Giant Raman scattering from J-aggregated dyes inside carbon nanotubes for multispectral imaging

Raman spectroscopy uses visible light to acquire vibrational fingerprints of molecules, thus making it a powerful tool for chemical analysis in a wide range of media. However, its potential for optical imaging at high resolution is severely limited by the fact that the Raman effect is weak. Here, we report the discovery of a giant Raman scattering effect from encapsulated and aggregated dye molecules inside single-walled carbon nanotubes. Measurements performed on rod-like dyes such as α-sexithiophene and β-carotene, assembled inside single-walled carbon nanotubes as highly polarizable J-aggregates, indicate a resonant Raman cross-section of (3 ± 2) × 10−21 cm2 sr−1, which is well above the cross-section required for detecting individual aggregates at the highest optical resolution. Free from fluorescence background and photobleaching, this giant Raman effect allows the realization of a library of functionalized nanoprobe labels for Raman imaging with robust detection using multispectral analysis. The use of Raman spectroscopy for high-resolution optical imaging is severely limited by the inherent weakness of the Raman effect. Now, a giant resonant Raman effect is demonstrated from J-aggregated dye molecules encapsulated in single-walled carbon nanotubes, and it is used to realize multispectral Raman imaging.

A lthough Raman spectra display narrow bands containing powerful information for chemical analysis and detection, its use for high-resolution optical imaging remains marginal due to its inherent lack of sensitivity. The Raman cross-section, which relates the scattered intensity of a molecule to the incident power density of the light, is typically 1 × 10 228 to 1 × 10 230 cm 2 per molecule, which is too low for imaging and detecting individual objects at high resolution. This disadvantage is further compounded by the strong fluorescence background and photodegradation of chromophores that can accompany Raman spectroscopy.
Two main enhancement strategies-resonant Raman spectroscopy (RRS) and surface-enhanced Raman spectroscopy (SERS)-have been pursued in the past to solve at least the sensitivity issue and to derive Raman nanoprobe labels (also called tags) for imaging applications 1,2 . However, each has shown significant limitations. With RRS, optical resonances in dye molecules such as Rhodamine 6G, which are conveniently located in the visible spectrum, can boost the cross-section to 1 × 10 224 cm 2 per molecule. However, making Raman dye labels is barely practical due to stability issues, especially as a consequence of photobleaching and the presence of a strong fluorescence background 2,3 . Approaches based on SERS can further increase the cross-section by coupling with plasmonic resonances in metallic nanostructures, leading to an immense but local enhancement of the electromagnetic field 1,2,4,5 . The success of this approach is demonstrated by the fact that the SERS crosssection of resonant dyes can reach up to 1 × 10 218 cm 2 per molecule, a value that is close to the cross-section of fluorescence. The SERS effect has been successfully implemented in the design of Raman nanoprobes, but the specific requirements in terms of size, shape (enabling hot spots by interparticle geometry) and composition (typically Ag, Au) of the SERS antennas add stringent design constraints and stability issues. Furthermore, the occurrence of hot spots in SERS leads to inhomogeneous enhancements that greatly complicate quantitative analysis.
Here, we report the discovery of giant resonant Raman effects from J-aggregated dyes encapsulated inside single-walled carbon nanotubes (SWNTs). The scattering intensities of the dyes combined with the suppression of background fluorescence and photobleaching provide robust spatial and multispectral detection of a single assembly in the limits of optical resolution. We elucidate the origin of this Raman effect and demonstrate its use in the design of a new library of stable, yet chemically tunable, Raman nanoprobes for experiments in Raman imaging and labelling. Multispectral Raman imaging demonstrations of dyes encapsulated in SWNT (dyes-SWNT) nanoprobes deposited on surfaces, attached onto yeast cells and functionalized with biotin to detect surface patterns of streptavidin are also presented.
For our study, well-known organic dyes such as a-sexithiophene (6T) and b-carotene (bcar) were encapsulated inside SWNTs with diameters of 1.3+0.2 nm (ref. 6). As schematized in Fig. 1, a three-step procedure was followed to fabricate the dyes@SWNT in solution: (1) purification (oxidation) and cutting of SWNTs; (2) encapsulation of the dyes inside the SWNT (giving 6T@SWNT); and (3) covalent functionalization (labelled 6T@f-SWNT). Only a detailed Raman study on 6T encapsulated in SWNTs (6T@SWNTs) is presented below, but the same conclusions can be drawn for bcar@SWNTs and other dyes tested so far. The detailed characterization of the dye@SWNT systems prepared by our method is consistent with results from other groups on 6T@SWNTs 7-11 and bcar@SWNTs 12,13 in bulk form and clearly indicates that the dyes have been encapsulated (see Supplementary Figs 1, 3 and 5 for details.) The presence of optical resonances was established by the acquisition of absorption spectra for bulk SWNTs, 6T@SWNTs and 6T@f-SWNTs dispersed in mineral oil (Fig. 2a). Absorption of the 6T dyes in the 6T@SWNT spectrum is clearly seen as an intense and structured band in the wavelength range between 350 and 700 nm. Additional weak SWNT absorption bands are observed, overlapping in part with the 6T band.
To avoid averaging effects and other optical phenomena such as energy transfer between adjacent 6T@SWNTs, Raman measurements at l ¼ 532 nm wavelength excitation were taken on isolated SWNTs deposited on a 100-nm-thick oxide grown on a silicon substrate (SiO 2 /Si). An isolated (in appearance) SWNT with a diameter of 1.4 nm, was located using atomic force microscopy (AFM) and probed using micro-Raman spectrometry ( Supplementary Figs 2 and 3). Raman spectra were then acquired using the same polarization and position conditions (located using markers patterned by optical lithography) after each step of the preparation, as in Fig. 1. Before encapsulation (top spectrum, Fig. 2b), a strong Raman peak at 1,590 cm 21 is observed and can be ascribed to the vibrational tangential modes of the SWNT. This tangential mode band, often called the G-band, has two sharp peaks, which are unambiguously ascribed to the circumferential TO (tangential optical) and axial LO (longitudinal optical) modes of a semiconducting nanotube 14 . After encapsulation (middle spectrum, Fig. 2b), another Raman feature appears at 1,450 cm 21 . This feature is unambiguously ascribed to the C= =C stretching mode vibrating along the main axis of the 6T molecules 15 .
Additional polarization experiments at 532 nm were carried out on the same individual 6T@SWNT (Fig. 3c), where the polarization angle V is defined by the polarization vector of the incident light and the orientation of the nanotube axis. Superposition of a cos 2 V dependency over the polarization-dependent intensity of the 6T mode at 1,450 cm 21 and the SWNT mode at 1,590 cm 21 demonstrates that the transition dipole moments of the molecules are aligned. Thus, a 6T aggregate is clearly formed with polarizability tensors a ij perfectly oriented along the nanotube axis 7 . This result is consistent with the structure of 6T@SWNT illustrated in Fig. 1, which was previously deduced from transmission electron microscope images and calculations 8,9 : the 6T molecules are stacked in pairs and aligned head-to-tail, thus forming J-type aggregates inside the nanotube.
The detection of a Raman signal from the encapsulated 6T molecules in an isolated SWNT is unexpected, because only a few hundred molecules (545 at maximum filling) are present within the 600 nm focus spot of our Raman instrument. Indeed, surfactant molecules and even large polymers ( Supplementary Fig. 14) attached to the nanotube sidewalls are generally not detected in Raman microspectrometry. For further comparison, 6T aggregates (diameter, 40 nm) located on SiO 2 /Si but far away from the SWNTs were probed using Raman spectroscopy and AFM. As expected, submicrometre aggregates and even a 50-nm-thick film of 6T gave much weaker signals and strong fluorescence backgrounds (Supplementary Figs 6,7 and Supplementary Table 1). We therefore conclude that the Raman signal from the 6T@SWNT is strongly enhanced and arises from an encapsulated 6T J-aggregate, and not from other 6T aggregates on the oxide surface. Significantly, only minimal fluorescence at l ¼ 532 nm was detected. Both results are consistent with previous work on bulk 6T@SWNTs in solution, although some fluorescence was detected but at shorter excitation wavelengths 8,9 . Other studies on different dyes inside SWNTs reported efficient fluorescence quenching by the SWNTs 12,13,16 , and this effect, seen in weakly bonded chromophore-SWNT complexes, was explored recently in the context of light harvesting applications 17,18 .
A rough estimate of the Raman cross-section of encapsulated 6T aggregates can be inferred using the SWNT signal as an internal  standard. Using laser excitation at l ¼ 532 nm (an RRS condition for both 6T and SWNTs) and the highest spatial resolution ( 600 nm) of our microscope, the Raman bands of the 6T at 1,450 cm 21 and of the SWNT at 1,590 cm 21 exhibited nearly the same integrated peak intensity (A 6T /A NT ¼ 0.7). Considering that the RRS cross-section of the G-band of an individual SWNT is 1 × 10 221 cm 2 sr 21 (refs 19-21), we estimated the crosssection of the 6T J-aggregate to be roughly the same. (Note that a cross-section between 1 × 10 222 and 4 × 10 222 cm 2 sr 21 is reported for the radial breathing mode (RBM), but the G-band is about ten times more intense.) To confirm this estimate, a separate experiment based on the substitutional method described in refs 19 and 22 (for details see Supplementary page 21) was devised using spectra from a large ensemble of individualized 6T@SWNT and graphene as an internal standard deposited in the same area of the SiO 2 /Si substrate. Using the literature values for the Raman efficiency of graphene ( 200 × 10 25 m 21 sr 21 ) 22 , we obtained a cross-section of (3+2) × 10 221 cm 2 sr 21 per 6T@SWNT with an average length of 0.6+0.1 mm. This value of the cross-section assumes that the signal ratio is A 6T /A NT ¼ 0.7 when the filling of 6T inside nanotube is at a maximum. Complete filling with the 6T J-aggregate is, however, unlikely, given the liquid-phase procedure used for the encapsulation in this study. This Raman cross-section value is thus only a lower bound (note that the cross-section of incoherent Raman scattering scales linearly with the number of molecules) and clearly demonstrates a giant Raman scattering effect from the 6T J-aggregate inside the SWNT. Using the RRS data mentioned above on amorphous aggregates and on thin films of 6T molecules on SiO 2 /Si, the normalized signal of the 6T@SWNT compared to other similarly close-packed 6T aggregates is 1 × 10 4 times more intense (Supplementary Table 1).
To understand the origin of the enhancement of the Raman cross-section, we investigated the influence of the SWNT resonances on the Raman scattering from the dyes and tested the hypothesis that strong optical resonances in the SWNT can produce an optical antenna effect similar to a SERS mechanism. That is, the scattering efficiency can be reduced or enhanced by the optical near-field at the interior of the SWNT. The strength of this effect was estimated by modelling the SWNT as a hollow cylinder with an infinitesimal wall characterized by an optical frequency conductance g (ref. 23). By applying Mie theory (see Supplementary page 14 for details; Supplementary Fig. 10), it can be shown that E int /E 0 , the ratio of the optical electric field in the interior of the SWNT to the illuminating plane wave, is given by where k ¼ 2p/l, a is the SWNT radius, Z is the impedance of free space, and the usual cylindrical Bessel function notation is used. With a tube of radius a ¼ 0.6 nm and a peak optical conductance g ¼ 8e 2 /h (that is, in resonance), the field at the interior of the SWNT at an incident wavelength l ¼ 633 nm is E int /E 0 ¼ 0.999, corresponding to a negligible screening of the optical near-field. This estimate shows that the small SWNT diameter, as compared to the optical wavelength, renders optical enhancement or screening ineffective in the visible spectrum. Our experiments support this calculation and rule out the antenna effect from the SWNT being the cause of the giant Raman effect from the dye J-aggregate. In our samples, two different families of nanotubes-semiconducting (s-SWNT) and metallic (m-SWNT) species-are present. The optical resonances can be excited with laser lines at l ¼ 633 nm for the m-SWNTs and at l ¼ 532 nm and 488 nm for the s-SWNTs. Using AFM, two individual SWNT specimens were selected and subsequently probed in situ using Raman microspectrometry before and after encapsulation. In Fig. 3, RRS spectra taken at l ¼ 633 nm, 532 nm and 488 nm for the two 6T@SWNTs confirmed that one nanotube is metallic (Fig. 3a) and the other is semiconducting (Fig. 3b). In other words, the semiconducting nanotubes present two sharp and distinct G-peaks, and the metallic nanotube features a broad and downshifted G-band, which is due to a Kohn anomaly in the nanotube bandstructure 14 . At l ¼ 532 nm and 488 nm, the individual 6T@m-SWNT spectrum showed an intense peak from the 6T molecules at 1,460 cm 21 and almost no signal from the m-SWNT. Consistent with the absorption spectra in Fig. 2a, the Raman signal of the m-SWNT is off resonance, whereas it is clearly in resonance for the 6T aggregates. At l ¼ 633 nm, the opposite situation is observed: the signal is weak for the 6T molecules and strong for the m-SWNT. The Raman scatterings of the m-SWNT and 6T are therefore uncorrelated and simply follow their respective resonance profiles. For the individual 6T@s-SWNT (Fig. 3b), the Raman signatures from both constituents are seen at l ¼ 488 nm and 532 nm, but not at l ¼ 633 nm, indicating that the 6T scattering intensity is only correlated to the resonance of the 6T aggregates. Whether the SWNT is metallic or semiconducting, in resonance or not, has no observable influence on the scattering of the encapsulated 6T. This finding indicates that the reported giant crosssection is a robust phenomenon that depends mainly on the optical properties of the confined dye aggregates. As discussed below, this is a key feature enabling a whole new set of applications with functionalized dyes@SWNT nanoprobes for multispectral Raman labelling and imaging. All our experiments point towards an enhancement mechanism that is driven by RRS and J-type aggregation induced by confinement. The Raman signal of the dyes was also found to be stable against photobleaching, even at high laser fluence, and free from fluorescence. By considering that 545 molecules of 6T is the maximum number that can fit inside a 600-nm-long SWNT (that is, the diameter of our laser spot) and assuming incoherent scattering, a RRS cross-section of 6 × 10 224 cm 2 sr 21 per 6T molecule can be deduced. Literature data for 6T were not found, but this value is similar to the RRS cross-section of other dyes such as Rhodamine 6G 3 and b-carotene 24,25 in solution and to the cross-section of Rhodamine 6G physisorbed on a graphene layer where fluorescence is also quenched 26 . However, the cooperative effect of many molecules aligned as a J-aggregate inside SWNT boosts the total cross-section to 1 × 10 221 cm 2 sr 21 per dyes@SWNT assembly in the laser focal point or, more conveniently, to a value of 1 × 10 224 to 1 × 10 223 cm 2 sr 21 for each nanometre of filled SWNT depending on the resonant dye. This remarkably large crosssection has important implications for Raman imaging because a shorter segment of the dye aggregate, at least shorter than the optical resolution limit of our instrument (,600 nm), can be readily detected using conventional Raman microspectrometry.
Another main requirement in the elaboration of a competitive nanoprobe for imaging is its covalent functionalization. We probed the influence of chemical functionalization of 6T@SWNTs using a radical reaction that covalently attaches bromophenyl moieties to the nanotube sidewall. This is an important step towards achieving the dispersion, chemical functionality or selectivity and biocompatibility of the 6T@SWNT nanoprobes. In Fig. 2a, the bulk 6T@f-SWNT products in mineral oil exhibit an absorption spectrum that is similar to that of 6T@SWNT, except for a significant weakening of the SWNT absorption bands. The Raman spectrum taken after functionalization of the individual semiconducting 6T@SWNT of Fig. 2b is shown at the bottom. After functionalization, the G-band of the SWNT at 1,590 cm 21 decreases in intensity, which is related to disorder induced into the carbon lattice by covalent bonding of the bromophenyl grafts 27 . The fluorescence quenching ability of the f-SWNT is preserved, as no fluorescence is observed in the spectra. More importantly, the Raman intensity of the 6T modes remains constant, which indicates that the SWNT sidewall acts as a barrier that protects the dyes 12 and their giant scattering cross-section. That is, the toughness of the Raman signal of the dyes to functionalization reactions, photobleaching (see Supplementary page 26) and even to harsh treatment in a Piranha solution, a strong oxidizing reagent composed of 1:3 H 2 O 2 and H 2 SO 4 , are further proof that the dyes are indeed encapsulated inside the SWNT. Experiments on a variety of rod-like dyes indicate that the giant cross-section observed in the 6T@f-SWNT system is common to a large family of chromophores. For instances, methylene violet B, 3,3 ′ -dimethylthiadicarbocyanine iodide (DTDCI) and oligothiophene derivatives such as 3,6-bis-[2,2 ′ ]bithiophenyl-5-yl-2,5-di-noc-tylpyrrolo [3,4-c]pyrrole-1,4-dione (DPP2) encapsulated inside SWNTs gave similar RRS responses but at a longer wavelength of l ¼ 633 nm (Fig. 4d). These general properties of dye@SWNT nanoassemblies thus make them well adapted for constructing Raman nanoprobes for imaging and tagging experiments. For the purpose of demonstrating the applicability of the Raman effect, we tested various nanoprobes in different imaging configurations. A submonolayer of 6T@SWNTs deposited on a SiO 2 /Si substrate was first patterned and then imaged using scanning electron microscopy (SEM; Fig. 4a) and micro-Raman mapping (Fig. 4b). The Raman acquisition time at 300 mW mm 22 laser power density was only 5 s per pixel. The Raman 6T signal correlates with the varying density of nanoprobes observed using SEM (Fig. 4c), indicating that the signal is probably proportional to the density of nanoprobes. In another experiment, shown in Fig. 5a, a mixture of isolated bcar@SWNT and 6T@SWNT nanoprobes was deposited on SiO 2 /Si and subsequently analysed by hyperspectral Raman imaging at l ¼ 532 nm excitation. Thanks to the sharp and specific vibrational signatures of each nanoprobe, the 6T@SWNTs (red) and bcar@SWNTs (green) fingerprints were readily and unambiguously detected and selectively localized on the surface without interference from each other. The presence of a third dye@SWNT in Fig. 5b, arising from phenazine molecules (Ph@SWNTs in blue, Supplementary Fig. 9), was also detected, further illustrating the powerful application of multispectral bands in identifying nanoprobes, even when they are overlapping at the same location.
It is important to mention here that SWNTs are already known as giant Raman scattering objects exhibiting RRS cross-sections between 1 × 10 222 and 1 × 10 221 cm 2 sr 21 depending on the bands [19][20][21] . They also produce no fluorescence in the visible spectrum, which is a key feature enabling Raman applications. In effect, Dai and co-workers have shown that functionalized SWNTs can be used as Raman labels for protein detection 28 and biological imaging of cancer cells 29 . As Raman nanoprobes, SWNTs are therefore interesting, but they provide a small number of Raman-active bands, which is clearly limiting in the context of imaging and multiplexing. Furthermore, SWNTs with different chiralities operate at different RRS energies, which implies that SWNT labels require fastidious sorting on the basis of their resonance energies. Adding chemical functions onto the SWNT sidewalls  by covalent reaction (to aid dispersion in liquids and biocompatibility) drastically reduces their Raman cross-section. In this context, the giant Raman cross-section reported here on J-aggregated dyes inside SWNTs is an important step forward that circumvents the hurdles of using the SWNT signal only for detection. Finally, multiplexing of the Raman signals as shown above is similar to that of Raman labels made with functionalized SWNTs with different contents of 13 C/ 12 C isotopes 28,29 . However, the number of combinations that can be derived with dye@f-SWNT is by far greater due to the large number of dyes and their narrow bandwidths in the Raman spectra.
Finally, we present preliminary results for the use of dyes@f-SWNT as contrast agents for living cells and for biomarker types of application. bcar@SWNTs were functionalized using poly(ethylene glycol) bis-(3-aminopropyl) (PEG) to allow their solubility in physiological saline solution and mixed with a culture of a yeast called Candida albicans (see Methods and Supplementary page 6 for details). Figure 5c (top) presents an optical image of aggregates of Candida albicans cells mounted on a glass slide after tagging with bcar@PEG-SWNTs and rinsing with saline solution. The Raman hyperspectral image taken at the same place (Fig. 5c, bottom) shows that the tagged yeasts scatter strongly at the characteristic Raman frequency of the bcar@PEG-SWNT nanoprobes. The uniform contrast across the cells indicates also that the bcar@PEG-SWNT nanoprobes have bonded to the cell membranes and some may have penetrated into the yeasts. Importantly, the tagged yeasts lived more than three weeks after their incubation with bcar@PEG-SWNTs, which is a good but preliminary indication of the low toxicity of the dyes@f-SWNTs. The dyes@f-SWNT can also be used as probes for protein recognition. As a proof of concept, streptavidin was patterned onto a surface using microcontact printing (see Supplementary page 7, Supplementary  Fig. 15), and its detection was carried out using the probe bcar@f-SWNT functionalized with PEG-biotin (bcar@PEG-Biot-SWNT). Using Raman hyperspectral imaging of the peak centred at 1,520 cm 21 , it was possible to determine whether the streptavidin was present either in the dots of the pattern, or surrounding the dots (Fig. 5d and inset).
In summary, the dye@f-SWNT nanoassembly gives access to a giant resonant Raman scattering effect that is powerful for multispectral Raman imaging experiments. Our study shows that the SWNT is in essence a capsule that not only quenches fluorescence, but acts also as a template for J-aggregation as well as a protection for the dye molecules from the environment. By adding chemical functions to the SWNT sidewalls and by expanding the variety of encapsulated dyes exhibiting such giant Raman scattering, outstanding opportunities can be created with the dyes@f-SWNT nanoassemblies for experiments requiring a large library of robust and nanometre-sized Raman tags.

Methods
The SWNTs used in this work were produced by laser ablation 6 and had a diameter distribution of 1.3+0.2 nm. The SWNTs were first purified in concentrated nitric acid, washed with water, dried and then dispersed in dimethylformamide (DMF). Encapsulation of the dyes was performed under reflux by mixing 10 mg of dye in 10 ml of an appropriate organic solvent with 10 mg of purified SWNTs. For the study of individual SWNTs, the SWNTs were dispersed in DMF and deposited onto SiO 2 /Si substrate coated with an aminopropyltriethoxysilane (APTES) layer. This substrate was annealed under vacuum at 500 8C for 1 h and then immersed in a solution of dyes (at 1 mg ml 21 ) in organic solvent. The solution was then heated under reflux for 24 h followed by rinsing in DMF. AFM and polarized micro-Raman spectroscopy were used to characterize these encapsulated SWNTs.
The sample for multispectral imaging was prepared by depositing Ph@SWNT, 6T@SWNT and bcar@SWNT on a SiO 2 /Si substrate coated with an APTES layer. The preparation of the Candida albicans tagged with bcar@PEG-SWNT and bcar@PEG-Biot-SWNT required the functionalization of bcar@SWNTs with poly(ethylene glycol) bis-(3-aminopropyl) 30 to obtain water-soluble nanoprobes. The solvated nanoprobes were then incubated with Candida albicans yeast broth 31 for 30 min, rinsed, and centrifuged before dispersing again in physiological saline.
The yeasts were finally mounted on a glass slide for imaging. Raman imaging was performed using a hyperspectral Raman imager RIMA from Photon etc. Inc. The streptavidin patterns were created by microcontact printing of albumin from bovine serum using a polydimethylsiloxane stamp. Experimental procedures and materials are described in detail in the Supplementary pages 1-8.