Single Molecule Raman Scattering

BY KATRIN KNEIPP AND HARALD KNEIPP WELLMAN CENTER FOR PHOTOMEDICINE HARVARD-MIT DIVISION OF HEALTH SCIENCES AND TECHNOLOGY HARVARD UNIVERSITY, MEDICAL SCHOOL BOSTON, MASSACHUSETTS 02114 Single Molecule Raman Scattering INTRODUCTION Detecting and identifying a single molecule represents the ultimate sensitivity limit in chemical analysis. Tracking and counting single molecules, characterizing their chemical structures, and monitoring their structural changes offer far-reaching opportunities in basic and applied research. The first spectroscopic detection of a single molecule was achieved using laserinduced fluorescence. 1 3 However, the amount of molecular structural information that can be obtained from a fluorescence signal is limited, particularly under ambient conditions. Ideally, one would wish for a spectroscopic tool that detects a single molecule and simultaneously identifies its chemical structure. Therefore, vibrational spectroscopy, such as Raman spectroscopy, would be a preferred method for single molecule studies due to the wealth of chemical information it generates. However, Raman scattering is a very weak effect. Extremely small crosssections, typically 10 30 to 10 25 cm 2, with the larger values occurring only under favorable resonance Raman conditions, require a large number of molecules to achieve adequate conversion rates from excitation laser photons to Raman photons, thereby precluding the use of Raman spectroscopy as a method for single molecule detection. For example, the number of Stokes photons can be estimated as the product of the Raman cross-section and the excitation intensity: Assuming a Raman line with a scattering cross-section of 10 29 cm 2 and 100 mw excitation light focused to 1 m 2, a single molecule scatters 10 4 photons per second, which means that one must wait more than an hour for a Stokes photon from a single molecule. Such simple estimates made single molecule Raman spectroscopy science fiction. This situation is dramatically improved if surface-enhanced Raman scattering (SERS) is used. The exciting phenomenon of a strongly increased Raman signal from molecules attached to a rough silver surface was discovered in 1977. 4,5 Within the next few years, strongly enhanced Raman signals were verified for many different molecules that had been attached to various metallic nanostructures. 6 For an overview of SERS, see Refs. 7 15. Estimated enhancement factors of the Raman signal started with modest factors of 10 3 to 10 5 in the first reports on SERS. Later, many authors claimed enhancement factors of about 10 10 to 10 11 for dye molecules in surface-enhanced resonance Raman experiments (SERRS). 16 22 At that time, it was already stated that single molecule detection using SERS should be possible and that this technique would provide some advantages over fluorescence, 21 which at that point was becoming a single molecule detection tool. About 20 years after the discovery of SERS, new methods for determining effective cross-sections resulted in unexpectedly large cross-sections on the order of at least 10 16 cm 2 per molecule, corresponding to enhancement factors of about fourteen orders of magnitude as compared with normal nonresonant Raman scattering. 23 These enhancement factors exceeded all former estimates by at least two orders of magnitude. Interestingly, these effective cross-sections were inferred for near-infrared (NIR) excitation that was not in resonance with electronic transitions in the target molecule. The experiments demonstrated that effective nonresonant SERS cross-sections can reach the size of fluorescence cross-sections of good laser dyes or may even exceed these numbers. Effective cross-sections of such size enable the detection of Raman scattering signals of a single molecule using NIR excitation. 24 26 Single molecule Raman spectroscopy can also be achieved using surface- 322A Volume 60, Number 12, 2006

molecules, or we will probe molecules exhibiting absorption in the ultraviolet range. These nonresonant single molecule Raman experiments can be understood in terms of an extremely high SERS enhancement factor that overcomes a gap of fourteen orders of magnitude between a nonresonant Raman cross-section and cross-sections on the order of 10 16 cm 2 typically required for single molecule spectroscopy. In the following section, we discuss the physics behind SERS at such an extremely high enhancement level, which is mainly based on strongly enhanced and spatially highly confined local optical fields in the vicinity of gold and silver nanostructures. The third section describes single molecule SERS experiments in two different configurations; one deals with single molecule Raman spectroscopy in solution using gold and silver nanoclusters as SERS active substrates, and the second describes single molecule detection on a dry fractal silver surface. The next section covers single molecule SERS spectroscopy at the anti-stokes side of the excitation laser. Due to the extremely high effective SERS cross-section, spontaneous anti-stokes Raman scattering becomes effectively a two-photon process. The last section describes potential applications of single molecule Raman spectroscopy. We briefly compare SERS and fluorescence as single molecule tools and conclude with a short review regarding the perspectives and limitations of single molecule Raman scattering. FIG. 1. (a) Schematic of surface-enhanced Raman scattering. 11 (b) Typical silver (left) and gold (right) nanoaggregates in different sizes. enhanced resonance Raman scattering (SERRS). 27 29 This Focal Point article will address single molecule Raman scattering. We will focus on single molecule SERS performed without additional contributions of intrinsic molecular resonance Raman enhancement of the target molecules, i.e., we will use near-infrared excitation light, which is nonresonant to the electronic transitions of the target PHYSICS BEHIND SINGLE MOLECULE RAMAN SCATTERING: SURFACE-ENHANCED RAMAN SCATTERING AT EXTREMELY HIGH ENHANCEMENT LEVELS Figure 1a shows the schematic of a SERS experiment, where molecules are in the close vicinity of a metallic nanostructure. 11 In the schematic, this nanostructure is a small aggregate built from spherical silver or gold nanoparticles; for compari- APPLIED SPECTROSCOPY 323A

son, see the electron micrograph of colloidal silver and gold particles in Fig. 1b. Two effects might be operative in the SERS system shown in Fig. 1a. (1) Electromagnetic enhancement: The excitation takes place in the enhanced local optical fields of the metal nanostructures. Additionally, the Raman emission is enhanced due to the nanostructure working as a resonant nanoantenna. (2) Chemical or electronic enhancement, first layer effect: A molecule in contact with a metal (nanostructure) exhibits a new Raman process at a larger cross-section than the Raman cross-section of a free molecule. For more explanation of the origin of electromagnetic and chemical SERS enhancement, see Refs. 7 15. The formula shown in Fig. 1a estimates the scattering signal in a SERS experiment. The total Stokes SERS signal P SERS ( S ) is proportional to the Raman cross-section RS ads, the excitation laser intensity I( L ), and the number of molecules N involved in the SERS process. A( L ) and A( S ) express enhancement factors for the laser and for the Raman scattered field, respectively. RS ads describes an enhanced cross-section of the adsorbed molecule due to electronic interaction with the metal compared to the cross-section RS free in a normal Raman experiment. From Fig. 1a we can define an effective SERS cross-section as SERS S RS ads A( L ) 2 A( S ) 2 (1) with P SERS N SERS S n L (2) where n L is the number of photons per cm 2 and the second corresponding I( L ). The total SERS enhancement factor is determined by the ratio of the effective SERS cross-section and the normal Raman cross-section RS : RS ads 2 2 SERS RS L S free free G A( ) A( ) (3) With A( L ) A( S ), Eqs. 1 3 show that the effective SERS cross-section and the electromagnetic SERS enhancement factor depend on the field FIG. 2. Stokes and anti-stokes Raman transitions in a molecular level scheme and the rate equation for population and depopulation of the first excited vibrational state by Raman scattering. enhancement A( ) to the power of four. In general, G SERS can be experimentally inferred from a comparison between surface-enhanced Raman signals and normal Raman scattering or fluorescence, taking into account the different numbers of molecules contributing to surface-enhanced and normal effects. 11 However, SERS enhancement factors were underestimated in all these experiments for many years because of the incorrect assumption that all molecules in a SERS experiment contribute to the observed SERS signal. In order to avoid this assumption, a different approach was invented in which both surface-enhanced Stokes and anti- Stokes Raman scattering were used to extract information on the effective SERS cross-section. 23 The idea behind this experiment was that a very strong Raman process should measurably populate the first excited vibrational level in addition to the thermal population. Figure 2 illustrates Stokes and anti-stokes scattering in a molecular level scheme. Population and depopulation of the first excited vibrational state by Raman Stokes and anti-stokes transitions can be described by the rate equation shown in Fig. 2. 11,23 In anti- Stokes Raman scattering, photons are scattered by molecules in the first excited vibrational state. Therefore, vibrational population pumping to this state by the strong Raman process should result in an unexpectedly high anti-stokes signal. For extremely high SERS enhancement factors, this higher anti-stokes signal can indeed be observed and the effective Raman cross-section operative in the vibrational pumping process can be estimated from the deviation of the anti-stokes to Stokes signal ratio from the expected Boltzmann population. Anti-Stokes SERS under the condition of vibrational pumping is described in more detail in another section below. At this point, using this approach, we inferred effective SERS cross-sections on the order of at least 10 16 cm 2. In order to make the large cross-section consistent with the observed Stokes signal, we must draw the conclusion that the number of molecules involved in the scattering process must be extremely small. 23 It should be noted that the extremely high level of SERS enhancement has been obtained at near-infrared excitation energies that are not in resonance with electronic transitions in the target molecule. Under these conditions, effective cross-sections on the order of 10 16 cm 2 imply SERS enhancement factors on the order of 10 14. However, under resonant conditions, the requirements for the surface enhancement effect can be greatly lessened. For example, normal resonance Raman scattering (RRS) that brings RRS crosssections to a level of about 10 26 cm 2 lessens the requirement for SERS enhancement in single molecule experiments to 10 10. The resonance contribution in single molecule Raman spectroscopy is particularly pronounced for single wall carbon nanotubes (SWNTs), which exhibit an extremely strong resonance Raman effect based on resonances of the excitation and/or scattered photons with the van Hove singularities in the electronic energy levels of these one-dimensional structures. 30 The 324A Volume 60, Number 12, 2006

high intrinsic resonance Raman cross-sections of SWNTs allow the detection of strong SERS signals from a single SWNT, exploiting relatively low electromagnetic SERS enhancement factors of 10 3, as can be obtained in so-called tip-enhanced SERS. 31 Here we want to focus on single molecule SERS without the contribution of molecular resonance and we discuss how the extremely high SERS enhancement factors required for these experiments can be achieved. Vibrational pumping experiments give evidence for the existence of such high SERS enhancement factors, but they do not provide information about the origin of the high enhancement level. In principle, electromagnetic and chemical effects can account for this effect. Interestingly, to our best knowledge, extremely high SERS enhancement and single molecule sensitivity has never been observed for isolated metal nanoparticles. The fact that extremely high SERS enhancement factors are always associated with composites formed by gold or silver nanoparticles provides strong indication that single molecule Raman spectroscopy is primarily a phenomenon associated with extremely high local optical fields in the vicinity of specific silver or gold nanostructures. 24 Mainly two kinds of composites of metal nanoparticles can give rise to giant electromagnetic SERS enhancement along with a high confinement of the local field, so-called hot spots. Extremely high field enhancement can be obtained for dimers and small aggregates formed from silver and gold nanoparticles. 32,33 Strong field enhancement exists particularly at the intersection between two nanoparticles. The other class of SERS active structures that provides an extremely high local optical field are fractal types of nanostructures. 34 40 Electromagnetic effects in self-similar structures can result in SERS enhancement factors up to 10 12 and even higher, leaving a factor of 10 100 that should be ascribed to a chemical enhancement effect. Other studies mainly use the chemical mechanisms involving ballistic electrons in order to explain extremely large SERS enhancement factors in single molecule SERS. 29,41 A strong chemical contribution in single molecule Raman experiments might also be supported by a recent report claiming that small silver clusters consisting of a few silver atoms can result in strongly enhanced Stokes and anti-stokes Raman signals. 42 On the other hand, recent studies show that combining plasmon resonances and photonic resonances can give rise to electromagnetic enhancement factors sufficient for single molecule Raman detection without chemical enhancement. 43 In general, the extent of chemical/electronic enhancement to single molecule SERS remains a subject of discussion. The key role of the electromagnetic enhancement mechanism is also strongly supported by experimental results obtained for nonlinear Raman effects such as the scaling of the enhancement factors for surface-enhanced Raman (SERS) and surface-enhanced hyper Raman (SEHRS) scattering, as they can be predicted for electromagnetic field enhancement. 44 So far, aggregates formed by silver or gold nanoparticles provide the highest enhancement level, where nanoparticles of both metals show totally different SERS enhancement factors depending on whether they are used as isolated particles or as aggregates. A gap of ten orders of magnitude exists between SERS enhancement factors for isolated gold nanoparticles measured at 514 nm excitation and gold nanoclusters measured at NIR excitation. 37 For silver nanoclusters, the SERS enhancement factor measured at 830 nm excitation has been found to be seven orders of magnitude higher than enhancement factors for isolated silver colloidal particles, even when excited at their plasmon resonance at 407 nm. 26 To study the dependence of the enhancement factor on the size of the aggregates, Fig. 3 compares Stokes and anti-stokes signals measured from clusters from 1 m in size and from silver clusters 200 nm in size. 26 Comparing only the Stokes (or anti-stokes) signals in both experiments shows that the 1 m cluster exhibits a higher signal level compared to smaller clusters. This could lead to the erroneous conclusion that when the number of particles in the aggregate increases, the relative SERRS activity also increases. 45 However, as Fig. 3 demonstrates, the ratios between anti- Stokes and Stokes SERS signals are constant within the accuracy of our measurement. That means that the vibrational pumping rate, or, in other words, the SERS enhancement factor in the hot spots, must be the same for both cluster sizes. The fact that SERS enhancement factors are independent of the size and shape of the clusters has important consequences for single molecule spectroscopy as it results in almost the same scattering signal observed from a single molecule. This has useful consequences for the capabilities of SERS as a single molecule tool, such as the potential for counting of single molecules. An interesting question involves the spatial dimensions of the hot spots on silver or gold clusters or aggregates. Since SERS takes place in the local fields of metallic nanostructures, the probed volume is determined by the confinement of these local fields. The spectroscopic selection of single SWNTs within a bundle shows that the SERS signal must have been collected from dimensions smaller than 5 nm in order to select a single or very few SWNTs from the neighboring tubes in the bundle. 46,47 The strongly confined hot spots provide the opportunity to probe volumes that can be two orders of magnitude smaller than the diffraction limit. Figure 4 illustrates that SERS can spectroscopically select single nano-objects or molecules within a larger population. 14 Despite the density of rhodamine 6G molecules on the fractal silver surface, which results in 100 molecules in the spot size of the laser, when scanning over the surface, there are a few APPLIED SPECTROSCOPY 325A

FIG. 3. Stokes and anti-stokes SERS spectra of 10 8 M CV adsorbed on silver clusters of different sizes measured at 830 nm excitation. FIG. 4. (Top) Many molecule and (bottom) single molecule SERS spectra of rhodamine 6G measured at different places on a fractal silver surface with approximately 100 molecules in the 1 m laser spot. points where the collected spectra are associated with a single molecule compared to the inhomogeneously broadened spectra usually measured from this sample. Enhanced and strongly confined fields are always associated with high field gradients. These large field gradients may also influence the Raman spectrum, leading to relaxed selection rules or changed depolarization ratios for single molecule spectra that are always measured in such high field gradients. 48,49 Of particular interest is the prediction that high local optical fields may also induce and increase the effect of Raman optical activity. 50 This was confirmed in recent experiments showing surface-enhanced Raman optical activity (SEROA). 51 Interesting and of direct importance for single molecule Raman detection is the observation that high field gradients may direct molecules to the hot spot and hold them there 326A Volume 60, Number 12, 2006

for detection. 14 This ensures that most of the molecules that are attached to a nanostructure will find a spot and will be detected in a SERS experiment. SINGLE MOLECULE SURFACE-ENHANCED RAMAN SCATTERING EXPERIMENTS As mentioned above, we focus on single molecule experiments where no resonance Raman effect for the target molecule is required. In the previous discussion we have shown that for colloidal silver and gold clusters strong SERS enhancement factors exist in the NIR wavelength range. Fortunately, high-intensity NIR diode lasers are easily available, making this region also attractive for compact, low-cost Raman instrumentation. Further, the development of low noise, high quantum efficiency multichannel detectors (chargecoupled device (CCD) arrays), combined with high-throughput singlestage spectrographs used in combination with holographic laser rejection filters, has led to high sensitivity Raman spectrometers. In general, a state-of-the-art Raman system has components that allow the performance of single molecule SERS experiments. Single Molecule Surface-Enhanced Raman Scattering on Silver and Gold Nanoclusters in Solution. For single molecule detection in solution, small silver and gold colloidal clusters in sizes between about 100 and 500 nm are very useful SERS active substrates. At very low analyte concentrations of the target molecule (approximately 10 11 M and lower), when the concentration of target molecules becomes comparable to or smaller than the concentration of the colloidal clusters, no analyte-induced cluster cluster aggregation occurs, and the SERS spectra show a very good reproducibility. The smallest colloidal clusters we used in single molecule SERS experiments are 150 nm in size, formed by only 5 10 individual nanoparticles. These nanoclusters provide at least one hot spot suitable for single molecule measurements. 14 No cold clusters were obtained as reported for dimers and trimers. Figure 5 shows a schematic of a typical single molecule SERS experiment performed in silver or gold colloidal solution. 24 26 Spectra are excited by 830 nm laser light. A microscope attachment is used for laser excitation and collection of the Raman scattered light. It is important, particularly for ensuring the free Brownian motion of the nanoparticles into and out of the probed volume, to keep the NIR excitation laser at relatively low intensities ( 10 6 W cm 2 ) in order to avoid any trapping effect for the nanoclusters. The effect of trapping is discussed in detail in Ref. 52. Here we avoid the effect in order to perform a correct statistical analysis of the scattering signals. Analyte concentrations on the order of 10 12 to 10 14 M and probed volumes whose sizes are on the order of femtoliters to picoliters result in average numbers of one or fewer target molecules in the focus volume. The Brownian motion of single-analyte-molecule-loaded silver or gold clusters into and out of the probed volume results in strong statistical changes in the Raman signals measured from such a sample in time sequence. Figure 5b shows typical unprocessed SERS spectra measured from a sample with an average of 0.6 crystal violet molecules in the probed 30 pl volume. Figure 5c displays the peak heights of one characteristic Raman line of the target molecule measured time sequence. The signals appear at different power intervals, which can be assigned 0, 1, 2, and 3 -molecule events. The statistical distribution of the 0.6 molecule SERS signal exhibits four relative maxima that are fit by the superposition of four Gaussian curves. The gradation of the areas of the four statistical peaks are consistent with a Poisson distribution for an average number of 0.5 molecules. This reflects the probability of finding 0, 1, 2, or 3 molecules in the scattering volume during the actual measurement. Comparing the measured Poisson distribution with the 0.6 molecule concentration/ volume estimate shows that about 80% of molecules are detected by SERS. In the meantime, these SERS experiments in solutions using small silver colloidal aggregates as SERS active substrates have been repeated by several groups, and the Poisson distribution of single molecule SERS signals has been corroborated in several studies. 53,54 In general, single molecule experiments should not suffer from the problem that one does not know how many molecules contribute to the SERS signal. Due to the Poisson statistics the measured signal comes from 1, 2, or 3 molecules and therefore these experiments should allow one to infer SERS enhancement factors in the correct order of magnitude in a straight forward way by comparing SERS signals of single molecules with normal Raman signals of a known large number of molecules. Figure 6 shows Raman spectra measured from an aqueous solution of silver nanoaggregates containing 10 14 M pseudoicocyanine (PIC) and 5 M methanol, which yields one PIC molecule and 10 14 methanol molecules in the probed volume. The experiment shows that the SERS signals of a single PIC molecule appear at the same signal level as the nonenhanced Raman signal of the 10 14 methanol molecules, confirming SERS enhancement factors on the order of fourteen orders of magnitude, in agreement with the results obtained from anti-stokes to Stokes signal ratios (see also the next section). As we discussed regarding Fig. 5, and also Fig. 6, Raman lines assigned to PIC appear at varying signal levels due to Brownian motion of the silver colloidal cluster carrying single PIC molecules into and out of the probed volume. Data analysis of the PIC SERS signals results in a Poisson distribution, whereas the Raman signal of the 10 14 methanol molecules remains statistically constant within a Gauss distribution (data not shown here, see Ref. 25). In addition to changes in single molecule SERS signal strengths that APPLIED SPECTROSCOPY 327A

FIG. 5. Single molecule SERS spectroscopy. (A) Schematic of the experimental setup. (B) Typical SERS spectrum of a single molecule of crystal violet attached to a silver nanocluster; the 1170 cm 1 Raman line used for the statistical analysis is marked. (C) Peak heights of the 1174 cm 1 crystal violet Raman line for the 100 SERS spectra collected in time sequence, 1 second each. (D) Statistical analysis of the SERS signals shown in (C). follow the expected Poisson statistics, SERS spectra collected at the single molecule level in time sequence can also show fluctuations and changes in the spectrum, such as the appearance of new Raman lines. Often the new spectral features do not correlate with the spectral signal strengths of the target molecule. These spectral fluctuations, or blinking, can have different reasons depending on the experimental situation. We will come back to the temporal changes in SERS spectra in the next section. In SERS experiments performed on silver or gold nanoaggregates in solution as described above, the new Raman lines can be ascribed to signals of impurities on the surface of the colloidal particles, which are probably introduced during the chemical preparation process. The impurity spectra change due to different colloidal clusters loaded with different impurities, which move into the focal volume. In SERS experiments using larger probed volumes and extremely low concentrations (in order to achieve one or less target molecules in the probed volume), SERS signals of the target molecule can vanish in a background of impurities. This observation was discussed in the first studies of SERS spectroscopy at the single molecule level. 21,55 Figure 7 demonstrates this effect for the SERS spectrum of rhodamine 6G in silver colloidal solution. Single Molecule Surface-Enhanced Raman Scattering on Fixed Fractal Silver and Gold Cluster Structures. Particularly high local optical fields can also exist on extended silver and gold fractal structures such as larger aggregates of colloidal particles or evaporated metal island films. 35,56 Here we demonstrate the capability of fractal silver surfaces for single molecule detection using the small protein enkephalin as a target molecule. 57 Enkephalin is a mixture of two pentapeptides, (Leu)enkephalin, and (Met)enkephalin. It was brought to the silver surface in a concentration resulting in an average of one molecule in the 1 m laser spot. The molecule can be detected based on the strongly enhanced ring- 328A Volume 60, Number 12, 2006

FIG. 6. Raman spectra measured from a probed volume that contains an average of one pseudoicocyanine molecule attached to a silver colloidal cluster and 10 14 methanol molecules, 100 mw cw ex-citation at 830 nm, 1 second collection time per spectrum. 25 FIG. 7. SERS spectra of rhodamine 6G in silver colloidal solution in concentrations between 10 11 and 10 16 M along with SERS spectra of impurities. breathing mode of phenylalanine ( 1000 cm 1 ), which is a building block of both pentapeptides. Measuring only one typical SERS line of the target molecule and using this Raman line as an intrinsic spectroscopic signature for the specific molecule is a useful tool for detecting a known molecule without the use of labels. The top panel of Fig. 8 shows selected single molecule SERS spectra measured in a time interval in a spectral window, which displays the 1000 cm 1 SERS line of phenylalanine and a line at about 750 cm 1, which can be ascribed to impurities on the silver colloids. The bottom panel of Fig. 8 displays the Raman signal measured at 1000 cm 1 shift from the 830 nm excitation from the same 1 m spot in a time sequence. Over the first 95 seconds, no SERS signal was measured. Obviously, the laser spot was in an area where statistically no target molecule is present. Then, a SERS signal appears relatively abruptly, stays for about 20 seconds at the same level, and vanishes again. Such behavior, namely, the appearance of the 1000 cm 1 SERS signal within a 10 30 second time window, has been observed in many measurements on single enkephalin molecules. If the enkephalin signal appears, it always appears at the same level. Spectra B through E in the top panel of Fig. 8 were measured in the time window between 95 and 115 seconds, while spectrum A shows a situation when the 1000 cm 1 SERS signal does not exceed the noise level. A possible explanation for these changes in scattering power is that a single enkephalin molecule is diffusing on the colloidal silver cluster and can only be seen when it enters a hot spot. Entering and leaving a spot can be explained by the discontinuous illumination: the laser was illuminating the sample during the one-second collection time and was closed by a shutter for about one second between the measurements. Fluctuations in scattering power and/or sudden spectral shifts and changes that appear as blinking of APPLIED SPECTROSCOPY 329A

the SERS signals has been reported by several authors. 27,28,58 Different effects can account for these spectral changes, such as thermally and nonthermally activated diffusion of molecules, 59 61 as well as real transformations, such as protonation and deprotonation. 62,63 Although blinking had been claimed as a hallmark of single molecule detection, it has become evident that this behavior is not necessarily connected to single molecule SERS. The effect has also been observed in lower concentration many-molecule SERS spectra. 64 PUMPED ANTI-STOKES SURFACE-ENHANCED RAMAN SCATTERING: A TWO-PHOTON EXCITED RAMAN EFFECT As described before, SERS can result in unexpectedly high anti-stokes to Stokes signal ratios associated with the population of the excited vibrational state due to vibrational pumping. 23 It should also be noted that anomalies in anti-stokes to Stokes SERS signal ratios could be explained by molecular or charge transfer resonance effects, which can be selectively efficient for Stokes and anti-stokes scattering. 65,66 Similar asymmetry effects can occur for a situation when Stokes or anti- Stokes have different electromagnetic enhancement conditions due to different resonances with surface FIG. 8. (Top) Selected single molecule SERS spectra of enkephalin from one fixed spot on a fractal silver surface showing the 1000 cm 1 SERS line of phenylalanine and a line at 750 cm 1, which can be ascribed to impurities on the silver colloids. (Bottom) Time series of the 1000 cm 1 phenylalanine signal measured from one fixed spot on a sample with an average of one enkephalin molecule in the laser spot. Spectra were observed over a time interval of 120 seconds, 1-second collection time each. The signal level of 100 cps represents approximately the background level. 330A Volume 60, Number 12, 2006

plasmons resulting in different effective cross-sections for Stokes and anti-stokes SERS. 23 All these abnormal anti-stokes to Stokes signal ratios are independent of the excitation intensity. However, this situation is different for vibrational pumping. Simple equations for the anti- Stokes (Eq. 4) and Stokes signals (Eq. 5) can be derived from Fig. 2 assuming steady state and weak saturation (exp( h M /kt) SERS S 1 n L K 1). Under steady-state and weak saturation conditions a continuous wave (cw) laser-excited Raman process populates the first excited vibrational state comparable to or higher than the Boltzmann population, but still far away from approaching equilibrium population between N 0 and N 1 : SERS h P (N e M/kT N SERS as 0 0 1n L) SERS n L (4) PSERS N SERS n (5) S 0 L where 1 is the lifetime of the first excited vibrational state, T is the sample temperature, and h and k are the Planck and Boltzmann constants, respectively. The first term in Eq. 4 describes the anti-stokes signal due to thermal population of the first excited vibrational state. The second term occurs due to vibrational pumping by the strong Stokes process. In normal Raman scattering this term can be neglected compared to the thermal population since for normal Raman scattering the product RS ( m ) 1 ( m ) is on the order of 10 40 cm 2 s. That means that about 10 38 laser photons cm 2 s 1 (approximately 10 20 Wcm 2 excitation intensity) are required to make the Raman pumping comparable to the thermal population of the first vibrational level. In order to account for the experimental observations in anti-stokes and Stokes SERS, the product of cross-section and vibrational lifetime in Eq. 4 must be on the order of 10 27 cm 2 s. Assuming vibrational lifetimes on the order of 10 ps, the surface-enhanced Raman cross-section is then estimated to be at least 10 16 cm 2, corresponding to SERS FIG. 9. Anti-Stokes and Stokes signal versus excitation intensity measured from crystal violet attached to fixed silver nanoclusters. enhancement factors on the order of 10 14. 23 As the second term in Eq. 4 shows, anti-stokes Raman scattering starting from a pumped vibrational level is a two-photon Raman process: one photon populates the excited vibrational state, while a second photon generates the anti-stokes scattering. The anti-stokes signal P as depends quadratically on the excitation laser intensity, whereas the Stokes signal P S remains linearly dependent (Eq. 5). Figure 9 shows quadratic and linear fits to the experimental data of SERS anti-stokes and Stokes signal powers of crystal violet versus excitation intensity, displaying the predicted quadratic and linear dependence. In general, a nonlinear dependence of the anti-stokes signal on the excitation intensity could also be caused by an increase of temperature due to laser heating. 23,46 In order to exclude this effect and to provide evidence that the quadratic dependence is due to vibrational pumping, SERS studies have been performed on single wall carbon nanotubes (SWNTs) on silver aggregate clusters. 46,67 SWNTs show a strong dependence of their Raman frequencies on temperature. This allows for the monitoring of temperature changes during the SERS measurement. The observed very small shifts in frequencies along with a quadratic dependence of the anti-stokes signal show that temperature changes in the sample are too small to explain the observed nonlinear dependence on the laser excitation intensity. Of course, vibrational pumping also exists in cases of surface-enhanced resonance Raman scattering. Under resonance conditions, possible heating (and photodecomposition) of the target molecule can result in various regimes of interferences between heating and pumping. 68,69 Considering only the second term, we can write Eq. 4 as: PSERS N 0 SERSn2 as,nl as,nl L (6) using an effective two-photon crosssections SERS with as,nl SERS SERS SERS as,nl S as 1 (7) Assuming a SERS cross-section of approximately 10 16 cm 2 and a vi- APPLIED SPECTROSCOPY 331A

FIG. 10. Potential application of single molecule SERS for DNA sequencing (see text). brational lifetime on the order of 10 picoseconds, effective two-photon cross-sections can be inferred to be about 10 43 cm 4 s. This cross-section can be understood in terms of the real intermediate state in this twophoton process formed by the excited vibrational state. 70 APPLICATIONS OF SINGLE MOLECULE SURFACE-ENHANCED RAMAN SCATTERING The observation of Raman signals at the single molecule level is of basic scientific interest as it provides insight into the intrinsic properties of a molecule or a nanostructure as well as structural changes without ensemble averaging. But detection and differentiation of single molecules using their specific SERS signatures is also of practical interest. Applications range from the use of Raman spectroscopic characterization of specific DNA fragments down to structurally sensitive detection of single bases without the use of fluorescent or radioactive tags. One of the most spectacular potential applications of single molecule SERS might be in the field of rapid DNA sequencing at the single molecule level. The idea is described in Fig. 10. The nucleotide bases show welldistinguished surface-enhanced Raman spectra, also shown in Fig. 10. Thus, after cleaving single native nucleotides from the DNA or RNA strand into a medium containing colloidal silver or gold clusters, direct detection and identification of single native nucleotides should be possible using the unique SERS spectra of their bases. SERS active silver or gold nanoclusters can be provided in a flowing stream of colloidal solution or onto a moving surface with silver or gold cluster structures, enabling the detection of the bases in order when the nucleotide-loaded nanoclusters move through the laser beam. 26 Effective SERS cross-sections on the order of 10 16 cm 2 can be inferred for adenosine monophosphate (AMP) and for adenine on colloidal silver clusters. NIR SERS spectra for both compounds are identical, indicating that sugar and phosphate bonds do not interfere with the strong SERS effect of adenine. It is interesting to estimate the detection rate of single nucleotides in such an experiment. Single molecule adenine spectra can be measured at good signal-to-noise ratios of 10 in 1-second collection time at 3 10 5 Wcm 2 excitation. Assuming a SERS cross-section on the order of 10 17 to 10 16 cm 2 and a vibrational lifetime on the order of 10 picoseconds, saturation of SERS will 332A Volume 60, Number 12, 2006

be achieved at 10 8 to 10 9 Wcm 2 excitation intensity. Extrapolation to saturation conditions shows that single molecule SERS spectra over the complete fingerprint region (approximately 700 1700 cm 1 ) should be measurable in one millisecond or at a detection rate of one khz. PERSPECTIVES AND LIMITATIONS OF SINGLE MOLECULE RAMAN SPECTROSCOPY Numerous experiments conducted during the past decade have demonstrated that SERS enables the measurement of Raman spectra of single molecules, one at a time. When the target molecule is attached to a silver or gold nanostructure, the nonresonant Raman signal can be enhanced up to 14 orders of magnitude, and effective Raman cross-sections can reach and even exceed fluorescence cross-sections of dyes used as fluorescence labels. In principle, electromagnetic and/or chemical effects can account for this enhancement level. There is evidence that single molecule SERS is primarily a phenomenon associated with strongly enhanced and geometrically confined local optical fields. As a single molecule tool, SERS opens up exciting capabilities as compared with fluorescence, another widely used single molecule technique. Currently, because of the high structural information content of a Raman spectrum, SERS is the only way to detect a single molecule and simultaneously identify its chemical structure. Additionally, measuring only one SERS line, which is characteristic for the target molecule, and using this Raman line as an intrinsic marker for the specific molecule is of interest for detecting and tracking single molecules without the use of fluorescence or radioactive labels. Raman scattering is a very general property of a molecule, and almost every molecule has Raman active molecular vibrations that can be seen by the Raman effect at a cross-section of at least 10 30 cm 2, whereas far fewer molecules show fluorescence. For the detection of a molecule by SERS, it has to be attached to a SERS active substrate in order to increase its effective Raman cross-section by 12 to 14 orders of magnitude. Due to the primarily electromagnetic origin of this enhancement, it should be possible to achieve an equally strong SERS effect for each molecule, thus making SERS a single molecule tool for a broad range of molecules. Another interesting aspect of SERS for single molecule detection involves the total number of photons that can be emitted by a molecule. This is determined by the maximum number of excitation emission cycles a molecule survives. In fluorescence experiments, this number is limited by the rate of photobleaching of the molecule. As a process that is electronically nonresonant to the target molecule, SERS avoids photodecomposition. Also in SERRS experiments, photodecomposition of the molecules adsorbed on the metal is strongly reduced as compared to molecules in solution. 27 Another number that is of particular interest for the rapid detection and screening of single molecules is the maximum number of photons that can be emitted by a molecule per time interval. Under saturation conditions, this number is inversely proportional to the lifetime of the excited molecular states involved in the optical detection process. Due to the shorter vibrational relaxation times as compared to the electronic relaxation times, a molecule can go through more Raman cycles than fluorescence cycles per time interval. Therefore, the number of Raman photons per time interval that can be emitted by a molecule under saturation conditions can be higher than the number of fluorescence photons by a factor of 10 2 to 10 3. 21 Limitations of SERS spectroscopy are attributed to the fact that target molecules must be attached to silver or gold nanostructures. However, chemically inert nanometer gold particles might be the favorite choice in several applications compared to fluorescent dyes or radioactive labels, particularly for biomedical applications. Moreover, in some applications, this attachment of the target molecule to a bigger nanoparticle can be an advantage for single molecule detection, as increased diffusion times minimize the possibility of the target molecule escaping from the focal volume of the laser too rapidly. ACKNOWLEDGMENTS This work is supported in part by DOD grant #AFOSR FA9550-04-1-0079 and by the generous gift of Dr. and Mrs. J. S. Chen to the optical diagnostics program of the Massachusetts General Hospital, Wellman Center for Photomedicine. 1. M. Eigen and R. Rigler, Proc. Natl. Acad. Sci. U.S.A. 91, 5740 (1994). 2. P. M. Goodwin, W. P. Ambrose, and R. A. Keller, Acc. Chem. Res. 29, 607 (1996). 3. S. M. Nie, D. T. Chiu, and R. N. Zare, Science (Washington, D.C.) 266, 1018 (1994). 4. D. L. Jeanmaire and R. P. V. Duyne, J. Electroanal. Chem. 84, 1 (1977). 5. M. G. Albrecht and J. A. Creighton, J. Am. Chem. Soc. 99, 5215 (1977). 6. H. J. Seki, J. Electron Spectrosc. Relat. Phenom. 39, 239 (1986). 7. A. Otto, in Light Scattering in Solids IV. Electronic Scattering, Spin Effects, SERS and Morphic Effects, M. Cardona and G. Guntherodt, Eds. (Springer-Verlag, Berlin, Germany, 1984), pp. 289 418. 8. M. Moskovits, Rev. Mod. Phys. 57, 783 (1985). 9. K. Kneipp, Exp. Techniques Phys. 38, 3 (1990). 10. A. Campion and P. Kambhampati, Chem. Soc. Rev. 27, 241 (1998). 11. K. Kneipp, H. Kneipp, I. Itzkan, R. R. Dasari, and M. S. Feld, Chem. Rev. 99, 2957 (1999). 12. M. Moskovits, J. Raman Spectrosc. 36, 485 (2005). 13. C. L. Haynes, C. R. Yonzon, X. Y. Zhang, and R. P. Van Duyne, J. Raman Spectrosc. 36, 471 (2005). 14. K. Kneipp, H. Kneipp, and J. Kneipp, Acc. Chem. Res. 39, 443 (2006). 15. K. Kneipp, M. Moskovits, and H. Kneipp, Eds., Surface-Enhanced Raman Scattering (Springer, Heidelberg, 2006). 16. A. Bachackashvilli, S. Efrima, B. Katz, and Z. Priel, Chem. Phys. Lett. 94, 571 (1983). 17. K. Kneipp, G. Hinzmann, and D. Fassler, Chem. Phys. Lett. 99, 503 (1983). 18. P. Hildebrandt and M. Stockburger, J. Phys. Chem. 88, 5935 (1984). 19. K. Kneipp, H. Kneipp, and M. Rentsch, J. Mol. Struct. 156, 331 (1987). 20. B. Pettinger and K. Krischer, J. Electron Spectrosc. Relat. Phenom. 45, 133 (1987). 21. K. Kneipp, Experimentelle Technik der Physik 36, 161 (1988). APPLIED SPECTROSCOPY 333A

22. B. Pettinger, K. Krischer, and G. Ertl, Chem. Phys. Lett. 151, 151 (1988). 23. K. Kneipp, Y. Wang, H. Kneipp, I. Itzkan, R. R. Dasari, and M. S. Feld, Phys. Rev. Lett. 76, 2444 (1996). 24. K. Kneipp, Y. Wang, H. Kneipp, L. T. Perelman, I. Itzkan, R. R. Dasari, and M. S. Feld, Phys. Rev. Lett. 78, 1667 (1997). 25. K. Kneipp, H. Kneipp, G. Deinum, I. Itzkan, R. R. Dasari, and M. S. Feld, Appl. Spectrosc. 52, 175 (1998). 26. K. Kneipp, H. Kneipp, V. B. Kartha, R. Manoharan, G. Deinum, I. Itzkan, R. R. Dasari, and M. S. Feld, Phys. Rev. E 57, R6281 (1998). 27. S. Nie and S. R. Emory, Science (Washington, D.C.) 275, 1102 (1997). 28. H. X. Xu, E. J. Bjerneld, M. Kall, and L. Borjesson, Phys. Rev. Lett. 83, 4357 (1999). 29. A. M. Michaels, M. Nirmal, and L. E. Brus, J. Am. Chem. Soc. 121, 9932 (1999). 30. M. S. Dresselhaus, G. Dresselhaus, A. Jorio, A. G. Souza, G. G. Samsonidze, and R. Saito, J. Nanosci. Nanotechnol. 3, 19 (2003). 31. A. Hartschuh, E. J. Sanchez, X. S. Xie, and L. Novotny, Phys. Rev. Lett. 90, 095503 (2003). 32. M. Inoue and K. Ohtaka, J. Phys. Soc. Jpn. 52, 3853 (1983). 33. H. X. Xu, J. Aizpurua, M. Kall, and P. Apell, Phys. Rev. E 62, 4318 (2000). 34. K. Li, M. I. Stockman, and D. J. Bergman, Phys. Rev. Lett. 91, 227402 (2003). 35. M. I. Stockman, V. M. Shalaev, M. Moskovits, R. Botet, and T. F. George, Phys. Rev. B 46, 2821 (1992). 36. E. Y. Poliakov, V. M. Shalaev, V. A. Markel, and R. Botet, Opt. Lett. 21, 1628 (1996). 37. K. Kneipp, H. Kneipp, R. Manoharan, E. B. Hanlon, I. Itzkan, R. R. Dasari, and M. S. Feld, Appl. Spectrosc. 52, 1493 (1998). 38. Y. Yamaguchi, M. K. Weldon, and M. D. Morris, Appl. Spectrosc. 53, 127 (1999). 39. Z. J. Wang, S. L. Pan, T. D. Krauss, H. Du, and L. J. Rothberg, Proc. Natl. Acad. Sci. U.S.A. 100, 8638 (2003). 40. M. Stockman, in Surface Enhanced Raman Scattering, K. Kneipp, M. Moskovits, and H. Kneipp, Eds. (Springer, Heidelberg, 2006), pp. 47 66. 41. A. Otto, J. Raman Spectrosc. 36, 497 (2005). 42. L. P. Capadona, J. Zheng, J. I. Gonzalez, T. H. Lee, S. A. Patel, and R. M. Dickson, Phys. Rev. Lett. 94, 058301 (2005). 43. S. L. Zou and G. C. Schatz, Chem. Phys. Lett. 403, 62 (2005). 44. K. Kneipp, H. Kneipp, I. Itzkan, R. R. Dasari, and M. S. Feld, Chem. Phys. 247, 155 (1999). 45. I. Khan, D. Cunningham, R. E. Littleford, D. Graham, W. E. Smith, and D. W. Mc- Comb, Anal. Chem. 78, 224 (2006). 46. K. Kneipp, H. Kneipp, P. Corio, S. D. M. Brown, K. Shafer, J. Motz, L. T. Perelman, E. B. Hanlon, A. Marucci, G. Dresselhaus, and M. S. Dresselhaus, Phys. Rev. Lett. 84, 3470 (2000). 47. K. Kneipp, H. Kneipp, M. S. Dresselhaus, and S. Lefrant, Philos. Trans. R. Soc. London, Ser. A 362, 2361 (2004). 48. E. J. Ayars, H. D. Hallen, and C. L. Jahncke, Phys. Rev. Lett. 85, 4180 (2000). 49. E. J. Ayars, C. L. Jahncke, M. A. Paesler, and H. D. Hallen, J. Microsc.-Oxf. 202, 142 (2001). 50. S. Efrima, Chem. Phys. Lett. 102, 79 (1983). 51. H. Kneipp, J. Kneipp, and K. Kneipp, Anal. Chem. 78, 1363 (2006). 52. F. Svedberg and M. Kall, Faraday Discuss. 132, 35 (2006). 53. A. R. Bizzarri and S. Cannistraro, Appl. Spectrosc. 56, 1531 (2002). 54. P. Etchegoin, R. C. Maher, L. F. Cohen, H. Hartigan, R. J. C. Brown, M. J. T. Milton, and J. C. Gallop, Chem. Phys. Lett. 375, 84 (2003). 55. K. Kneipp, Y. Wang, R. R. Dasari, and M. S. Feld, Appl. Spectrosc. 49, 780 (1995). 56. V. A. Podolskiy and V. M. Shalaev, Laser Phys. 11, 26 (2001). 57. K. Kneipp, H. Kneipp, S. Abdali, R. W. Berg, and H. Bohr, Spectr.-Int. J. 18, 433 (2004). 58. M. Futamata, Y. Maruyama, and M. Ishikawa, J. Mol. Struct. 735 736, 75 (2005). 59. D. B. Lukatsky, G. Haran, and S. A. Safran, Phys. Rev. E 67, 062402 (2003). 60. Z. J. Wang and L. J. Rothberg, J. Phys. Chem. B 109, 3387 (2005). 61. S. R. Emory, R. A. Jensen, T. Wenda, M. Y. Han, and S. M. Nie, Faraday Discuss. 132, 249 (2006). 62. A. Weiss and G. Haran, J. Phys. Chem. B 105, 12348 (2001). 63. A. R. Bizzarri and S. Cannistraro, J. Phys. Chem. B 109, 16571 (2005). 64. C. J. L. Constantino, T. Lemma, P. A. Antunes, P. Goulet, and R. Aroca, Appl. Spectrosc. 57, 649 (2003). 65. T. L. Haslett, L. Tay, and M. Moskovits, Chem. Phys. 113, 1641 (2000). 66. A. G. Brolo, A. C. Sanderson, and A. P. Smith, Phys. Rev. B 69, 045424 (2004). 67. P. V. Teredesai, A. K. Sood, A. Govindaraj, and C. N. R. Rao, Appl. Surf. Sci. 182, 196 (2001). 68. E. C. Le Ru and P. G. Etchegoin, Faraday Discuss. 132, 63 (2006). 69. R. C. Maher, L. F. Cohen, E. C. Le Ru, and P. G. Etchegoin, Faraday Discuss. 132, 77 (2006). 70. K. Kneipp, H. Kneipp, I. Itzkan, R. R. Dasari, and M. S. Feld, J. Phys.: Condens. Matter 14, R597 (2002). 334A Volume 60, Number 12, 2006