IMM Report Number 17

STM Fabrication

Scanning probes such as the scanning tunneling microscope (STM) today give us the most direct method for fabricating atomically precise structures, albeit one atom at a time. This paper describes an advance in one of these techniques.

Writing in [Nature 404:743-745 13Apr00], T.W. Fishlock, A. Oral, R.G. Egdell, and J.B.Pethica describe moving Br atoms on a Cu (001) surface at room temperature with an STM in UHV. Previous work from Gimzewski et al. showed atomically precise movement of adsorbates attached to a surface via Van der Waals forces at room temperature. Van der Waals bonds are fairly weak, while Cu-Br bonds are true chemical bonds, with a strength of 3.4 eV.

The authors found that when imaging the Br atoms with 3 nA, the atoms were pushed across the image, perpendicular to each individual scan line. In other words, the Br atoms were given a sideways push when an STM tip scanned by them, shoving them in the direction of the next scan line. The atoms moved along {110} directions between adjacent sites on the surface. This displacement was sensitive to the tunneling current, and did not occur at a lower imaging current of 1 nA.

The authors attempted to move atoms by moving the STM tip directly toward them, but found that at room temperature drifts in tip position could cause “either an undesired sideways jump or even a complete miss.” The authors solved this problem, but at a cost. As they brought their STM tip towards the atom to be moved, they dithered the tip position perpendicularly to the approach vector, which “locally replicates the effect of the image line scan on a single atom.” The dither movement makes atom movement less sensitive to STM tip drift at the cost of preventing atoms laterally closer than the dither distance from being independently moved. At the moment this distance is a few lattice spacings, but the authors hope to reduce it with a more stable STM.

The authors did several experiments to determine the mechanism of Br atom movement. They ruled out pure repulsion from the tip, and forces from the electric field in the STM. They attribute it to tunneling current induced heating of the Cu-Br bond vibrations, similar to earlier work on the atomically precise desorption of hydrogen from Si-H bonds on a silicon surface, which was also found to be induced by tunneling current.

This work advances our ability to form atomically precise structures by dragging adsorbed species across surfaces at room temperature. It allows us to construct structures built of stronger bonds than were previously accessible this way. This work also makes the use of tunneling current induced heating of adsorbate bonds look more generally useful than earlier work suggested. In the Si-H experiments, even a change from hydrogen to deuterium was enough to reduce the desorption rate by orders of magnitude, so it looked like possibly very few surface/adsorbate combinations would exhibit it. Seeing a second working example of atomically precise fabrication using this interaction in such a different surface/adsorbate combination promises much wider applicability for it.

Diagnostics: Vibrational Spectroscopy

The spectroscopy of molecular vibrations has been useful for many decades in determining molecular structures. In molecular scale systems it also lets us directly check mechanical properties such as stiffness or drag. The papers summarized below describe two advances in this field.

Writing in [C&EN 78:41-50 7Feb00], S. Borman presents an overview of work in multidimensional vibrational spectroscopy from several research groups.

Traditional vibrational spectroscopy looks at absorption of infrared light in a sample as a function of one frequency. For small molecules, it is possible to interpret the peaks in the absorption as being due to specific vibrational modes in the molecule. This has been used to determine structural features in organic molecules for decades. For large molecules, such as proteins, however, far too many absorptions at similar frequencies overlap, preventing us from extracting useful information from these spectra.

Multidimensional vibrational spectroscopy illuminates the sample with several sources of infrared light at several frequencies. The interaction is not just a simple sum of the interaction of the individual sources with the sample because one or more of the sources used is intense enough to significantly perturb the sample. For example, one IR source may start vibration of all of the C=O bonds in a sample, then another source may probe the absorption of C-N bonds while a substantial fraction of the C=O bonds in the sample are still vibrating. The final C-N absorbtion is a function of both the frequency at which the C=O bonds were illuminated and the frequency at which the C-N bonds were illuminated, making this a two-dimensional measurement.

In general, multidimensional spectroscopy takes advantage of these interaction to break up overlapping, uninterpretable bands in one dimension into discrete peaks in several dimensions. A wide variety of these techniques have been used in NMR spectroscopy for decades. Borman quotes R.M. Hochstrasser as saying: “I don’t believe that two-dimensional IR will ever be a serious competitor to two-dimensional NMR for the determination of large structures. However, to look at structural changes…on a faster than NMR timescale [should be very useful].” The timescale seen by NMR is typically milliseconds, while the time scale seen by these vibrational techniques can extend down to picoseconds. There are a variety of nonlinear vibrational spectroscopies, some where the experimental variables are laser frequencies, others in which the variables are delays between laser pulses, and others where both types of variables are used. In principle they all extract the same type of information, but in practice different techniques are preferable in different situations.

From the viewpoint of nanotechnology, refinements in vibrational spectroscopy are useful not just for structural analysis but in probing the mechanical properties of subsystems. These techniques probe the coupling between vibration modes. In a machine, the coupling between the intended motion of the machine and thermal vibration is friction that we need to observe and minimize.

Writing in [Nature 403:405-407 27Jan00], S. Komiyama et al. demonstrate the ability to detect individual photons in the far infrared (175-210 µm).

The authors’ detector is a special type of single-electron transistor (SET). The central island of their SET is a 700 nm by 700 nm quantum dot in a GaAs/AlGaAs heterostructure. This structure is placed in a strong magnetic field (3.4-4.0 Tesla), which splits the energy levels in the quantum dot into two Landau levels. The Landau levels are physically separated, with the upper level lying towards the center of the quantum dot and the lower level in a circle around it. The lower, outer, level is weakly connected to source and drain electrodes and current transport through this SET occurs through it. The higher, inner, level acts like a SET gate, affecting the current through the outer level electrostatically.

Excitation of an electron from the lower level to the upper results in its transfer to the inner portion of the dot. Decay back to the lower level is limited by tunneling across hundreds of nm and is therefore slow, taking from a millisecond to 1000 seconds (increasing rapidly with magnetic field). When this detector absorbs a photon, the excited electron sits in the inner dot till it decays, and it affects the current through the SET during this whole period.

Due to the SET amplification “one absorbed photon leads to a current of 10⁶-10¹² electrons through the quantum dot.” Due to the low energies involved, this device is operated at a very low temperature, 0.05 K, though the authors “have studied the influence of elevated lattice temperatures up to 0.4 K in additional experiments, and confirmed that no conductance switches [the signature for an excitation] as observed here are induced.” So thermal noise is tolerable to at least that point.

The authors’ detector is also frequency selective, since only photons at the resonance for a transition to the upper Landau level are absorbed. This frequency is tunable by varying the magnetic field. This detector might be used in nanotechnology for taking diagnostic spectra of individually constructed molecular machines or other very scarce molecular parts in this spectral region.

Carbon Nanotubes

Carbon nanotubes are very promising components for nanoscale systems. Their stiffness and strength makes them desirable mechanical components. The electrical conductivity of nanotubes of the proper chirality makes them desirable wires for molecular electronics. The papers summarized in this section advance our understanding and control of these materials.

Writing in [Science 287:637-640 28Jan00], M.-F. Yu et al. report the first measurements of the tensile strength of isolated nanotubes.

In their experiments, multi-walled carbon nanotubes (MWCNT) were attached to two AFM cantilever probe tips and stressed to failure by moving the probes apart.

This loading was done within a scanning electron microscope (SEM). The SEM images let the authors measure the diameters of the MWCNTs, the stress applied to them, and the lengths and shapes of the fragments after the MWCNTs broke. The SEM also provided the means to attach the MWCNTs to the probe tips. In an imperfect vacuum, certain organic contaminants are dissociated by an SEM’s electron beam and deposit a hard carbonaceous deposit on the beam’s target. The authors produced a 100 nm square of this deposit at both probe attachment points. This was sufficiently strong that the MWCNT usually broke before either attachment.

The tensile strength of 19 MWCNTs were measured. The MWCNT’s typical diameters were from 13-33 nm. The distances between the AFM tip attachment points were ~5-10 µm. After the MWCNTs broke, the total length of the two fragments was much longer than the length of the original MWCNT. The authors attribute this to a “sword-in-sheath” failure, where the outer layer of the MWCNT broke, but the inner layer remained intact and pulled out of the outer layer. In one case the separation of the fragments was not completed by the rebound of the cantilevers after MWCNT failure, so the authors were able to pull the inner tube out the rest of the way by further movement of the probe tips. They found that the force required to slide the inner tube out of the MWCNT was below 10 nN. The forces at failure ranged from 390-1340 nN.

Since only the outer layer of the MWCNTs appeared to fail, the authors used an effective cross-section of just this outer shell in their computations of tensile strength and stiffness (Young’s modulus E). They found tensile strengths from ~11 GPa to ~63 GPa and stiffnesses from ~270 GPa to ~950 GPa. Strains at failure ranged from ~2% to ~12%. The authors plan to attempt similar experiments on single-walled nanotubes.

Two recent articles describe effects of adsorbed gases on the electronic properties of single-walled nanotubes (SWNT), [Science 287:622-625 28Jan00] by J. Kong et al. and [Science 287:1801-1804 10Mar00] by P.G. Collins et al.

The Kong paper reports the effects of NH₃ and NO₂ on SWNTs, while the Collins paper reports the effects of O₂. Both groups measured the resistances of SWNTs as a function of their environments. Both also measured current as a function of an additional imposed voltage, effectively obtaining carrier density spectra. Kong et al. did this by measuring current in an isolated SWNT as a function of a substrate bias voltage, while Collins et al. did so by measuring tunnel current into a SWNT as a function of bias voltage.

Both groups concluded that adsorbed electronegative gases, O₂ and NO₂, gave p-type doping of their SWNTs. Kong et al. saw conductivity reductions from NH₃ exposure, which “effectively shifts the valence band of the nanotube away from the Fermi level, resulting in hole depletion and reduced conductance.” Collins et al. found that SWNTs with narrow bandgaps, 175-500 meV, could be doped to the point of acting as metallic conductors by O₂, while tubes with wider bandgaps, 550-750 meV, had their bandgaps reduced but did not become metallic conductors.

Both groups saw partial or complete reversibility of these effects on removing the relevant gas. They both interpreted this as desorption of the gas from the SWNTs (possibly including gas adsorbed on the interior of open-ended SWNTs). Time constants for these processes ranged from seconds to hours depending on sample topology (isolated tubes vs. films or mats) and temperature.

The short term implications of this work seem to mostly call for care in the preparation of nanotubes for electrical measurements. Over the long term, this does not appear to bar electronic applications of nanotubes in electronics, since they can be packaged in inert environments if needed. The time constants for desorption of these surface molecules appear to be too short to allow them to serve as intentional dopants. More reactive dopants forming solid covalent bonds such as carbenes appear to remain preferable for long-term electronic modification.

Writing in [C&EN 78:39-42 28Feb00], R. Dagani summarized work presented at NASA-sponsored NanoSpace 2000. One group, S.R. Wilson, A.N. Kirschner, B.F. Erlanger, and B.C. Branden, have elicited antibodies to C₆₀ and shown that the antibody “binds very strongly to single-walled nanotubes [SWNTs].”

The group has “determined the gene sequence that codes for the anti-C₆₀ monoclonal antibody” and has also found the 3D structure of the binding region in the Fab fragment using X-ray diffraction studies. Wilson attributes the binding to both C₆₀ and SWNTs to recognition of highly curved hydrophobic surfaces. The group has done simulations which suggest that building antibodies which selectively bind to SWNTs of a specific diameter or chirality should be feasible, though experiments along these lines haven’t started yet.

If this work is successful it could significantly boost the utility of SWNTs in nanotechnology, both because it would enable us to separate samples of SWNTs with a precise atomic pattern; and it would allow us to incorporate these SWNTs into larger structures through atomically precise binding sites.

Proteins

Proteins and DNA currently provide us with our best capabilities for building large numbers of fairly large atomically precise 3D structures. These papers below report advances in our knowledge of how proteins fold into their 3D structures and bind to substrates and each other to perform chemical reactions or construct yet larger structures.

Writing in [Nature 405:39-42 4May00], D. Baker presents an overview of recent work on the dynamics of protein folding. It has been found that, although the final 3D structure of a protein can be quite sensitive to fine details of its amino acid sequence (“removal of several buried carbon atoms can completely destabilize a protein”) the speed at which a protein folds is dependent on much coarser properties.

Folding speed appears to be largely determined by contact order: “the average separation along the sequence of residues in physical contact in a folded protein, divided by the length of the protein.” A high contact order means that residues from far apart on the original linear chain must find each other in the folded structure. Folding speed correlates well with contact order, dropping exponentially as it rises. This relationship holds across a wide variety of protein shapes and across a million-fold ratio in folding speed.

Baker also cites evidence that wide changes in amino acid sequence which do not affect the topology of residue contacts (and therefore do not affect contact order) have only weak effects on folding rate, “usually less than a tenfold effect”. This is true both of structurally related natural proteins and of mutated proteins. Sequence changes that don’t disrupt the folded structure don’t radically change the folding rate.

These findings remove a concern from protein design that finding a viable design requires solving two problems: first, ensuring that the desired folded structure was the energetically favored structure, and second, ensuring that there was a reasonably fast folding pathway to that structure. The dependence of folding speed on contact order lets us ensure reasonable speed very early in the design process. Contact order can be evaluated from the general shape of the protein’s backbone, before a single amino acid is chosen.

Writing in [J.Am.Chem.Soc. 122:1241-1242 16Feb00], A. Medek, P.J. Hajduk, J. Mack, and S.W. Fesik describe a new NMR technique for determining how a protein binds a ligand. M. Rouhi also commented on this work in [C&EN 78:30-31 21Feb00].

The authors started with an NMR spectrum of a known protein, FKBR, with a known structure and known assignments of NMR lines to atoms in the protein. FKBR binds a small molecule (FK506, a 43-carbon polycyclic ring structure). An NMR spectrum of the FKBR-FK506 complex shows shifts in more than 90% of the signals when compared to uncomplexed FKBR. These shifts are too numerous to permit interpretation, short of fully solving for the structure of the complex. Part of the problem is that there are significant global changes in the protein’s conformation on binding its ligand, and this produces spectral shifts even in atoms well removed from the binding site.

The authors prepared a series of ligands with small chemical differences (e.g. removing a single oxygen atom from one hydroxyl group) and looked at the differences between the NMR spectra of the related complexes. These spectral differences were much smaller than the differences between the free protein and either complex.

The handful of spectral differences between the two related complexes could be identified with a small set of protein atoms which were now known to be adjacent to the binding site. In fact, they were known to be adjacent to the particular atom in the ligand that was modified, constraining the orientation of the ligand. Four modifications of FK506 were examined by the authors, allowing them to find the binding locations for two regions in this molecule and to show a third moiety was outside the binding site.

They also applied this technique to a 16mer peptide, using mutations to produce the modified peptides. The authors expect to use this technique “especially in cases where conventional methods for structure determination fail”, suggesting that “this approach could be used to characterize the quaternary structure of even very large (>100 kDa) molecular assemblies.”

There are three different ways that this technique could aid in the development of molecular nanotechnology:

1. Better tools for characterizing and constructing protein/protein interface can help us build up larger atomically precise structures purely from proteins.
2. Better knowledge of how ligands bind to proteins can help us build non-protein structures to tie proteins together, using, for instance, the knowledge of what part of a substrate is outside the protein binding site to tell us where to attach an exterior structure to the substrate.
3. Better knowledge of enzymes’ active sites can help us use them to form bonds and build structures that cannot be built with classical techniques.

Writing in [Science 287:1279-1283 18Feb00], W.L. DeLano, M.H. Ultsch, A.M. de Vos, and J.A. Wells describe the binding of immunoglobulin G to four natural proteins and to a 13-amino acid peptide that they constructed.

It had been known from earlier work that immunoglobulin G bound to four proteins at the same “consensus binding site”, though the proteins had “radically different folds.” The authors wished to determine if the common binding site was determined by the physical characteristics of immunoglobulin G or whether it was due to constraints from its biological function.

They built a large (4 x 10⁹) library of random peptides and screened them for binding to an Fc fragment of immunoglobulin G. These peptides were not constrained to bind to any particular site on the fragment. Initial screenings led to two 20-mer peptides, which were further optimized, leading to a 13-amino acid peptide, Fc-III. This peptide bound remarkably tightly. The authors found that “the binding affinity of Fc-III to Fc was only twofold weaker than that of the domains from Protein A and protein G, which are each about four times larger and bind with K_d‘s of around 10 nM.”

The authors found that, despite the lack of biological constraints on their peptide, it bound to the same area as the natural proteins. Even the details of which bonds formed were remarkably close, with Fc-III, protein A, and protein G “all mak[ing] the same buried hydrogen bond with the backbone amide proton of Ile-253 on Fc.”

The authors attribute the convergence to a common binding patch to three factors:

1. The binding site is more exposed than random comparison patches, with more of its area accessible to solvent or binding partners than comparable areas.
2. The binding site is more hydrophobic than comparison patches. This makes its burial energetically favorable and also places “fewer specific geometric constraints on binding partners” from hydrogen bond formation.
3. The binding site is more “adaptive” than comparable sites. It “undergoes a series of conformational changes in order to complement the distinct surface of each binding domain.”

This work has implications for construction of atomically precise structures with this type of inter-protein docking.

1. The ability of one binding site to bind to multiple domains may reduce our ability to find unique interactions that can self-assemble structures without undesired binding.
2. On the other hand, the ability to reuse a binding domain while radically changing distant regions of a protein suggests that hierarchical design which treats binding regions as independent building blocks may work reasonably well.
3. The inability to bind to other surfaces of the Fc fragment may indicate that it is difficult to build connections on several surfaces of a protein. This is consistent with the hydrophobicity of the binding site, while stable folding of the protein requires a largely hydrophilic surface.
4. The “adaptability” of the binding site raises a possible red flag about its mechanical properties. Do multiple stable conformations imply an unacceptably soft interface? Unfortunately, this work did not estimate the spring constant for the binding site.

Writing in [Nature 403:617-622 10Feb00], M.M. Altamirano, J.M. Blackburn, C. Aguayo, and A.R. Fersht describe using targeted evolution to convert indole-3-glycerol-phosphate synthase (IGPS) into a phosphoribosylanthranilate isomerase (PRAI).

In the wild, PRAI produces the substrate that IGPS acts on, so they both bind this intermediate compound. Both steps are needed in the biosynthesis of tryptophan. Both enzymes share an alpha/beta-barrel fold.

The authors write that “The active site of most alpha/beta-barrel enzymes is located at the bottom of a funnel-shaped pocket created by the loops connecting the carboxy-terminal end of the beta-strands with the amino-terminal end of the alpha-helices that form the barrel. Nature has evolved the alpha/beta barrel to have the ‘substrate-binding’ residues predominantly within the barrel itself, and the ‘catalytic’ residues predominantly within the connecting loop regions.”

Since the barrel portion of IGPS bound the right substrate, the authors mutated the connecting loop region in their search for a new enzyme with PRAI activity. Roughly 3 x 10⁴ mutants (including varied loop lengths) were assayed for PRAI activity by insertion into a PRAI-deficient E. coli strain and the strain grown on a low tryptophan medium. About 500 colonies were selected at this phase. The genes from the colonies were shuffled and they were again selected for ability to synthesize tryptophan. The PRAI gene was extracted from one of the more successful variants, the gene sequenced, the enzyme expressed and purified and assayed for PRAI activity. It proved to have six times the specificity constant of wild-type PRAI. The amino acid sequences in the barrel continued to largely match IGPS, not wild-type PRAI, so the directed evolution left the substrate binding barrel mostly unchanged, as expected.

From the perspective of nanotechnology enzymatic catalysis is perhaps the closest bulk analog to eutactic synthesis. The separation of function in the alpha/beta barrel enzymes is exactly analogous to the distinction between cutting heads and workpiece clamps in macroscopic machinery. Extending the range of enzymatically catalyzed reactions may provide routes to exotic molecular building blocks and may suggest techniques for building self-aligning catalytic tips in nanotechnology.

It would be nice if enzymes that occupy smaller solid angles about their substrates can be designed or located. If an enzyme is to help build a structure larger than itself, it would ideally clamp on to a roughly flat surface of the workpiece, perhaps with a ring-shaped binding site, catalyze a reaction at the center of the ring, then peel off.

Jeffrey Soreff is an IMM research associate.