Accurate Quantification and Imaging of Cellular Uptake Using Single-Particle Surface Enhanced Raman Scattering

Abstract

Understanding the uptake, distribution, and stability of gold NPs in cells is of fundamental importance in nanoparticle sensor and therapeutic development. Single nanoparticle imaging with surface enhanced Raman spectroscopy (SERS) measurements in cells is complicated by aggregation dependent SERS signals, particle inhomogeneity, and limited single particle brightness. In this work, we assess the single-particle SERS signals of various gold nanoparticle shapes, and the role of silica encapsulation on SERS signals, to develop a quantitative probe for single particle level Raman imaging in living cells. We observe that silica encapsulated gap enhanced Raman tags (GERTs) provide an optimized probe that is quantifiable per voxel in SERS maps of cells. This approach is validated by single particle inductively coupled mass spectrometry (spICP-MS) measurements of NPs in cell lysate post-imaging. spICP-MS also provides a means of tag stability measurements. This analytical approach can be used not only to quantitatively assess nanoparticle uptake on the cellular level (as in previous digital SERS methods), but also to reliably image the subcellular distribution and to assess the stability of NPs in cells.

Graphical Abstract

Gold nanoparticles (NPs) have utility in biology as both therapeutics and sensors due to their biological stability, low cytotoxicity, and biocompatibility. For example, gold NPs are currently under clinical trial for use as anticancer agents via photothermal therapy and drug delivery (e.g., AuroLase Therapy by Nanospectra, CYT-6091 (Aurimune) by CytImmune Sciences). Additionally, gold NPs have potential as cellular sensors of physiologically relevant conditions, such as pH, and of specific biomarkers.^1–4 Gold NPs have been explored as uniquely suited probes for multimodal and intraoperative imaging of tumors.^5–7 Due to the wide interest in gold NPs, it is important to be able to monitor and track them to understand how these NPs interact within cells. Quantitative imaging of gold NPs allows for a better understanding of the spatial distribution of these particles, with relevance to their development as cellular sensors and therapeutics. Assessing how effectively NPs are internalized and where NPs accumulate in cells has been a topic of ongoing research.^8–10 Methods to assess gold NP internalization by cells include flow cytometry and ICP-MS. However, these techniques are destructive and provide no spatial information about nanoparticle location in cells. In addition, functional assays that measure mRNA delivery or cell death are not direct measures of the number of internalized NPs.^11–13 TEM provides absolute quantification of NPs via direct imaging but is low throughput and requires cell fixation and microtoming, which prevents measuring NP uptake dynamically and may introduce artifacts.^{14, 15} There persists a need for quantitative imaging of gold NPs in living cells.

Surface enhanced Raman scattering (SERS) has the potential to provide quantitative imaging of gold NPs in living cells with exquisite sensitivity, multiplexing capacity, and high photostability. However, SERS imaging of gold NP distribution in cells is challenged by intrinsic difficulties to obtaining reproducible signals. NP aggregation can increase SERS signals by orders of magnitude, making aggregate detection facile, but also causing large intensity fluctuations and challenges for quantifying NPs in cells.^{1, 16, 17} For this reason, gold NPs are often functionalized with a Raman active molecule and a protecting layer to form SERS tags, where the protecting layer inhibits plasmon coupling and avoids the interactions between NPs that result in variations in SERS intensities.¹⁸ NP shape can provide higher local electric field strengths and therefore higher Raman scattering.^{19, 20} Even among highly homogenous NPs, polarization dependent SERS signal originating from anisotropy of the NP shape can result in inconsistent signal from particle to particle. This has led to synthesis of more symmetric particles with increased facet density to maintain signal strength without polarization dependence.^{21, 22} Recently, digital methods for detecting SERS have circumvented the intensity variation challenge and enable dynamic measurement of NPs in cells, but this often sacrifices spatial resolution to obtain accurate quantification.^{23, 24} In digital SERS, multiple measurements must be made across an area for quantification of particle number within the total region measured.²³

Advances in NP design have enabled bright and reproducible SERS signals, leading to greater potential for quantitative SERS imaging of NPs. Previously synthesized and characterized ultrabright NPs have been shown to be useful for rapid imaging applications.^{5, 25–27} Herein, we build upon SERS imaging by assessing the ability to image and quantify individual bright NPs. We first assess different particle geometries for reliability of single particle SERS signal in solution at low concentrations and on carbon films in correlated SERS and TEM imaging. Our investigations indicate that petal covered gap enhanced Raman tags (GERTs) NPs have more intense single particle Raman signals than star and bipyramid shaped particles. The GERTs used consist of a ~40 nm spherical core, onto which hundreds of ~5 nm petals are grown (Figure S1). Nitrobenzene thiol (NBT) spaces the petals and functions as a Raman reporter molecule. The role of silica shells is also shown to increase the reproducibility and intensity of the SERS signal within a synthesized batch. The high intensity and reproducibility of single GERT SERS signal enables particle counting on a commercially available confocal Raman microscope. By quantifying the number of particles detected in each focal volume, rather than averaging over regions of multiple measurements, this work enables high resolution quantitative SERS imaging in living cells, opening the door to quantitative subcellular gold NP tracking by SERS imaging. Inductively coupled mass spectrometry (ICP-MS) has previously been used to quantify NPs in cells, and can provide validation for non-destructive measures of NP uptake.^{23, 28, 29} A single particle adaptation (spICP-MS) dilutes the sample in order that the mass of each particle is detected individually. spICP-MS of cell lysate provides not only validation of NP concentration, but also additional particle composition and aggregation information.³⁰

Results and Discussion

Gold NPs, ca. 40 nm diameter, are readily synthesized via the Turkevich/Frens method,^{31, 32} functionalized with a Raman reporter molecule, and silica encapsulated. However, these particles can only be detected as aggregates (Figure S2, Table S2), leading to challenges with quantification in cells readily attributable to the relative brightness of particles in different aggregation states (Figure S3). Challenges in purification, gap distance, and polarization dependent signals make aggregated spheres difficult to use for single particle measurements.^{33, 34} We therefore synthesized and compare three different gold NP geometries: stars, bipyramids, and GERTs, which are reported to give increased SERS signals, to assess their ability for single particle SERS (Figure 1).^{27, 35, 36} The NPs were functionalized with a Raman reporter molecule. 4-mercaptobenzoic acid was chosen as a reporter for bipyramids and stars, as it renders the particle surfaces vitreophilic, facilitating silica encapsulation.^37–39 Formation of internal gaps on GERTs is facilitated by nitrobenzene thiol (NBT), which then serves as a reporter molecule for this particle geometry. In this case, silica encapsulation is likely directed by surfactant.^{26, 40}

Figure 1.

The representative extinction spectra and transmission electron microscopy of different types of nanoparticles with and without silica encapsulation are shown. Scale bar widths on all TEM images are all 100 nm. (A-C) are results for bipyramid particles, (D-F) are nanostars, and (G-I) are from GERT particles. (J) shows ensemble SERS signal per particle determined from each particle type shown.

Assessing particle types for reliable single-particle SERS

Particles were characterized with and without silica encapsulation (Figure 1 A–I). The ensemble average signal per nanoparticle can be used to quickly identify intense SERS signals appropriate for single particle, quantitative SERS imaging. Nanoparticle tracking analysis (NTA) was used to determine particle concentrations, with sample data shown in Table S3 and Figure S4. The average SERS intensity per particle is determined in ensemble measurements of the bipyramid, star, GERT gold nanoparticle geometries (Figure 1J) based on the nitro stretch of NBT or the v8a ring breathing of MBA. The average SERS brightness for the GERTs is an order of magnitude brighter than the other particle types. This is primarily due to the enhancement of large numbers of reporter molecules for this particle geometry. In addition to the geometry being more conducive to bright signals, the 638 nm excitation wavelength is better aligned with the LSPR of the GERTs than the stars or bipyramids, as evident in the particles’ extinction spectra (Figure 1 A, D & G).. The brightness of this particle geometry is unsurprising given previous work with similar particles.^{26, 41} In addition to the difference in intensity between particle geometries, there is a difference in the signal intensity between the encapsulated and unencapsulated particles (Figure 1J, Figure S5).

This difference from silica encapsulation was small for the bipyramids and stars, but large for the GERTs, and is attributed to silica encapsulation shifting the LSPR relative to the 638 nm laser used in these experiments (Figure 1 A, D & G). This ensemble signal per particle is complicated by aggregation dependent local field enhancements.^{16, 25, 42, 43} Before applying Raman imaging to quantify the number of gold NPs, nonlinearities between the SERS signal intensity and nanoparticle concentration need to be understood. The effect of plasmon coupling on the SERS signal intensity can be further assessed using time series acquisitions of NPs in solution and comparing the particle-to-particle signal heterogeneity (Figure 2). SERS signal acquisitions of NPs diffusing through solution are made at 785 nm for the bipyramid and star NPs and 633 nm for the GERTs (closest available lasers to the respective extinction maxima as shown in Figure 1A, D & G, using similar laser powers). Concentrations within an order of magnitude, optimized to achieve a low percentage of spectra with signal, were chosen for each particle type. These acquisitions are processed in MATLAB as described in Figure S6 and Figure S7. Briefly, the raw data are reduced to direct classical least squares scores for the relevant Raman reporter molecule (Figures S6 and S8) before being baseline corrected in time by adaptive iteratively reweighted penalized least squares (airPLS).⁴⁴ Events in time are determined by thresholding the signal with the mean plus three standard deviations of the blank. The local maxima of the resulting time series are identified to count particle events. The intensity distribution and frequency of events per run are quantified for each concentration measured. At low concentrations, it is expected that the number of events scales directly with concentration (this is the basis of digital SERS measurements).⁴³ The distribution of the event signal intensities corresponds to the distribution of the Raman scattering intensities of the sample of particles and, with sufficient sampling, is representative of the population. Data for number of events with different particle concentrations are shown in Figure 2A and Figure S9, the number of events per run increases with concentration but scales less directly and with higher relative error for bipyramid and star particle geometries. Bipyramids have few events relative to particle concentration, potentially only generating signal during particle collisions within the focal volume and indicating poor suitability for single particle SERS. Stars have slightly more events, and GERTs have many events. In contrast, the frequency of SERS events of silica encapsulated particles is higher at lower concentrations. Silica encapsulated bipyramids and stars exhibit a tendency to aggregate during encapsulation, leading to more detected events. The difference in encapsulated aggregates among particles used can be seen in Figure 1C, F & I. This is consistent with the effect of aggregation during encapsulation observed for spherical particles (Table S2). Encapsulation of the GERTs produces little change in the frequency of signal events with concentration relative to the other particle types, as aggregation of gold cores during encapsulation is greatly reduced in the presence of a surfactant. The observed silica encapsulated GERT signal is consistent with single particle detection (Figure S10).

Figure 2.

Frequency and intensity of signal events from time series acquisitions of SERS of different particle types for both bare functionalized particles and silica encapsulated functionalized particles in solution. The frequency of signal events relative to particle concentration is shown in (A). The baseline corrected classical least squares scores for events in (A) for B) GERTs, and C) bipyramids, and D) stars with and without silica encapsulation.

The variance of the event intensities of unencapsulated particles increases with particle concentration (Figure 2B–D, Figure S9). This is contrary to what would be expected from increased sampling, but consistent with dynamic plasmon coupling as particles interact in the measurement volume leading to higher intensity variation in event signals.^{16, 17} The average signal intensities of silica encapsulated particles show decreased variance at higher concentrations, as more events leads to increased sampling and less uncertainty of the mean particle intensity for each run. With silica encapsulation, dynamic plasmon coupling is prevented, resulting in more consistent event intensities when scaling concentration. Preventing dynamic coupling improves the quantitative response from low concentrations of NPs, and better ensures signal intensity will scale linearly with particle concentration.^{16, 17} The effect of dynamic plasmon coupling without particle encapsulation will be even more significant in cells, where particles can be at higher local concentrations than what is measured in the time series.

The effect of aggregation state of silica encapsulated particle signal is further examined for each particle geometry. The shape of the encapsulated particles and the orientation of aggregation are both know to affect signal intensity from aggregates.^{17, 45} The effect of the number of cores encapsulated in silica on particle intensities is examined for each particle geometry by correlative TEM and SERS experiments (Figure 3, Figure S11, Figure S12). The laser power used for correlative Raman and TEM images is limited to 0.25 mW by damage to the TEM grid in Figure 3A – D & E – H. A Raman map (Figure 3A) and the corresponding areas of a TEM image for silica encapsulated bipyramids (Figure 3C), demonstrate many bipyramids are unable to provide detectable signal. The signals that are visible arise predominantly from encapsulated aggregates. The signal intensity of particles is highly variable, as seen by the difference in the intensities of spectra 1 and 2 in Figure 3B, and is likely dependent on the number, orientation, and spacing of the particles (Figure 3D). A Raman map (Figure 3E), sample spectra (Figure 3F), and corresponding areas of a TEM image for silica encapsulated stars (Figure 3G & H), demonstrate many stars appear to be on the verge of detection (when considering the spatial distribution of the Raman map signal brightness, rather than viewing the spectral peaks alone) with aggregates having a brightness on the same order of magnitude as single stars. This is consistent with a linear increase in signal with particle number in the aggregate, demonstrating star signal intensity is less sensitive to aggregation state, as seen previously.¹⁶ Polarization dependent SERS enhancement due to anisotropy in the NP shape also contributes to particle-to-particle variability in nanostar and bipyramid detection on the grid. A Raman map (Figure 3I), sample spectra (Figure 3J), and TEM image displaying the location of particles in the map (Figure 3K) demonstrate GERTs consistently provide detectable signal. Nearly all GERTs measured generate sufficient signal for detection. Given the uniform brightness of the silica encapsulated GERTs, higher magnification TEM images were obtained to show the aggregation state of each particle (Figure S12). Singly encapsulated GERTs provide signal intensities like those of encapsulated aggregates: an encapsulated aggregate is measured in spectrum 1 and a singly encapsulated particle in spectrum 2 (Figure 3J & L). The pixel in the Raman map of highest intensity for each well separated encapsulated particle/aggregate was used to generate a signal intensity distribution for the GERTs (Figure S13). One out of 27 well separated single GERTs gave insufficient signal for detection (not included in the Figure S13 distribution as intensity could not be determined). The undetected particle is smaller than 99% of the GERT size distribution (Figure S14), suggesting a possible link to its lack of SERS intensity. The symmetry of the round etched Au NPs (Figure S1) used as seeds for GERTs preparation result in Raman signals from encapsulated GERTs that are remarkably consistent. Figure S13 shows the GERT aggregates have similar signal intensity to singly encapsulated particles. Signal intensity tends to decrease slightly with increased aggregation. The intensity distribution of signal events in solution is compared to the intensity distribution of signal events on the grid in Figure S15. Measurements in solution provide much better sampling of the GERT population, but the distributions appear qualitatively similar. Adjusting for power density and collection efficiency, the on-grid measurements provide a higher average signal intensity than the solution measurements (Table S4). This is likely due to particles not residing in the center of the focal volume for the full duration of signal acquisitions in solution.

Figure 3.

Raman maps (A), selected spectra (B), TEM images corresponding to Raman map area (C), and zoomed TEM images showing locations of selected spectra (D) for silica encapsulated bipyramids. Raman maps (E), selected spectra (F), TEM images corresponding to Raman map area (G), and zoomed TEM images showing locations of selected spectra (H) for silica encapsulated stars. Raman maps (I), selected spectra (J), TEM images corresponding to Raman map area (K), and zoomed TEM images showing locations of selected spectra (L) for silica encapsulated GERTs. Raman maps are signal intensity integrated over a range including a prominent peak of the reporter molecule. Raman maps are signal intensity integrated over a range including a prominent peak of the reporter molecule. For bipyramids (A) and stars (E), 1058-1099 cm-1 was integrated, for GERTs (I), 1310-1360 cm-1 was integrated. Scale bars are 10 microns.

Single-particle SERS imaging in cells

The consistent, intense, and aggregation insensitive single particle SERS from silica encapsulated GERTs indicates a sufficient system to quantify these particles in Raman maps of cells. Parameters for cell mapping experiments are optimized with some additional preliminary experiments. Confocal volumes for the objective used are measured using a GERT on glass as a point source (Figure S16). The focal volume is approximated by a 3D gaussian with an 1/e2 profile of 1.8 x 1.8 x 4 micrometers in x,y, and z, respectively. Mapping parameters for live cell imaging are based on these parameters so that one particle would not be significantly excited in multiple acquisitions, but that dips in excitation intensity between focal volumes in the point mapping would be minimized. A laser power of 0.3 mW was used to avoid photodamage seen at higher laser powers in measurements on glass. An acquisition time per point of 0.1 s was sufficient to detect particles in the cell maps. Dark field imaging provided a reference for at what depths the cells become out of focus, and for visualization of cell outlines.

The number of particles present in each focal volume is calculated as described in the methods, with plots of intermediate data processing steps in Figure S17. The sorted signal above the threshold is used to determine the average single particle intensity. A processed sample Raman x-y slice and corresponding dark field image are shown in Figure 4A. The full volume map of this result is shown in Figure 4B, where 15 NPs per cell are observed. Correlation with a reference NBT spectrum is mapped in Video S1. Quantification of particles per cell from multiple different maps are examined in Figure 4C. Control cells average less than one predicted particle per cell, representative of noise or low-level false positives. Cells with nanoparticle treatments have 12 +/− 4 particles per cell on average. Two data points were excluded as outliers using a generalized extreme studentized deviate test (Figure S18). spICP-MS was used to validate the number of particles counted in cells by SERS. spICP-MS counted 14 +/− 5 particles per cell, consistent with the SERS particle count.

Figure 4.

(A) An overlay of a dark field image acquired at a depth of 13 microns, the approximate middle of the cells, with the nearest X-Y slice (Z=15) obtained from the SERS volume map used for particle quantification. (B) The 3D Raman volume map showing the number of particles in each focal volume. (C) The bar chart plots particles per cell +/− standard deviations observed from control cells without particles as determined by SERS, cells with particles as determined by SERS, and cells with particles as determined by spICP-MS.

In addition to understanding the particle distribution in cells, we show the agglomeration of silica encapsulated particles in cells can be characterized using spICP-MS. TEM is used to characterize the NPs prior to NP treatment of cells, showing the distribution of GERT sizes (Figure 5A), and the percentage of GERTs that are encapsulated as aggregates (74+/−8% single, 17+/−6% dimer, and 9+/−3% higher order aggregates) in the NP solution. After cell treatment, detachment, and lysis, ICP-MS data of the resulting lysate was acquired in a particle counting mode, where mass of gold detected during each event corresponds to a single particle or multiple particles stuck together. This provides a distribution of particle diameters (Figure 5B). The aggregation and agglomeration of silica encapsulated NPs can be estimated from the particle size distribution. These data can be compared to the aggregation of the GERT sample prior to cell incubation (Figure 5A) to understand the reduction in particle stability for particles incubated with cells.

Figure 5.

(A) Distribution of particle volumes before cell treatment by TEM. (B) distribution of particle volumes after cell treatment by spICP-MS. (C) distributions of particle nitro peak SERS intensities for single particles on the TEM grid, and particles in cells mapped with small or large step sizes.

Particle counting by SERS is also examined in smaller step size Raman mapping as shown in Figure S19, which corre sponds to the same cells mapped in Figure 4B, with 18 particles per cell counted in the higher resolution cell map. Correlation with a reference NBT spectrum is mapped in Figure S19A and Video S2. In this case, sampling is per formed at about half the full width at half maximum of the focal volume. This ensures each particle was near the center of a focal volume, narrowing the distribution of signal that could be achieved from a single particle (Figure 5C). Particles were then detected in multiple focal volumes, as expected by sampling theory. Local maxima of the signal are used to determine signal intensity of single particles, in a way analogous to time series measurements, picking out the focal volume centered nearest to the particle. Adjusting for power density and collection efficiency, these measurements provide a similar single-particle signal intensity as the on-grid measurements (Figure 5C and Table S4). The under-sampling of the large step size map leads to slightly lower average particle signals (Figure 5C), and some particles may be far from the center of any focal volume and remain undetected (Figure S20).

Our work to count particles from their SERS signals can be compared to a previous description of challenges in achieving countable single molecule detection by SERS.⁴⁶ Using Raman tags with high symmetry and loaded with many molecules (Figure S1), we avoid significant influences of hot-spot inhomogeneity, blinking effects, and polarization dependent signals. Plasmon coupling is prevented as demonstrated in Figure 2 and Figure 3. Cell-bound particles can be considered static under our measurement parameters (Figure S19). With control of these confounding factors, appropriately stepped Raman maps can count particles in cell volumes. Importantly, we show both the two SERS methods and the spICP-MS method yield similar results when counting particles in cells, indicating this direct particle counting method can scale to different spatial resolutions. The upper limit of pixel size is determined by the signal intensities of particles farthest from the center of any focal volume, while the minimum useful pixel size is determined by sampling theory. The error of the particle counting determination may be approximated by the error of the time series experiment (Figure S9), the percent of NPs that gave signal on the TEM grid (Figure S12), the control cell particle count rate and spICP-MS determination of number of particles per cell (Figure 4C), and the difference between the large and small step size Raman maps.

Conclusion

SERS imaging shows potential to quantitatively assess gold NP drug carrier or bio-sensor uptake and intracellular movement in living cells. We demonstrate that silica encapsulated GERTs show bright and uniform SERS intensity. Reducing nanoparticle aggregation, and using a particle geometry for which signal intensity is less sensitive to aggregation, we have demonstrated high spatial resolution SERS based quantification of gold NPs in living cells. This method of particle counting is a different approach to digital SERS in cells, enabling scalable and high-resolution particle quantification. This compliments digital SERS approaches that rely on heavily under sampling and determining the percent of spectra with signal, which are useful for analysis of larger volumes such as whole cells.^{23, 47, 48} This method is suitable for applications of particle counting in small volumes/subcellular regions limited by the confocal volume typically used in point mapping. spICP-MS provided insight into NP agglomeration in cells, which may be useful to understand important application dependent parameters related to drug release from mesoporous silica or the effective lifetime of a biosensor in cells.