A publication of the Foresight Institute
the University of Michigan, Joel Gregory grabs a molecular
rod with both hands and twists. It feels a bit weak, and a ripple
of red reveals too much stress in a strained molecular bond
halfway down its length. He adds two atoms and twists the rod
again: all greens and blues, much better." Unbounding the
Future, Drexler, Peterson and Pergamit.
Joel Gregory's virtual reality rig doesn't exist yet (although Joel himself probably does, bouncing around some Detroit day care center) but the Joel scenario illustrates how useful molecular modelling software is to nanotechnology. Joel's software uses the Schroedinger equation to approximate the Dirac equation to approximate quantum electrodynamics to approximate how atoms behave in real life...but Joel doesn't know what these equations are, nor does he need to. Thanks to software, this bright engineering student can design nanomachines without having to understand the work of dead geniuses six abstraction layers north.
However, few scientists today rely on molecular modelling software to get their work done. Most get kudos for a test tube of finished product, not a text file of the recipe. So a lot of shake-and-bake types distrust nanotechnologists' heavy use of modelling software. ("It's very pretty, Dr. Drexler, but how do you know it'll work?")
And their indignation is not unjustified. Because synthetic chemists and molecular biologists work with loose and floppy organic molecules: a million ways exist to brew up any complex structure, and fifty amino acids have more than mere trillions of ways to fold up into a protein. With all those approximations, doing it all in software just isn't possible yet.
But nanotechnology is engineering, not chemistry. Nanotechnologists are free to choose strong, stiff materials whose atoms won't jig and flop about. Materials like tetrahedral covalent carbona.k.a. diamond. The atoms in diamondoid structures have little "conformational freedom" (Nanosystems, section 1.4.2) within a very wide energy band, and even gross approximations of their activity can be accurate. So while today's software can't model a protein crumpling, it can model a nanomachine. That means engineers can design nanomachines and pronounce them valid today, before it's possible to build them in real life. One package that lets you do it is CambridgeSoft's Chem3D Pro, which I bought as part of the ChemOffice Ultra (Windows) bundle for $899. (Editor's note: the software suite is also available for the MacOS.)
The ChemOffice Ultra CD includes Chem3D Pro 3.5, ChemDraw (a 2D structure drawing program known to most chemists) and a chemical structure database or three, plus a hand-holding video. The three components aren't integrated very well and are best used separately. Two semi-empirical molecular calculation programs are included (MOPAC and MM2) and Chem3D's user interface gives them both friendly faces. If you need Gaussian (an extra $750 from its vendor) there's a free user interface for it at CambridgeSoft's Web site. The box (with excellent manuals) fits nicely on a bookshelf.
Chem3D's user interface is simple: a drawing window with a grab-and-drag arrow to rotate your structure at each corner, a toolbox of chemical bonds, and a too-short toolbar. (I wanted far more functions on the toolbar, since I soon got sick of navigating through menus; unfortunately you can't customize it.) You place atoms with a cursor, entering atom types into a text box. The text box recognizes atom names as well as chemical symbols, plus some common molecular fragments. It takes about an hour to get used to. I'd have liked a drag-and-drop toolbox of atoms.
The mechanics of building molecules in Chem3D are the most intelligent I've seen in this sort of program: they second-guess many actions, leaving settings like current atom type in the text window so you don't have to retype them. The program can also adjust bond lengths and angles as you work. It "rectifies" (caps off dangling bonds with hydrogen) too, then deletes these hydrogens when you draw in additional bonds.
|Disassembling Drexler and Merkle's planetary gear to see how it worked was easy.|
Select an atom and change the text box, and you've replaced
that atom with whatever you typed in the text box, even if what
you typed was the name of a nanomachine part you built yesterday.
(This feature was probably intended to make life easier for
molecular biologists changing R groups on amino acids, but it's
just as useful to nano tinkerers.) You can select single atoms,
fragments and substructures, and complete molecules. Building and
adding fragments in this way I managed to make a copy of Drexler
and Merkle's 2568-atom
fine-motion controller in under three hours.
Next comes the fun part: relaxing your molecule to an energy-minimized state. MM2 gives you local minima only, so it's best to check the bond integrity of your structure first with the Cleanup tool. Watching the atoms shake, rattle and roll their way to a minimum is hypnotic, and as wasteful of your time as it is of your PC's, so it's best to switch off the "display every iteration" option if you're the type who needs to eat occasionally. The more complex MOPAC engine (handling AM1, MNDO, MINDO/3, and PM3 potential functions, whatever they are) is slower and not recommended for anything over 200 atoms. Molecular dynamics computationsbasically, how the atoms in your structure bubble about as a function of heatare just as easy.
So easy in fact that many of my nanomachine parts hit the nanogarbage can at this point, minimizing to shapeless lumps of dough. (One criticism of molecular machine designs is that they look too suspiciously simple to actually work; Drexler and Merkle might improve nanotechnology's standing if next to their beautiful pumps, gears, and manipulators they'd put some of the designs that minimized to sludge!) Non-chemists who'll get the most out of Chem3D are those who learn best by making mistakes.
Nanomachines are mechanical devices, and mechanical devices depend on surfaces moving against each other. So spacefill rendering (intersecting spheres representing Van der Waals radii) gives the best impression of which part does what. (Try viewing the .pdb files on the Institute for Molecular Manufacturing web site as ball-and-stick and you'll see how difficult it is to see function when the machine doesn't look solid.) Yes, it is tough on your graphics subsystem, and movies are even tougher. Moviesanimations of your structure rotating around any axisneed serious hardware. A 36-frame movie starring Drexler's fine-motion controller, rotating 10 degrees each frame and rendered in spacefill without perspective, couldn't finish on a 32MB Pentium 166; halfway through it started caching to disk, slowing redraws to a crawl and taking up so many resources I couldn't move the cursor. The same movie almost fit into the same system with 80MB, and on a 128MB AMD-K6 workstation it worked fine.
(By the way, the rendering itself isn't anything that will win a graphic design competition. This is workaday science software, after all; its strength is in the way it crunches numbers, not how it displays your molecules. If you want your spacefills raytraced and Gourauded, look elsewhere.)
It's easy to pull numbers out of Chem3D, everything from bond lengths and strains to close contacts and ring closures. It can also generate lots of tables for your nanostructure, although in the Windows version this part is surprisingly clunky: you click on ATYPES.TBL, not "Table of atom types." All other tableslists of substructures, elements, bonds, it's all hereare the same throwbacks to DOS I thought I'd left behind years ago. When you decipher the file names, though, the tables are comprehensive.
There are some other rough edges. The move-object-to-centre command doesn't work properly, maximizing a window results in your model redrawing three times, and copying an unminimized molecule to the clipboard often pastes back a mangled blob. Also, the number of fatal errors that occurred during MM2 minimizations was unacceptably high; save your work often. (This lack of robustness is common among PC programs originally written for the Mac, as anyone using Adobe software on a PC will attest. The Mac version of the program is stable and robust.)
Chem3D supports over a dozen molecular file formats including .pdb and .mol, but it wouldn't open any of six .pdb files I downloaded from the Protein Data Bank websitethe program seems to be fussy about syntax and won't ignore extraneous lines on a text file. So be prepared to do some editing to get .pdb files to open properly.
Also, remember Chem3D was created for chemists, not nanotechnologists, so there's a long list of nanotechy things it can't do. It can tell you if a finished nanostructure is likely to be stable; it can't tell you if there's any synthetic route to it, or if any step in its construction would be impossible for mechanosynthesis. It can't model machines in action; move rods or spin gears and they won't move or spin any part they're attached to. It can't model chemical reactions like hydrogen abstraction or dimer deposition. But it does let you design molecular machine parts, tell you if they're stable, and help you if they're not. In other words, it lets you studyhow bond lengths differ from atom to atom, how different forces hold atoms apart and together, why molecules do the wonderful things they do.
I've spent a lot of hours with Chem3D now, spaced out by the spacefills, slack-jawed and drooling in wonder at how these tiny machines are going to change our lives. It's giving me a real feel for how atoms interact...and it's making me ask intelligent questions of my bookshelf, replacing the wow-factor that first interested me in molecular nanotechnology. It's increasing my understanding of the basic science in Nanosystems. But most of all, it's making it more fun.
Chris Worth is a technology writer and Foresight Senior Associate based in Singapore. He can be reached by email at email@example.com.
|Foresight Update 30 - Table of Contents|
As the ability to fabricate nanometer scale structures
improves, opportunities arise for applying this capability. The
two papers described below report applications in sensing and in
separation techniques respectively.
The recent announcement of the biosensor based on ion-channel switching from the Cooperative Research Centre for Molecular Engineering & Technology directed by B.A.Cornell has received a good deal of press coverage. From the viewpoint of nanotechnology development, the sensor is an impressive application, combining several large atomically precise components with a lipid bilayer tailored with synthetic organic compounds to yield a sensor with the specificity of antibodies, an inherent gain mechanism, and sufficient stability to last for months.
B.A.Cornell et. al. published a description of their sensor in [Nature 387:580-583 5Jun97--MEDLINE Abstract]. There are two basic variations on the sensor. Both versions control an ion current through gramicidin ion channels embedded in a lipid bilayer. When the channels are free to diffuse, channels in the top and bottom layers of the bilayer line up to form dimers. These dimers then conduct a current through the bilayer, and this current is sensed by an electrode underneath the bilayer. The biosensor controls this current by tethering the ion channels, restricting when they can form dimers by restricting their movement. The ion channels in the lower layer of the bilayer are always tethered to the electrode in both types of sensors. Whether a current flows depends on whether the ion channels in the upper layer are free to diffuse to the positions of the lower layer channels. At this point the two types of sensors differ.
In the form where current is increased by the presence of the analyte, the upper level ion channel is tethered to an analyte molecule. This analyte molecule is in turn bound to an antibody Fab fragment. The Fab fragment is tethered to the electrode surface, so neither it, nor the analyte molecule, nor the upper level ion channel is free to move. At this point no ion channel dimers form so no current flows. When free analyte molecules are added, they displace the analyte molecule from the Fab fragment. This frees the ion channel, still tethered to the analyte molecule, to diffuse in the upper layer of the lipid bilayer. The upper ion channel then encounters a site with a tethered ion channel in the lower bilayer, forms a dimer with it, and permits an ion current to flow.
In the form where current is decreased by the presence of the analyte, the upper level ion channel is tethered to an antibody Fab fragment that binds to one site on the analyte. Another antibody Fab fragment, complementary to a different site on the analyte, is tethered to the electrode surface. In the absence of the analyte the upper level ion channel can diffuse freely, forming a dimer with the lower level ion channel and permitting a current to pass. In the presence of the analyte, the analyte forms a cross-link between the Fab tethered to the electrode and the Fab tethered to the upper ion channel, preventing the upper ion channel from forming a dimer and cutting off the current.
There are some additional complications to this picture, notably the use of membrane spanning lipids to improve the stability of the membrane. This development effort has taken about 10 years and about $21 million. Since the selectivity of the sensor is set by Fab fragments, versions can be built to sense a wide variety of analytes: "uses might include cell typing, the detection of large proteins, viruses, antibodies, DNA, electrolytes, drugs, pesticides, and other low-molecular weight compounds." Strictly speaking, the breakup of the ion channel dimers by the analyte cross-linking can't be considered mechanochemistry, since the bonds between the dimers are weak enough that they can diffuse thermally. Perhaps the best analog is to compression of a gas piston, since the cross-linking effectively compresses the 2D gas of upper lipid bilayer ion channels into a smaller area than they would occupy in the absence of the analyte.
The authors write that "The switch has a high gain; a single channel facilitates the flux of up to a million ions a second." In a sense, there are two stages of gain built into the analyte cross-linking device, because "the antibodies on the mobile [upper lipid layer] channels scan an area of the order of 1 µm2 in less than 5 minutes. Thus with a low density of channels and a high density of immobilized antibodies, each channel can access up to 103 more capture antibodies than if the gating mechanism were triggered by a directing [sic] binding of analyte to the channels."
|..."industrial kidneys" for heavy metal recovery from wastewater...|
In [Nanotechnology, 7:177-182 Sep96],
S.L.Gillett suggested that, amongst other applications,
"industrial kidneys" for heavy metal recovery from
wastewater would be an early application of nanotechnology. A
paper by X. Feng et. al. in [Science 276:923-926
Abstract] appears to be a substantial step in this
direction. This paper describes the fabrication of "a
cross-linked monolayer of mercaptopropylsilane [which] was
covalently bound to mesoporous silica and closely packed on the
surface." The mesoporous silica was "synthesized in
cetyltrimethylammonium chloride/hydroxide (CTAC/OH), silicate,
and mesitylene solutions." The silica itself is ordered on
the nanometer scale, with 55 Å pores and "a surface area of
900 m2g-1." This is not, however, an
atomically precise material. The actual functional monolayer,
however, has well defined -SH moieties on its surface, and also
has fairly well controlled lateral organization from the close
packing of the monolayer. The monolayer was synthesized by mixing
mesoporous silica (which had been calcined at 540°C, then
partially rehydrated by refluxing with water) with
tris(methoxy)mecaptopropylsilane in an organic solvent and
refluxing. The methoxy groups hydrolyzed off, leaving the silane
bound to the silica. This system was studied at surface coverages
ranging from 10% to 76%. The authors studied the adsorbed layer
with 29Si and 13C NMR. The carbon NMR
showed that "at higher population densities...The molecules
have a higher degree of ordering that narrows the linewidths in
the 13C spectrum and allows better resolution of the
peaks for all three carbons." The silicon NMR shows a
corresponding shift from isolated and terminal to cross-linked
siloxane groups at high coverage.
This material was designed to bind heavy metals. When binding mercury, its "distribution coefficient, Kd, has been measured to be as high as 340,000. [Kd is defined as the amount of adsorbed metal (in micrograms) on 1 g of adsorbing material divided by the metal concentration (in micrograms per milliliter) remaining in the treated waste stream.]" The chemical bonding of the mercury to the material has been studied with EXAFS, which uncovered a chelate-like cyclic structure, -S-Hg-O-Hg-S-, binding two mercury atoms to two adjacent thiol groups. Part of the high affinity for mercury is therefore due to the close packing of the organic monolayer. The authors write that: "Beyond its immediate applications in environmental cleanup, FMMS [functionalized monolayers on mesoporous supports] provides a unique opportunity to introduce molecular binding sites and to rationally design the surface properties (for example, wettability and charge density distribution) of mesoporous materials." From the perspective of nanotechnology development, this class of materials provides an application architecture that might employ increasingly complex and sophisticated self-assembled structures.
|Foresight Update 30 - Table of Contents|
The ability to design and synthesize atomically precise 3D
structures is central to nanotechnology. One route to this
capability is the synthesis of nonperiodic linear polymers such
as proteins, which then fold into the desired 3D structure. This
strategy is not limited to naturally produced polymers such as
proteins and nucleic acids. The exponential space of design
possibilities for this technique is available for any polymer
which has a reliable stepwise coupling chemistry and a diverse
selection of monomers. These polymers have been dubbed
"foldamers" by professor S.H.Gellman. The papers
described in this section report some recent advances in this
C.L.Wysong et. al., writing in [Chemtech 27:26-33 Jul97] describe some recent advances in the use of unusual amino acids to control peptide structure. In particular, they describe the use of ,-disubstituted amino acids to produce a helical structure for the peptide. Normal amino acids have a single substituent on their carbons. They have a H2N-CHR-COOH structure. The disubstituted amino acids have a H2N-CRR'-COOH which replaces the hydrogen on the carbon with a second alkyl group. These disubstituted acids restrict the angles that can exist between their side chains and the adjacent peptide bonds. As a result, "these short peptides [containing the disubstituted acid residues] are highly helical, more so than would be expected based on their length and the helix-promoting effects of the [normal, singly substituted] proteinogenic amino acid residues alone." Oddly enough, these ,-disubstituted amino acids were initially discovered in natural products, in fungal peptides containing a high proportion of the simplest possible disubstituted amino acid, -aminoisobutyric acid (Aib, H2N-C(CH3)2-COOH).
If the only accessible disubstituted acid were Aib, it would not provide a powerful extension to our control over peptide structure, despite the improvement in the predictability of the secondary structure. General synthetic procedures for these compounds have in fact been known since 1911, but, until recently, linking them into peptides was difficult. The authors write: "The additional alkyl (or aryl) group reduces the number of peptide backbone conformations [hence their value], but the greater steric requirements make the coupling inherently more difficult." Recently, Carpino introduced the use of acid fluorides of ,-disubstituted amino acids as an activated form of these monomers which are suitable for peptide synthesis. They are sufficiently reactive that coupling occurs under mild conditions, "mild enough to allow the use of a broad range of protection schemes." In the authors' laboratory, they "have synthesized peptides containing up to 80% ,AAs [amino acids] with as many as three sequential ,AAs using standard Fmoc SPPS [solid phase peptide synthesis] with preformed acid fluorides."
The general procedures for preparing ,AAs are the Bucherer-Bergs and Strecker syntheses. Both convert readily available ketones into the ,AA in two steps. In addition to the simple case where the two substituents are separate alkyl groups, they can form part of a ring. Cyclic ,AAs from cyclopropyl to cyclooctyl have been incorporated into peptides. These cyclic side groups both provide additional positions at which design modifications can be made and constrain the conformational changes that thermal motion can cause. The authors themselves refer to methyl sulfide substituted and phenyl-substituted cyclopropyl ,AAs as "conformationally restrained cyclic analogues of methionine and phenylalanine."
Peptides containing ,AAs "offer conformational stability and resistance to enzymatic hydrolysis, which are two major shortcomings that have hampered the development of efficient peptide drugs." While the enzymatic stability is not relevant to nanotechnology, the conformational stability is directly relevant to building stable, predictable building blocks for nanoscale machinery. The use of ,AAs is specifically cited in Drexler's Nanosystems as a useful tactic in building stably folding proteins.
In Update 28, this column covered recent advances in the use of -amino acids to construct short peptides with predictable helical structures. More recently, D.H.Appella et. al. have reported in [Nature 387:381-384 22May97--MEDLINE Abstract] on the design and synthesis of another, closely related peptide with a different helical structure. The authors had previously synthesized trans-2- aminocyclohexanecarboxylic acid (trans-ACHC). They had linked this into a peptide and found that it formed stable helices "defined by interwoven 14-membered-ring hydrogen bonds." They examined 8 cycloalkane-containing -amino acid peptides in 6 possible helices and found that, while the trans-ACHC peptide in the 14-helix secondary structure was the most stable, a trans-2- aminocyclopentanecarboxylic acid (trans-ACPC) peptide in a novel helix containing 12-membered rings was nearly as stable. This molecular mechanics prediction motivated the synthesis of the trans-ACPC peptide. The octamer was shown to have the predicted structure via x-ray diffraction, and a solution of the hexamer in pyridine-d5 was shown to have the predicted structure via NMR measurements. The structural change from the 14-helix to the 12-helix changes some important structural features, inverting the direction of the helical hydrogen bonds with respect to the location of the C-terminal and N-terminal ends of the peptide. The authors write that: "The predictable residue-based conformational control offered by -peptides suggests that this class of unnatural foldamers will be well suited to molecular design efforts, such as the generation of novel tertiary structures and combinatorial searches for selective biopolymer ligands."
Writing in [C&EN 32-35 16Jun97], S.Borman also surveys recent -peptide work. This article also contains a sidebar citing a wide variety of foldamers that have been synthesized in recent years, including:
From the perspective of nanotechnology development, any of
these chemistries might serve as a source of useful machine
parts. We and the pharmaceutical industry share the need for
predictable, controllable secondary structure. We are less
concerned with toxicity, and are not concerned with in-vivo
degradation mechanisms. We generally need larger structures, so
are more concerned with maximum chain length. It isn't clear
which group needs a larger variety of monomers, and which will be
more price sensitive. In general, the needs of both groups appear
to be fairly similar.
Jeffrey Soreff's Technical Progress column is continued on the next page.
From Foresight Update 30, originally published 1 September 1997.