SERS Mapping in Langmuir–Blodgett Films and Single-Molecule Detection

INTRODUCTION

The detection of a single-molecule spectrum ¹² with surface-enhanced spectroscopy ³ has been a central issue in plasmonics. ⁴ The electromagnetic enhancement of surface-enhanced (resonance) Raman scattering (SERS or SERRS) is achievable thanks to the localized surface plasmon resonances (LSPR) of metallic nanostructures. The magnitude of the local field created by the plasmonic nanostructure depends on parameters such as size, morphology, arrangement, and local environment of the nanostructure. ⁵ By fine-tuning these parameters, electromagnetic hot spots can be created to facilitate ultra-sensitive SERS, SERRS, and surface-enhanced fluorescence, used for single-molecule detection (SMD) with nanostructures.^4,6 The average SERS enhancement factor normally lies between 10³ and 10⁶, but the intense electromagnetic field or hot spots can increase this factor up to 10¹⁰. ⁷

SMD by using SERS, or more commonly SERRS, is now well established, exploiting the enhancement at hot spots and the new technologies in micro-Raman spectrographs.^8,11 The first reports were published independently by Nie et al. ² and Kneipp et al., ¹ who used Ag colloids to achieve SERS for rhodamine 6G and crystal violet molecules, respectively. Later on, Constantino et al. ⁹ extended SMD to Langmuir-Blodgett (LB) monolayers containing the target molecule embedded in a fatty acid host matrix deposited onto an Ag island film (LB-SERS). This method with LB films offers the highest degree of control over the homogeneity and average spatial distribution of the target molecules in the sample. ¹² SMD of phospholipids^13,14 and biomolecules such as amino acid oligomers with dye attachment have been reported.^15,16 It is not trouble-free to prove the single-molecule character of the spectra of a single dye. In particular, when a new SERS substrate is studied, rare events at ultralow concentrations cannot be considered expressions of single-molecule sensitivity, especially in highly inhomogeneous SERS substrates such as colloidal systems. The bi-analyte SERS technique ¹⁷ was developed to solve the problem of reliability in single-molecule spectra. The latter is achieved by providing a contrast signal to the one we want to detect and studying their relative occurrences. ¹⁰ Indeed, using substrates coated with Ag island film and one LB monolayer, it is possible to establish a regime (ultralow doping) in which single-molecule events can be observed.

In this study, the spectral data is obtained from a homogeneous distribution of a phenothiazine derivative (cationic methylene blue [MB]), which is electrostatically attached to a Langmuir film host matrix containing the anionic phospholipid dipalmitoyl phosphatidyl glycerol (DPPG). Different degrees of doping are attained by controlling the concentration of MB in the subphase. MB exhibits a high cross-section for SERS and SERRS and is commonly applied in photodynamic therapy for battling tumor cells, viruses, and bacteria.^18,–22 Micro-Raman maps are recorded for a series of concentrations down to one molecule, on average, in the field of view of the microscope objective. For data analysis, we also include a new computational method to extract, in real time, the SERS data that can be assigned to single-molecule spectra. The latter is achievable with some certainty by using the LB technique; but it could require further analysis in other SERS or SERRS experiments. ¹⁷

EXPERIMENTAL

The anionic phospholipid DPPG {1,2-dipalmitoil-sn-3-glicero-[fosfo-rac-(1-glicerol)} was purchased from Avanti Polar Lipids, Inc. The phenothiazine MB (tetra methyl-diamino-di-phenyl-thiazine) and Ag shot (1-3 mm in size and 99.99% pure) were acquired from Sigma-Aldrich Co. All chemicals were used without further purification. The molecular weights of DPPG and MB are 745 and 319 g/mol, respectively. Both DPPG and MB stock solutions were prepared at 1 mM by dissolving the powder in methanol:-chloroform (1:9 in volume) and chloroform, respectively. All the solvents used were high-performance liquid chromatography grade. Ultrapure water (18.2 MΩ cm) was obtained with a Milli-Q water purification system, model Simplicity. Ag island films 9 nm in mass thicknesses were vacuum evaporated with homemade equipment under ca. 10∼ ⁶ torr onto Corning glass microscope slides kept at 100 °C (kept for 1 h after evaporation).

The Langmuir and LB films were fabricated with a Nima model 302M Langmuir trough. The Langmuir films were characterized by surface pressure versus mean molecular area (π–A) isotherms at 20 °C, using the Wilhelmy method. The cationic MB interacts specifically with the anionic polar head group (PO₄⁻) of the DPPG phospholipid.^23,–25 Therefore, one can take advantage of this DPPG-MB interaction to fabricate Langmuir and LB films containing different surface concentrations of MB embedded in a DPPG host matrix. Co-spreading solutions were prepared from stock solutions of DPPG and MB, and the monolayer was symmetrically compressed under a constant barrier speed of 10 mm/min, with the subphase containing ultrapure water. The DPPG-MB ratio of the co-spreading solutions was calculated to achieve 1, 10, and 100 MB molecules to 2.5 × 10⁶ DPPG molecules within 1 μm ² , which corresponds to the focused area of the laser beam in the experimental setup. The number of DPPG molecules was calculated based on π-A isotherms of neat DPPG. The mean molecular area in the latter is ca. 40 Å ² , which means 2.5 × 10⁶ of DPPG molecules/μm ² (10⁸ Å ² ), as indicated in Fig. 1. Also shown is the isotherm for the Langmuir film containing DPPG co-spread with MB in a ratio of 2:1 (in number of molecules). The π–A isotherms are shifted toward larger areas with increasing MB concentration, since MB molecules remain at the air–water interface due to their electrostatic interaction with DPPG,^23,–25 as illustrated in the diagram in Fig. 1. Such strong interaction in DPPG Langmuir films is consistent with results by Hidalgo et al. ²⁵ for two other phenothiazine derivatives, namely trifluoperazine and chlorpromazine. Then, Langmuir monolayers containing different concentrations of MB molecules were transferred from the air-water interface onto the 9 nm Ag island films, forming the LB films. The surface pressure was kept at 25 mN/m, and the upstroke dipping speed varied from 0.5 to 3.0 mm/min. Under such conditions, Z-type LB films with a transfer ratio close to unity were obtained.

OPEN IN VIEWER

Fig. 1.

π–A isotherms of neat DPPG and co-spread DPPG+MB in a ratio of 2:1 (in number of molecules), recorded for the ultrapure-water subphase. The inset shows a schematic of MB interacting with a condensed DPPG monolayer at the air–water interface. Also shown is the DPPG molecular structure.

The ultraviolet-visible absorption spectra were recorded for MB solutions and 9 nm Ag island films with a Varian spectrophotometer Cary 50. The absorption and fluorescence spectra of 5 × 10⁻⁶ M MB aqueous solutions and the surface plasmon absorption of a vacuum-evaporated 9 nm Ag island film can be seen in Fig. 2a. Atomic force microscopy (AFM) images were collected in the contact mode by using a Digital Instrument, model Nanoscope IV, with a tip of silicon nitride. All images were obtained with high resolution (512 lines per scan) at a scan rate of 0.5 Hz. Topographical (height) images were processed with the software Gwyddion 2.19 and are exhibited in Figs. 2b and 2c. The micro-Raman scattering experiments were conducted with a micro-Raman Renishaw spectrograph, model inVia, with laser excitation at 514.5 nm and powers of 10-20 μW at the sample. The system is equipped with a Leica microscope, whose 50X (NA 0.75) objective lens allows for collecting the spectra with ca. 1 μm ² spatial resolution. Single-point spectra were recorded with 4 cm⁻¹ resolution and 10 s accumulation time, whereas two-dimensional (2D) mapping results were collected through the rastering mode by using a computer-controlled motorized stage (XY) and 1 s of accumulation time.

OPEN IN VIEWER

Fig. 2.

(a) Plasmon absorption of a 9 nm Ag island film and the absorption and emission spectra for a MB solution (5 × 10⁻⁶ M). The emission spectrum was recorded by exciting the MB solution with a 633 nm laser line. The inset shows the MB molecular structure. AFM images of a 9 nm Ag island film in (b) 2D and (c) 3D.

RESULTS AND DISCUSSION

To obtain the reference ensemble average SERS spectra of MB excited with the 514.5 nm laser line, Langmuir films with a 2: 1 ratio of DPPG : MB, leading to a DPPG monolayer containing a maximum of ca. 10⁶ MB molecules/μm², were transferred onto 9 nm Ag island film. Since MB is a water-soluble molecule, the spontaneous formation of Langmuir films onto an air–water interface is not possible. It is also known that phenothiazine pharmacological derivates tend to interact with cell membrane models made with phospholipids.^23,25,–31 The interaction between MB and the anionic phospholipid DPPG in vesicles ²⁷ and monolayers^23,31 has been studied with layer-by-layer and LB films, respectively. The cationic MB was found to interact preferentially with the anionic polar head group of the DPPG phospholipid. Therefore, taking advantage of this DPPG-MB interaction Langmuir and LB films can be formed, containing different surface concentrations of MB embedded in a DPPG host matrix, including the single-molecule regime.

The point-by-point mapping results obtained by collecting spectra along an area of 100 × 100 μm with a 3 μm step, i.e., leading to a total of 1156 spectra, are shown in Fig. 3a. The relative intensity of the band centered at 1625 cm⁻¹ was used to create the spectral map. The band at 1625 cm⁻¹ is assigned to C = C + C–N = C–stretching vibration of the MB molecule.^24,26

OPEN IN VIEWER

Fig. 3.

(a) MB ensemble average SERS spectra collected with the 514.5 nm laser line for an LB monolayer containing a maximum of 1.25 × 10⁶ MB molecules/μm ² , deposited onto a 9 nm Ag island film. The Raman spectrum for a MB cast film is also given as reference. (b) Representation of the LB monolayer containing MB within a DPPG matrix deposited onto 9 nm Ag island film, and the same system in the direction of single-molecule regime.

The main findings inferred from the results in Fig. 3a are as follows: The signal in the area mapping arises from the ensemble average SERS spectra of MB, since DPPG has a much lower cross-section for SERS. ²⁴ The Raman spectrum for an MB cast film, using the slot-data limits for the purpose of comparison in Fig. 3a, has a profile similar to the SERS spectrum for the LB film, in terms of band position and relative intensities. This indicates that MB does not form a complex with the Ag nanostructures. MB molecules are distributed all over the area mapped in Fig. 3a, since it was possible to collect the SERS spectra from all spots within similar Raman intensity. This distribution is homogeneous in the micrometer range, as denoted by the relatively uniform intensity of the MB 1625 cm⁻¹ band. Therefore, the electrostatic interaction between DPPG and MB, established during the Langmuir film formation at the air–water interface, is sufficiently strong to withstand transfer onto LB films. In summary, for LB monolayers containing a large number of MB molecules, 1.25 × 10⁶ MB molecules/μm² are deposited during LB fabrication, assuming that all co-spread MB molecules could be transferred due to specific electrostatic interaction between MB and the phospholipid at the air–water interface. The latter also confirms the detection of attomole quantities in LB-SERS. To test the detection limit, it is necessary to dilute the surface concentration of MB molecules within the host matrix of DPPG, as illustrated in Fig. 3b.

Single-Molecule Detection for Methylene Blue in Langmuir-Blodgett Films. For SMD, the ideal condition is to place, on average, only one MB molecule within the probing area of the micro-Raman objective. However, it is important to keep in mind the statistical restriction for the observation of the single-molecule spectrum; it is the rarity of spatial coincidence between electromagnetic hot spots in the Ag nanostructured film and the target molecule in the LB monolayer. Remarkably, the LB technique is suitable to provide molecular-level control in 2D surface structures that can be directly probed by the micro-Raman objective. The accumulated experience working in LB-SERS-SERRS ¹⁰ indicates that the single-molecule regime is achieved in LB monolayers containing, on average, 100 or fewer target molecules/μm², and the 2D maps from the micro-Raman system confirmed that among the thousands of spectra recorded, only very few single-molecule spectra are observed.

In this work, Langmuir films with an average of 100, 10, and 1 MB molecules/μm² embedded in a host matrix of DPPG were formed for the SMD experiments. At these surface concentrations, only monomers of MB are the potential source of the measured Raman signal. The samples were mapped at low laser powers (20 μW) to avoid photodegradation. The spectra were collected every 3 μm in a total area of 100 × 100 μm (total of 1156 spectra for each map) to avoid contribution from overlapping areas. Maps such as the one shown in Fig. 3 were constructed with Wire 3.0 software from Renishaw for the vibrational band at 1625 cm⁻¹. After generating the SERS mappings from MB molecules, each spectrum was examined with, as reference, the MB ensemble spectrum for the elimination of false positives. Figure 4 shows the SERS mappings collected for samples containing 100, 10, and 1 MB molecule/μm², where brighter points correspond to the stronger SERS signals. Single-molecule spectra were extracted from the mappings and compared with the MB reference spectrum shown at the bottom of Fig. 4.

OPEN IN VIEWER

Fig. 4.

SERS area mapping for LB monolayers containing 100, 10, and 1 MB molecule/μm² embedded in a DPPG host matrix. Areas of 100 × 100 μm were mapped with 3 μm steps by using the 514.5 nm laser line. The bright spots correspond to the intensity of the MB band at 1625 cm⁻¹.

The number of spots containing MB SERS spectra decreased when the number of MB molecules/μm² was reduced, as expected. In addition, the SERS spectra for 100, 10, and 1 molecule/μm² show fluctuations when compared with the ensemble average SERS spectrum. Thus, band parameters including bandwidth, band shape, Raman shift, and absolute and relative intensity deviate from those of the average, which is typical of the single-molecule regime. This variation is attributed to differences in the local molecular environment, ³² highlighting the potential for the probe molecule to report differences in its vicinity.

Visual Exploration of Single-Molecule Spectral Identification. In the procedure above, the identification of single-molecule spectra was done manually, using only one characteristic peak of the MB spectrum. However, this time-consuming procedure is not efficient when tens of thousands of spectra are to be scanned. A unique computational tool was tested here to identify the spectrum of the single molecule among those thousands of spectra in a map and to attain visualization of the information in real time. In this task, the first step is to process all the spectra and compare them against a reference spectrum. The procedure comprises four steps: (i) filtering, (ii) detrending, (iii) smoothing, and (iv) similarity calculation.

The first step transforms the spectra to remove spurious values that could affect comparison with the reference spectrum. We accomplished this step by using a mean filter. ³³ Let s = {y₁, y₂, …, y_m} represent a given spectrum, where y_i is the Raman scattering intensity at wavenumber i, and m corresponds to the highest wavenumber in the spectrum. Each y_t value is verified with a window of length w << m (w is an odd value), centered on y_t, calculating the mean

on this window, and verifying if the criterion

is satisfied. s_t is the standard deviation of y t - w / 2 + ⋯ + y t + ⋯ + y t + w / 2 , and L is a threshold. If the inequality is not satisfied, measurement y_t is assumed as an outlier, being replaced by m_t.

The second step, detrending, is a statistical or mathematical operation for removing trends from the series. ³⁴ It is often applied to remove a feature that distorts or obscures relationships of interest. We normalized the spectra, making

with y_max = max(y₁, y₂, …, y_m) and y_min = min(y₁, y₂, …, y_m), and subtracted the mean for each value in the series. All the spectra were then smoothed to remove noise in the third step of the procedure. We used two techniques for smoothing, the Savitzky–Golay first-order filter, ³⁵ which performs a least-squares fit of a small set of consecutive data points to a polynomial and then calculates the central point of the fitted polynomial curve as the new smoothed data point, and the moving averages technique, ³⁶ which smoothes a spectrum signal by replacing a value by the mean, considering a window centered on it, similar to the outlier-removal approach described previously. The result of both techniques is not only the removal of high-frequency signal, but also the preservation of the shape of the spectrum. The images shown in this section were generated with the Savitzky–Golay technique, as it delivered results closer to the manual analysis, considered here as a control. This operation is necessary, because high-frequency noise affects the dissimilarity calculation, described next.

The last step before visualization is the dissimilarity calculation. Let s₁ = {y₁, y₂, …, y_m} and s₂ = {x₁, x₂, …, x_m} be two different spectra. The Euclidean distance between s₁ and s₂ can be computed as:

where D_t = (x_t − y_t). Although this is the most common form of distance calculation, due to shifts on the spectra, it can fail to correctly capture the dissimilarity. In order to handle this, we adapt it to take shifts into account. Consider a window of size k << m (k is an odd value) centered on D_t. We replace D_t on d(s₁, s₂) by max(D_{t − k/2}, …, D_t,… D_t + _k/2). Hence, small shifts on the spectra are ignored. We have tested different strategies such as dynamic time warping, ³⁷ which matches two curves represented by sequences of data points so that the sum of the squared Euclidean distances between the matched data points is minimized, but this adaptation of the Euclidean distance attained results more compatible with the manual analysis.

As a proof of concept, SMD visual maps are generated based on matching the spectra against a representative MB ensemble spectrum used as reference. In this case, the single-molecule identification takes into account the similarity between the entire spectral window (1048-1651 cm⁻¹) of both MB reference and collected map spectra. Each SERS spectrum from the map recorded from 100, 10, and 1 MB molecule/μm² is examined by the computational method for the elimination of false positives. Basically, only spectra with high similarity to the reference spectrum are selected. Here, a spectrum is classified as “high similar” if its similarity to the reference is at least 90% of the highest similarity found on the measurements. The analyzed spectra are visualized in a grid with the same dimension and spectral location of the original surface Raman mapping (Fig. 4). The positions of the MB molecules (100, 10, and 1/μm²) are highlighted on the black grid in Fig. 5. Clearer spots refer to best matches between the MB reference and MB spectra from 100, 10, or 1 molecule/im² LB samples. In the right panel in Fig. 5 are shown a few MB spectra, classified based on the MB reference spectrum after filtering, detrending, and smoothing. In almost all cases, the positioning of the MB molecules was found in agreement with those in Fig. 4. Note, however, that in the middle left panel (corresponding to 10 molecules/μm²) in Fig. 5, there is a molecule missing from the analogous panel in Fig. 4. We inspected this spectrum and observed that even though the most important band coincided with the reference spectrum, there were considerable differences in other parts of the spectra. Therefore, it is possible that the analysis in Fig. 4 led to a false positive in the manual analysis.

OPEN IN VIEWER

Fig. 5.

Data exploration for 100, 10, and 1 MB molecule/μm² positioned in a grid with the same dimension of the raw Raman mapping of Fig. 4. A few spectra after filtering, detrending, and smoothing, classified as MB single molecule, are shown in the right. Finally, in real samples with more than one SERS-active molecule, one can use several reference spectra or even the combination of distinct spectra. Because the comparative analysis between the data being considered and the reference spectra is made with quantitative computational methods, the identification of relevant features should be much more efficient.

In order to illustrate the usefulness of the computational tool, we present data from two other case studies, for which a manual analysis would be prohibitively costly. In the first case, we applied the information visualization technique to SMD detection of Texas Red anchored (chemically attached) to a phospholipid dispersed into a fatty acid matrix. ¹³ The analysis of 6561 spectra collected in an area Raman mapping is given in Fig. 6. An even more demanding application is the second case study, where 37 848 spectra were collected, covering larger areas of a fatty acid matrix containing octadecylrhodamine B (R18) target molecules.^38,39 This molecule was selected because it has been used as reference for the streamline Raman mapping. The analysis of R18 SMD data with information visualization led to correct identification of single molecules, as illustrated in Fig. 7. The latter demonstrates the potential of information visualization to identify single molecules without false positives, taking only a few seconds rather than the time-consuming, tedious manual analysis. In terms of computational cost, the technique is very fast. It is a linear technique, i.e., the processing time is proportional to the number of signals to be compared. So if it takes 1 ms to compare one signal, it will take 30 s to compare 30 000 signals.

OPEN IN VIEWER

Fig. 6.

(a) Information visualization for a mapping containing 6561 spectra. (b) Molecular structure of arachidic acid- and Texas red-tagged phospholipid (TRP). (c) A few TRP single-molecule spectra extracted from (a).

OPEN IN VIEWER

Fig. 7.

(a) Information visualization for a mapping containing 37 848 spectra. (b) Molecular structure of octadecylrhodamine B (R18). (c) A few R18 single-molecule spectra extracted from (a).

It should be stressed that other statistical methods have been employed to analyze SERS data aimed at SMD,^40,–42 but a direct comparison with our approach is not possible for the following reasons. Sasic et al. ⁴⁰ used principal component analysis (PCA) to compare spectra from distinct spots and to determine the number of chemically distinguishable species contributed to the spectra, rather than doing a direct comparison with a reference, as we did here. Dos Santos et al. ⁴¹ employed a modified PCA in order to interpret fluctuations in anti–Stokes-to-Stokes ratios, which were taken as correlating to resonances at the plasmonic hot spots in SERS spectra. In our approach, fluctuations could affect the estimated distances between spectra, but they were not explored directly. Etchegoin et al. ⁴² described a framework designed to analyze the statistics of single-molecule SERS signals, also providing an example of data from two analytes in the same sample. Again, PCA was used to distinguish between the analytes, and this feature can be added in our approach when comparison is to be done with spectra from samples with more than one analyte.

CONCLUSION

Single-molecule spectra were obtained for a biochemical system by using the pharmacological drug MB (SERS probe) attached to the DPPG phospholipid in a Langmuir monolayer that was then transferred onto Ag nanostructures, forming LB films. The homogeneous doping of the Langmuir films with MB was achieved by electrostatic interactions. The 2D SERS mapping of LB contained, on average, less than 100 target molecules in the field of view, warranting the acquisition of single-molecule spectra. The implementation of a data exploration algorithm in SMD experiments was successfully demonstrated.