IMM Report Number 9
In conjunction with Foresight Update 36
Recent Progress: Steps Toward Nanotechnology
By Jeffrey Soreff
Carbon nanotubes are potentially important components in nanoscale machinery. They are stiff and strong. Some of them are good conductors, while others can form electrical components such as diodes. The papers summarized below advance our knowledge of the mechanical properties of nanotubes on the one hand and our ability to synthesize them on the other.
M. R. Falvo et al., writing in [Nature 397: 236-238 21Jan99] describe experiments where nanotubes were rolled by AFM tips on graphite surfaces.
By applying force from an AFM tip, the authors were able to roll a 500 nm long, 27 nm diameter nanotube. They alternated pushing the tube with imaging its topography. The tube had an asymmetrical conical cap which made the rotation clear in the images. The lateral forces measured by the AFM tip also changed with a period matching the circumference of the nanotube. The rolling wasn’t smooth, but rather showed slip-stick peaks of 20-50 nN. The authors attribute this to peeling apart the tube and substrate surfaces, noting that peeling apart the entire contact area would require 800 nN, so this upper bound is consistent with the forces they observe. The authors found that “rolling behavior has been accompanied by a preferential, threefold, in-plane orientation that indicates intimate nanotube/graphite contact, and perhaps lattice registry.”
This work demonstrates a nanometer-scale rack and pinion gear. This technique allows controlled rotation of nanometer-scale objects around an axis in the plane of the substrate. This permits lifting planar workpieces out of the plane of the substrate for the construction of 3D objects.
S. Fan et al., writing in [Science 283:512-514 22Jan99] describe a technique for synthesizing blocks of aligned multiwalled nanotubes on a porous silicon substrate. The nanotubes have diameters of 16 ± 2 nm.
In their work, the authors anodically etched silicon wafers to form a porous surface layer, evaporated a 5 nm thick iron film on to the wafers through shadow masks with square openings with sides of 10-250 microns, annealed (and oxidized) the silicon and iron catalysts at 300 C in air, then grew blocks of nanotubes on the surface at 700 C from ethylene in argon. The blocks’ height ranged from 30 to 240 microns for growth times of 5 to 60 minutes. The nanotubes adhere to each other well enough that “no nanotubes are observed branching away from the blocks” and that “high-resolution SEM images show that the nanotubes within each block are well aligned along the direction perpendicular to the surface.”
The authors interpret the nanotubes as growing from catalyst iron oxide particles embedded in the porous silicon, with the ethylene feedstock diffusing through the porous layer to feed the growing tubes from below. They also write that strong interactions with the porous silicon “prevent catalyst particles from sintering at elevated temperatures”
The authors consider their technique significant as an enabler of integrated nanotube/semiconductor applications. They demonstrated that their nanotubes perform well as electron field emitters, showing stable emission over a 20 hour test and I(V) curve similar to arc-discharge synthesized nanotubes. They note that “the synthesis process should be entirely compatible with existing semiconductor processes, and should allow the development of nanotube devices integrated into silicon technology.”
From the viewpoint of nanotechnology, this technique provides a method to hold catalyst particles immobile and separate during synthesis of nanotubes. Perhaps some variant on the composition of the catalyst or of the carbon source will permit synthesis of nanotubes with precisely defined geometry. Certainly the environment is more homogeneous than in the arc-discharge or laser-pulse technique. Perhaps catalyst particles with well-defined sets of metal atoms could be made from organometallic compounds or metalloproteins.
Protein synthesis is currently one of our most powerful routes towards construction of atomically precise 3D structures. It provides a rich design space, both functionally and geometrically. The papers summarized below advance our ability to predict and control the 3D geometry of the proteins we build.
S. Borman, writing in [C&EN 51-55 24Aug98] reported on advances presented at the Symposium of the Protein Society in San Diego in July 1998. One advance that could particularly aid nanotechnology is a technique for 3D protein structure prediction called Convex Global Underestimation (CGU) invented by A. T. Philips, J. B. Rosen, and K. A. Dill. In one of the authors’ papers, in From Local To Global Optimization, P. M. Pardalos et al. eds. (Kluwer Academic Publishers, 1988), available at http://www.cs.uwec.edu/~phillips/papers/tuy-volume.pdf, they describe their basic idea as avoiding “kinetic traps” which other computational methods fall in to.
An n-residue protein has a minimum of 2n degrees of freedom (corresponding to rotations around single bonds in the backbone) plus possibly others, depending on how complex a model of side chains is used. Other current methods, such as Monte Carlo, Simulated Annealing, Genetic Algorithms, and Molecular Dynamics can easily get stuck in local minima in the full energy landscape of the protein. The number of these local minima can grow exponentially with the length of the protein to be folded. In effect, the other current methods treat each of these local minima as a plausible contender for the global minimum. Each local minimum becomes a “kinetic trap” that the other methods must climb out of.
What CGU does is to use local minima as evidence for the global shape of the protein’s energy landscape, then jump to the predicted minimum of that global shape. It has a three phase procedure:
- Sample local minima throughout the region thought to contain the global minimum (initially the whole space). At least 2d+1 conformations are sampled for a d-dimensional problem.
- Find the global energy function which is below all of the known minima, but below them by the smallest possible amount. The global energy function is extraordinarily simple, just a sum of independent quadratic terms, one for each dimension.
- Jump to the predicted location of the global minimum, and restrict the search region to include it and the lowest local minimum seen so far.
|“Proteins (and possibly other foldamers) provide one of the most accessible routes to atomically precise 3D structures today. Advances such as this one in 3D structure prediction help us design sequences to build desired structures.”|
The process repeats until the predicted global minimum matches the lowest local minimum seen so far.
The authors found that, even with the initial random sampling of the configuration space, CGU folded proteins consistently, finding the same global minimum from 100 different starting points per protein. It consistently found lower energy minima than simulated annealing did, finding twice the binding energy for a 36-mer peptide.
CGU has consistent speed, scaling as n4 regardless of the particular peptide being folded. It is also easily parallelized, since the sampling of local minima accounts for 99% of the CPU time and can be spread across many machines. It also provides information on local minima found before the global one, helping to establish how robustly a design folds.
Proteins (and possibly other foldamers) provide one of the most accessible routes to atomically precise 3D structures today. Advances such as this one in 3D structure prediction help us design sequences to build desired structures.
P. B. Harbury et al., writing in [Science 282:1462-1467 20Nov98] describe the design and construction of a group of peptides with a novel backbone geometry.
Two major structural motifs in proteins are alpha helices and beta sheets. In alpha helices, the backbone wraps around the helical axis with a rotation of 100 degrees per amino acid residue. One class of natural proteins consists of bundles of left-handed supercoiled alpha helices. The bundling and the supercoiling are closely related. In a bundle of helices, one side of each helix is on the inside of the bundle, touching the other helices, and the other side is exposed to water. The bundle gets its stability by having hydrophobic amino acids on the bundle-interior side of the helix and hydrophilic amino acids on the bundle-exterior side of the helix. If each turn of the helix was exactly an integral number of residues, a peptide could put hydrophobic residues on one side just by repeating one sequence of residues on a helix with a straight axis.
For example, if the rotation was 120 degrees per amino acid residue, then a repeated three residue pattern of hydrophilic-hydrophobic would allow bundles to form. Because the rotation doesn’t neatly fit three residues into 360 degrees, the axis of the alpha helix winds up twisting around the center of the bundle. In natural bundles, a repeating pattern of seven residues almost matches two full turns, losing 20 degrees per repeat. The left-handed supercoiling keeps the hydrophobic residues in the repeated heptads aligned toward the center of the bundle.
In the authors’ peptides, repeating patterns of 11 residues were built. These patterns gain 20 degrees over three full turns per repetition of the pattern, so the helices twist into a right-handed supercoil.
Previous protein design work has often changed amino acids while retaining the positions of the backbone atoms. The authors designed their peptides as a demonstration of a more powerful design technique that let them vary the backbone, as well as the amino acid side chains. For each candidate amino acid sequence, they examined side chain rotamer conformations. For each rotamer conformation “main-chain coordinates were determined by exploring a parametric family of superhelix backbones” to find the lowest energy structure, thus adjusting the backbone. Low energy packings were sought in symmetrical dimeric, trimeric, and tetrameric bundles, leading to three optimal amino acid sequences. More precisely, only the 3 residues (those closest to the bundle’s interior) in the 11 residue pattern were varied during the optimization. Even so, 8 days of computer time were used.
The final peptides were 33-mers, containing 3 copies of the 11 residue sequence. The peptides contained two unnatural amino acids, norvaline (with an n-propyl side chain) and alloisoleucine (“the stereoisomer of isoleucine with inverted chirality at the Cbeta carbon”). They were synthesized by solid-phase synthesis. Circular dichromism measurements showed that all three peptides folded, with stability increasing from the dimer to the tetramer. The tetrameric peptide was crystallized, and x-ray structures agreed with the predicted core residue structures to within 0.2 angstroms.
From the perspective of nanotechnology, this work broadens the utility of protein design techniques. The ability to design a protein without requiring a natural model with the same backbone enlarges the set of global target structures that can now be chosen for protein design.
While the 3D structure of proteins is important, it is also important to understand and control how proteins that we wish to use as components in nanoscale designs contort as they are put under mechanical stress or as they catalyze a reaction. The papers summarized below extend our knowledge in these areas.
R. Merkel et al., writing in [Nature 397:50-53 7Jan99] describe experimental measurement of a sequence of barriers in the breaking of biotin-avidin and biotin-streptavidin bonds.
The authors measured the forces on the bonds at the point of rupture as a function of how rapidly tension was applied to the bonds. A piezoelectric deflector was used to control the movement of one surface, and force was measured with a “biomembrane force probe”, a weak spring built from a red blood cell on a micropipette.
When a biotin/(strept)avidin bond is stressed slowly, thermal fluctuations have more opportunity to push the biotin molecule out of its binding site. The rupture strength is therefore lower at lower loading rates. If there were a single barrier to breaking the bond, the rate at which these bonds break would depend exponentially on the applied force. There is some displacement x from the equilibrium position with no applied force to the position at the top of the bond-breaking barrier. Applying a force f effectively lowers the barrier by fx, speeding bond breakage by efx/kT.
Over small ranges in bond breaking rates, the authors saw this simple exponential dependence. Over larger changes in force, they were able to see several exponential regions in their data, which they attribute to a sequence of barriers in the bond-breaking process. For the biotin-streptavidin bond they saw barriers at 0.12 nm and 0.5 nm, while for the biotin-avidin bond they saw barriers at 0.12 nm, 0.3-0.5 nm, and at 3 nm. Detecting these barriers required examining bond breaking over a million-fold range of loading rates, from 10-2 to 104 pN/sec. Previous AFM measurements had measured rupture strength at high loading rates.
Early nanomachines may be built of biomolecules held together by noncovalent forces like the bonds probed in this paper. The thermomechanical failure of these bonds under load is an important constraint on these structures. This paper probes this process, and probes its load dependence in the long lifetime regime important to the construction of stable structures.
Another class of potential components for nanoscale structures is derived from polymers outside the realm of biotechnology. Some of these polymers self-assemble into nanoscale structures which are useful for special purposes, or they illuminate general techniques that may also be useful in building atomically precise structures. The papers summarized below describe advances in these techniques.
E. R. Zubarev et al., writing in [Science 283:523-526 22Jan99] describe the cross-linking of non-covalently bound self-assembled clusters of oligomers into covalently bound objects.
The authors’ clusters (summarized previously in Update 29) consist of roughly triblock polymers built from a (roughly 9 unit) polystyrene section, a (roughly 9 unit) polyolefin (here polybutadiene) section, and a biphenyl ester trimer. The first two sections are disordered, while the last forms a crystal. The mismatch between these behaviors creates mushroom-shaped clusters a few nanometers in size. In the current paper, the authors heat these clusters at 250 C for several hours. This cross-links the remaining double bonds in the polybutadiene section within a cluster, connecting the initial non-covalent cluster into a single molecule. The cross-linking “is accompanied by a 70% loss of unsaturated bonds” as they are converted into single bonds crosslinking the original units.
As expected, the covalently linked clusters are sturdier than the originals. The original non-covalent clusters melt into an isotropic phase at 330 C. The covalently linked clusters remain in a liquid crystal phase till they decompose at 430 C.
In order to cross-link their clusters, the authors “synthesized and studied more than 20 analogous molecules with varying chemical structure”, ultimately needing to terminate their biphenyl ester trimer section with a -CF3 group. Other triblock polymers were very prone to cross-links across cluster boundaries and gave “products that are either insoluble or have very broad polydispersity.”
The authors’ technique for crosslinking their structures into a covalent whole looks promising as a stiffening technique for self-assembled nanomachines. One clearly wants to isolate such cross-linking joints from other components that might extend the cross-links beyond the desired region. As it stands, the temperature needed could be tolerated by nanotube-based machinery, but not by biomolecule-based machinery. Perhaps lower temperature cross-linking mechanisms such as photochemistry may be feasible.
A. Harada and K. Kataoka, writing in [Science 283:65-67 1Jan99] and S. A. Jenekhe and X. L. Chen, writing in [Science 283:372-375 15Jan99; press release] have described the construction of self-assembled structures from block copolymers.
In Harada and Kataoka’s experiments, micelles were built with an ionic core and a neutral shell. These authors synthesized two types of block copolymers. A negatively charged polymer consisted of a negatively charged poly-aspartic acid block and a neutral poly(ethylene glycol) block. A positively charged polymer consisted of a positively charged poly-lysine block and a neutral poly(ethylene glycol) block. Mixing the two polymers yielded micelles with the core/shell structure Both sets of charged blocks formed the core of each micelle, while the neutral poly(ethylene glycol) blocks formed a shell around the core.
The authors demonstrated that this self-assembly process requires that the lengths of the charged sections of the two polymers match. They showed that a lysine18 block will selectively associate with an aspartic acid18 even in the presence of an aspartic acid78 block.
The aspartic acid18/lysine18 micelles were 31.5 nm in diameter, containing 40 chains of the two copolymers, with roughly a 5% variation in cluster size. The aspartic acid78/lysine78 micelles were 40.5 nm in diameter, containing 180 chains of the two copolymers, with roughly a 3% variation in cluster size.
In Jenekhe and Chen’s experiments, hollow micelles were built from a rod-coil block copolymer. Their copolymers consisted of a relatively stiff poly(phenylquinolene) block and a flexible polystyrene block. In CS2 the polystyrene block is soluble, forming a shell around a hollow core of the poly(phenylquinolene) blocks. The micelles have diameters of 3-5 microns. At a higher level of organization, the authors were able to crystallize 2-D and 3-D crystals of their micelles. They propose that these crystals may be useful as optical band gap structures, “of interest for applications in photonics, optoelectronics” and other areas.
These papers bring up a general question of how useful statistically controlled materials are to nanotechnology. My view is that these techniques are rather like “top-down” techniques. They do produce structures with nanometer scale dimensions, and the structures are useful in applications where fluctuations of a few bond lengths are acceptable. To really advance MNT, we need structures that can control positions to roughly a bond length. We need this in order to control synthetic reactions with the precision of enzymes, attaching reactants together at selected sites, even though chemically equivalent ones are present one or two bond lengths away. Structures built from polymers with variable numbers of monomers in their blocks appear to me to be too imprecise for this task. Perhaps the authors’ copolymers can be fractionated to the point where individual oligomers are separated, or perhaps analogous oligomers can be built by deterministic Merrifield-like synthesis. In any event, both sets of authors have demonstrated stable self-assembly motifs which can plausibly be exploited by atomically precise structures which are locally similar to these authors’ structures.
Jeffrey Soreff is an IMM research associate.