Quantitative LSPR Imaging for Biosensing with Single Nanostructure Resolution

Abstract

Localized surface plasmon resonance (LSPR) imaging has the potential to map complex spatio-temporal variations in analyte concentration, such as those produced by protein secretions from live cells. A fundamental roadblock to the realization of such applications is the challenge of calibrating a nanoscale sensor for quantitative analysis. Here, we introduce a new, to our knowledge, LSPR imaging and analysis technique that enables the calibration of hundreds of individual gold nanostructures in parallel. The calibration allowed us to map the fractional occupancy of surface-bound receptors at individual nanostructures with nanomolar sensitivity and a temporal resolution of 225 ms. As a demonstration of the technique’s applicability to molecular and cell biology, the calibrated array was used for the quantitative LSPR imaging of anti-c-myc antibodies harvested from a cultured 9E10 hybridoma cell line without the need for further purification or processing.

Introduction

The use of the localized surface plasmon resonance (LSPR) observed in metallic nanostructures for label-free biosensing is relatively recent but its applicability has already proven to be far reaching. Early studies were primarily proof of principle, demonstrating techniques that had the sensitivity to detect the binding of well-characterized receptor-ligand pairs such as streptavidin and biotin (1–6). More applied studies followed, such as the detection of liposomes and Alzheimer’s-related antibodies (7–9). The applications have grown in sophistication such that LSPR has now been applied to plasma-enhanced enzyme-linked immunosorbent assay (ELISA) (10), interferometry-based biosensing (11), cell-based assays (12), and the measurement of protein conformational changes (13) to name a few (14–19).

Advances in instrumentation and analysis now allow for many of these measurements to be made on individual nanostructures, opening the door for new imaging applications in which hundreds or thousands of nanostructures are measured in parallel (10,20,21). Thus, LSPR imaging has the potential to take advantage of each sensor’s nanoscale dimensions to map complex spatio-temporal variations in analyte concentration, such as those encountered in live-cell applications (22,23). In particular, this technique is well suited for measuring protein secretions from individual cells. Such secretions play a critical role, for example, in wound healing (24,25), immune response (26,27), and the building of the extracellular matrix (28). Patch clamp and electrode probe measurements also map out secretions from individual cells but are limited to those molecules that are readily oxidized (i.e., neurotransmitters) (22). As a binding affinity-based technique, LSPR imaging would be able to measure molecular secretions, which are inaccessible to such electrical current-based probes while retaining the advantage of being label free. As such, these nanoplasmonic sensors are potentially the next generation of biophysical instruments for quantitative single-cell secretion measurements.

Before such applications can be realized, fundamental questions regarding the capabilities of LSPR imaging must be answered. First, what are the limits of detection in terms of time, space, and analyte concentration? Here, we demonstrate a new, to our knowledge, LSPR imaging technique capable of detecting antibody concentrations on the order of 1 nM with a spatial resolution determined by the size of a single nanostructure and with a temporal resolution of 225 ms. Second, we asked whether these results could be quantified and interpreted to give meaningful biophysical insight. We show that indeed individual nanostructures can be calibrated to determine the time-dependent fractional occupancy of surface-bound receptors, f(t). The fractional occupancy is a key parameter in the quantitative analysis of analyte binding in that once known it can be used to determine kinetic rate constants if the concentration of analyte is known or, conversely, for determining concentration if the rate constants are known.

The key features of our design are 1), the fabrication of the arrays by electron beam lithography for the production of highly uniform nanostructures, as confirmed by both size and spectral characterizations; 2), the simultaneous measurement of spectra and imagery; and 3), the combination of the spectral and imagery data into an analysis formalism that enables the determination of f(r,t), where r denotes the location on the substrate. It is important to note that the calibration takes place in an imaging, or batch mode, which allows for simultaneous data collection over an entire array of nanostructures. This is essential because the sequential calibration of hundreds or thousands of individual nanostructures is time consuming and impractical.

Using an array of 400 nanostructures, we first demonstrate that our technique allows for the qualitative detection of commercially available anti-c-myc antibodies with single nanostructure resolution using only a charge-coupled device (CCD) camera. Using the same array of nanostructures, we then detail the calibration methodology that enables the quantification of the CCD-based measurements for the determination of f(r,t). As a demonstration of the technique’s applicability to molecular and cell biology, the calibrated array was used for quantitative LSPR imaging of anti-c-myc antibodies that were harvested from the hybridoma cell line 9E10 without the need for their further purification or processing. All experiments were conducted in the same serum-free medium (SFM) used for cell culturing applications.

Experimental Section

Fabrication and functionalization of the nanostructures

The arrays of nanostructures were fabricated on No. 1.5, 25 mm diameter borosilicate glass coverslips by electron-beam lithography (EBL) (20). The bases of the nanostructures were circular in cross section with diameters of 70 ± 5 nm and the heights were 75 ± 2 nm. Each square array consisted of 400 nanostructures (20 × 20) with a spacing of 400 nm between nanostructures. The chip was cleaned by plasma ashing in 5% hydrogen, 95% argon mixture, and then functionalized by immersion in a two-component thiol solution (0.5 mM), consisting of a 3:1 ratio of SH-(CH₂)₈-EG₃-OH (SPO)/SH-(CH₂)₁₁-EG₃-NH₂ (SPN), for 18 h (Prochimia, Sopot, Poland) (29). The SPN component of the self-assembled monolayer (SAM) was first reacted with a 10 mg/mL solution of the heterobifunctional cross-linker sulfo-N-succinimidyl-4-formylbenzamide (Sulfo-S-4FB, Solulink, San Diego, CA) in 100 mM phosphate buffered saline (100 mM PBS, Thermo Scientific, Rockford, IL) and then conjugated to the c-myc peptide (HyNic-c-myc-tag, Solulink) according to the manufacturer’s instructions. Anti-c-myc secreting hybridoma cells (MYC1-9E10.2, ATCC, Manassas, VA) were adapted to growth in SFM containing 1% antibiotic/antimycotic by the sequential, stepwise reduction in fetal bovine serum (BSA) content over a 1-month culture period. The secretion of the 9E10.2 monoclonal antibodies (anti-c-myc) was confirmed and quantified by ELISA using a BSA-c-myc peptide conjugate (∼7 c-myc peptides per BSA) as the immobilized antigen (coated overnight at 5 μg/mL in 100 mM PBS at 4°C). The details of the nanostructure fabrication by EBL, plasma ashing procedure, and application of the SAM layer have been previously described (20). Additional information on the materials used, chip functionalization procedures, hybridoma culturing, and the ELISA assay can be found in the Supporting Material.

Optical setup

CCD-based LSPR imaging, as well as LSPR spectra, was collected in a reflected light geometry with an inverted Zeiss microscope using Koehler illumination, an infinity-corrected 63X, 1.4 numerical aperture oil-immersion objective and crossed polarizers to reduce the background contribution from substrate-scattered light. Imagery and spectra were obtained simultaneously by placing a beam splitter at the output port of the microscope (Fig. 1 a) and a longpass filter with a 593 nm cutoff wavelength was placed before the CCD camera. For the spectral measurements, the focused image of the entire nanostructure array was projected on to the end of a 600 μm diameter optical fiber and the spectra were subsequently measured with a thermoelectrically cooled, CCD-based spectrophotometer (QE65000, Ocean Optics, Dunedin, FL). The spectrophotometer integration time was 4 s. For image acquisition, the focused image of the array was projected onto a thermoelectrically cooled CCD camera with 6.45 × 6.45 μm sized pixels (ORCA R², Hamamatsu, Bridgewater, NJ) and a frame integration time of 225 ms. Details of how the previous setup was optimized for high contrast imaging of the gold nanostructures are described elsewhere (20). Analyte was introduced under continuous flow conditions using a custom-made microfluidic cell at a flow rate of 10 μL/min. The microscope stage was equipped with a temperature-controlled insert, which kept the stage temperature and optical light train at 28.0 ± 0.04°C (PeCon GmbH, Erbach, Germany). Under these conditions, the drifts in the x, y, and z directions were <3 nm/min. For data analysis, all frames were aligned in x and y using a commercially available image processing alignment algorithm (Axiovision, Zeiss, Thornwood, NY).

Figure 1.

(a) Imaging and spectroscopy setup in which P1 and P2 are crossed polarizers, BS is a 50/50 beam splitter and LP is a longpass filter with a 593 nm cutoff. The inset shows an atomic force microscopy scan of a witness array fabricated on the same chip. (b) Two spectra from a specific binding study in which 200 nM of anti-c-myc was introduced over the c-myc functionalized array (10 μL/min). The f_initial spectrum (black) was taken before the anti-c-myc was introduced and the f_final spectrum (red) after 1 h of exposure. (c) Normalized spectra from 18 individual nanostructures taken in air on a separate darkfield microspectroscopy setup. Individual spectra are superposed (gray curves) and compared to the ensemble average (black curve).

Results and Discussion

Qualitative biosensing with single nanostructures

The imagery as well as LSPR spectra were simultaneously acquired by passing the reflected light through a 50/50 beam splitter as shown in Fig. 1 a. Fig. 1 b shows two spectra from a specific binding study in which 200 nM of anti-c-myc was introduced over a c-myc functionalized array at a flow rate of 10 μL/min. The f_initial spectrum (black) was taken before the anti-c-myc was introduced and the f_final spectrum (red) was taken after 1 h of exposure. The characteristic red shift of the peak position and the corresponding increase in counts from 605 to 750 nm are indicative of a local change in the dielectric constant of the medium caused by the specific binding of the anti-c-myc antibody (1,13). Although monitoring the peak shift is currently the most common method of detecting analyte binding, we have shown (20) that it is the increase in the scattered intensity over such a large portion of the resonance spectrum that allows for the fractional occupancy of the array to be determined spectroscopically, while simultaneously enabling LSPR imaging via the CCD camera.

The calibration of hundreds of nanostructures in batch mode requires that the spectral properties of individual nanostructures closely resemble that of the array ensemble average. Topological studies of our nanostructures by atomic force microscopy revealed small variations in nanostructure shape, particularly with regards to the surface roughness at the tips of the nanostructures (Fig. 1 a, inset). To investigate the corresponding spectral variations among individual nanostructures, a separate setup designed for single-nanostructure spectroscopy (based on darkfield microspectroscopy) was used for characterizing a witness row of 18 nanostructures fabricated on the same chip as the array used for biosensing. The data were collected in air, which accounts for the fact that the resonance peaks are blue shifted relative to those in Fig. 1 b, which were collected in SFM. Further details of this technique are given in the Supporting Material. The results, summarized in Fig. 1 c, show that although small variabilities in the shape, amplitude, and resonant wavelength could be discerned, the spectrum of nearly every nanostructure fell close to the ensemble average. In the CCD imagery, these distributions can be manifested as a distribution of intensities, as shown in the Fig. 2 insets, which have been contrast enhanced to highlight the intensity variations.

Figure 2.

LSPR imaging time-course measurement from a 200 nM anti-c-myc specific binding study. (a) Mean intensity for the entire array (square light blue ROI, 84 × 84 pixels) (b) Mean intensity of a single nanostructure (square red ROI, 4 × 4 pixels). (c) Comparison of the normalized responses of the whole array and the single nanostructure. Also plotted in (a) and (b) are the results of a drift study that preceded the introduction of analyte (black circles). The inset images in (a) and (b) shows a contrast-enhanced CCD image of the 8 × 8 μm array that highlights the variations in nanostructure intensities. All studies were conducted in SFM at a flow rate of 10 μL/min.

In Fig. 2, we detail the time course of a 200 nM anti-c-myc specific binding study as measured by LSPR imagery and demonstrate a straightforward image analysis technique for qualitatively monitoring the kinetics down to the single nanostructure. Fig. 2 a shows the enhanced counts from binding for the entire array (84 × 84 pixels) as calculated from the mean intensity of the pixels bounded within the light blue region of interest (ROI) square:

$$I\left(r_i,t_n\right)=\frac{1}{m_i}\sum _{\overrightarrow{x}\in r_i}{I}_{image}\left(\overrightarrow{x},t_n\right),$$ (1)

where $m_i$ is the number of pixels in the ROI denoted as r_i and t_n is the time point. Also shown is a drift study that preceded the introduction of analyte (black circles) in which SFM flowed over the array for 30 min at 10 μL/min. In contrast to simple buffers, SFM typically contains anywhere from 50 to 1000 mg/L of additional proteins such as albumin, transferrin, and insulin, which can potentially biofoul the sensors. Despite this presence, our measurements showed minimal drift and sensitivity to analyte that was retained. Additional studies showing minimal nonspecific binding studies between the antibody and the SAM-functionalized surface were conducted using a Bio-Rad XPR36 surface plasmon resonance instrument and are presented in the Supporting Material. In these studies, we compared the signal from the nonspecific binding of c-myc blocked anti-c-myc to the specific binding signal given by unmodified anti-c-myc (control). The blocked anti-c-myc gave a signal of 1% or less compared to its corresponding control study (Fig. S3), thus showing minimal nonspecific binding.

Fig. 2 b plots the same two experiments but with the ROI now composed of only a single nanostructure, as selected by a 4 × 4 pixel (410 × 410 nm) square ROI shown in red near the center of the array. The relative response of the nanostructure is compared directly to that of the entire array in Fig. 2 c by plotting the normalized counts:

$$\frac{[I\left(r_i,t_n\right)-I\left(r_i,t_o\right)]}{[I\left(r_i,t_f\right)-I\left(r_i,t_o\right)]},$$

where $I\left(r_i,t_o\right)$ is the average of the first 20 time points and $I\left(r_i,t_f\right)$ is the average of the last 20 time points. This same straightforward ROI approach can be applied to any nanostructure in the array, giving 400 independent and label-free nanosensors within the 8 × 8 μm area. A separate specific binding experiment in which 70 nM of anti-c-myc was introduced over c-myc functionalized gold nanostructures is shown in Fig. 3. This figure highlights the distribution in the normalized response of individual nanostructures about that of the entire array by plotting the response of five representative nanostructures within the array. The five nanostructures are indicated in the inset CCD image by a red box and labeled A–E from left to right.

Figure 3.

LSPR imaging time-course measurement from a 70 nM anti-c-myc specific binding study. Time course plots from five representative nanostructures coplotted with the response of the entire array. The insets show a contrast-enhanced CCD image of the array in which the five nanostructures are outlined by the red box. The nanostructures are labeled A–E from left to right in the image. The study was conducted in SFM at a flow rate of 10 μL/min.

Fig. 2 c and Fig. 3 highlight a key observation in this work, namely, that the response within the nanoscale ROIs are typically in excellent agreement with that of the entire array. Thus, if the entire array can be calibrated for the determination of the fractional occupancy, f(t), this homogeneous response in principle can be used to simultaneously calibrate the individual nanostructures. This at first can seem surprising given the likelihood that the surface-bound receptors on any given nanostructure will have an inhomogeneous spatial distribution and response to analyte (30,31). It is reasonable, however, given that both the theory of random sequence adsorption (32) and experimental estimates using similarly functionalized gold surfaces (10,33) are in agreement that nanostructures of this size can accommodate hundreds of proteins, thus averaging out the effect of such inhomogeneities. We now discuss how this qualitative observation can be expanded into a data analysis formalism, which allows for the quantitative determination of the fractional occupancy within the imagery.

Quantitative biosensing with single nanostructures

We have previously reported (20) a methodology that allows for the determination of the fractional occupancy of the entire array from the spectral data, denoted here as $f_S\left(t\right)$. In short, the number of counts at wavelength, $\lambda$, accumulated by the spectrometer during the time interval t_n can be written as a linear response model with Poisson counting noise, η_Poisson, in terms of the $f_S$ for a specifically bound monolayer perturbing the localized surface plasmon resonance:

$$N_{\lambda }\left(t_n\right)=g\left(t_n\right)·\left[a_{\lambda }{f}_S\left(t_n\right)+b_{\lambda }\right]+{\eta }_{Poisson}.$$ (2)

The model parameters $a_{\lambda }$ and $b_{\lambda }$ represent the wavelength-dependent dielectric response caused by the bound analyte and the initial background of the LSPR array, respectively. The overall time-dependent coefficient, $g\left(t_n\right)$, is initially set to one at the beginning of the experiment but can account for drift and jump processes that cause variations in the scattered light intensity with no wavelength dependence. Generally, $g\left(t_n\right)\approx 1$ and can be ignored in most situations. For a given nanostructure array and experimental conditions the $b_{\lambda }$ is determined when no analyte is present at the beginning of the experiment and the $a_{\lambda }$ is determined by a saturating injection of a known concentration of analyte at the end of the experiment. The time-dependent functions $g\left(t_n\right)$ and $f_S\left(t_n\right)$ can be determined by nonlinear regression within a Poisson noise model for the counts at each time interval, $t_n$.

The question of interest in this study is whether the imagery of single nanostructures can encode the same information as the ensemble measurement of the entire array as measured spectroscopically. In other words, if the array is subjected to a uniform spatial distribution of analyte, can we calibrate the optical response of nanostructure-sized ROIs to that of the ensemble fractional occupancy found by spectroscopy? It is not obvious that this should be possible because the image sums up all the spectral information and individual nanostructures can be subject to stochastic processes that average away when the entire array is used. To assess whether the imagery of single nanostructures can capture the fractional occupancy information found in whole-array spectroscopy, we propose a simple generative model for how the image data is formed, similar to that used in analyzing the spectroscopy data:

$$I\left(r_i,t_n\right)=A\left(r_i\right)·f_I\left(r_i,t_n\right)+B\left(r_i\right)+{\eta }_{Poisson}.$$ (3)

Here, the model parameters $a_{\lambda }$ and $b_{\lambda }$ are analogously represented by the spatially dependent parameters $A\left(r_i\right)$ and $B\left(r_i\right)$ for the determination of the fractional occupancy from the image data, $f_I\left(r_i,t_n\right)$. Because the size of the array is small compared to the diffusion length of the analyte over the exposure time of the CCD camera, the concentration is effectively uniform. This allows us to determine $A\left(r_i\right)$ and $B\left(r_i\right)$ via multivariate linear regression (see the Supporting Material for details) by setting $f_I\left(r_i,t_n\right)=f_S\left(t_n\right)$, thus, calibrating the entire array via imagery. Once the array is calibrated, inhomogeneous fractional occupancy can be estimated as

$${\hat{f}}_I\left(r_i,t_n\right)=\frac{I\left(r_i,t_n\right)-B\left(r_i\right)}{A\left(r_i\right)}.$$ (4)

To determine if this relatively simple treatment of the imagery data is effective, we can calculate the variance in the local response, ${\hat{f}}_I\left(r_i,t_n\right)$, of the nanostructure array from $f_S\left(t_n\right)$ over the ROIs:

$${\sigma }^2\left(r_i\right)=\frac{1}{N}\sum _{n=1}^N{\left|{\hat{f}}_I\left(r_i,t_n\right)-f_S\left(t_n\right)\right|}^2.$$ (5)

The resulting image map will show what parts of the array are capable of being calibrated for the determination of the local fractional occupancy.

As an example of the applicability of this approach to molecular and cell biology, we demonstrate the determination of fractional occupancy versus time at the nanoscale to the secreted antibodies contained within the supernatant of cultured MYC1-9E10.2 hybridoma cells. The harvested antibodies experiment was conducted by simply centrifuging the cells at 3000 rpm for 5 min, collecting the supernatant and applying that solution to the nanostructures via the microfluidic setup at a flow rate of 10 μL/min. The concentration of secreted anti-c-myc antibodies in the supernatant was independently determined by ELISA to be 9 nM. The nanostructure calibration was conducted as described previously by introducing a known concentration of commercial anti-c-myc (250 nM) over the nanostructures immediately following the harvested antibody experiment.

Fig. 4 displays the $B\left(r_i\right)$ and $A\left(r_i\right)$ response maps of a 12.4 × 12.4 μm area centered about the array for every pixel in the image (ROI: 1 × 1 pixel). A grid of 4 × 4 pixel (410 × 410 nm) squares has been superimposed over each map to demarcate the location of each nanostructure from the imagery. In Fig. 4 a, the dark red regions of the coefficient map for the background term, $B\left(r_i\right)$, are highly correlated with the brightest nanostructures in the CCD image (Fig. 2, inset), as is to be expected for the background contribution to the fit. The coefficient map for the linear response term, $A\left(r_i\right)$, is shown in Fig. 4 b. Again the strongest positive responses are located within the squares of the grid and thus are correlated with the locations of the nanostructures. We then repeated the optimization using a sliding ROI window of the same size as the grid (4 × 4 pixels), which had the advantage of closely approximating the size of the diffraction-limited image of the nanostructures. In addition, this larger ROI averaged out the presence of a slight drift of ∼2 pixels (1.7 nm/min), which occurred over the course of the 2 h run.

Figure 4.

(a) Background response map, $B\left(r_i\right)$, and (b) linear response map, $A\left(r_i\right)$, of the array for the anti-c-myc harvested antibody study, calculated for every pixel in the image (ROI: 1 × 1 pixel). The calibration was conducted immediately following the harvested antibody study by injecting 250 nM of commercial anti-c-myc in SFM. A grid of 4 × 4 pixel squares has been superimposed to demarcate the location of the nanostructures in the imagery.

The results of the 4 × 4 pixel sliding ROI calibration and analysis are summarized in Fig. 5. Fig. 5 a shows the error estimate map of ${\sigma }_i$, which presents the deviations between $f_I$ and $f_S$ as a grayscale map. The shape of the array is clearly reproduced on the map because it is only within the array area that $f_I$ and $f_S$ are within reasonable agreement. In fact, by setting the scale of ${\sigma }_i$ on the map to be between 0 and 0.1 we show that over 75% of the area encompassed by the array can be well calibrated with the spectroscopically determined fractional occupancy. To illustrate the deviations between $f_I$ and $f_S$ associated with a given ${\sigma }_i$, the data from specific ROIs are plotted in Figs. 5, b–e, with ${\sigma }_i=0.03$, ${\sigma }_i=0.05$, ${\sigma }_i=0.10$, and ${\sigma }_i=0.42$, respectively. The color of the ROI data points are matched with that of the ROI square label on the response map and the vertical dashed line indicates the end of the harvested anti-c-myc antibody run, at which point 250 nM of commercial anti-c-myc was injected. There is excellent agreement between the two for ${\sigma }_i\le 0.05$ (red and green ROIs), which deteriorates somewhat at ${\sigma }_i=0.1$ (dark blue ROI), whereas there is no statistically meaningful correlations for the light blue ROI located outside of the array.

Figure 5.

(a) Error estimate map of ${\sigma }_i$ for the anti-c-myc harvested antibody study, as calculated by a 4 × 4 pixel (410 × 410 nm) sliding ROI window, plotted for deviations between $f_I$ and $f_S$ ranging from 0 to 0.1. Four plots of select ROIs are shown in (b–e) to illustrate the deviations associated with a given ${\sigma }_i$. The color of the ROI data points are matched with that of the ROI square label on the response map with (b) ${\sigma }_i=0.03$ (c) ${\sigma }_i=0.05$, (d) ${\sigma }_i=0.10$, and (e) ${\sigma }_i=0.42$. The vertical dashed line indicates when the 250 nM of commercial anti-c-myc was injected.

Conclusions

The results in Fig. 5 show that for the majority (over 75%) of the array, $f_I$ can be determined to within 10% of $f_S$ using ROIs of a similar size to that of the diffraction-limited image of each nanostructure. Even with the reproducibility of fabricating by EBL, however, it was not possible to calibrate the entire array area to within this range of error. This is not an impediment though because a great advantage to our approach is that those ROIs, which do not calibrate to within a set specification can simply be ignored, whereas those that do can be used for the quantitative analysis. As such, our LSPR imaging technique allows for label-free and quantitative characterization of cell supernatant with minimal preparation and nanomolar sensitivity, using hundreds of nanostructures independently calibrated to within the user’s specification. In its current form, this technology sets the stage for future applications in high density proteomics arrays as well as for imaging analyte concentration gradients in complex live cell environments. Future work will focus on implementing a nonprotein-based calibration procedure by using more easily replaced mixtures of glycerol and water (34) and the inclusion of nonlinear terms in the analysis.