Single-cell analysis with surface-enhanced Raman scattering

Spectroscopy analyzes the interaction between matter and radiant energy, allowing scientists to study our surroundings at scales ranging from the atomic to those of distant stars and galaxies. Spectroscopic techniques¹ enable us to obtain valuable information about the chemical composition, structure, and physical properties of matter. For instance, Raman spectroscopy, where we collect and analyze inelastically scattered light from illuminated matter, makes it possible to understand the content of chemical bonds in a molecule by observing its rotational and vibrational modes.² This enables convenient detection of interactions between molecules, or changes in a single one. Consequently, Raman spectroscopy is useful in forensic, environmental, and materials science, and in pharmaceutical and biomedical research.^3–7 However, the method's drawback is that the signal intensity is generally not high enough to efficiently analyze samples, especially biological ones. To address this, we can modify the technique using a noble metal surface such as gold or silver as a substrate, which enhances Raman scattering ×10¹⁴ through two simultaneous mechanisms: electromagnetic and chemical.^{8, 9} This technique is then called surface-enhanced Raman spectroscopy (SERS) and is widely used in biomedical research, particularly for its capability in single-molecule detection.¹⁰

Our research aims to extract spectral information from cells through Raman spectroscopy or SERS to relate this information to cellular processes. Our rationale is that the molecular dynamics in the small restricted area of the cell decides the function of larger units such as organs. Despite major developments in the field, we remain far from a complete understanding of molecular dynamics in a single cell when it receives a stimulus from its environment.^11–13 To enhance our understanding, we undertook two experiments.

In the first, we introduced citrate-reduced 50nm gold nanoparticles (AuNPs)—see Figure 1(a)—into a cell culture medium. The process is very simple. However, determining the location and behavior of the AuNPs is complex, because only the molecular structures that are in contact or very close to the AuNPs can provide useful information.

Figure 1. Transmission electron microscope images of 50nm nanoparticles (NPs) of (a) gold and (b) silver. The scale bar is 100nm.

Existing methods to determine the toxicity of NPs in cells are neither reliable nor accurate, and in most cases are also labor-intensive. The NPs' unique physicochemical properties can interfere with conventional toxicity assay ingredients and lead to false results.¹⁴ Therefore, to assess nanotoxicity at the single-cell level, we treated two cell lines (a lung cancer cell line, A549, and human dermal fibroblasts, HDF) with commonly used NPs: single-walled carbon nanotubes, and zinc oxide and titanium dioxide NPs. We incubated these with the AuNPs for 24 hours, after which we rinsed the cells, seeded on calcium fluoride (CaF₂) slides, with phosphate-buffered saline, and added a few drops of serum-free cell medium to the slides to enable mapping with Raman microscopy. We acquired SERS spectra using an 830nm laser with 2s exposure time and 100% laser power. We preprocessed the resulting spectra (∼100 spectra per cell) by baseline subtraction, cosmic ray removal, and smoothing, and obtained an average spectrum from a single cell. When we compared the average spectra of treated and untreated cells, we observed changes in protein and lipid content-related peaks (see Figure 2) that can be counted as indications of toxic response. Moreover, we collected several spectra from cells to perform principal component analysis (PCA), a multivariate statistical method used widely for data dimensionality reduction in chemometrics. According to our results, it is possible to differentiate untreated and NP-treated cells.

Figure 2. Average surface-enhanced Raman spectroscopy (SERS) spectra from A549 cells treated with NPs and positive control chemical reagents. The gray rectangles highlight the main peak changes. SWCNT: Single-walled carbon nanotubes. TiO₂: Titanium dioxide. ZnO: Zinc oxide. DMSO: Dimethyl sulfoxide. H₂O₂: Hydrogen peroxide. a.u.: Arbitrary units.

In our second experiment, for substrates we used silver NPs (AgNPs) reduced with citrate: see Figure 1(b). In contrast to the previous approach, we allowed the cells to dry at their fixed position before adding the colloidal AgNPs, and we dried the substrates in an overturned position to enable a better distribution of the NPs.¹⁵ We mapped each cell line, cancerous and noncancerous, for SERS spectra. Figure 3 shows the average of the spectra obtained from the cancerous and the healthy cells, where their spectral differences related to the biochemical modifications.

Figure 3. Mean SERS spectra from cancer cells (ACHN) and healthy cells (HEK 293).

We reduced the high-dimensional SERS spectral data set obtained from cells to new variables and principal component scores (PCs) using PCA. Figure 4 shows the 3D scatter plot of PCA based on SERS spectra collected from cancerous and healthy cells. Then, we used a linear discriminant algorithm (LDA) to differentiate the cell types: metastatic kidney cancer (ACHN), immortalized human kidney (HEK), and primer cancer cells (A-498), and obtained the classification results using the leave-one-out cross-validation method, with a sensitivity and specificity of 88 and 84%, respectively.¹⁵ Therefore we can use the LDA based on PCs of SERS spectra as a diagnostic method to differentiate cancerous and healthy cells.

Figure 4. 3D principal component analysis scatter plot of SERS spectra acquired from ACHN and HEK 293 cells.

In summary, in conjunction with multivariate statistical tools, SERS can provide valuable information to track the molecular status of a living cell. We can detect the response of a cell to a toxic material or to the altered molecular processes caused by diseases such as cancer. Our preliminary results indicate that nanotoxicity determination using SERS will be possible in the near future. To accomplish this, we will focus on LDA and other statistical methods, such as hierarchical clustering of the data obtained from SERS spectra.

The authors acknowledge financial support from TUBITAK (project 113Z554) and Yeditepe University.