00:04:37.00 it turns out you can make these objects
00:04:40.01 of the size and complexity of a DNA Origami.
00:04:42.20 No long strand required.
00:04:45.03 Here's another representation of that motif,
00:04:48.07 an abstract representation
00:04:50.01 that more closely resembles a Lego brick.
00:04:53.01 So what we have are, again, four domains.
00:04:56.07 We have Domain 1, 2, 3, and then 4.
00:05:01.11 And we can Domain 4 of one of these tiles
00:05:03.17 is now interacting with Domain 2
00:05:05.29 of another one of the tiles,
00:05:07.23 and we have the double helices
00:05:09.06 are running the lower left-hand corner
00:05:11.27 and now up to the upper right.
00:05:16.00 There's a stereospecificity that's indicated
00:05:19.15 by the shape of the key and the hole,
00:05:22.06 so we can hopefully understand
00:05:23.28 that the key and the hole can only interact in one orientation,
00:05:26.18 and that enforces this coplanarity of the tiles.
00:05:32.13 And the actual physical basis, of course,
00:05:34.18 is that each one of those interactions is
00:05:36.23 one complete turn of the double helix.
00:05:38.19 That forces it to be coplanar.
00:05:41.16 And once you understand those principals,
00:05:43.19 hopefully you can see that if you had a bunch of these tiles,
00:05:46.15 each with a different sequence,
00:05:48.02 each with those sequence complementarity
00:05:50.01 between plugs and holes that I described,
00:05:51.25 you can now self-assemble very large tapestries,
00:05:53.29 in principle,
00:05:55.09 where every location in the tapestry
00:05:57.00 is occupied by a unique tile.
00:06:00.13 And what Bryan and his colleagues demonstrated was,
00:06:02.15 remarkably, that this works
00:06:04.15 - no long strand required.
00:06:07.06 So here what they've done is
00:06:07.28 they've self-assembled a structure with something
00:06:10.11 on the order of a few hundred unique tiles,
00:06:13.00 each one with a stereotyped design position
00:06:15.16 within the tapestry,
00:06:17.03 and they found that they could make these
00:06:18.19 in fairly high yield.
00:06:20.13 So on the left we can see an agarose gel
00:06:22.22 where we can monitor roughly the formation of the object.
00:06:25.00 U means unpurified,
00:06:27.00 and we have the initial building blocks in the bottom.
00:06:29.15 You just cook these for a while,
00:06:31.29 again you do this annealing profile
00:06:33.02 where you heat to 65°C and you cool to room temperature
00:06:35.12 over the course of a day or so,
00:06:37.03 and then when you look on a gel after a day,
00:06:39.09 you can see a large fraction of these building blocks
00:06:41.17 have self-assembled into an object of discrete size.
00:06:45.06 Of course there's some mis-assemblies as well,
00:06:46.26 that's where the smears are coming from,
00:06:48.29 but then you can now cut out that band from the gel
00:06:51.28 and then you have a population of molecules
00:06:55.03 that are enriched for the one that you really want.
00:06:57.16 And in their case, they then looked at these objects
00:06:59.18 using atomic force microscopy,
00:07:02.07 and they say that they were making rectangles
00:07:04.07 of the desired shape and size.
00:07:08.01 First of all it's just amazing that,
00:07:10.12 to a lot of us, that this works.
00:07:12.12 You just throw all these sequences together,
00:07:14.00 there's no sequence design,
00:07:15.19 all of the plugs and holes were designed
00:07:17.23 using a random sequence generator,
00:07:19.20 and the method just works.
00:07:21.26 One of the remarkable aspects of this method
00:07:23.27 is that you can now generate new structures
00:07:26.19 simply by repipetting the strand sets
00:07:28.11 and leaving out strands.
00:07:30.09 So for example, if we imagine pipetting
00:07:32.23 that rectangle but we just leave out the strands corresponding
00:07:36.00 to eyes and the mouth,
00:07:38.03 then now we could generate something
00:07:40.00 like this smiley face.
00:07:44.29 And one could imagine, again, just having...
00:07:47.25 repipetting these strands
00:07:49.22 with different subsets
00:07:51.10 and you can now generate different shapes
00:07:53.02 in this way.
00:07:54.26 You can either pipet manually,
00:07:56.05 which becomes tedious if you try to build
00:07:58.20 something like the hundred objects that Peng Yin's group demonstrated.
00:08:02.00 What will be more efficient,
00:08:03.22 which they eventually implemented,
00:08:05.10 is if you have some pipetting robot
00:08:07.07 that actually does all the pipetting for you.
00:08:10.17 So maybe with the standard pipetting robot
00:08:13.01 to pipet the pools to build a hundred different objects
00:08:16.09 might take a couple of days,
00:08:18.02 but then that can be basically unsupervised.
00:08:23.29 And then comes currently a lot of hard work of the imaging,
00:08:27.04 so far Peng's group and my group
00:08:29.20 -- I'm not aware of any group that has
00:08:31.03 an automated imaging platform for these objects --
00:08:34.05 but after a lot of labor on the atomic force microscope,
00:08:37.18 one can see that something over
00:08:40.07 95% of the designed objects
00:08:42.02 actually were able to fold as predicted.
00:08:45.24 So we can see different letter,
00:08:47.00 we can see numbers,
00:08:48.17 Chinese characters,
00:08:49.22 emoticons,
00:08:51.27 we can see a journalist, Ed Jong,
00:08:53.24 was inspired so in Photoshop
00:08:55.27 he cut out some of these letters and made a message
00:08:58.01 that says "Wyss Institute for
00:08:59.22 Biologically Inspired Engineering at Harvard University".
00:09:03.29 So in the future we'd like to be able
00:09:05.15 to assemble the letters into this kind of arrangement
00:09:08.11 on their own, without the use of Photoshop,
00:09:11.03 but for now we think it's already an advance
00:09:13.04 that we can at least make the letters.
00:09:16.03 So here's a movie prepared
00:09:17.15 by Gael McGill that illustrates
00:09:19.29 how we imagine the self-assembly might occur.
00:09:23.06 Again, each one of these tiles has four nearest neighbors,
00:09:26.09 and at some point it's going to have to nucleate,
00:09:29.01 and then once you form a nucleus,
00:09:30.21 we believe that that will then grow to the larger structure.
00:09:34.24 Actually we think that the key to the success
00:09:36.24 of this method
00:09:38.12 is that we designed
00:09:41.27 it in a way that the nucleation is very slow
00:09:44.22 and the growth is very fast.
00:09:46.24 And in that way it's kind of like population control.
00:09:49.20 That any time you form a seed,
00:09:52.03 then it's going to have an abundant supply
00:09:54.07 of food or building blocks
00:09:56.03 in order to grow to its full size.
00:09:58.08 I mean imagine a situation
00:09:59.19 where nucleation was fast and growth was fast.
00:10:02.21 Then you'd basically get nuclei and seeds forming
00:10:05.21 everywhere,
00:10:07.06 and very quickly you'd deplete the pool
00:10:09.16 of building blocks
00:10:12.15 and at that point you'd be in trouble
00:10:13.25 because a lot of the seeds would have grown into
00:10:15.28 partial structures.
00:10:17.26 In order to complete their growth,
00:10:19.14 because there's no more building blocks,
00:10:20.22 they would have to start cannibalizing each other.
00:10:23.08 So we think that a robust design principle
00:10:26.01 for programmable self-assembly
00:10:29.00 is to try to build your system so that
00:10:31.28 nucleation is slow or controlled.
00:10:34.24 So we can see with DNA origami,
00:10:36.08 we can now envision those long scaffolds as controlled seeds,
00:10:40.03 that if we're adding in an excess of the staple strands,
00:10:43.06 then we know the number of seeds
00:10:45.01 is basically the number of those long strands
00:10:47.11 that we're adding.
00:10:48.20 And in that way you never run out of the building blocks.
00:10:51.25 In this case with the single-stranded tiles,
00:10:53.24 it's because that nucleation event is slow
00:10:56.27 and the growth is fast.
00:11:02.03 Alright, so I was just in the peanut gallery
00:11:04.27 watching this amazing work going on in
00:11:07.02 my colleague Peng Yin's lab.
00:11:10.00 Yonggang Ke is a postdoctoral fellow in my group.
00:11:12.08 Luvena Ong is a graduate student in Peng Yin's group.
00:11:15.28 And Yonggang and Luvena decided
00:11:17.09 they wanted to collaborate with Peng
00:11:20.11 and extend this into 3 dimensions.
00:11:23.10 So that's the work I'm going to tell you about next.
00:11:25.13 So just like we were able to extend 2-dimensional DNA Origami
00:11:28.15 into 3-dimensional solid structures,
00:11:30.17 Yonggang and Luvena were able to do this
00:11:32.26 using single-stranded bricks.
00:11:37.19 It turns out the principle for
00:11:39.24 converting from 2 dimensions to 3 dimensions
00:11:42.09 is extremely simple
00:11:44.29 - in principle, if it works.
00:11:46.14 So on the upper left-hand corner
00:11:47.29 we have the diagram that I showed you previously
00:11:50.09 - the 2-dimensional single-stranded tiles,
00:11:53.13 where since each one of these plugs and holes
00:11:56.08 is exactly one turn of the double helix,
00:11:58.16 that enforces a stereospecific geometry
00:12:01.21 between the tiles such that they're coplanar.
00:12:06.06 But if you think about it,
00:12:08.00 you could get something that's not coplanar
00:12:09.24 just by changing the length of those plugs and holes,
00:12:12.29 so that they're no longer integral numbers
00:12:14.17 of turns of the double helix.
00:12:15.21 So for example, here what Yonggang did
00:12:18.25 was he designed these plugs and holes
00:12:20.13 to be only 8 basepairs instead of 10.
00:12:24.01 And so now 8 basepairs
00:12:25.23 is roughly three quarters of a turn,
00:12:28.23 and because it's three quarters of a turn
00:12:30.16 then the stereospecific interaction between these bricks
00:12:33.14 is now going to form a dihedral angle of 90 degrees.
00:12:37.02 And we illustrate that
00:12:38.25 with the following arrangement of plugs and holes
00:12:41.15 so that you can see, again,
00:12:43.04 the key and the keyhole are only going to fit together
00:12:45.20 making that dihedral angle of 90 degrees,
00:12:48.11 and that's in physical reality enforced by the fact
00:12:51.01 that it's only three quarters of a turn,
00:12:53.14 8 basepairs interacting.
00:12:56.00 So now let's go through a thought experiment
00:12:57.25 that further elaborates this idea
00:13:00.04 that this 90 degree dihedral angle
00:13:01.27 allows the self-assembly
00:13:03.16 of a 3-dimensional solid cuboid structure.
00:13:06.27 Imagine that we have in our CAD program
00:13:09.01 a bunch of these single-stranded bricks,
00:13:11.28 and then the first thing that we do is
00:13:12.24 we lump together a bunch of these bricks
00:13:15.11 into these planar groupings.
00:13:17.11 And in this representation,
00:13:18.28 the bricks are not actually interacting with each other
00:13:21.06 with any base pairing,
00:13:22.24 we're just grouping them together in our CAD program
00:13:24.13 for explanatory purposes.
00:13:27.02 The next step is we generate another planar grouping of these bricks,
00:13:30.26 where we've now rotated the orientation of the bricks
00:13:33.02 by 90 degrees counterclockwise.
00:13:34.28 So hopefully by looking at the
00:13:37.25 orientation of these keyholes,
00:13:39.19 you can see that we've rotated the orientation
00:13:41.04 of the bricks by 90 degrees counterclockwise.
00:13:44.19 We can now repeat the process,
00:13:46.11 another 90 degrees counterclockwise,
00:13:48.06 another 90 degrees counterclockwise,
00:13:50.21 and then another 90 degrees counterclockwise.
00:13:54.25 Now the next step is we program those plugs and holes
00:13:57.16 to have unique sequence complementarity.
00:13:59.29 So for example, this plug here is going to be complementary
00:14:02.26 with this hole here, etc, etc.
00:14:05.18 Again, each one of these single-stranded bricks
00:14:07.13 has a unique sequence,
00:14:09.08 has four unique nearest neighbors,
00:14:11.17 and has the desired base complementarity
00:14:15.07 between those nearest neighbor domains.
00:14:18.07 And if you do that hopefully you can see
00:14:19.25 how you could self-assemble these different planes
00:14:22.17 into this cuboid structure,
00:14:24.26 in fact you'd just be throwing all those single-stranded bricks
00:14:27.08 together into a pool
00:14:28.25 and having them self-assemble just like before,
00:14:30.25 but now in 3 dimensions.
00:14:34.28 Furthermore, we can abstract this
00:14:36.17 in terms of the design process,
00:14:38.23 in terms of a 3-dimensional canvas,
00:14:41.23 a 3-dimensional cuboid canvas,
00:14:44.02 where each one of these volume elements, or voxels,
00:14:46.23 is 2.5 nm x 2.5 nm x 2.5 nm.
00:14:50.18 So in this case,
00:14:51.19 the double helices again are running
00:14:52.25 from the lower left-hand corner
00:14:54.17 to the upper right-hand corner.
00:14:58.13 And each one of these, again, it's 8 basepairs,
00:15:00.24 that represents one domain
00:15:03.09 from each of those bricks interacting with each other.
00:15:06.05 So in design space what we do is we
00:15:07.11 start from this 3-dimensional cuboid canvas,
00:15:10.09 we start removing voxels
00:15:12.10 until we end up with a 3-dimensional object that we want.
00:15:15.19 Then we have a computer program
00:15:17.10 that will compile this abstract voxel element representation
00:15:22.11 into the brick representation,
00:15:24.27 so the program will ask,
00:15:26.07 "OK, what series of bricks do I need to remove
00:15:29.00 in order to allow me to remove
00:15:31.12 individual volume elements."
00:15:34.22 Then whatever series of bricks
00:15:36.21 that are remaining for us to pipet,
00:15:38.20 that's now translated into instructions to the pipetting robot,
00:15:41.20 which will then go and pipet subsets of strands
00:15:44.04 corresponding to whatever kind of object
00:15:46.06 that we want to build.
00:15:48.24 So again, Peng loves the number 100,
00:15:52.08 so Yonggang and Luvena
00:15:55.01 strove to build over 100 different objects,
00:15:57.26 just like before but now in 3 dimensions.
00:15:59.27 This represents the different designs that can be created,
00:16:02.23 now we have letters that are in 3-dimensional relief,
00:16:06.03 we have Chinese characters
00:16:07.25 that are inscribed into 3-dimensional bricks/blocks,
00:16:11.03 same thing with numbers.
00:16:13.01 In this row here, it's an interesting representation
00:16:15.18 where now the solid is supposed to represent
00:16:19.13 bricks that we left out of the assembly,
00:16:22.22 and the translucent represents bricks that we left in.
00:16:26.10 So what this means is that this is
00:16:27.06 supposed to self-assemble into a solid object
00:16:30.09 with a completely enclosed cavity
00:16:32.15 that has a toroidal-type arrangement.
00:16:37.09 And then some pipetting was done by a pipetting robot,
00:16:40.10 so just feed the instructions to the robot,
00:16:41.26 come back in two days, and again,
00:16:43.21 we don't yet have an automated imaging platform,
00:16:45.21 so then there was a lot of work involved
00:16:49.16 in generating this figure
00:16:51.14 where we have electron micrographs now,
00:16:53.18 of these different objects.
00:16:54.21 These are projection images,
00:16:57.27 for example here we can see a little spaceship
00:17:00.24 that we were trying to design.
00:17:04.09 Here's an animation from Gael McGill
00:17:05.27 at Harvard Medical School
00:17:07.15 that is illustrating what we think
00:17:09.28 the dynamics of the structure might be.
00:17:15.28 So now I'm going to go through a series of
00:17:17.24 examples of different kinds of structures,
00:17:19.17 just to give you, again, a feeling of the generality.
00:17:22.03 Again, you start from this 3-dimensional canvas,
00:17:24.02 you start removing volume elements,
00:17:26.08 whittle it down until you get the object you want.
00:17:31.23 So here's an object with that cavity on the inside
00:17:35.08 I just mentioned.
00:17:36.28 So now when you image this in transmission electron microscopy,
00:17:40.01 you're going to get projection images
00:17:42.00 -- they're kind of like X-rays --
00:17:43.27 so if you look at the particles in different orientations,
00:17:46.13 then you'd expect to see different images.
00:17:49.27 So for example,
00:17:51.06 you'd expect to see the "O"
00:17:52.25 if you're looking from the top down,
00:17:54.21 but if you're looking from the side,
00:17:56.04 then you'd actually expect to see
00:17:57.26 something more like this.
00:18:03.03 Here's an object, it's a 3-dimensional smiley face.
00:18:06.13 Again, each one of these volume elements
00:18:08.11 is 2.5 nm x 2.5 nm x 2.5 nm, 8 base pairs.
00:18:13.28 And looking down from the top
00:18:15.11 we can see the smiley face,
00:18:16.22 looking from the side
00:18:17.23 then we see a different kind of image.
00:18:23.25 Here's an object that's designed to form
00:18:27.05 kind of like a 6-sided die
00:18:29.29 except it's a cheating die
00:18:31.26 and it only has 3 numbers,
00:18:33.16 so it has different crisscrossing channels
00:18:36.07 through the object and, again,
00:18:37.24 depending on which face lands on the grid,
00:18:39.15 you expect to see different images.
00:18:42.14 So, 1, 2, 3.
00:18:44.22 All the same object,
00:18:45.29 just landing in different orientations on the grid.
00:18:51.13 Here's an object that when you look from the top
00:18:53.17 is supposed to look like the letter "B"
00:18:55.15 and when you look from the side
00:18:56.21 is supposed to look like the letter "A".
00:18:58.21 And again, that's something that we can see.
00:19:03.28 Here's another object,
00:19:05.02 looks like "C" from the top,
00:19:06.15 and "D" from the side.
00:19:13.09 Here's an object with basically a channel in the top,
00:19:18.03 and if we look from the top
00:19:19.28 then we can see this characteristic channel pattern,
00:19:23.16 again if we look from the side,
00:19:25.10 then we can see we only removed strands
00:19:27.07 for part of the top of the object,
00:19:29.16 the bottom of the object remains solid.
00:19:34.12 For the 2-dimensional structures,
00:19:36.02 Peng's group developed some software
00:19:38.05 that allows them to quickly design
00:19:39.24 any shape they want.
00:19:41.15 So you can start from some kind of image
00:19:43.12 that you upload into the software,
00:19:45.14 the software will do edge detection
00:19:47.28 and then figure out where the boundaries
00:19:48.23 of the object are,
00:19:50.20 and then based on that algorithm
00:19:53.25 the program can automatically determine
00:19:55.24 which strands to include in the self-assembly,
00:19:57.29 which ones to leave out.
00:20:03.23 For the 3-dimensional structures,
00:20:05.27 this is something that's still in process,
00:20:07.11 but what Yonggang did was he took
00:20:10.01 his favorite 3-dimensional rendering program,
00:20:12.24 told it to render this series of volume elements,
00:20:15.22 and then just what he's doing now in real time
00:20:17.18 is he's carving channels into this cuboid structure,
00:20:22.20 so he's just removing channels.
00:20:25.23 In real time we can see him create two crisscrossing channels
00:20:29.00 that are orthogonal.
00:20:32.24 And this gives you a feeling that,
00:20:34.15 within just minutes,
00:20:36.06 you can now design any structure you want,
00:20:38.12 very much analogous to what a sculptor is doing.
00:20:41.04 But then it's going to take some time for the pipetting robot
00:20:43.08 to pipet all the strands,
00:20:44.23 for the folding to occur,
00:20:46.09 and then for the imaging,
00:20:48.02 it's going to be somewhat time-consuming
00:20:50.08 until we have an automated platform for that.
00:21:01.01 So just to recap,
00:21:02.12 we have a design phase
00:21:04.04 where we start from our canvas
00:21:05.24 -- 2-dimensional/3-dimensional canvas --
00:21:08.01 we figure out which of the bricks
00:21:11.05 we want to include/exclude.
00:21:13.06 That gets converted by software into pipetting instructions to the robot.
00:21:16.19 Robot does it's thing.
00:21:18.16 And then we heat up and cool down the strands
00:21:20.28 over the course of a day,
00:21:22.09 or longer for the more complicated objects,
00:21:25.09 and then we look at them using atomic force microscopy
00:21:28.07 or transmission electron microscopy.
00:21:33.06 It's always nice to have some more movies
00:21:34.19 so we can see the pipetting robot in action.
00:21:42.01 And we can envision hopefully
00:21:44.21 a day not too far from now where everything is automated,
00:21:47.21 so we can just design the objects
00:21:49.28 and then everything else will be handled automatically,
00:21:52.20 including the imaging.
00:21:53.29 It might make a wonderful resource for students that,
00:21:58.04 if they can go online,
00:21:59.18 submit their designs online
00:22:01.01 and then maybe there's a chance
00:22:02.28 that a lab will actually build the object in the laboratory
00:22:06.00 and then the student can see their object,
00:22:08.23 an electron micrograph
00:22:10.15 or an atomic force micrograph
00:22:11.24 of the object they designed.
00:22:14.02 To summarize up to this point,
00:22:15.22 now we have a second method that allows us to
00:22:18.14 generate objects that are
00:22:20.20 roughly twice the mass of a ribosome or larger,
00:22:23.20 that was just published in the last year
00:22:25.02 from Peng Yin's lab,
00:22:27.29 DNA tiles and bricks now,
00:22:29.29 that doesn't require a long single strand.
00:22:33.06 And for some applications you can imagine
00:22:35.02 with this overlapping capability,
00:22:37.05 you could arbitrarily choose which one you want to select.
00:22:40.22 However, when we look closer
00:22:42.04 we could imagine that the independent methods
00:22:44.13 might have different advantages depending on the application.
00:22:48.18 So for example, with DNA Origami,
00:22:50.18 we've noticed so far that the assemblies
00:22:52.15 seem to be faster.
00:22:53.27 So although that long strand doesn't seem to be absolutely necessary,
00:22:57.07 we could imagine it does help to speed things up
00:22:59.24 by grabbing the individual strands
00:23:01.08 and bringing them together more quickly.
00:23:04.08 A second advantage is that
00:23:05.22 we believe that the DNA Origami,
00:23:07.24 at least how it's currently constituted,
00:23:09.12 could be thermodynamically more stable
00:23:11.11 to have this long strand running through the entire object.
00:23:14.06 We could imagine the thought experiment of
00:23:15.11 what if we started from the DNA tiles
00:23:17.19 and then just started ligating some of those tiles or bricks together
00:23:20.15 to make a long strand.
00:23:21.28 Then it should be more stable,
00:23:24.05 so in that way we imagine DNA Origami
00:23:26.05 has more linkages between the strands,
00:23:27.29 longer strand,
00:23:29.08 then it should be more stable,
00:23:30.29 at least currently.
00:23:32.15 And finally we can imagine that DNA Origami
00:23:34.20 probably can offer greater mechanical strength,
00:23:37.20 that if you have that long scaffold strand
00:23:39.22 is crisscrossing throughout the entire structure,
00:23:42.17 now you have to break covalent bonds, probably,
00:23:44.26 in order to really disrupt the object.
00:23:46.25 Whereas with the DNA tile object,
00:23:48.22 now if you could imagine creating a facet, a breakage
00:23:52.22 without having to sever any covalent bonds.
00:23:56.05 So what are the potential advantages of
00:23:57.25 DNA tiles or bricks over Origami?
00:24:00.04 Well, one is that the design is more modular,
00:24:02.08 it corresponds more to our intuition of how Lego bricks
00:24:06.17 can be designed.
00:24:08.23 It's conceptually simpler
00:24:10.13 and that's usually something that is desirable.
00:24:13.12 Often times when the design process is simpler
00:24:15.12 then it's going to be more versatile
00:24:17.24 and more powerful.
00:24:19.11 It'll be better for teaching to students how this works.
00:24:23.06 And then finally the DNA tiles offer the
00:24:26.05 advantage of synthetic diversity,
00:24:28.05 because all of these elements are short strands
00:24:31.10 and they're accessible through synthetic chemistry,
00:24:34.05 which means we can put any kind of nucleoside analogue
00:24:36.11 that we want in there,
00:24:38.06 assuming it still base pairs,
00:24:39.27 whereas with the DNA Origami,
00:24:41.07 because it's relying on this long single strand,
00:24:44.10 currently our only way to generate these very long single strands
00:24:46.17 is enzymatically,
00:24:48.19 and therefore we're limited to those nucleoside triphosphates
00:24:51.06 that are recognized by DNA polymerases.
00:24:54.25 So where could this potentially be advantageous?
00:24:56.24 So let's say that you're trying to
00:24:58.06 self-assemble a drug delivery vehicle.
00:25:01.16 Maybe if you built it with DNA Origami,
00:25:03.11 you'd start to worry about, well,
00:25:05.14 maybe nucleases are going to digest my long strand.
00:25:08.14 Maybe my long strand is going to
00:25:09.28 trigger an innate immune response.
00:25:12.21 But if that's my concern,
00:25:14.11 then maybe I should think about designing the same kind of structure,
00:25:17.13 but with DNA bricks,
00:25:18.25 where I can use let's say mirror-image building blocks
00:25:21.15 that are nuclease resistant
00:25:23.18 and that are not recognized by the innate immune response.
00:25:27.14 What we found is that, again,
00:25:28.21 for these discrete objects,
00:25:30.13 maybe the performance of the two methods is similar,
00:25:32.26 but where the DNA brick method really seems to shine
00:25:36.00 is in building periodic structures.
00:25:38.13 So what we've done here is...
00:25:40.02 what Yonggang has done is
00:25:41.19 he's programmed the right-hand side
00:25:43.15 of this lightly shaded unit cell
00:25:45.20 to have complementary sticky ends
00:25:47.24 to the left hand side,
00:25:49.17 or complementary plugs and holes I should say,
00:25:51.25 and complementary plugs and holes
00:25:53.02 from the front end and the back end.
00:25:55.21 And so now what will happen is
00:25:57.09 that that unit cell won't stop with a discrete object,
00:25:59.19 it'll actually polymerize
00:26:01.23 into a 2-dimensional lattice.
00:26:03.20 Furthermore, it's not...
00:26:05.12 we don't think that it's forming hierarchically
00:26:07.02 - it's not that you form a bunch of unit cells
00:26:08.22 and those unit cells assemble.
00:26:10.13 Rather, we believe that the assembly
00:26:12.03 is growing piece by piece.
00:26:14.01 So each individual brick
00:26:15.06 is adding on one by one.
00:26:17.24 And in that way, looking at this periodic assembly,
00:26:20.04 actually, if you think about it
00:26:22.02 -- a thought experiment --
00:26:23.27 the definition of the unit cell now
00:26:25.25 is a little bit arbitrary,
00:26:27.13 because we could just as easily draw
00:26:28.29 a unit cell connecting these four corners.
00:26:31.06 It's equivalent with these periodic structures.
00:26:34.28 Anyway, the important thing is that
00:26:36.09 this single-stranded brick method
00:26:37.29 seems to give us better performance
00:26:39.18 in the test tube
00:26:40.23 in making these periodic structures.
00:26:43.08 So this is a quite remarkable design
00:26:45.14 that was developed by Yonggang,
00:26:47.19 where the helices are pointing up
00:26:49.28 out of the plane of the DNA crystal
00:26:52.11 and the unit cell has dimensions of
00:26:54.21 6 helices x 6 helices,
00:26:56.12 so about 15 nm x 15 nm.
00:27:01.16 And in this particular example,
00:27:03.01 he designed a cavity within the unit cell
00:27:05.19 of a 2x2 helix, helices that are missing.
00:27:09.10 And then what this is going to do is now
00:27:10.18 self-assemble into a crystal that,
00:27:13.03 where again the unit cell has dimensions of about 15 nm,
00:27:15.20 the holes dimensions of about 5 nm,
00:27:17.27 and the entire crystal can grow to
00:27:19.29 multiple microns in dimension.
00:27:22.24 We believe that these kinds of structures
00:27:24.15 could have application as template
00:27:26.29 for perhaps growing inorganic materials
00:27:29.16 to make molecular wires
00:27:31.06 and plasmonic devices.
00:27:32.29 We think that it might also have application in biology
00:27:35.16 for something like the host-guest crystallography
00:27:38.04 that Ned Seeman envisioned.
00:27:40.11 In this example it would be two dimensional,
00:27:42.24 so what if we could get membrane proteins
00:27:45.03 to assemble into stereotyped orientations
00:27:47.26 and locations
00:27:49.05 within these cavities
00:27:50.27 and use the DNA crystal
00:27:52.14 in order to impose that crystalline order
00:27:53.18 on those proteins.
00:27:55.10 That could be a way to accelerate structural biology research.
00:28:00.00 So this is just some more examples of
00:28:01.29 periodic 2-dimensional crystals.
00:28:03.21 In this case, what Yonggang is doing is
00:28:05.27 he's polymerizing in the direction of the helices,
00:28:09.10 so again, every cylinder is a double helix,
00:28:14.24 and we can see these precise channels.
00:28:16.11 It's the same story as with the discrete objects,
00:28:18.15 he just starts from a solid cuboid unit cell
00:28:21.15 and then removes strands in order to create
00:28:23.22 the cavity features,
00:28:25.14 and in this way can create
00:28:27.01 an extremely diverse set of crystals
00:28:29.04 with intricate features.
00:28:31.06 That is basically not accessible
00:28:33.28 using any other known method.
00:28:35.27 So this is an interesting example where what he did was
00:28:37.23 he made a very think crystal that was only
00:28:40.19 I believe 32 basepairs in height,
00:28:43.07 and now it turned out with his design,
00:28:46.11 the structure no longer wanted to be planar,
00:28:48.21 but instead had a tendency
00:28:50.16 to wrap around to make a tube.
00:28:53.19 And we can see these nanotubes
00:28:55.07 that have an appearance that's somewhat reminiscent
00:28:57.27 of biological assemblies such as...
00:28:59.25 this is a Tobacco mosaic virus.
00:29:02.29 Of course, this object is made entirely out of DNA
00:29:04.29 - it's not infective.
00:29:10.07 Yonggang and Wei Sun in Peng Yin's lab
00:29:12.16 have gone on to use these crystals
00:29:15.00 to template the self-assembly
00:29:16.25 of gold nanoparticles onto them.
00:29:18.16 Again, potentially for electronics
00:29:20.05 or photonics-type applications.
00:29:22.16 And so what they've done here is they've decorated
00:29:23.24 5 nm gold particles with single-stranded glue,
00:29:30.08 and then they have the complementary glue
00:29:31.24 that's lining the inside of these channels,
00:29:34.15 and in that way they're able to get high densities
00:29:36.12 of these gold particles into those channels.
00:29:39.10 Here, what they've done is they've just
00:29:41.09 coated the entire surface with a high density
00:29:44.04 of these 5 nm gold nanoparticles.
00:29:52.03 I should mention that although DNA Origami
00:29:54.08 is not as good as DNA bricks
00:29:56.08 for making these 2-dimensional structures,
00:29:58.01 it does have some ability to do that.
00:30:00.07 So this is work from Yonggang
00:30:01.23 that we didn't publish
00:30:03.08 where he built these honeycomb building blocks,
00:30:08.00 hexagonal building blocks that self-assemble
00:30:10.08 into a hexagonal crystal
00:30:11.24 that has similar dimensions as what I showed you before
00:30:14.01 - a couple microns x a couple microns.
00:30:16.24 And Ned Seeman's group published a very nice work
00:30:19.21 in which they designed a building block
00:30:21.17 that looks kind of like a two layer
00:30:24.13 Rodeman-style Origami
00:30:26.29 and we able to self-assemble this
00:30:29.11 into a rectangular array,
00:30:30.23 again a couple microns x a couple microns.
00:30:33.07 But I'd like to emphasize that with DNA Origami
00:30:35.22 it's just a couple of idiosyncratic cases
00:30:39.00 where we've been able to succeed
00:30:40.22 to build these very large crystals,
00:30:42.24 but with the single-stranded bricks,
00:30:44.12 it seems like most of the things we try work,
00:30:46.29 and it's just much, much easier to design.
00:30:48.24 You just leave out some strands
00:30:49.29 and then now you have a new crystal.
00:30:52.28 Thus far what we've observed
00:30:54.17 is that the DNA brick crystals seem to be more robust
00:30:57.29 than the scaffolded DNA Origami crystals.
00:31:00.16 With the Origami crystals,
00:31:02.01 we just have a couple of cases
00:31:03.06 where it seems to have worked.
00:31:04.17 With the DNA brick crystals,
00:31:05.22 it's very simple for us to just
00:31:07.00 leave out some of the strands
00:31:08.14 and make a new crystal,
00:31:10.05 and something that's more rigid, higher quality.
00:31:13.01 So hopefully in the future
00:31:15.00 we can develop methods for improving DNA Origami crystals,
00:31:18.03 but in the meantime we can speculate about why, currently,
00:31:22.09 the DNA brick crystals are forming better.
00:31:24.24 So we can do the thought experiment that maybe,
00:31:27.04 for the DNA Origami crystal,
00:31:28.24 you could imagine either pre-forming the unit cells...
00:31:32.25 you could imagine pre-forming the unit cells
00:31:34.17 and then now you mix them together,
00:31:36.04 and the problem is that because the unit cells are so large,
00:31:39.07 it can be very difficult to get reversible assemblies.
00:31:41.29 So you make so many contacts with the growing lattice
00:31:44.03 that it's hard to dislodge yourself.
00:31:45.24 And note that this the same kind of difficulty
00:31:47.21 that plagues macromolecular crystallography,
00:31:50.20 that it becomes very difficult to crystallize large complexes
00:31:53.15 for this reason, among others.
00:31:57.18 Let's contrast that with the DNA brick crystal growth,
00:32:00.05 where the growth is occurring
00:32:02.29 through these very short elements
00:32:04.23 that are only 32 bases long.
00:32:06.20 And because they're so short,
00:32:07.27 it's very easy for them to come in, come out.
00:32:10.02 If there's an error, it has a chance to leave.
00:32:12.27 But because there's many different kinds of bricks,
00:32:15.22 you can still achieve a very complicated unit cell.
00:32:19.03 So it's almost like you can have your cake and eat it too.
00:32:22.04 You can have a very complex unit cell,
00:32:24.04 but it's assembling one subcomponent at a time,
00:32:27.06 so you still get that reversible self-assembly
00:32:30.01 that seems to be critical for robust growth of a crystal.
00:32:38.19 So in conclusion,
00:32:39.29 what we've seen from the lab of Peng Yin
00:32:42.03 over the last year or two,
00:32:43.29 we also collaborated to help them on this,
00:32:46.12 is a fantastic new method for self-assembling structures
00:32:50.10 that are the size of a ribosome or maybe even larger.
00:32:53.17 We can build them in 2-dimensions,
00:32:54.27 we can build them in 3-dimensions.
00:32:57.03 Right now what it looks like is there's
00:32:58.21 a special advantage with these single-stranded bricks
00:33:01.07 in growing periodic structures,
00:33:03.13 and we believe this could have important applications
00:33:06.01 ranging from molecular electronics and photonics
00:33:08.20 to structural biology.

00:00:07.17 Welcome to the third part of this lecture
00:00:09.13 on structural DNA nanotechnology.
00:00:12.22 My colleague George Church said,
00:00:15.08 "The problem with your field is that it looks like
00:00:17.16 you're having too much fun."
00:00:20.00 And the reality is that learning how to build
00:00:23.13 with these structures is fun,
00:00:24.28 but we also firmly believe that the power behind it
00:00:28.07 also has to do with what kinds of applications
00:00:31.17 might be possible.
00:00:33.05 And in particular, we're very interested in applications
00:00:35.29 in molecular biophysics
00:00:37.26 and future therapeutics,
00:00:39.16 so I'm going to share with some of the research directions
00:00:42.04 in my laboratory trying to find useful applications
00:00:45.03 for these DNA nanostructures.
00:00:49.23 So the first application is...
00:00:52.01 we were able to create a tool that allows
00:00:55.04 the NMR structure determination
00:00:57.12 of an alpha helical membrane protein.
00:00:59.09 This is work that was done in collaboration
00:01:02.07 primarily with James Chou's group at Harvard Medical School.
00:01:07.11 So the story starts, again, with Ned Seeman,
00:01:09.18 the person who started the field
00:01:11.22 of structural DNA nanotechnology.
00:01:13.21 You might recall from the first part
00:01:15.18 that we had this vision of flying fish,
00:01:18.01 of the host-guest crystal,
00:01:20.04 and that maybe this would make structural biology
00:01:22.14 much easier if we had access to these host-guest crystals.
00:01:26.10 Ned Seeman's group made an important landmark discovery
00:01:29.20 along the road to this eventual goal.
00:01:32.11 In 2009, they reported this nice paper
00:01:34.17 where they rationally designed a unit cell
00:01:37.19 -- they called it a tensegrity triangle --
00:01:40.05 that basically has three crisscrossing double helices
00:01:43.17 that define three different axes in 3-dimensional space.
00:01:46.11 And if they program with sticky ends to self-assemble,
00:01:49.06 then they could actually make a macroscale crystal
00:01:52.08 that has dimensions of about 0.2 mm per side
00:01:56.06 and diffracted X-rays down to 4 Ångstrom resolution.
00:02:00.08 So this is an important first step.
00:02:02.01 A next step will be to improve the resolution of the crystal,
00:02:05.08 and then another difficult step on top of that
00:02:07.22 will be to get the target proteins to dock into very ordered,
00:02:11.06 stereotyped positioned
00:02:12.23 within each unit cell.
00:02:14.23 So I think this is a fantastic goal for the field,
00:02:18.05 maybe because it's very hard.
00:02:20.06 I think it's important to have hard goals to reach for,
00:02:23.00 but it may still take a while before we have a workable object
00:02:26.11 that will actually help us with 3-dimensional protein crystallization.
00:02:31.28 In the meantime what James Chou's group and my group have been able to do
00:02:35.28 is to do something that's technologically much more modest,
00:02:39.29 and yet it achieves the same end goal of
00:02:42.05 enabling atomic resolution protein structure determination.
00:02:45.24 And this is a method that's know as weak ordering.
00:02:49.04 It's something that I'll explain in a little bit,
00:02:51.03 something that's been known about for a number of decades now,
00:02:54.13 but hasn't really been well applicable
00:02:56.28 to membrane proteins until recently,
00:02:59.09 we believe because of our tool.
00:03:02.05 So host-guest crystallization,
00:03:04.24 you can think of that as a very strong sort of ordering,
00:03:07.09 that you want to force the protein
00:03:09.07 into a very stereotyped translational position,
00:03:12.01 very stereotyped rotational orientation.
00:03:15.06 And if that becomes too messed up,
00:03:17.24 that's going to destroy the process of structure determination.
00:03:24.09 In contrast, with weak ordering what we're doing
00:03:26.13 is we're just barely trying to change the population
00:03:29.21 of solution tumbling molecules away from isotropic.
00:03:32.15 So what does that mean?
00:03:34.08 So in a nutshell, we have this animation to explain the concept.
00:03:37.24 So imagine that the little red dots floating around
00:03:39.27 represent our target protein of interest
00:03:42.06 that we're trying to solve the structure of.
00:03:44.27 And what we want to do is to introduce weak order
00:03:47.26 into those proteins by mixing it into a dilute liquid crystal
00:03:51.27 of these long molecular rods.
00:03:55.11 So we can zoom in here - the notion is that
00:03:57.12 -- this is supposed to represent a membrane,
00:03:59.17 this is the detergent micelle solubilizing the membrane protein --
00:04:02.27 and the idea is that most proteins are not perfectly spherical in their shape,
00:04:07.12 and they'll have a higher tendency to bump into these molecular trees
00:04:11.25 if their long axis is perpendicular
00:04:13.23 to the long axis of the molecular trees -
00:04:17.16 is perpendicular.
00:04:19.18 And so, therefore, if you could
00:04:21.12 now mix your protein with an aligned sample of molecular trees,
00:04:24.19 that should now provide a slight orientational bias
00:04:28.18 so that the proteins tend to spend a little bit longer
00:04:31.25 with their long axis pointing parallel to the aligning material,
00:04:36.01 than perpendicular.
00:04:38.05 So in this case, usually you want these long trees
00:04:41.14 to be about 1-2% weight:volume,
00:04:44.11 and furthermore for the NMR experiment,
00:04:47.07 you want them to have a magnetic susceptibility
00:04:49.16 so that if we put them in a magnetic field,
00:04:51.22 then they'll basically globally align.
00:04:55.03 And it turns out if you can now get your proteins to be partially aligned,
00:04:59.12 you can now extract out information
00:05:01.09 that otherwise would be invisible,
00:05:03.12 and that information can be very precise
00:05:05.03 and allow you to calculate the atomic resolution structure of the protein.
00:05:09.26 So I mentioned that this weak alignment method
00:05:11.19 has been around for a while,
00:05:13.15 unfortunately it's only been accessible to people
00:05:15.29 studying soluble proteins,
00:05:18.07 because the most popular alignment media
00:05:21.02 turns out to be a natural bacteriophage
00:05:23.06 related to the M13 bacteriophage.
00:05:26.11 And these work quite well for soluble proteins,
00:05:29.05 we can make large amounts of them in bacteria,
00:05:32.01 they have magnetic susceptibility,
00:05:34.04 they'll line up with each other, works great.
00:05:36.19 But the problem is that these bacteriophage
00:05:40.19 will denature in the detergents that you need to
00:05:42.16 solubilize membrane proteins,
00:05:44.21 and therefore we haven't been able to use the bacteriophage
00:05:47.15 to weakly align membrane proteins,
00:05:49.19 because they just fall apart.
00:05:51.21 That's where we came in.
00:05:53.10 We decided we wanted to build DNA nanostructures
00:05:57.03 that would be a shape mimetic of these filamentous bacteriophage,
00:06:02.16 but because they're self-assembled from DNA,
00:06:04.14 they should be impervious to denaturation by the detergents,
00:06:07.27 and therefore we should now make this method of weak alignment
00:06:10.28 available to detergent-solubilized membrane proteins.
00:06:18.28 On the upper left-hand panel what we have are
00:06:21.10 electron micrographs of our DNA nanotubes.
00:06:24.01 In this case, we designed M13s
00:06:26.20 to fold into a 6-helix DNA nanotube
00:06:28.29 that's about 400 nm long and about 7 nm in diameter.
00:06:32.16 We actually programmed two of them to come together
00:06:34.15 to make a structure that is almost a micron in length. I
00:06:38.15 t turns out if you can make the structures longer
00:06:41.11 then they'll do a better job as these molecular trees.
00:06:45.07 They'll actually line up more easily.
00:06:47.21 And what we find is that when we concentrate them
00:06:50.13 to about 2% weight:volume,
00:06:52.18 then in fact they do start lining up.
00:06:54.26 So this is an entropic phenomenon
00:06:57.01 that's behind liquid crystal formation.
00:06:59.17 And one signature of liquid crystal formation is birefringence,
00:07:03.13 so if we look at our DNA nanotubes under crossed polars,
00:07:07.11 then we see this beautiful birefringence pattern
00:07:09.18 which is indicating that we're forming the liquid crystal.
00:07:12.05 So when the sample is much more dilute you don't see anything,
00:07:14.24 but then now when the sample is concentrated
00:07:16.16 it forms a liquid crystalline phase
00:07:18.11 and then you can see this nice pattern.
00:07:20.25 But most importantly, now if we take a test protein of known structure,
00:07:24.23 so this is a transmembrane domain from the
00:07:27.16 -- zeta transmembrane domain from the T-cell receptor --
00:07:30.17 we know what the structure is and based on that known structure
00:07:33.04 what we can do is we can calculate what
00:07:36.02 the kind of magnetic response will be
00:07:38.05 for a partially aligned structure,
00:07:40.20 so these are called dipole-dipole couplings.
00:07:43.17 And then we can compare those predicted couplings
00:07:46.07 against the ones that we experimentally measure
00:07:49.28 when we now mix the protein with the dilute liquid crystal
00:07:53.27 in the magnetic field.
00:07:55.28 And then on the lower left-hand side
00:07:57.25 what we're doing is we're comparing
00:08:00.19 on the y-axis the predicted couplings,
00:08:03.21 and on the x-axis the observed couplings,
00:08:06.08 and what we get is a very nice correlation between what we predict
00:08:09.10 and what we observe,
00:08:10.17 with high signal-to-noise.
00:08:13.00 We published this result in 2007,
00:08:15.11 we were very excited about it because we knew that this meant that
00:08:18.12 we had a tool that really works.
00:08:19.28 That we could use this to solve the structure
00:08:22.08 of membrane proteins.
00:08:24.28 However, there's of course many different hurdles
00:08:28.21 that still have to be overcome in order to solve the structure,
00:08:30.28 but just to review,
00:08:32.10 what we want to do is have these molecular trees,
00:08:34.14 our DNA nanotubes,
00:08:36.06 that when you concentrate them to 2% weight:volume
00:08:37.19 they start lining up.
00:08:39.17 We put them in an external magnetic field,
00:08:41.06 so we get global lining up.
00:08:43.07 We mix that with our protein of interest,
00:08:44.26 the protein bounces off the rods,
00:08:46.19 and we get that weak alignment.
00:08:48.10 So we get that bias in the orientation of the population of molecules.
00:08:52.28 Introducing that bias allows us to measure these
00:08:55.00 dipole-dipole couplings that encode precise
00:08:57.29 structural information about the protein,
00:09:00.18 which we can then throw into a computer program that,
00:09:03.14 if it has enough data, it can calculate the atomic resolution structure,
00:09:07.13 at least of the backbone chain.
00:09:09.18 Currently, we are not able to use this method
00:09:12.07 to experimentally measure the configuration of the sidechains.
00:09:16.01 But even if you can just generate the atomic resolution backbone,
00:09:19.25 then there are several very nice algorithms
00:09:22.13 that will allow you to predict
00:09:23.20 how the sidechains are going to pack onto that backbone.
00:09:29.12 So Marcelo Berardi in James Chou's lab
00:09:33.13 was able to use our DNA nanotubes
00:09:35.26 in order to solve the structure of a protein from the UCP family.
00:09:40.22 So this is the uncoupler proteins
00:09:42.28 that exist in the inner mitochondrial membrane,
00:09:45.12 and they're known to have an activity of translocating protons
00:09:49.23 back into the mitochondrion.
00:09:53.25 UCP1 i the most famous member of this family,
00:09:57.13 it's present in brown fat, and what it's doing is it's
00:09:59.04 just leaking protons across that membrane
00:10:01.19 and that generates heat.
00:10:03.10 So it's a mechanism to generate heat in brown fat.
00:10:07.04 Marcelo solved the structure of UCP2,
00:10:09.21 it's a related family member that's thought to be involved
00:10:11.27 in energy source selection,
00:10:15.01 so whether fatty acids vs. amino acids vs. pyruvate
00:10:19.26 should be metabolized for energy, as a source of energy.
00:10:23.29 So he wanted to solve the structure,
00:10:25.16 he tried for a long time using crystallography,
00:10:27.18 using conventional NMR,
00:10:29.15 but wasn't having a lot of luck.
00:10:31.14 Of course, for any kind of structural biology problem,
00:10:33.07 you first need to solve the problem of
00:10:35.21 being able to overexpress your protein,
00:10:38.00 being able to fold it to high homogeneity.
00:10:40.25 That's always going to be difficult
00:10:42.12 and it took many years to do that,
00:10:44.06 but to make a long story short,
00:10:45.21 once he was able to generate a large amount of the protein
00:10:48.17 in a homogeneous state,
00:10:50.15 then he was able to mix it with our DNA nanotubes,
00:10:53.18 weakly align the protein,
00:10:55.13 use that in order to measure the dipole-dipole couplings,
00:10:58.25 and then use that to calculate the atomic resolution
00:11:01.11 backbone structure of the protein.
00:11:03.10 And so this is here the crystal structure of this protein,
00:11:06.04 it's a 6-transmembrane helix protein.
00:11:08.22 Looking at the structure doesn't immediately tell you
00:11:11.18 the mechanism of proton transport,
00:11:14.09 but now James Chou's group is very interested in
00:11:18.09 using this structure as a foundation
00:11:20.15 for further structure/function exploration of
00:11:23.09 what the mechanism might be.
00:11:24.26 So their current hypothesis is that
00:11:26.22 you actually have proton transported by these fatty acids
00:11:30.05 that are flipping through the membrane, so when it's neutral...
00:11:32.18 when it's protonated it can flip through the membrane,
00:11:35.00 but now when it's deprotonated on the other side it can't flip back,
00:11:37.16 so you're not actually going to get net proton transfer.
00:11:40.09 So the idea is that the ionized fatty acid
00:11:42.28 can now diffuse into the inside
00:11:45.20 of this 6-helix barrel
00:11:47.14 and then you have a hydrophilic environment on the inside of that
00:11:49.14 where the fatty acid can flip back,
00:11:51.11 and therefore this might provide a mechanism
00:11:53.17 for multiple turnover of transfer of protons
00:11:57.07 across the membrane by fatty acids.
00:12:02.09 So that was a very satisfying experiment
00:12:04.14 because we were able to demonstrate utility
00:12:06.21 to a very urgent need in structural biology,
00:12:10.18 and we're looking forward to further developments
00:12:13.16 in the technology to make it more general,
00:12:16.24 both for NMR experiments, also for cryo-EM,
00:12:20.07 maybe for X-ray crystallography as well.
00:12:22.18 So the next area of applications that I'd like to describe
00:12:26.17 are involving single molecule biophysics.
00:12:29.12 So this is work making rigid handles for optical tweezing experiments
00:12:33.25 that was done in the lab of Hendrik Dietz.
00:12:36.24 He actually initiated this experiment when he was
00:12:39.21 doing his postdoctoral training with my group at Harvard,
00:12:42.29 and now finally was able to carry the project to fruition.
00:12:47.11 It's going to be a great tool and he was kind enough
00:12:48.20 to include me on the author list.
00:12:51.29 So the basic idea is let's say that you want to
00:12:54.27 use your optical traps in order to monitor single molecule
00:12:58.27 conformational changes of your molecule of interest.
00:13:02.17 So let's say that it's a DNA hairpin
00:13:04.27 that is opening and closing, and you want to be able to observe this.
00:13:08.06 So ordinarily this could be kind of hard to watch,
00:13:11.11 but you also want to be able to watch this
00:13:13.14 as a function of the force that you're applying
00:13:15.13 on the ends of the hairpin.
00:13:17.08 So you want to be able to dial in greater and greater amounts of force
00:13:19.23 on my elbows to peel it away
00:13:21.26 and watch what the effect is of that force
00:13:24.27 on the binding and peeling and unpeeling kinetics
00:13:28.09 of this DNA hairpin.
00:13:30.28 So the way that you actually observe this
00:13:32.24 is you now want to place your molecule of interest
00:13:36.21 in between two large microspheres,
00:13:41.06 and it turns out for technical reason you can't have these two microspheres
00:13:44.16 too close together,
00:13:46.02 so you need to give you some space,
00:13:48.20 and in order to create that space,
00:13:50.08 people like to use these double-stranded tethers
00:13:52.29 that are on the order of 300 nm long.
00:13:56.03 And so what you do is you have your beads, your tethers,
00:13:59.03 and then your molecule of interest,
00:14:00.18 and then you start to pull the beads apart,
00:14:02.26 that generates a force on the molecule in the middle.
00:14:05.25 And so if we now pull the beads further and further apart
00:14:07.21 that generates a greater and greater force
00:14:09.05 on the molecule in the middle,
00:14:10.27 and in principle we can watch the opening and closing
00:14:13.12 of that molecule in the middle
00:14:15.26 by watching how the distance between the beads changes.
00:14:18.24 So for example, we can imagine that
00:14:21.20 if you have the molecule now opens up,
00:14:24.28 now the beads should move further apart.
00:14:27.15 If the molecule in the middle now snaps shut,
00:14:29.17 then those two large beads should move closer together.
00:14:32.13 And so by measuring the distance between the beads,
00:14:34.28 in principle we can infer the conformational state
00:14:37.14 of the molecule in the middle.
00:14:39.15 Alright, sounds good, so where's the problem?
00:14:42.14 The problem is that you have Brownian motion.
00:14:44.29 So these large microspheres are actually
00:14:47.09 undergoing a lot of motion,
00:14:49.04 and it can be difficult to tell what is
00:14:51.06 just random motion and what is reporting on
00:14:53.15 something that's actually opening up in the middle.
00:14:56.03 And the problem becomes especially bad
00:14:58.11 when you have a very low force,
00:15:00.01 because at low force these double-stranded tethers
00:15:03.06 are going to be very floppy
00:15:04.23 and you're going to get lots and lots of Brownian motion.
00:15:07.24 So for example, on the lower right-hand trace,
00:15:12.05 what we see is a time trace of the distance between the beads
00:15:16.18 for the example on the top in B,
00:15:19.17 where we're looking at a DNA hairpin
00:15:21.18 that presumably is opening and closing at some specific force.
00:15:26.20 And I'm not going to explain
00:15:28.24 what's going on in the lower left-hand corner,
00:15:30.24 I encourage you to check out the publication,
00:15:32.28 but suffice it to say,
00:15:34.23 just looking at this bottom trace in E,
00:15:36.25 you can't really tell when that hairpin is opening or closing.
00:15:40.14 You can just see a lot of noise.
00:15:44.17 In contrast, if we now replace those long tethers
00:15:47.25 with a DNA Origami bundle of helices,
00:15:51.00 it's going to be much, much more rigid,
00:15:53.02 so even at those lower forces,
00:15:55.01 the Brownian noise is going to be suppressed.
00:15:57.16 And so if we look at the same force,
00:15:59.12 at the opening and closing of this object in the middle,
00:16:02.07 so D represents with these DNA Origami handles,
00:16:05.28 now you can hopefully tell that the noise is greatly suppressed,
00:16:09.09 and we have some more confidence
00:16:10.27 about assigning when the hairpin
00:16:13.13 is opening and closing.
00:16:16.06 So we think that this is a very nice application
00:16:18.15 of these rigid DNA nanorod elements
00:16:22.17 that will be useful for force spectroscopy
00:16:25.13 looking at single molecule dynamics
00:16:27.13 and energetics of biomolecules.
00:16:32.01 So the next area of application
00:16:34.01 I like to call Breadboard Biochemistry,
00:16:36.08 the notion that we can constrain the position of
00:16:38.29 many different protein actors in a molecular play
00:16:42.13 in order to understand and tease out
00:16:44.07 their individual roles in the process
00:16:46.00 - how are they interacting with each other?
00:16:47.26 What are the stoichiometric requirements?
00:16:49.21 What are the geometric requirements of this process?
00:16:53.12 So the first example of this Breadboard Biochemistry
00:16:56.00 I'd like to discuss is work that was
00:16:58.04 led in the lab of Sam Reck-Peterson,
00:17:00.13 my colleague at Harvard Medical School.
00:17:02.10 It was done primarily by two students:
00:17:06.15 Nate Derr and Brian Goodman.
00:17:08.11 Nate is a student that was co-advised by myself and by Sam.
00:17:12.15 And here we were interested in studying
00:17:15.06 how ensembles of cytoskeletal or microtubule motors
00:17:18.18 could antagonize each other
00:17:20.16 in terms of determining direction of motion
00:17:23.11 on a microtubule.
00:17:25.04 And the motivation is that it's been observed
00:17:27.07 that vesicles often times
00:17:28.16 bear both kinesins and dyneins,
00:17:31.08 which of course move in opposite directions along a microtubule,
00:17:34.05 and yet within a cell one can often observe that
00:17:36.24 the vesicles will choose one direction to go and then,
00:17:39.23 remarkably, will often times switch directions,
00:17:41.25 but what they don't typically do is just to stall.
00:17:44.22 So how can we try to study this in a reduced system,
00:17:47.06 an in vitro system?
00:17:49.05 And as a first step, what we did was we self-assembled
00:17:51.20 a chassis that is a 12-helix DNA nanotube
00:17:55.02 that's about 200 nm in dimensions.
00:17:58.16 And what we did was we decorated this chassis
00:18:01.00 with single-stranded DNA handles
00:18:03.00 that would come at regular intervals,
00:18:04.26 at 7 different positions.
00:18:07.07 And we can control and have any sequence we want
00:18:08.28 come out at any one of these positions.
00:18:11.19 In this particular case, what we did was
00:18:13.07 we had two different sequences:
00:18:15.01 one sequence for capturing dynein,
00:18:17.04 and another sequence for capturing kinesins.
00:18:19.15 In this example, 2 dyneins and 5 kinesins.
00:18:22.03 And the way that we capture
00:18:23.18 is that we express the protein,
00:18:26.00 let's say dynein, with a SNAP-tag
00:18:28.02 and then the SNAP-tag is used
00:18:30.00 to capture an oligonucleotide with a SNAP-tag ligand,
00:18:33.12 and in that way we're able to generate a protein-DNA conjugate.
00:18:38.05 And we purposely choose the sequence of that conjugate
00:18:40.23 so that it will be complementary
00:18:42.18 with the single-stranded DNA sequence
00:18:44.13 that comes out of the DNA Origami.
00:18:47.05 So in this example, we're creating
00:18:48.21 a chain gang of 2 dyneins and 5 kinesins,
00:18:51.13 and we wanted to see what happens
00:18:53.15 when we put this on a microtubules
00:18:55.03 - which way is it going to go?
00:18:57.05 And what Nate and Brian found is that
00:18:59.21 they basically stalled:
00:19:01.14 there was an irreversible tug-of-war,
00:19:03.14 at least in this first study,
00:19:06.02 and furthermore they were able to demonstrate
00:19:08.00 if they introduced photocleavable elements
00:19:10.14 either to release the dynein dynamically,
00:19:13.07 or else to release the kinesins dynamically,
00:19:15.27 then they could resolve this tug-of-war.
00:19:18.05 So for example, in one experiment,
00:19:20.09 if they released the dyneins,
00:19:22.09 then now that stall would be relieved
00:19:24.09 and the chassis would move towards the
00:19:25.19 plus end of the microtubules.
00:19:27.24 Likewise, if they cause the kinesins to be cleaved off,
00:19:31.17 then now the complex would no longer stall,
00:19:33.23 and move towards the negative end of the microtubule.
00:19:37.17 So we think this is a promising experimental platform
00:19:41.06 for now further trying to find out:
00:19:43.21 what else do we need to add in order to recapitulate
00:19:46.08 the very interesting behavior we see inside of a cell,
00:19:48.23 where the vesicles, they don't just stall,
00:19:50.17 but in fact they can move in one direction or the other,
00:19:52.22 and even more interestingly, change in direction.
00:19:56.02 So here's another application of Breadboard Biochemistry
00:19:59.05 where we're trying to study SNARE-dependent membrane fusion.
00:20:02.10 So in this process,
00:20:04.19 we have cells that are trying to fuse the vesicles
00:20:07.24 with let's say the plasma membrane,
00:20:09.27 and it's known that there are these transmembrane proteins
00:20:12.04 that are mediating this called SNARE proteins.
00:20:14.16 They have one domain that's...
00:20:18.05 let's say this is the vesicle membrane,
00:20:19.24 they have a transmembrane and they have a cytoplasmic domain
00:20:22.10 that's a coiled-coil,
00:20:24.07 and then in the target membrane you have something analogous,
00:20:26.09 you have a transmembrane domain and then a complementary
00:20:28.15 coiled-coil domain.
00:20:30.02 There's two other helices that are involved as well,
00:20:33.02 and so you can actually zipper up to form a 4-helix bundle,
00:20:36.25 and that's thought to provide the energy
00:20:39.01 for driving a vesicle in close proximity
00:20:41.29 to the plasma membrane,
00:20:43.13 because otherwise that's an energetically unfavorable process.
00:20:47.16 And then once the two vesicle membranes
00:20:49.00 are brought close together,
00:20:50.19 then some other process happens
00:20:52.25 -- that's not very well understood --
00:20:54.14 causing the membranes to fuse.
00:20:57.08 And there's some outstanding biophysical questions about this process
00:21:01.10 that can be very simply articulated
00:21:02.29 but difficult to nail down very precisely.
00:21:06.04 For example, how many different SNARE proteins
00:21:08.21 does it take to trigger the fusion event?
00:21:11.13 And then, more subtly,
00:21:13.10 how does the geometry of these proteins
00:21:15.06 affect the kinetics of this process?
00:21:18.07 And there have been some different measurements
00:21:20.21 of the number of SNAREs required,
00:21:23.06 and depending on the context it seems
00:21:25.07 to vary between 1 and 10.
00:21:27.08 In our case, what we're interested in
00:21:29.00 is generating a more robust method
00:21:30.29 for measuring these stoichiometric requirements,
00:21:33.02 and therefore we think we can use our system
00:21:35.04 to study this problem at higher resolution.
00:21:39.20 So here's the idea:
00:21:40.26 imagine you have a supported bilayer with your t-SNAREs down there,
00:21:45.19 and you're trying to now down a vesicle on there,
00:21:50.14 but you want to only have a controlled number of t-SNAREs
00:21:53.15 that could possibly participate in the reaction.
00:21:56.29 So what if we were to create a molecular corral
00:21:59.17 where we had 3 and only 3 of the t-SNAREs
00:22:02.01 in the corral.
00:22:03.26 And now what we could ask,
00:22:05.11 "Well, is 3 SNAREs enough for membrane fusion?"
00:22:07.25 So the idea here is that we chemically synthesize,
00:22:11.18 we chemically link to the N-terminus of this SNARE protein,
00:22:14.23 this white oligonucleotide in one test tube,
00:22:17.23 and then in another test tube we self-assemble this
00:22:20.09 red DNA nanostructure.
00:22:22.14 And the red DNA nanostructure again is decorated with these
00:22:24.14 single-stranded DNA handles
00:22:26.18 that are complementary to the white anti-handles.
00:22:29.06 Now when we mix the two together,
00:22:30.26 we should get a controlled stoichiometry
00:22:33.09 of the greens and whites onto the red,
00:22:35.09 in this case three.
00:22:36.29 And now we can say,
00:22:38.09 "Well, what happens when a vesicle tries to dock into the molecular corral?
00:22:42.01 What happens when we have 3 SNAREs in the corral?
00:22:44.06 What happens when we have 2?
00:22:45.25 What happens when we have 1?"
00:22:47.12 So certainly if we have no SNAREs in the corral,
00:22:49.12 then you wouldn't expect anything about background,
00:22:51.05 and then as we start to increase the number of SNAREs in the corral,
00:22:54.08 hopefully the proteins now will be able to cooperate
00:22:56.18 to trigger the membrane fusion above background.
00:23:00.24 So I should mention that this is work that's still in progress.
00:23:04.12 It's a collaboration between our lab
00:23:06.15 and the lab of Jim Rotheman.
00:23:09.00 Chenxiang Lin was a postdoctoral fellow in the group,
00:23:11.11 he initiated the project
00:23:13.07 along with Weiming Xu in Jim's lab.
00:23:15.15 Now Chenxiang is an assistant professor at Yale,
00:23:17.21 and in our lab the project has been taken over
00:23:20.07 by Bhavik Nathwani at the time of this taping.
00:23:30.27 So again, what's thought to happen is that
00:23:32.20 you have engagement of the t-SNAREs and the v-SNAREs
00:23:36.15 and then that brings the membranes close together,
00:23:39.00 some magic happens,
00:23:40.16 membranes fuse.
00:23:42.12 And the details of that biophysical process are very interesting,
00:23:45.06 but we're starting off by asking a simpler questions
00:23:47.16 of just, how many SNAREs are involved and how does the geometry affect that?
00:23:51.28 So just to prove our ability to
00:23:53.20 decorate our DNA rings with different guests,
00:23:56.01 we started with gold nanoparticles
00:23:57.22 instead of SNARE proteins,
00:23:59.20 they're just easier to see in the electron microscope.
00:24:02.09 And we can demonstrate that we can now
00:24:04.18 decorate our DNA rings with 3 gold particles on the side,
00:24:07.18 3 on the outside,
00:24:09.07 6 on the inside,
00:24:11.06 4 on the inside,
00:24:12.28 8 on the inside.
00:24:15.06 And what we've observed so far is that we can get
00:24:17.20 something on the order of 90% occupancy of our sites.
00:24:21.13 We'd like to get higher than that,
00:24:22.29 something more like 99.99%,
00:24:25.12 it's something we're working on,
00:24:27.21 but currently we think even with a 90% occupancy rate,
00:24:30.14 this is still useful because it provides an upper bound
00:24:33.11 on the number of SNARE proteins, or whatever guest,
00:24:36.00 that could be bound in our assembly.
00:24:40.02 So here what we're doing is we're actually
00:24:41.06 binding SNARE proteins now, detergent-solubilized SNARE proteins
00:24:44.02 instead of gold nanoparticles.
00:24:46.14 And again, what we're doing is
00:24:48.29 we have these green SNARE proteins
00:24:51.00 conjugated to a white oligonucleotide
00:24:53.10 that are now self-assembling with this complementary red oligonucleotide
00:24:56.21 being displayed on the outside of this DNA nanostructure.
00:25:00.06 And in the electron microscope,
00:25:01.26 we can see enough contrast from the proteins
00:25:03.21 to count the number of guest molecules
00:25:05.22 on the DNA nanostructure.
00:25:08.18 And then what we do is we,
00:25:10.11 so in this case what we're trying to do is create liposomes
00:25:13.09 with controlled numbers of SNARE proteins
00:25:15.04 and see how they behave in a fusion assay.
00:25:18.13 And so the next step here is that we mix our DNA rings
00:25:22.00 with protein guest
00:25:23.28 with giant liposomes
00:25:25.23 in the presence of slight amounts of detergent,
00:25:28.14 and through some process we don't quite understand,
00:25:30.24 these DNA nanostructures with hydrophobic groups
00:25:34.20 somehow take a bite out of the giant liposomes
00:25:37.11 and end up with smaller liposomes
00:25:39.19 basically filling the interior of the ring.
00:25:43.18 So in this way, through a process we don't quite understand,
00:25:45.24 we're able to capture liposomes on the inside of our DNA nanorings,
00:25:49.25 with controlled numbers of SNARE proteins.
00:25:53.01 Then here's a fusion assay that we use,
00:25:55.18 it was developed by Erdem Karatekin at Yale.
00:25:58.21 It's a microfluidic assay where we're
00:26:01.11 fluorescently labeling these vesicles that we've captured
00:26:04.14 in the DNA ring
00:26:06.18 and then when we get fusion,
00:26:08.18 what will hopefully happen is that those dyes
00:26:10.25 will then diffuse out into the supported bilayer.
00:26:13.28 So initially what happens is the dyes
00:26:15.22 are quenching each other somewhat,
00:26:17.28 the first thing is actually the spot should grow brighter
00:26:20.16 as the dyes de-quench,
00:26:22.13 but then as the dyes diffuse away from each other,
00:26:24.11 then you should now get rapid
00:26:27.12 dilution of the fluorescence response.
00:26:30.03 And so we have here a microfluidic setup
00:26:32.04 where we have an Eppendorf tube
00:26:34.05 with our labeled vesicles
00:26:36.18 that are being pulled through this microfluidic chamber.
00:26:39.13 We're observing it using TIRF microscopy,
00:26:41.20 and then trying to see whether or not those vesicles
00:26:45.03 can fuse to the supported bilayer,
00:26:46.27 as a function of number the number of SNARE proteins.
00:26:49.15 So this is just an example of the assay in action.
00:26:53.18 Your eye might be drawn to this giant blob on the bottom,
00:26:56.16 but I'd like you to try to ignore that.
00:26:58.06 Instead, focus on this red circle here, where we can see
00:27:01.19 -- it's on a loop --
00:27:03.01 so we can see a vesicle docking and shortly after docking
00:27:05.24 then we get fusion.
00:27:07.12 So that's the kind of event that we're trying to score,
00:27:10.06 so far we're still in the process of trying to see
00:27:12.25 whether or not our scaffolded DNA rings
00:27:15.02 actually can increase the rate of liposome docking and fusion.
00:27:19.05 Hopefully we'll be able to make some progress on that
00:27:21.05 for the next lecture.
00:27:25.15 So the other class of applications
00:27:27.15 I'd like to discuss with you involve
00:27:29.20 potentially using DNA nanostructures
00:27:32.07 as therapeutic delivery devices.
00:27:34.19 And as a first step towards that,
00:27:36.25 Franziska Graf, who was a student co-advised
00:27:39.13 by myself and Don Ingber,
00:27:41.07 we set out to do some pilot studies
00:27:43.06 to look at how the shape of DNA nanostructures
00:27:45.18 might affect their uptake into a model cell line.
00:27:48.23 In this case, these are umbilical vein endothelial cells.
00:27:52.29 So how the shape of DNA Origami
00:27:54.16 affect their tastiness to these cells?
00:27:56.27 And if you think about it, you could come up with
00:27:58.05 a long list of potential descriptors
00:28:00.27 that might make a difference,
00:28:03.06 and it would require very a systematic study
00:28:07.01 to look through all of these,
00:28:08.16 but we just started out with a pilot study
00:28:10.04 looking at aspect ratio.
00:28:12.17 So in this case, we wanted to ask the question:
00:28:15.07 if we have bunch of particles of the same mass,
00:28:17.22 but then we made them in different shapes,
00:28:20.08 then what would be the relatively uptake?
00:28:22.13 So let's first look at these 3 blue shapes
00:28:24.18 that are highlighted in yellow.
00:28:26.13 So they're the same mass,
00:28:27.22 they're about 5 megaDaltons,
00:28:29.11 but they're made into a very long spindly rod that's a 6-helix bundle,
00:28:32.23 and then we have this 24-helix bundle,
00:28:35.22 and this is a 48-helix bundle,
00:28:37.14 that are successively more compact.
00:28:39.24 So if we predict that structures
00:28:41.17 with a longer aspect ratio
00:28:43.13 should get into cells better,
00:28:44.29 then we'd predict that this longer one
00:28:47.15 should do the best in terms of getting into the cell,
00:28:49.19 being taken up.
00:28:51.11 In contrast, if we hypothesize
00:28:53.11 that the more compact structure should get in better,
00:28:55.16 then we expect to observe the opposite behavior:
00:28:58.07 that this more compact structure
00:29:00.00 should get in more quickly.
00:29:01.22 And we also made some other shapes.
00:29:03.08 So here we made a wireframe octahedron,
00:29:06.13 we also made three different orange shapes that are analogues,
00:29:10.06 but now just 40% the mass.
00:29:12.25 And so we can say,
00:29:14.01 "Well, with this starting panel of eight structures,
00:29:16.06 what's the relative rate of uptake into these HUVEC cells?"
00:29:22.04 And what we observed
00:29:24.13 -- so I have a very busy slide here
00:29:27.05 but I'll try to give you the overview --
00:29:30.04 so first thing is if you kind of squint your eyes
00:29:33.22 you might notice that the bars,
00:29:35.21 which represent some kind of uptake,
00:29:37.19 tend to be taller on the right-hand side of the slide
00:29:41.09 than one the left-hand side of the slide.
00:29:44.15 And also, if you look at it for a little bit longer,
00:29:47.14 you might notice that the structures
00:29:49.15 that are on the right-hand side
00:29:51.11 tend to be more compact
00:29:53.03 than the structures on the left-hand side.
00:29:54.24 So just from a simple eyeballing of the figure,
00:29:57.10 we can get the feeling that,
00:29:59.06 at least for this class of particles,
00:30:01.01 when the structures are more compact,
00:30:03.10 then they get in more easily into cells.
00:30:06.25 So the second order piece of information that we learned is -
00:30:10.05 so you might notice here that there's hollow bars and solid bars,
00:30:12.03 so what does that mean?
00:30:13.27 So the hollow bar represents the amount of
00:30:16.22 fluorescently labeled nanostructures
00:30:18.21 that were taken up by the cell
00:30:20.25 after some incubation time, 16 hours,
00:30:24.04 and then the solid bar represents the same experiment,
00:30:27.19 but what we did was we treated the cells with DNase I
00:30:31.21 before we did the fluorescence analysis.
00:30:33.27 So what that's going to do is to remove
00:30:35.22 any membrane-bound DNA nanostructures
00:30:38.18 and now the assay will only report on those structures
00:30:41.05 that actually have been internalized,
00:30:43.07 that are now protected from DNase digestion.
00:30:46.04 And so what we observe is something quite interesting.
00:30:48.11 That for these compact structures,
00:30:50.01 the height of the bars is very similar,
00:30:52.03 so that says that most of the particles
00:30:53.27 that stick to the cells immediately go in,
00:30:58.03 or are getting in quite efficiently.
00:30:59.25 But for these extended structures,
00:31:02.00 you might notice that the hollow bar
00:31:03.26 is actually much taller than the solid bar,
00:31:06.13 and what that suggests is that if you're really extended,
00:31:09.02 now maybe you can resist getting internalized,
00:31:11.28 which we can rationalize by saying for this class of cells,
00:31:14.27 they're going to have a hard...
00:31:16.08 they don't really have good mechanisms
00:31:17.23 for engulfing structures that are 400 nm long.
00:31:22.00 So this might present an interesting opportunity therapeutically,
00:31:24.16 that we could design structures
00:31:26.19 that have the tendency to persist on the outside of the cell,
00:31:29.29 that can resist internalization.
00:31:32.08 It might be useful for creating
00:31:34.18 a sentinel or a beacon for other nanoparticles
00:31:37.01 to then deliver their contents to that particular cell,
00:31:39.18 because if you just get swallowed,
00:31:41.04 then you're not going to be able to act as that sentinel.
00:31:46.07 Here what Franziska has done is she's taken one of the structures,
00:31:49.06 this nanocylinder,
00:31:50.27 and she's decorated the outside of the nanocylinder
00:31:53.23 with a bunch of ligands, in this case
00:31:55.20 cyclic RGD ligands that are known to bind the
00:31:58.17 alpha V beta 3 integrin receptors
00:32:00.03 that are overexpressed on this cell line.
00:32:02.18 So the prediction is that if we decorate our nanoparticles
00:32:06.04 with a high density of these ligands,
00:32:08.08 then we're going to increase the rate of internalization.
00:32:12.06 And then we can also do a control with cyclic RAD peptides
00:32:15.23 that should not interact with those receptors.
00:32:18.25 And in fact that's basically what we observe,
00:32:22.01 that when we have the cyclic RGD-labeled structures
00:32:26.11 we get an order of magnitude greater uptake of these particles
00:32:31.07 compared to controls that had no peptide
00:32:33.29 or controls that have the mock peptide sequence cyclic RAD.
00:32:39.14 And so what this says is that we have the ability
00:32:42.05 to make these particles of different shapes
00:32:44.02 to either help their uptake
00:32:46.07 or help prevent their uptake.
00:32:47.28 We can decorate these particles with a high density, controlled density,
00:32:51.28 of ligands to help further modulate that process,
00:32:54.21 either get faster uptake or not.
00:33:01.12 So far I've shown you naked DNA particles.
00:33:06.20 Again, we might be worried about things
00:33:08.18 like nuclease digestion of these particles,
00:33:11.10 we might have a desire to encapsulate
00:33:13.21 soluble factors in these structures.
00:33:15.24 So if we try to make a wireframe cage
00:33:18.18 and then we put soluble factors on the inside of the cage,
00:33:20.27 then those soluble factors
00:33:23.11 might just diffuse out through those windows.
00:33:25.08 So how do we keep those factors on the inside?
00:33:27.19 So Steve Perrault in the group
00:33:29.08 has pioneered a method for encapsulating
00:33:31.19 these DNA nanostructures within liposomes
00:33:34.11 to create something that's structurally analogous
00:33:36.15 to an enveloped virus.
00:33:38.14 So the way that he's done this is
00:33:40.06 he again self-assembles a DNA octahedron
00:33:42.19 with single-stranded DNA handles coming,
00:33:45.14 and then he hybridizes on a complementary oligonucleotide anti-handle
00:33:49.26 that has a lipid conjugate covalently linked to it,
00:33:52.14 solubilized by detergent.
00:33:54.18 And so through base pairing interaction,
00:33:56.07 he basically gets these detergent solubilized lipids
00:33:59.28 to cover his DNA octahedron,
00:34:02.18 he has something like on the order of 50 copies of these lipids
00:34:05.09 covering his octahedron,
00:34:07.18 solubilized by detergent.
00:34:09.27 The next step is that he mixes this...
00:34:11.19 he dilutes this into a solution of giant liposomes
00:34:14.04 and then dialyzes out the remaining detergent.
00:34:17.07 Through a process that we still don't quite understand,
00:34:19.23 we get shrink-wrapping of the liposomes
00:34:21.28 around our DNA nanostructures,
00:34:24.03 creating something, again, that resembles
00:34:25.20 under the transmission electron microscope,
00:34:28.05 envelope viruses.
00:34:29.18 So here we can see on the left the naked DNA octahedra,
00:34:33.09 they're about 50 nm in diameter,
00:34:36.11 and then on the right
00:34:38.07 we can see the liposome-encapsulated DNA octahedra.
00:34:42.25 Again, very reminiscent of an envelope virus.
00:34:46.04 So we think this is an important step
00:34:48.09 towards the versatile use of DNA nanostructures
00:34:51.19 for delivery, for example, of soluble factors,
00:34:54.18 or if we wanted to simply protect the DNA from nucleases
00:34:58.16 or factors that would try to digest it.
00:35:00.21 We of course still want to be able to
00:35:03.02 decorate the surface of the structures
00:35:05.02 with functionalities,
00:35:06.27 so now we're working on the ability
00:35:08.13 to present transmembrane features,
00:35:10.16 starting from the inside,
00:35:12.25 going through the membrane,
00:35:14.13 and then basically controlling the spatial orientation of ligands
00:35:17.04 through the puppet master on the inside.
00:35:21.27 And then the final thing that I'd like to show you
00:35:23.24 is actually not from my laboratory,
00:35:26.03 it's from Shawn Douglas,
00:35:28.09 when he was a postdoctoral fellow in George Church's group,
00:35:31.05 and he did this fascinating...
00:35:33.04 along with Ido Bachelet,
00:35:34.19 they did this fascinating pilot study
00:35:36.16 where they generated something they called
00:35:37.25 a DNA Origami nanorobot,
00:35:40.19 that was designed, at least in this test tube example,
00:35:44.08 to specifically recognize cancerous lymphocytes and kill those off
00:35:50.00 -- program them to commit suicide --
00:35:52.10 while leaving the healthy cells alone.
00:35:54.12 So how is this supposed to work?
00:35:55.28 The idea is that Shawn designed this DNA barrel
00:35:58.25 that's about 60 nm in diameter,
00:36:01.28 and he placed on the inside
00:36:04.05 antibodies that are known to bind to receptors
00:36:07.24 and cause receptor clustering
00:36:09.13 that then leads to apoptosis.
00:36:13.04 So kind of like what a natural killer cell might do.
00:36:15.21 But in this case, because the antibodies are sequestered
00:36:17.22 on the inside of the barrel,
00:36:19.06 they're not actually accessible to the cells.
00:36:21.26 The cells have fat fingers, if you will,
00:36:23.21 so they can't actually reach in
00:36:26.02 and touch those antibodies.
00:36:28.08 And the notion is that Shawn and Ido
00:36:29.27 wanted to program this robot
00:36:32.00 to open up when it encountered the cancerous cell
00:36:34.27 but not when it encountered the healthy cell.
00:36:37.23 And if the robot were to open up,
00:36:39.22 then now the antibodies that trigger apoptosis
00:36:42.17 could now be accessible to the cell surface.
00:36:45.12 So how do they get this robot to open up
00:36:47.00 only in the presence of the cancerous cell
00:36:49.01 but not in the healthy cell?
00:36:50.22 So what they did was they designed
00:36:52.08 this lock-and-key mechanism
00:36:54.03 where they had single-stranded DNA
00:36:56.13 on the two ends that would hybridize together,
00:36:59.20 and they designed this sequence
00:37:01.05 with something called a structure-specific aptamer.
00:37:05.15 And what happens is this aptamer
00:37:08.18 recognizes some protein ligand,
00:37:12.02 and so basically that protein ligand
00:37:13.22 is competing for interaction between the partner strand,
00:37:19.11 so in other words these two strands will interact with each other,
00:37:23.04 but that can be competed off by some specific protein ligand.
00:37:26.20 And they used sequence on one of the locks
00:37:29.04 that was derived from the literature,
00:37:30.20 a sequence that somebody else had isolated
00:37:33.06 through a SELEX experiment,
00:37:35.04 that sequence was known to bind to EGFR,
00:37:37.23 which is overexpressed on their sample cancer cell line.
00:37:41.06 And then they found in the literature another sequence
00:37:44.14 that was found to recognize something that's enriched
00:37:47.14 on their cancer cell line
00:37:48.28 but not on the healthy cell line.
00:37:50.28 So they're able to do a logical AND statement here.
00:37:53.27 Only if the target cell can open both locks
00:37:57.02 would the shell open up
00:37:59.03 and then reveal the antibodies to the cell surface.
00:38:03.00 So in this way you could argue that
00:38:04.21 it's an intelligent robot
00:38:06.19 in that it can do this more complicated logical argument.
00:38:10.29 And they were able to show, at least in the test tube,
00:38:13.01 that in fact this nanorobot
00:38:14.27 seems to trigger the apoptosis
00:38:16.29 a couple of orders of magnitude more easily
00:38:19.06 than in the healthy cell.
00:38:21.16 Of course, moving this into an actual therapeutic environment
00:38:24.27 requires several hurdles,
00:38:26.29 because if you wanted these things circulating in your blood,
00:38:29.22 you'd want them to avoid nuclease digestion,
00:38:32.08 you'd want them to avoid clearance by the immune system.
00:38:36.07 So, of course, there are going to be many hurdles to overcome
00:38:39.00 in order to translate this to the clinic,
00:38:41.11 but we think this is a fascinating first step in that direction.
00:38:48.04 So to conclude, DNA nanotechnology
00:38:51.28 is more than just smiley faces, we think.
00:38:54.26 In particular, we're very interested in two classes of applications.
00:38:59.13 One is as tools for molecular biophysics,
00:39:01.29 whether it be in tools for structural biology
00:39:04.10 or especially single molecule biophysics.
00:39:06.29 And secondly, we are exploring the notion that
00:39:10.00 these DNA nanostructures might be useful
00:39:13.07 as therapeutic delivery devices,
00:39:15.14 and of course there's many different platforms that
00:39:18.07 scientists are trying to explore for delivery,
00:39:20.24 but we're motivated by the insight
00:39:23.16 that our immune systems, in fact,
00:39:25.26 can be thought of as very complicated nanotechnologies.
00:39:29.17 They can process lots of information,
00:39:31.25 they can actuate all kinds of things,
00:39:33.13 they can punch holes into cells,
00:39:35.12 they can program each other to expand,
00:39:37.18 they can squeeze through tight spaces,
00:39:39.14 and the only way to achieve this really diverse,
00:39:42.19 sophisticated behavior
00:39:44.09 is by creating complex nanoscale objects.
00:39:47.03 At the moment, we believe that
00:39:49.12 the DNA nanotechnology platform
00:39:51.17 provides a very powerful method in that direction.
00:39:55.14 So I'd like to thank the following sources for support,
00:39:59.02 especially NIH and the Office of Naval Research.
00:40:02.10 We also got some support from the Wyss Institute
00:40:03.26 for Biologically Inspired Engineering.
00:40:06.07 I've tried to acknowledge
00:40:07.26 the different folks who have been doing the work
00:40:10.00 on the slides describing the work.
00:40:12.02 Thanks a lot.