In our roadmap we classified the parts needed by an assembler into several kinds:
We also described how basic design decisions about gripping method and joining method are coupled: one scheme is for the assembler's gripper to hold a new MBB in place while a joining tool catalyzes the formation of a strong bond to the workpiece, allowing the gripper to release the MBB simply by pulling away with enough strength; another scheme is for the gripper to reduce its affinity and retract even while the MBB is only weakly joined to the workpiece, and later to strengthen many MBB-workpiece bonds at once, perhaps by altering the chemical conditions in the solution.
The choice of permanent bonding method between the MBBs is the design issue of greatest uncertainty at the moment. In our internal planning that means we should work on it first, but to explain the choices we've already made, I should start by explaining the easier ones.
1 (structure). We think that pure-DNA structures of the kind pioneered by Seeman are sufficient by themselves to form most or all structural components of a primitive assembler. They are certainly able (in principle) to produce a sufficient variety and complexity of forms, so the only remaining issue is whether their stiffness is adequate to reliably separate the internal parts or objects they are holding which should not touch each other. (Some floppiness is fine, and will actually help the bonding of MBBs to the workpiece, as long as it is not so much that a new MBB can be bonded to the wrong existing MBB in the workpiece being formed.)
The tests of stiffness of the DX and TX motifs, published by Seeman, already support the idea that they are stiff enough. (These tests consist of producing long beams, whose ends could in principle form covalent bonds to one another if they came near enough, and testing whether they do in fact form those bonds. Similar tests can be performed on substructures intended to be parts of an assembler.)
The general principle that thicker beams are much stiffer also will apply in this case.
1b. If, for some reason in spite of the above arguments, pure-DNA structures are not stiff enough, we would want to augment our set of MBBs by some kind of much stiffer rod. (And, eventually, we'd want to do this anyway, to open up new applications requiring more stiffness in the assembler than whatever it needed to reproduce itself.)
There are natural protein rods that are quite stiff, including some virus coats (which are also quite thick, but would still be usable). The one with the longest persistence length is actin, a linear polymer of two protein subunits which has a persistence length of more than a micron, and is quite thin. It's part of muscle, and is also used inside cells for various purposes related to motion and stiffness. To use actin as the basis of short rod-like MBBs for nanomachines, we'd need to develop a way to make actin rods of a desired length, preferably with specific DNA sequences attached at the ends and maybe at specific points along the rod. We have some ideas for doing this - contact us if you are interested.
It is also conceivable to stiffen DNA structures by incorporating other organic molecules into them which can produce covalent crosslinks. Bergstrom has designed and incorporated custom organic molecules into DNA with the idea of linking neighboring dsDNA domains (double helices), though I don't believe he actually demonstrated linking them (and he told me his grant applications to pursue this were rejected -- that was several years ago); Glick has incorporated bases containing -SH groups into DNA, which when oxidized produce a "disulfide base pair" (whose overall shape is compatible with that of a standard Watson-Crick base pair, so the dsDNA is not distorted) which makes a single dsDNA domain stiffer. Glen Research, which offers a catalog of about 60 modified bases that can be incorporated into DNA, told me it could synthesize a custom one (from an existing published protocol, like that of Glick's) for about $1500.
1c. If stiffness is still an issue, there are several other potential ways to improve it, but they're all harder. But we'd be surprised if the above-described methods weren't sufficient. Once we have a chance to produce a detailed design of a prototype assembler on which to do a mechanical analysis and/or simulation (after seed funding, but long before trying to build one), we'll have much better evidence for this claim.
2. For flexible joints, we can in most cases just provide fewer connections between structural parts, leaving them with some freedom of motion, since the individual connections will not be very rigid.
Whenever a single tether-like connection is used (whether short or long), we have a fully flexible joint.
For the assembler to work, its main operation is to bring some molecular surface into some range of possible positions and orientations, letting the natural affinity of the structure of that surface provide the finest positional control, so "tethering" (with short tethers) will often be adequate.
Single-stranded DNA is a very flexible tether. Several other flexible linear polymers can be chemically joined to DNA if they make better tethers for some purposes (for example, if the negative charge of DNA makes it undesirable).
So we don't consider flexible joints to be a problem.
3 (actuators). Our DNA actuator design will be sufficient for controlling the first assembler, since it provides an almost unlimited number of independent degrees of freedom (by means of variants which can be independently controlled), each with a modulatable tensile force several times higher than that required to pull apart dsDNA (and which can be increased by using more than one actuator in parallel), and an estimated reaction speed of no worse than 30 seconds (the demonstrated speed of Bernie Yurke's related actuator), which is compatible with performing hundreds of sequential operations on a reasonable timescale, and which can probably be highly optimized.
Ultimately we would also like to develop faster actuators, and there are many possible ways to do this. One of the simplest was reported by John Gaynor at the most recent Foresight Molecular Nanotechnology conference, in which a specific protein's affinity for DNA is modulated by a small organic ligand. There are also possibilities for actuators controlled by something other than a dissolved chemical species. But none of these will be a requirement for making the first functioning assembler.
4 (grippers). It is likely that DNA hybridization alone will be sufficient for making an assembler per se, which assembles MBBs with specific DNA sequences already attached, which are used only for allowing the assembler to grip them in a controlled orientation.
For many applications, we'd like to grip a more general class of object, such as the small organic molecules which have been developed as possible electronic switching components. The most obvious (and best developed) method is to use a protein-based receptor, or possibly an RNA-based receptor, with complementary shape and charge to the object to be gripped. It has been possible for many years to induce animals to develop antibodies to a variety of targets, and more recently several other methods have been developed for creating peptides with high affinities to specific targets, such as phage display. This seems to be a mature technology, and is within the area of expertise of one of our informal advisors, and his company.
Recent announced developments have included peptides with a high and specific affinity for silicon and for certain metals (a different peptide specific for each one). These would be directly useful for certain applications, and they add to the long list of reported successes in development of molecules with specific desired affinities.
More speculatively, but quite feasibly, it should be possible to use similar methods to develop peptides whose affinity for some target was high only when, for example, some metal ion was present (as is the case for natural "zinc finger" proteins, and for the "6-His" tags which are used commercially for separating synthetic proteins from mixtures in a nickel- dependent way). This could probably be used to develop peptides which gripped specific targets with a modulatable affinity, if this capability was necessary.
5 (joining tools). There are several possible solutions for the system used to permanently join MBBs in the first assembler and in its first products, and we don't yet know which one will prove simplest. For any of them, we would need to augment our set of scientific advisors by at least one chemist, and one specialist in whatever technique we would use for developing the joining tool. (In the case of artificial evolution of protein enzymes, such a specialist has already informally offered to be our scientific advisor.)
a. The ideal solution for many applications would be to form covalent bonds between specific organic moieties (molecular fragments) designed into the MBBs we wanted to assemble, or naturally present in them, and to catalyze the formation of these bonds by an enzymatic "tool" held in place by the assembler, so that no special provisions would be required to keep the rate of "accidental" bonding of MBBs very low, even when new MBBs introduced into solution had not yet been adsorbed onto the gripper's receptor for them.
One strategy for this would be to develop a new protein enzyme which catalyzed the formation of a suitable bond. To do this, we'd study the known natural enzymes which catalyze related reactions, and with the help of expert advice from a chemist, choose some which might serve our purposes. Then we would contract with a company which artificially evolves improvements to natural proteins by generating and testing lots of modifications to the gene that codes for them. I have asked the CEO of one such company how feasible that sounds to him, and what it would cost. He says that, once the natural protein to modify is chosen and a test for its effectiveness exists, optimizing it by artificial evolution takes 6 months and costs $500,000. If the protein already catalyzes the desired reaction on small substrates but has to be modified in order to remove steric constraints so it will work on extended structures, then he is not aware of this kind of modification having been tried, but he sees no fundamental obstacle to its working, so if several were tried, it it reasonable to suppose that one would work.
In fact, just a few months ago there was a rationally designed modification to a catalytic protein reported here which modified its substrate specificity in a similar way. This modification consisted of changing the shape of a "pocket" into which the substrate should fit, near the part of the substrate modified by the reaction. This can be thought of as "proof of concept research" along these lines.
b. There have recently been several "dna enzymes" produced by artificial evolution of DNA sequences which hold a metal ion in a way which forms a catalytic center. The developer of these thinks they can be applied as very sensitive detectors of the metal cofactor. The same technique could presumably also be applied to produce metal-containing DNA enzymes to catalyze a desired reaction which was already known to be catalyzed by that metal in a less specific way under other conditions.
c. We could choose organic groups which readily form covalent bonds under suitable conditions, with no need for catalysis. There are several possible choices, including disulfide bonds (though we'd want a better chemist advisor to help us pick the best choices to actually try). The problem here is that the reactions are slow, so we'd want the assembler to have some way of temporarily fixing several MBBs into place before joining all of them at once. That way might consist of any of the weaker fast-forming bonds I'll mention next.
d. Especially for the first assembler, we might not need the strength of covalent bonds, either because weaker ones were sufficient for the expected operating forces and design lifetime, or because we would use several in parallel when greater strength was required. (It is worth noting that most large structures in living organisms are held together with non-covalent bonds.)
There are several kinds of non-covalent bonds to make use of, some of which form very quickly. I've already mentioned bonds between peptides which form around coordinated metal ions such as nickel. There is also DNA hybridization itself, which can be quite permanent for sufficiently long DNA sequences in the right solutions. There are variants of DNA, namely PNA, which form stronger bonds which (unlike DNA) don't depend on the salt concentration for their strength. And there are high-affinity bonds between peptides or proteins and other proteins or a variety of other molecules, the strongest of which is between biotin (a small molecule often built into synthetic DNA) and avidin (a natural protein readily available commercially).