Photothermal raster image correlation spectroscopy of gold nanoparticles in solution and on live cells

Raster image correlation spectroscopy (RICS) measures the diffusion of fluorescently labelled molecules from stacks of confocal microscopy images by analysing correlations within the image. RICS enables the observation of a greater and, thus, more representative area of a biological system as compared to other single molecule approaches. Photothermal microscopy of gold nanoparticles allows long-term imaging of the same labelled molecules without photobleaching. Here, we implement RICS analysis on a photothermal microscope. The imaging of single gold nanoparticles at pixel dwell times short enough for RICS (60 μs) with a piezo-driven photothermal heterodyne microscope is demonstrated (photothermal raster image correlation spectroscopy, PhRICS). As a proof of principle, PhRICS is used to measure the diffusion coefficient of gold nanoparticles in glycerol : water solutions. The diffusion coefficients of the nanoparticles measured by PhRICS are consistent with their size, determined by transmission electron microscopy. PhRICS was then used to probe the diffusion speed of gold nanoparticle-labelled fibroblast growth factor 2 (FGF2) bound to heparan sulfate in the pericellular matrix of live fibroblast cells. The data are consistent with previous single nanoparticle tracking studies of the diffusion of FGF2 on these cells. Importantly, the data reveal faster FGF2 movement, previously inaccessible by photothermal tracking, and suggest that inhomogeneity in the distribution of bound FGF2 is dynamic.


Introduction
The direct observation of individual molecules by optical microscopy [1] in living cells [2][3][4][5][6]  spatial distribution of nanoparticles and extracting diffusion dynamics at a wide range of timescales. We present, as a proof of principle, measurements made on gold nanoparticles in solution. The method is then applied to probe the diffusion of a gold nanoparticle-labelled protein, fibroblast growth factor 2 (FGF2), in the pericellular matrix of live fibroblast cells.

Material and methods
2.1. Preparation of single 8.8 nm gold nanoparticle samples using poly-L-lysine A rectangular coverslip (22 × 44 mm, Leica Surgipath, Leica Microsystems, Milton Keynes, UK) was incubated with poly-L-lysine solution (MW 70 000-15 0000 Da; 0.01% (v/v), Sigma Aldrich, Gillingham, UK) for 40 min to adhere. The coverslip was washed three times with Milli-Q ultrapure water. A 1-2 pM solution of 8.8 nm gold nanoparticles (nominally 10 nm, BBI Solutions Ltd., Cardiff, UK) was then added to the coverslip and left for another 40 min. The coverslip was washed again three times with Milli-Q water and mounted in 80% glycerol (glycerol: Fisher Scientific, Leicester, UK). A second coverslip was added on top with a Parafilm 'M' spacer, and the chamber sealed by melting the parafilm with a soldering iron.

Preparation of 8.8 nm gold nanoparticle samples in glycerol : water
Nanoparticles (as above) were diluted in the appropriate glycerol : water mixture (indicated in text). The solution was introduced by capillarity into a homemade fluidic channel formed with Parafilm 'M' between two coverslips.

Nanoparticle functionalization with FGF proteins (1 : 1)
The synthesis and subsequent conjugation of FGF2 protein to maleimide nanoparticles was previously described by Nieves et al. [44]. Briefly, maleimide functionalized gold nanoparticles were incubated with a 35 times molar excess of recombinant FGF2 (produced according to Ke et al. [45]) for 3 h at room temperature. The conjugation of FGF2 to the maleimide functionalized gold nanoparticle proceeds via thiol-Michael addition with the thiol side chain of the exposed cysteine (C-95) at the surface of the FGF2 (FGF2-NP). The mixture was then centrifuged for 80 min at 13 000 g and 4 • C. The supernatant containing the excess FGF2 was removed and the particles were resuspended in 200 µl of 1× PBS Tween-20 0.005% (v/v). The centrifugation was performed four times in order to remove all excess FGF2, and the pellet was finally resuspended in 1× PBS Tween-20 0.005% (v/v).

PHI set-up
All images were acquired using a homebuilt photothermal confocal microscope (a schematic is presented in figure 1). The excitation laser (523 nm; frequency-doubled ND : YAG, Ventus Laser Quantum, Germany) was modulated at a frequency of 459.5 kHz using an acousto-optical modulator (Isomet Corporation, UK). This excitation beam was 'cleaned' using a spatial filter. This was done to remove ellipticity generated by the modulation and generates a Gaussian beam profile. Excitation laser power for all imaging was 2 mW. The excitation beam was overlaid with a non-resonant probe laser (633 nm, 10 mW; JDS Uniphase Corporation) via a cold mirror (ThorLabs). The superimposed beams were focused onto the sample via an oil immersion objective (Zeiss Plan-Apochromat 63×, numerical aperture (NA) 1.4). The sample was placed on a piezo scanning stage (MCL502385, MadCity Labs, Madison, WI, USA), which allows movement of the sample in three dimensions (x, y and z) over the fixed laser spot. Scanning by a piezoelectric stage driver (MCL NanoDrive 85, USA) under the control of the Nanonis RC4 module and Nanonis program (Specs-Zurich, Zurich, Switzerland) was used to move the sample over the fixed laser spot. The transmitted and forward scattered light was collected by a second oil objective (Zeiss NEOFLUAR 40×, NA 1.3) and passed through a red-pass filter (ThorLabs) to block the excitation laser. The red component was focused upon one photodiode of the balanced photoreceiver (Model 2107 10 MHz adjustable photoreceiver, New Focus, USA). A lock-in amplifier (DSP 7260, Signal Recovery, Oak Ridge, TN, USA) was used to identify the scattered component of the probe beam that corresponds to the modulation frequency or 'beat-note' (i.e. 459.5 kHz). A Nanonis SC4 Acquisition Module (Specs-Zurich) was used for signal acquisition. The signal was averaged (pixel dwell time indicated in text) and a greyscale pixel value was generated. The values along a scan path, i.e. photothermal signal intensity at each position, were then converted into a photothermal image. The images were saved in a .sxm format.

Acquisition parameters for PhRICS
The sample was raster scanned across the detection volume multiple times and the images saved. The images were 160 × 128 pixels with a pixel size of 50 nm (thus, the ROI was 8.0 × 6.4 µm) and a pixel dwell time of 60 µs. A rectangular image was taken to allow settling of the stage at the beginning of the scan line. In addition, the return of the stage to the next scan line after acquisition of the previous occurred at the same speed of acquisition. the bottom objective was focused at the coverslip. For imaging of diffusion in solution, it was focused 8-10 µm into solution, whereas for live cell experiments it was focused 1 µm above the coverslip.

Lateral dimension of the detection volume (1/e 2 radius)
Images of 8.8 nm gold nanoparticles immobilized on poly-L-lysine were taken using the PhRICS imaging parameters. The images were converted from .sxm to .txt files using GWYDDION (http://gwyddion.net/) [47]. The images were then processed using IMAGEJ (http://imagej.nih.gov/ij/) to give the final 128 × 128 pixel PhRICS image and saved as a 16-bit tiff image. Line profiles of the peaks were fitted with a Gaussian curve using ORIGIN 8.6. From the fit, the full width at half maximum (FWHM) was derived and the 1/e 2 radius calculated using the following equation; where w is the full width of the volume at 1/e 2 .

Analysis of PhRICS data
Images were first converted from .sxm to .txt files using GWYDDION. The images were then were cropped, i.e. 32 pixels removed at the beginning of each scan line, using IMAGEJ to give the final 128 × 128 pixel PhRICS image and saved as a 16-bit tiff image sequence. All the images that underpin the results presented here are available through Figshare (see Data accessibility). The sequence was then loaded into the RICS module of the SIMFCS software [48]. Firstly, a moving average of the image sequence (a window of 10 images) was taken. The moving average was subtracted from the image sequence to remove any immobile features from the images, i.e. signal that persists at the same position throughout the image sequence (described by Digman et al. [25]), and the spatial autocorrelation function applied. The surface of the resulting autocorrelation image was then fitted using the SIMFCS program [48] with the following fixed parameters: time per line (19.2 ms), 1/e 2 radius of the detection volume (0.240 µm) and pixel size (50 nm). The fit resulted in extraction of the diffusion coefficient for mobile features in the image sequence.

Imaging parameters for PhRICS
Typically, PHI of gold nanoparticles uses pixel dwell times of the order of 1-10 ms [18,[49][50][51]. Correlation spectroscopy both of fluorescent and non-fluorescent objects [25,[39][40][41] indicate that shorter dwell times, of the order of microseconds, are required. Thus, we first evaluated whether such pixel dwell times would still provide sufficient signal-to-noise ratio (SNR). Encouragingly, PHI imaging with a pixel dwell time of 80 µs has been achieved recently using a galvanometric laser scanning system [52]. In most PHI systems, including our own, a piezoelectric stage is used for scanning [38,50,53]. Thus, the sample is moved over a fixed laser spot, as opposed to the galvanometric method that moves the laser spot over the sample. While piezo-scanning simplifies laser alignment, speed is limited by stage stability and response time. Photothermal images of single 8.8 nm gold nanoparticles immobilized on poly-L-lysine acquired using pixel dwell times of 1 ms and 60 µs were acquired ( figure 2a,b). Single 8.8 nm gold nanoparticles were detected at both 1 ms and 60 µs dwell times ( figure 2a,b). However, for 60 µs pixel dwell times the imaging area had to be extended in the scan direction (to 8.0 µm and cropped to show the same area, as described §2.8) to allow for settling of the stage after every scan line. Comparison of the photothermal signal profile of the same nanoparticle (figure 2c) showed that the peak pixel value is very similar, but the SNR goes from over 200 to approximately 10 when the dwell time is reduced from 1 ms to 60 µs (SNR was calculated by dividing the average photothermal signal of the peak by the standard deviation of the background noise). This SNR is high enough to be used for single nanoparticle tracking [34], thus, the acquisition of rapid raster scan images with a piezo-stage PHI microscope for RICS analysis is possible. RICS fitting of spatial correlations requires knowledge of the 1/e 2 radius of the detection volume. Therefore, images of single gold nanoparticles were acquired with the same parameters, as in figure 2b, and the FWHM of the curve derived by Gaussian fitting. The 1/e 2 was then calculated from the FWHM and was found to be 240 ± 24 nm (n = 23).

Determination of the diffusion coefficient of gold nanoparticles in solution by PhRICS
The rationale for the RICS methodology has been extensively discussed [25,29,54], and the ability to extract information on diffusion dynamics from raster scan images has demonstrated its power for probing the movement of biomolecules in live cells [29,30]. Initially, the technique was verified by measurements performed upon samples of molecules/objects of known diffusion coefficients as a validation of the method. For example, RICS was able to effectively measure the diffusion speed and concentration of fluorescein and monomeric EGFP in solution, which compared favourably with FCS measurements of the same sample [29,55]. To determine whether accurate diffusion coefficients of nanoparticles can be extracted by PhRICS, 8.8 nm gold nanoparticles were suspended in glycerol : water (v/v) of different viscosities (20%, 50% and 80%, all v/v). Similar conditions have been used previously to establish autocorrelation spectroscopy of gold nanoparticles [39,41]. The experimental workflow for PhRICS is presented in figure 3. Exemplar images and the resulting spatial correlation and fitting are shown in figure 4. The diffusion of nanoparticles through the detection volume during the raster scan results in a characteristic 'streaking' pattern, which has previously been observed for mobile nanoparticle-labelled molecules [18,34]. As glycerol concentration (and, therefore, viscosity) was increased (figure 4a-c), the number and length of streaks observed per line decreased, and the persistence of gold nanoparticles from one scan line to the next became more common, which is consistent with a reduction in diffusion speed. The correlation properties described qualitatively above are reflected in the spatial correlations for each glycerol concentration (figure 4d-f ), which decrease in length along the x-direction and begin to become broader in the y-direction. PhRICS measurements were performed on three different samples for each glycerol concentration and the extracted diffusion coefficients from the fitting (figure 4g-i) were averaged (table 1;   glycerol (b,e,h) and 80% glycerol (c,f ,i) were imaged by PhRICS (a-c) and the corresponding average spatial correlation for nanoparticles was calculated (d-f ), as well as fitting result for each spatial correlation (bottom surface; g-i). The differences between the fitting model and data for g-i are also plotted (top surface).  32.2 K, respectively. The measured D values by PhRICS for each glycerol concentration were in good agreement with the range expected, and it appears that the diffusion of 8.8 nm gold nanoparticles is not thermally enhanced at this range of surface temperatures. The corresponding hydrodynamic diameter of the nanoparticles observed for 20%, 50% and 80% glycerol are 9.5 ± 1.0 nm, 9.6 ± 0.9 nm and 9.2 ± 0.7 nm. The slight increase in diameter as compared from TEM value might result from the observation of clusters of few nanoparticles during the PhRICS acquisition. Owing to the intensity of these clusters being higher than the monodisperse population, it is possible that they can contribute to the final average correlation function, as has been reported before for ICS measurements [56].

Probing the diffusion of gold nanoparticle-labelled FGF2 proteins on live fibroblast cells
As an exemplar application of PhRICS, we measured the diffusion of FGF2 protein in the extracellular matrix of live rat mammary fibroblast cells (Rama 27). This system has been analysed previously by photothermal tracking [18], and FGF2 was observed to have a heterogeneous diffusion behaviour, which was attributed to the heterogeneous distribution of its binding sites on the glycosaminoglycan heparan sulfate in the pericellular matrix. Recombinant FGF2 protein was labelled with gold nanoparticles (FGF2-NP), at a stoichiometry of one FGF2 protein to one nanoparticle, as described previously [44]. When Rama 27 cells are incubated with control nanoparticles bearing no FGF2 (described in §2.3), i.e. nanoparticles that do not possess the reactive group required for FGF2 conjugation, no photothermal signal from the nanoparticles is seen (figure 5a). The only weak signal is that from the mitochondria present within the cell, which are detected without the need for labelling in PHI [18,57]. However, when incubated with 600 pM FGF2-NP, a very strong signal is detected ( figure 5b). This signal is localized to the pericellular matrix of the Rama 27 cells, as FGF2 protein binds to heparan sulfate [18]. Stacks of PhRICS images were acquired in different areas of the cell membrane, away from the perinuclear region where mitochondrial signal is often high (boxes 1-5 in figure 5b). For all the areas observed in figure 5b, diffusion of FGF2-NPs was observed and the deduced diffusion coefficient for each box is presented in table 2. Here, it can be seen that there is indeed variability in the measured diffusion coefficient depending on the area being probed (table 2; reduced χ 2 -values for the fits range from 0.000009 to 0.000021). The diffusion coefficients range from 0.03 to 0.28 µm 2 s −1 . The PhRICS data and analysis from box 5 are presented in figure 6, as an example. Firstly, FGF2-NPs movement during image acquisition is indicated by the streaking pattern observed (figure 6a), which is consistent with previous studies [18]. There are also FGF2-NPs that do not move within a single image, as indicated by a circular intensity profile. These would not contribute to the RICS analysis, as they are treated as an 'immobile' feature and removed by the moving average that is subtracted from all the images in the stacks before applying the spatial correlation. These features are seen in all the areas observed in figure 5, with distribution being variable in those areas. The spatial correlation of the image stack acquired from this area (figure 6b) was notably broadened in the y-direction, as opposed to the other spatial correlations shown within this    4d,e and i). This is consistent with increased correlation from one scan line to another, which could be a reflection of the confined movement observed previously of FGF2 in the pericellular matrix [18]. Fitting of the spatial correlation from box 5 yielded a diffusion coefficient of 0.16 µm 2 s −1 (figure 6c and previously, photothermal tracking has a time resolution of the order of milliseconds and cannot access very fast movement [25]. Indeed, in the data presented in the electronic supplementary material, table S2 from Duchesne et al. [18], only approximately 4% of the mobile fraction of FGF2 has a diffusion coefficient comparable to the values measured here by PhRICS (boxes 1, 2, 4 and 5, table 2). This suggests that the number of 'fast' FGF2-NPs is considerably greater than previously measured by photothermal tracking [18].

Conclusion
The development of a new photothermal imaging technique for probing the diffusion dynamics of gold nanoparticles, PhRICS, is described. Imaging of single gold nanoparticles at short pixel dwell times (60 µs) with an SNR of 10 is achieved with a piezo-scanning PHI microscope. The ability to take images at this speed enables RICS analysis. Imaging of nanoparticles of a known size in mediums of varying viscosity confirms that quantitative measurements of diffusion coefficients across a wide dynamic range can be obtained. The method was then applied in the setting of a live cell, whereby the diffusion of gold nanoparticle-labelled FGF2 protein was explored. PhRICS was able to effectively extract diffusion measurements of FGF2 on live cells, showing that the fraction of mobile FGF2 in the pericellular matrix of fibroblasts is likely to be greater than previously determined. This has important repercussions for our understanding of cell communication and demonstrates the potential for the investigation of the movement and diffusion of proteins labelled by gold nanoparticles in live cells by PhRICS.
Data accessibility. All the datasets supporting this article are available at Figshare using the following doi addresses.