Optimization of shell synthesis
The type of structures formed by silica, and the rate at which they form, is known to be sensitive to pH, catalyst, the presence of salts, temperature, the solution composition, and the choice and concentration of precursors . Due to the sensitivity of the J-aggregates to solution conditions , the pH range that could be used was limited. The J-aggregates are oxidized under acid conditions, resulting in a loss of absorbance and fluorescence. This makes acidic conditions unattractive. However, the J-aggregates are stable under basic conditions up to about pH 10 .
When added at neutral pH, the positively charged amine groups on APTES associate with the sulfonate groups of C8S3 and can start to hydrolyze within the few minutes before TEOS is added. The hydrolyzed APTES can then cause condensation of TEOS directly on the aggregate surface without gel formation. However, APTES is also highly reactive with water. It is necessary to avoid even atmospheric moisture before addition to the J-aggregate solution to avoid hydrolysis of APTES and the formation of dimer and trimer oligomers before adding to the aggregate solution . If oligomers of APTES do form, this can cause rapid flocculation and gelation of the aggregate solution, similar to the addition of a bivalent salt. Silanization around pH 9, with ammonia used as the base, is known to generally result in isolated particles instead of networks [21, 26]. Silica surfaces formed under basic conditions tend to have a negative surface charge, which stabilizes them as colloids .
In order to find the optimum pH range for the preparation of a homogeneous silica shell without gelation or precipitation, we varied the pH in a range acceptable for the aggregates. Samples were prepared by adding APTES to the aggregate solution with near neutral pH, then adding TEOS and finally ammonia to adjust the pH. This was carried out for pH 7, 8, 9, and 10. Sample preparation with a final pH of 9 was found to be the most reliable for avoiding gelation or flocculation. Samples prepared under all four pH conditions were imaged in standard TEM, and typical images are shown in Fig. 2. Only short tubule segments were found for samples prepared at pH 7 and 10, and these short segments appear to be connected by silica forming network structures. Samples prepared at pH 8 still had longer intact tubular structures, and the structures tended to be more isolated. Samples prepared at pH 9 typically had isolated, long, uniform looking tubules. If a sample prepared at pH 9 was stable after a few hours, then it would not gel or show serious changes in TEM images for weeks after preparation. Otherwise, the silica and aggregates would form a gel at the interface between the container wall and the solvent within a few hours.
The network formation at pH 10 is likely due to TEOS hydrolyzing too rapidly. If a silica precursor molecule, such as TEOS, undergoes hydrolysis multiple times before condensation, then it has multiple points available for condensation at the same time and can easily form networks. Networks could form directly in solution or starting from the molecules associated with the aggregates. The end result is a loosely networked structure with sections of aggregates embedded. The similar formation of a network of short aggregate bundles at pH 7 is probably due to TEOS hydrolyzing too slowly. Since unhydrolyzed TEOS is very poorly soluble in water, if the hydrolysis step is slow, then the TEOS phase separates from the aqueous solution. This would then leave the aggregates with a partial coating of APTES. Unhydrolyzed APTES molecules encourage bundling by partially masking the aggregate surface charge, and APTES can rapidly dimerize with molecules associated with other aggregates. Cross-linking between aggregates that are not stiffened by a full silica coating could result in short pieces being broken off if there is any solution agitation. As TEOS gradually hydrolyzes, it can react with the APTES on the surfaces of these bundles and cross-linked aggregate segments.
The samples prepared at pH 8 and 9 probably have long, isolated, coated aggregates because the rate of TEOS hydrolysis is neither too fast nor too slow. TEOS can hydrolyze rapidly enough to be available for reaction with APTES before it causes bundling. With an intermediate rate of hydrolysis, only one or two active sites may be available on a TEOS molecule at any time, leading to more linear growth instead of networks. Sufficiently rapid growth of the TEOS shell may stiffen the aggregates so that short regions do not break off as frequently. Apparently, the condensation of TEOS is slower than that of APTES, allowing hydrolyzed TEOS to keep the surface charge negative without resulting in cross-linking when two growing shells come near each other. The results for pH 8 and pH 9 were quite similar, but samples prepared at pH 9 generally had fewer short segments.
The protocol with pH adjusted to 9 that most consistently produced stable solutions with what appear to be long, intact, isolated J-aggregate tubules in the TEM images was used for all the further characterizations described in this paper, if not mentioned otherwise. The details of this protocol are given in the experimental section. The samples prepared according to the optimized protocol will be referred to as “coated” for concision. Dry samples were prepared by allowing 30 μL to dry in the dark overnight on a roughly 2 × 2 cm2 piece of freshly cleaved mica.
While it was clear from the TEM images that the morphology and general shape of the aggregates was not disturbed, absorption and emission spectra were taken, because the shape of the spectra is very sensitive to any changes of molecular packing and hence the internal structure of the aggregates [27,28,29]. The spectra quasi serve as a “fingerprint” to probe the integrity of the aggregate structure. The optical spectra of aggregate solutions (see Fig. 3, the spectra of bare aggregates) can be explained reasonably well by a structural model where the aggregate tube is formed by a double layer of dyes [27, 29], as it was confirmed also by molecular dynamics simulations . In each dye layer, also called tube wall, the dyes are assumed to be arranged in a herringbone structure that wraps helically up to form the cylindrical wall. The excitonic coupling gives rise to at least two Davidov split bands in each wall, one polarized along the aggregate axis, the other polarized perpendicular . The transitions with parallel polarization appear as the absorption peaks at 589 nm for the outer and at 599 nm for the inner wall. The perpendicular polarized transitions are among the other peaks and shoulders between 520 and 580 nm (see Fig. 3a) that carry much less oscillator strength and will not be considered further here. The emission spectrum (see Fig. 3b, bare aggregates) essentially shows emission from the two intense exciton peaks with almost zero stokes shift. The intensity at the peak at 589 nm however is much less than the intensity at 599 nm and appears merely as a shoulder which is the result of energetic relaxation due to excitonic coupling between these two states . It is noteworthy to mention that the observation of a clearly recognizable shoulder at 589 nm in emission is an unambiguous signature for structural integrity  and more easy to detect for small amount of aggregates than an absorption signal. It is therefore a significant observation that for solutions of silica coated aggregates, not only the absorption but also the emission spectra are very similar compared wih the bare aggregate solutions. Especially, the two strong absorption peaks at 589 nm and 598 nm are still present with almost unchanged spectral width and peak position. Only a very small red shift of less or approximately equal to 1 nm is observed for the peak intensity at 599 nm, see Fig. 3 a.
Spectra were also taken of samples allowed to dry on mica for at least 12 h, see dotted lines in Fig. 3. For the absorption measurements, a bare piece of mica with the same thickness was used as a reference. The absorption spectrum of the bare aggregate then has only a single broad peak centered at 591 nm. In contrast, the coated aggregate samples still had the two separate peaks centered at 588 nm and 599 nm in the absorbance spectra, though they were significantly broadened which is taken as an indication for disorder of the molecular packing .
The emission spectra are presented in Fig. 3 b for the solutions and the dried aggregates as well. For silica coated aggregates the two peaks at 589 and 599 nm are well recognizable in the dried state, albeit somewhat broadened. For bare dried aggregates, the broadening of the peaks is so large that the two distinct peaks cannot be resolved any more. This is in accordance with the broad absorption band in this case. The excitonic band structure is mostly destroyed, although excitonic coupling between the dyes is retained, which causes the red shift. In general, the spectra indicate that the arrangement of the dye molecules within the aggregates was not significantly disturbed, and that the coated aggregates retain their structure when dried.
It is remarkable that these aggregates can be imaged by standard transmission electron microscopy at room temperature. Typical images had already been shown in Fig. 2. In Fig. 4, the TEM image of bare aggregates (a) and silica-covered ones (b) are shown in a magnified view. The outer diameter of the coated aggregates was typically 22 ± 2 nm. The images demonstrate the possibility to investigate single, isolated aggregates outside the aqueous environment (as used for cryo-TEM) and in vacuum.
Protection against chemicals
The rough appearance of the silica structure on the TEM images (see Fig. 4) gives rise to the suspicion that the coating is porous. It is therefore important to check if the coating can protect the aggregates against changes in solution composition. For this, we make use of the fact that the aggregates lose optical activity when the pH is lowered below 4 due to protonation of the dye molecules . Absorption spectra were taken of solutions of bare and coated aggregates before and after adding aqueous HCl to lower the pH to 2. For this, 45 μL of 21 mM HCl was added to 100 μL of bare or coated aggregate solution. The resulting spectra are shown in Fig. 5. The bare aggregate solution immediately became colorless and, as expected , the solution shows no absorbance at this acidic pH. The coated aggregate solution did not have any visible color change and was allowed to sit for 10 min before a spectrum was taken. The absorption spectrum of the coated aggregates at pH 2 is corrected to account for the change in concentration after addition of HCl. It exhibits the typical absorption spectrum of the aggregates, but with slightly reduced intensity of the sharp exciton peaks at 589 nm and 599 nm, which is accompanied by a slight broadening of these bands. This implies that the silica shell can protect most of the dye molecules from protonation and preserves the aggregate structure in acidic conditions.
The preservation of the typical aggregate spectra upon drying (Fig. 3) implies that the coating protects the aggregates from structural deformations. But it is also apparent from the TEM images that the silica-covered structures are quite straight with almost no curvature. In other words, the persistence length and hence the stiffness of the initially flexible tubular aggregates appears to be increased. However, from the TEM images, one cannot conclude about the strength of the silica shell. It could be gel- or rubber-like, but it also could be brittle like glass. Therefore, attempts were made to break individual units, with and without silica shell for comparison. For this purpose, bare and coated aggregates were deposited on HOPG surfaces by spin coating at very low rotation frequencies (about 600 to 1200 rpm). SFM was used to first image the aggregates in Tapping Mode™ and then also to manipulate individual aggregates in contact mode.
Typical SFM images recorded in Tapping Mode are shown in Fig. 6. Aggregate structures can be found for both bare and coated aggregates after depositing on HOPG and are easily identified by the thread-like structures. While the structures can vary in width and height, there are many structures with identical width and height. This common width and height are also smaller than for any of the variably sized structures. Structures with the minimum width and height are identified to be single aggregates. The larger objects are bundles, which have been previously observed in solutions of this dye . Besides that, there are islands of flat objects with constant and regular height visible. Those are probably due to adsorption of monomers or remaining non-adsorbed silica. In case of the bare aggregates, large areas (dimensions of hundreds of nanometers) of a homogeneous background film were observed. It seems that in this case, either more monomers are remaining in the solution or aggregates also disintegrate upon adsorption at the HOPG surface—a possibility we do not investigate further here.
The measured height for bare aggregates is 8 nm, which is smaller than the 13 ± 0.5 nm diameter measured for bare aggregates in solution. Such a low height, less than the real tube diameter, was also observed in previous SFM investigations on aggregates dried on quartz substrates . It is reasonable to assume that the aggregates are structurally destroyed upon drying. It could be that the double layer tubes simply collapse with the walls lying flat on top of each other. The width of one double layer wall is about 4 nm, so a collapsed structure would be in the order of 8 nm. Such a structural destruction could explain the changes of the absorption and emission spectra as shown in Fig. 3, where only an overall red shift of spectra is seen but no features of the exciton band structure of the double wall tubular structure are recognizable. However, one also has to bear in mind that the height of the aggregates in Fig. 6 is only the relative height between the top of an aggregate and the surrounding substrate. Since the substrate is covered by a background film of molecular monomers, as mentioned above, it could also be that the aggregate is partially immersed into this layer, and the actual diameter of the aggregates would be larger than the measured height. Furthermore, the different interaction between the SFM tip and the aggregate or the substrate may cause height anomalies in this intermittent contact mode .
For the silica coated aggregates, the measured height is around 25 nm, in good agreement with the typical diameter found from TEM imaging. As seen from the height profile given in Fig. 6, also the larger object in Fig. 6 b has the same height. Apparently, this is a double stranded aggregate with both tubes lying parallel. The surface structure of the silica coated aggregates looks much rougher than in the case of the bare aggregates. This is in accordance with the TEM images.
Manipulation of single aggregates with the tip of an SFM is demonstrated in Fig. 7. Tapping Mode images of bare and coated aggregate samples before and after manipulation mode are shown. The manipulation consists of dragging the tip in contact mode with increased vertical force along the line marked with a dotted arrow. This manipulation of bare aggregates results in a gap in the aggregate structure and the appearance of a large deposit of material at the end of the manipulation path. This lump of material was collected by the tip during the manipulation and deposited there upon retraction of the tip. The remaining aggregate does not seem to have any significant plastic deformation or change in position. Additionally, the path of the tip can be clearly identified by deep (black) scratches crossing the aggregates. These scratches are removals of the surrounding layer of monomers or other loosely adhered material. The removed material contributes to the lump at the end of the manipulation path. Overall, one notices a decrease of the size of bare aggregates with time which becomes obvious from comparison of the images before and after manipulation. Based on our experience, we attribute this behavior to progressive drying out of the structures.
In contrast, the manipulation of silica coated aggregates shows a very different behavior. Dragging the tip across coated aggregates causes segments of the aggregates to break off. No bending or plastic deformation is seen, as reported for other semi-flexible materials such as DNA  or dendronized polymers . The aggregates are apparently brittle, and the segments are sufficiently rigid to be pushed around on the surface. This is further demonstrated in the Supporting Information. No loose material is removed from the surface along the tip path. This implies that the coated aggregates are truly laying on the HOPG surface rather than a bed of other dye molecules. The small islands of material that can be seen between the aggregates may be due to free dye molecules or silica precursors. These manipulation experiments provide strong evidence for enhanced mechanical stability of the coated aggregates. The mechanical rigidity combined with the fact that they can also span free-standing holes make these structures excellent candidates for further measurements of the mechanical elasticity by means of bending tests on individual aggregates similar to experiments performed by Kis et al. .
TEM characterization of freely suspended aggregates in vacuum
The enhanced mechanical stability of the coated aggregates provides new opportunities for studying these aggregates with minimal background signal from the surroundings. One interesting approach is using TEM grids with holey carbon films as a substrate. Aggregates suspended across the holes can then be studied in vacuum without any interference from the substrate. As can be seen in Fig. 8, the silica-covered aggregates are able to span holes. This was never observed for bare aggregates because they pull back to the edges of the holes due to the drying process, see Fig. 8 a and b. This behavior is further demonstrated using fluorescence microscopy in the Supporting Information.
As one application of the robustness against vacuum drying in a TEM, we characterized the aggregates’ chemical composition by energy dispersive X-ray spectroscopy and related STEM imaging. In Fig. 9, a HAADF-STEM image (Fig. 9a) is shown together with respective EDXS elemental maps for silicon (Fig. 9c) and oxygen (Fig. 9d). The full EDX spectrum is also shown (Fig. 9b). In the EDX spectrum, signals not only for Si and O are found, but also for sulfur (S), chlorine (Cl), and carbon (C). These peaks result from the dyes and clearly show the presence of the molecules within the silica tubes. Additionally, copper (Cu) signal is visible due to the copper TEM grids. A quantitative analysis of the signals is difficult, since the molar ratio of the silica to dye is unknown. However, the intensity ratio of silicon to oxygen can be compared with a signal obtained from a thick film of SiO2. The data are shown in the Supporting Information. If both spectra are normalized to the peak intensity of the Si signal, it is found that the silica-covered aggregates have a slightly higher oxygen signal. This higher amount can be explained by oxygen present either from the fact that there may be remaining silanes that have not reacted according to Scheme 2 or from the sulfonate groups of the dyes. Overall, the excess of oxygen compared with a solid SiO2 film is less than 20%, and this is just triple the intensity of the sulfur peak. Although intensities of different elements cannot be compared directly, it tells that the oxygen excess is in the order of magnitude of what is expected for the oxygen present in the dye.
The high stability of freely suspended silica-covered aggregates makes them promising candidates for spectroscopic investigations using electron beams in a scanning TEM. As an example, we demonstrate the measurement of energy electron loss spectra (STEM-EELS) on single aggregates spanning holes of the carbon film. EELS measures energy loss of inelastically scattered electrons due to excitations of electrons of the material that is investigated. In the case here, we looked for inner shell ionizations (K and L edge) of relevant elements, such as silicon, oxygen, and carbon. For future times, it would be desirable to extend this technique to low energy electrons for the detection of electronic transitions in the visible range of few electron volts .
The main results for inner shell silicon and oxygen excitation are summarized in Fig. 10. An aggregate spanning a hole of the carbon film was identified by HAADF imaging, and EELS spectra were recorded while scanning the beam across the aggregate. The path of the beam is indicated in Fig. 10 a by the dotted arrow. The EELS spectra were deconvoluted using multivariate analysis software to produce element-specific intensity scan profiles . Two of these intensity profiles, the one for the L2,3-edge of silicon and the other for the K-edge of oxygen, are plotted in Fig. 10 b. More information about the spectra and data collection is given in the Supporting Information. One can see that the profiles for silicon and the oxygen signals have the same extension and a similar shape. The intensities are not quantitatively comparable due to the different reaction cross sections. The shape can be explained assuming a simple cylindrical model similar to the one described in detail in the Supporting Information of Ref. . Under the assumption that the signal strength is proportional to the thickness of the structure transmitted by the electron beam (which is justified, since this is proportional to the reaction volume of the beam), the profile is obtained from simple thickness calculation along the electron beam path. Assuming the tubular aggregate is only covered by a homogeneous silica shell, the calculated EELS profile for the silicon signal looks like the profile that is indicated in Fig. 10b by the solid line. It has the typical peaks at the boundaries of the inner diameter and a circular trough in the center. The depth of this trough depends only on the thickness of the shell and is material-independent. For the fit in Fig. 10b, an inner diameter of 13 nm and an outer diameter of the shell of 25 nm was assumed. The intensity was adjusted manually for approximate fit of the data. If the inner space of the tube is filled with a homogeneous solid silicon cylinder of same density with diameter of 6 nm, then the dotted curve shown in Fig. 10b appears. The data points are between the empty and filled tube, which at least indicates that there is silica filling. This finding is in accordance with previous observations on accidentally found pieces visible from the front end of the tubes . Overall, as with EDX, the silicon and oxygen profiles correlate well to the original tube thickness. However, additionally one can conclude from the EELS scans that the aggregates are not only covered by a silica shell but also are somehow filled in the inner core with silica.