Assemblers that have been designed to be able to make exact copies of themselves have been termed replicators.
A key capability of assemblers will be the ability to do mechanosynthesis. Drexler (1992) has defined mechanosynthesis as "Chemical synthesis controlled by mechanical systems operating with atomic-scale precision, enabling direct positional selection of reaction sites ..."
The assemblers envisioned by Drexler for use in molecular manufacturing would be very strong, stiff structures that would work in a vacuum, be able to handle very reactive chemical moieties, such as free radicals, and be able to position such species to a precision of better than 0.1 nm. It is difficult to see how such complex devices could be manufactured directly except by already existing, more primitive assemblers.
The obvious bottleneck to the development of nanotechnology is the identification of the first generation of replicating, programmable assemblers that can be fabricated using current laboratory technology, or incremental developments of current technology, which will not require for their fabrication already existing replicating assemblers. To be useful, such proto-assemblers must be able, with appropriate programming, to fabricate second generation assemblers with enhanced synthetic capabilities.
It has been suggested (Drexler, 1992) that early generations of assemblers will deal not with very reactive single atoms or small groups of atoms in a vacuum, but rather with larger molecular building blocks (MBBs) in solution. Drexler considered two approaches: (1) Brownian assembly of fairly large MBBs, each of which is a polymer designed to fold into a specific shape, and (2) mechanosynthetic assembly of small MBBs guided by a scanning force microscope.
Building molecular devices (nanodevices) by Brownian assembly of engineered proteins or other polymers, designed to fold and form matching complementary surfaces with which they bind noncovalently to each other in specific arrangements, presents very formidable design challenges, which seem unlikely to be solved within the next few years at current rates of progress.
As for the second approach, initial analysis (Krummenacker, 1994) reveals that the choice of appropriate MBBs and methods of linking them together in precise three-dimensional networks is a very difficult problem in organic chemistry.
Therefore, there seems to not be an obvious way to make even proto-assemblers starting with current technology.
Drexler discussed the engineering of protein-like molecules having complementary surfaces for self-assembly, and proposed that designing proteins to fold in a predictable fashion was much easier than predicting the folding pattern of natural proteins. He suggested methods for increasing the stability of the desired conformation of designed proteins. This approach can almost certainly eventually be made to work, but it entails a formidable challenge in protein design, as stated earlier.
Drexler pointed in this direction in general terms in 1992 ("Nanosystems"
p.446, after discussing the work of Ned Seeman (1991)):
"The ability of complementary segments of DNA to bind to one another provides a powerful, readily controllable mechanism for guiding the Brownian assembly of molecules. Hybrid structures of DNA and other macromolecules may prove particularly attractive."
Protein-DNA conjugates, with different DNA sequences attached at specific points on the protein surface, thus become the MBBs to be assembled by Brownian motion to build nanodevices. One of the advances in Smith's proposal beyond what Drexler suggested in 1992 is that "loose" binding of complementary DNA strands on different protein molecules confers specificity on subsequent rigid, covalent binding between protein MBBs by holding the MBBs accurately enough that each covalent binding site is accessible only to its intended partner. The resulting structures will be strong enough to use as components of molecular machines able to actively position similar MBBs, and to perform some forms of positionally-controlled chemical synthesis. A second advance is how to achieve a defined spatial arrangement of several different DNA sequences on the surface of the protein, even though it is not feasible to use a different attachment chemistry at each attachment site on the protein.
The modifications required of the proteins for permanent bonding are thus much simpler than if the proteins had to be redesigned to have complementary surfaces. An additional advantage of this approach is that the library of potential MBBs includes the vast collection of proteins produced by biological evolution (many of which have known structures) as well as synthetic variants thereof. Thus a wide variety of MBBs with specific functions (structural, ligand-binding, and catalytic) can be included in these first generation nanodevices.
One well-known scheme to join proteins together to form complex structures (Drexler 1981, 1986, 1992) is to learn to engineer folding polymers with many different specified surface structures, and to use this ability to make "bricks" which self-assemble in solution into a specific arrangement due to complementary surfaces. The protein design issues to implement this scheme are very challenging; there has been steady progress, but the level of success required for this application still appears to be several years away. In contrast, the DGAP method requires much less innovation to develop from already-demonstrated technologies.
The DGAP invention is a method for assembling almost-natural molecular building blocks (MBBs) into specific, almost-arbitrary arrangements, potentially with hundreds of distinct block locations in each assembly. Blocks at different locations in one assembly are allowed to be the same, which means that a small library of distinct block types could suffice for many designs. The assemblies of blocks could be joined recursively into larger structures.
MBBs could be joined in a fairly rigid fashion. Protein-based MBBs could be held together in a compact fashion with covalent tethers at several locations around the joined surfaces. Since the overall connectivity in a network of MBBs could be essentially arbitrary (thus 3-D and polycyclic), larger structures could be reasonably rigid as well. They are likely to be able to be made rigid enough for positional control of chemical reactions, though not strong enough, at least in the first generation, to apply the significant forces needed for some uses of mechanosynthesis. (See Drexler 1992, pp.462-463, for a rigidity and strength analysis of a similar system.) They will certainly be rigid enough for making "robot arms" which could move other similar building blocks into contact with the correct partners, and thus build similar devices more directly than by unassisted self-assembly.
Besides rigid joints, rotating or tethered joints between MBBs are also possible, either via different choices of chemistry for covalent bonding, or by the use of building blocks which already incorporate such joints.
Since slightly modified natural proteins could be used as building blocks, the resulting devices could incorporate biomolecules of interest in particular applications, such as antibodies or enzymes.
Specifically, one could attach to each building block (at particular sites on its surface, here called "C sites" for "complementary") a few different single-stranded DNA molecules, and (at "P sites", for "permanent") a few non-specific permanent binding sites. (The P sites will probably be sites capable of forming covalent crosslinks, though other choices are conceivable; several choices of chemistry for covalent crosslinking are already in wide use in molecular biology. Also, it will sometimes be convenient for an assembly consisting of a DNA strand and one or more P sites to be added to each C site already on the protein, rather than for the protein to start out with separate P sites and C sites.)
During assembly of several MBBs into one structure, the P sites are prevented from reacting initially by maintaining a chemical environment preventing reaction until the DNAs are hybridized, after which each P site is only able to contact that P site with which it is meant to react. The chemical environment is then changed to permit reaction (e.g. by the addition of appropriate crosslinking reagents, active only at the P sites).
That is, DNA hybridization (joining of very-easy-to-design, very-specific complementary "surfaces") guides self-assembly, allowing non-specific connection sites designed solely for their chemical/mechanical properties to be uniquely paired.
The problem of making a MBB by attaching a different DNA sequence to each of several specific sites on a protein is the only part of the DGAP invention without a direct precedent in current molecular biology. We have developed several possible methods of solving it (the best of which we are keeping under nondisclosure), and this is the main focus of our present research; fully developing one of these methods and putting it into practice will require some research and development in the laboratory. The kinds of equipment and procedures required for performing the R&D on any of our proposed DNA-attachment methods are standard in molecular biology.
For making large structures out of networks of linked MBBs, the average MBB must be directly linked to more than two neighbors, preferably at least 3 or 4 if the structures should be rigid. The number of C sites and P sites on each MBB needed to accomplish this depends on the specific design of an assembly of MBBs; for ease of designing assemblies without this connectivity being a limiting factor, we should be able to add different DNA sequences to at least 6 different sites on one protein (taking advantage of the possibility of connecting the C site to the middle of the DNA strand, so there would be 12 different DNA sequences leaving the MBB). The methods we are investigating should be able to accomplish this; if they can't, as few as 4 or even 3 separate C sites per protein would still be useful for many applications, with more design effort required for some of them.