Plasmonic metallic nanoparticles are commonly used in (bio-)sensing applications because their localized surface plasmon resonance is highly sensitive to changes in the environment. Although optical detection of scattered light from single particles provides a straightforward means of detection, the two-photon luminescence (TPL) of single gold nanorods (GNRs) has the potential to increase the sensitivity due to the large anti-Stokes shift and the non-linear excitation mechanism. However, two-photon microscopy and spectroscopy are restricted in bandwidth and have been limited by the thermal stability of GNRs. Here, we used a scanning multi-focal microscope to simultaneously measure the two-photon excitation spectra of hundreds of individual GNRs with sub-nanometer accuracy. By keeping the excitation power under the melting threshold, we show that GNRs were stable in intensity and spectrum for more than 30 min, demonstrating the absence of thermal reshaping. Spectra featured a signal-to-noise ratio of >10 and a plasmon peak width of typically 30 nm. Changes in the refractive index of the medium of less than 0.04, corresponding to a change in surface plasmon resonance of 8 nm, could be readily measured and over longer periods. We used this enhanced spectral sensitivity to measure the presence of neutravidin, exploring the potential of TPL spectroscopy of single GNRs for enhanced plasmonic sensing.

The unique optical properties of gold nanorods (GNRs) have found multiple applications in research and industry. A weak plasmon-induced optical luminescence signal of bulk gold was first reported by Mooradian.1 Mohamed et al. later observed that the luminescence of gold could be enhanced by >106 when exciting rod-shaped gold nanoparticles in resonance with their surface plasmon wavelength, typically in the near-infrared (NIR) part of the spectrum, increasing the quantum yield to ∼10−4.2 This field enhancement yields a signal intensity similar to that of quantum dots,3 being bright enough for straightforward detection of individual particles. GNRs, however, do not blink and gold provides excellent biocompatibility, making them more appealing for biological applications. In combination with the reduced absorbance and scattering of NIR light in vivo, GNRs form attractive labels for in vivo imaging,3,4 but GNRs are also used in cancer therapy, fluorescence enhancement, and bio-sensing.5–13 

The longitudinal surface plasmon resonance (SPR) wavelength of a GNR is almost independent of the rod diameter for diameters < 50 nm, but scales linearly with its aspect ratio,14 yielding resonances between 600 and 1000 nm for aspect ratios between 2 and 5. Because the electric field of the GNR is confined to several tens of nanometers from the tips of the GNR,12 it creates a very local excitation volume, down to several zeptoliter, which can be readily exploited for single-molecule bio-sensors, even in high concentration solutions.15–17 The distinctive merits of single-molecule detection, i.e., avoiding population and temporal averaging, provide an attractive new range of applications of GNRs.18 

Metallic nanoparticles can be used for (bio-)sensing in two manners, as reviewed by Taylor and Zijlstra.18 First, the intensity of fluorescently labeled analytes is highly enhanced when entering the proximity of the nanoparticle by increases of both the excitation rate and the quantum yield of the fluorophore. This fluorescence enhancement requires sufficient spectral overlap between the plasmon and the spectrum of the label. Various examples of single-molecule detection have been reported.15,19 When labeling of analytes is not desired, GNRs can also be used as label-free detectors. In this mode, a change in the dielectric properties of the surroundings, as analytes enter the proximity of a GNR, induces a shift in the plasmon resonance.13,16,20 This plasmon shift can amount up to several nanometers but is highly dependent on the nanoparticle geometry, binding position, and size of the analyte.

Fluorescence enhancement and SPR shifts can readily be measured in bulk.21 However, bulk analysis yields the average properties of a potentially complex solution, and the kinetics can only be extracted indirectly. In the case of GNRs, the measured signal is further convoluted by the poly-dispersity of the particles. Although GNRs can be synthesized with fairly small poly-dispersity,14 the strong geometry-dependence of the SPR compared to its linewidth typically broadens the bulk spectrum as compared to single GNRs. Optimal sensitivity, therefore, requires analysis of individual GNRs, rather than bulk-averaged signals.

Microscopic techniques provide access to spectroscopic properties of individual GNRs, sparsely distributed on a transparent substrate. Optical signals from GNRs can be divided into scattering, one-photon luminescence, photo-thermal scattering, and two-photon luminescence. GNRs have a high scattering cross section and detection can straightforwardly be multiplexed by wide-field imaging using dark-field excitation.19 Scattering from other sources than GNRs may be difficult to discriminate in complex environments, limiting the sensing applications to relatively clean samples. Confocal imaging provides better rejection of undesired scattering, at the cost of lower throughput. Because the illumination beam focusses one single GNR at a time, spectrometric analysis can easily be introduced in the emission path of the microscope, resolving both the transversal SPR at ∼500 nm and the longitudinal SPR at NIR wavelengths.22,23 As particles get smaller, the absorption cross section becomes dominant over the scattering cross section. One-photon luminescence is based on detecting the photons that are emitted after radiative relaxation of optically excited GNRs.23,24 The associated Stokes shift of the emitted light allows for spectral filtering of luminescence from scattering. Note that the physical mechanism of the two-photon luminescence is still under debate. Following Molinaro et al., it appears that the local field enhancement of the GNRs boosts electron–hole generation and relaxation of these electron–hole pairs proceeds subsequently via excitation of the plasmon modes of the nanorod to radiative relaxation.25 Thus, like scattered photons, the luminescent photons feature an emission spectrum that is defined by the plasmon resonances of the nanorods and is independent of the excitation wavelength.

Next to radiative relaxation, there is non-radiative relaxation that results in heating of the GNR. This is exploited in photo-thermal imaging, in which the increased scattering induced by the thermally induced elevated refraction index of the medium surrounding the GNR is detected through a second non-resonant laser beam.15 In this paper, we will focus, however, on two-photon luminescence of single GNRs, which will be further discussed below. Next to detailed insight into their geometry and fundamental photo-physical properties, spectroscopic analysis of single GNRs also is key to design sensors with desired sensitivity and kinetic response.18 

Two-photon luminescence (TPL) of GNRs has the potential to enhance the sensitivity of single-particle sensing applications by orders of magnitude.26 TPL results from successive absorption of two photons, and subsequent rapid radiative decay, which is strongly enhanced in nanoparticles relative to bulk gold. The quadratic dependence of the TPL signal on excitation power produces sharper plasmon resonance peaks compared to single-photon microscopy. Wang et al., for example, measured the TPL intensity of single GNRs with a fixed excitation wavelength of 820 nm, which indeed resulted in a narrower linewidth of the SPR compared to the bulk absorption spectrum.3 Zijlstra et al. and Molinaro et al. also showed TPL excitation spectra of a single GNR.27,28 However, these spectral measurements were rather limited in both sample size and spectral resolution due to manual laser tuning and the confocal nature of typical two-photon setups.

Another major obstruction for using TPL of GNRs in sensing applications is their limited thermal stability.30–33 As exploited by photo-thermal imaging, on-resonance excitation leads to heating of GNRs, which can result in temperature increases of several hundred degrees.28 Thermally induced diffusion of surface atoms, which happens at temperatures far below the bulk meting temperature of gold, reduces the aspect ratio of GNRs and, therefore, drastically changes the longitudinal plasmon resonance.33 The high-intensity pulsed laser excitation required for TPL results in such reshaping at powers of several mJ/cm2. Next to reducing the stability of the optical signal, elevated temperatures are also detrimental for the stability of the biomolecules, so reduction of the photo-thermal effects in TPL is crucial for biological applications.

Here, we present wide-field two-photon spectroscopy of single GNRs, using an automated tunable Ti:Sa laser in combination with a multi-focal microscope. Scanning an area of 60 × 60 µm2 with 625 focused laser beams at a framerate of several frames per second yielded rapid, high signal-to-noise, multiplexed optical detection of hundreds of GNRs. By tuning the excitation wavelength, we measured individual TPL excitation spectra and showed that these were stable for more than 30 of min. Finally, we resolved changes in the refractive index of the surrounding medium similar to those expected from single-molecules to illustrate the potential of TPL of GNRs for bio-sensing applications.

A tunable near-IR Ti:Sa laser (Coherent, Chameleon Ultra) was coupled into a home-built two-photon multi-focal microscope, depicted in Fig. 1(a). A diffractive optical element (DOE, custom made by Holo-eye) diffracted the laser beam into an array of 25 × 25 foci. A fast-scanning mirror (Newport, FSM-300-1) driven by an Archimedean spiral rapidly scanned the beams yielding a fairly homogeneous wide-field illumination, as characterized before.4 The laser beams were focused using a 60X NA 1.49 TIRF objective (Nikon, CFI Apochromat TIRF 60XC Oil), illuminating area of 60 × 60 µm2, shown in Fig. 1(b). A single period of the spiral scan took 100 ms and was synchronized with the camera integration time. Based on the scan pattern and point spread function, we estimate the duty cycle for illumination of a single spot in the sample to be around 2%. The wavelength for excitation was automatically tuned over the range of 720–950 nm. Polarization of the excitation light was controlled by inserting a half wave plate (Thorlabs, AHWP05M-980) mounted on a stepper-motor stage. For circularly polarized light, the half wave plate was removed and only a quarter wave plate (Thorlabs, AQWP05M-980) was included. TPL was collected by the same objective, filtered with a dichroic mirror (Semrock, 700dcxr) and a 720 nm short pass filter (Semrock, FF01-720-SP), and focused on a 512 × 512 pixel EM-CCD camera (Photometrics, QuantEM 512SC). Using home-written software in LabVIEW (National Instruments), the laser, scanning mirror, stepper motor, and camera were controlled in synchrony, to record time traces in which either polarization or wavelength were scanned, see Fig. 1(c). For measurements in which the refractive index of the medium was varied and for neutravidin-sensing, we used a 25X NA 1.1 water dipping objective (Nikon, CFI75 Apochromat 25XC W) and a back-illuminated 2048 × 2048 sCMOS camera (Photometrics, PRIME BSI), resulting in a 145 × 145 µm2 field of view (FOV).

FIG. 1.

Multi-focal two-photon laser scanning microscope for high-throughput microscopy and spectroscopy. (a) Schematic overview of the setup. The laser beam is diffracted in an array of 25 × 25 foci by a diffractive optical element (DOE). A scanning mirror creates a homogeneous time-averaged excitation profile allowing for wide-field epi-detection using an EM-CCD camera (b) A flat wide-field excitation pattern is generated by spiral scanning the 2D array of foci generated by the DOE pattern within the exposure time of the camera. (c) A stack of 2D images of GNRs dispersed on a glass slide. The polarization (θ) or excitation wavelength (λ) are changed per slice, resulting in spectral characterization of the GNRs. Scale bar = 30 µm.

FIG. 1.

Multi-focal two-photon laser scanning microscope for high-throughput microscopy and spectroscopy. (a) Schematic overview of the setup. The laser beam is diffracted in an array of 25 × 25 foci by a diffractive optical element (DOE). A scanning mirror creates a homogeneous time-averaged excitation profile allowing for wide-field epi-detection using an EM-CCD camera (b) A flat wide-field excitation pattern is generated by spiral scanning the 2D array of foci generated by the DOE pattern within the exposure time of the camera. (c) A stack of 2D images of GNRs dispersed on a glass slide. The polarization (θ) or excitation wavelength (λ) are changed per slice, resulting in spectral characterization of the GNRs. Scale bar = 30 µm.

Close modal

Glass coverslips were rinsed with ethanol and dried in a stream of nitrogen. 10 µl of GNRs (Nanopartz, A12-10-808) diluted in 100 µl of distilled water was spin-coated in three steps: 200 rpm for 5 s, 600 rpm for 15 s, and 1000 rpm for 60 s. After deposition, the samples were treated in an ultraviolet ozone (UVO)-cleaner (model No. 42A-220, Jelight Company) for 30 min.

High intensity peaks in the images were attributed to single (clusters of) GNRs. Regions of interest of 5 × 5 pixels around each peak were summed and background subtracted, yielding TPL signal I. The TPL intensity as a function of the excitation wavelength λ was fitted to a squared Lorentzian,

(1)

where λSPR is the SPR, ω is the Full Width at Half Maximum (FWHM) of the SPR, I0 is the maximum intensity, and C is an offset corresponding to the residual background signal.

The TPL signal as a function of the polarization angle of the excitation light θ was fitted to

(2)

with θ0 being the orientation of the GNR relative to the polarization angle of the laser.

For scanning electron microscopy (SEM) measurements, GNRs were spin-coated on an indium tin oxide (ITO) coated cover glass (SPI Supplies, 06480-AB), following the same cleaning and deposition procedure as for regular slides. Using a diamond tipped pen, a cross was scratched in the glass surface to provide a reference point for correlating the two microscopy modalities. The sample was first imaged in the two-photon microscope, after which it was placed in the SEM (FEI, Nova NanoSEM).

Sucrose solutions in HPLC grade water were used to control the refractive index of the medium. The refractive index of each sucrose solution was measured using a refractometer (Carl-Zeiss). To facilitate exchange of solutions, measurements were performed in flow cells made from two glass coverslips that sandwiched a piece of double-sided tape with cut out flow channels into which fluids were pipetted.

Neutravidin sensing measurements were performed in similar flow cells as the sucrose experiments. One-step passivation of the glass coverslip was performed, following the protocol of Gidi et al. with a few modifications.39 Glass coverslips were cleaned with an UVO cleaner for 10 min. After UVO treatment, coverslips were put in a petri dish containing desiccant (i.e., calcium chloride) and pre-heated in an oven at 90 °C for 5–10 min to further ensure water-free conditions. To graft the silane polyethylene glycol (PEG) azide (SPA) on the coverslip’s surface, 1 ml of 0.1% SPA in anhydrous dimethyl sulfoxide (DMSO) solution was evenly spread on the coverslip’s surface and heated at 90 °C for 30 min. Coverslips were rinsed with 1 ml HPLC grade water and dried under a continuous stream of N2 to remove excess SPA, after which the flow cells were assembled.

Four oligo sequences were purchased (Integrated DNA Technologies, USA), the specific sequences of which can be found in Table I. All oligo’s were prepared in Tris buffer (pH = 7.2).

TABLE I.

The sequences of the oligo’s used for the neutravidin sensing experiments.

Oligo sequence no.Sequence and base pair (bp) length
OS1 /5DBCOTEG/AAATTATAACTATTCCTA (18 bp) 
OS2 TAGGAATAGTTATAAAAA/3DTPA/(18 bp) 
OS3 GGTGGTGGTGGTTGTGGTGGTGGTGGAAAAAAAAA/3DTPA/(35 bp) 
OS4 ACCACCACCACCAAA/3Bio/(15 bp) 
Oligo sequence no.Sequence and base pair (bp) length
OS1 /5DBCOTEG/AAATTATAACTATTCCTA (18 bp) 
OS2 TAGGAATAGTTATAAAAA/3DTPA/(18 bp) 
OS3 GGTGGTGGTGGTTGTGGTGGTGGTGGAAAAAAAAA/3DTPA/(35 bp) 
OS4 ACCACCACCACCAAA/3Bio/(15 bp) 

GNRs (Nanopartz, A12-10-808) were conjugated with OS2 and OS3, following a conjugation protocol from Li et al.35 In short, GNRs were mixed with 0.02% sodium dodecyl sulfonate (SDS), 1× tris-borate-EDTA (TBE), and 500 mM NaCl (pH = 3.5). Oligo OS2 and OS3 were added with a 1:1000 GNR–oligo ratio and the whole mixture was incubated for 30 min mounted on a rotator at room temperature. SDS screens the positive charge of the cetyltrimethylammonium bromide (CTAB) and assists in the functionalization of the GNRs with the thiolated oligo’s. The GNR–oligo pellets were separated from the supernatant by centrifuging at 8000 rpm for 30 min. The GNR–oligo was reconstituted in Tris buffer and stored at 4 °C for future use.

50 µl of oligo OS1 (3.33 µM) was incubated in the channels of the flow cells for 1 h at the room temperature to allow the copper free click chemistry of DBCO and azide. The DBCO at the 5′ end of the oligo covalently binds to the azide of the SPA grafted on the coverslip’s surface. Next, the flow cells were incubated with 50 µl of the GNR–oligo OS2, OS3 solution for further 15–30 min to allow the annealing of oligo OS1 and OS2. After that, the flow cells were rinsed with 200 µl Tris buffer. The flow cells were either subsequently used or stored at 4 °C for future experiments.

Neutravidin was mixed with oligo OS4 at a ratio of 1:4 in Tris buffer (pH = 7.2). Finally, oligo OS4 was annealed with oligo OS 3 on the GNRs and the biotin at 3′ end of the oligo was conjugated with the neutravidin (a final concertation of 25 nM).

The average of the selected time traces was normalized (Inorm) and weighted according to

(3)

where In(t) is the signal intensity at time t for trace n and sn is the sensitivity of trace n at the excitation wavelength as measured from the excitation spectra prior to the time trace measurement.

The standard deviation (SD) of the normalized intensity SDnorm was defined according to

(4)

where the signal per trace In at time t is subtracted by the average signal of all selected traces Ī at time t.

To identify and characterize the TPL signal of single GNRs, we first compared TPL imaging with scanning electron microscopy (SEM) images. SEM readily showed the location, geometry, and orientation of single GNRs. Figures 2(a) and 2(b) show that the positions of the GNRs as identified by SEM and TPL imaging are highly correlated. The orientation of individual rods was confirmed by TPL imaging as a function of the polarization angle of the excitation beams, following Ref. 36. The corresponding polarization spectra, an example is shown in Fig. 2(c), follow Eq. (2) and readily resolve the orientation of each rod with a standard error of fit of less than 1°. Differences between the orientation as obtained by TPL and from the SEM images were within 10° and may originate from optical aberrations in the excitation and/or imaging path. Nevertheless, the good agreement between the SEM and TPL imaging confirms that, indeed, single GNRs can be identified using TPL, and that successive wide-field two-photon imaging can accurately resolve the orientation of these.

FIG. 2.

Correlated light and electron microscopy (CLEM) images of single gold nanorods. (a) Scanning electron microscopy image of single GNRs dispersed on an ITO-coated glass substrate. (b) Two-photon image of the same GNRs as in (a). (c) Orientation of the rods in two-photon microscopy is measured by rotating the polarization of the linear excitation light. (d) GNR close-ups from the electron microscopy image and comparison of the orientation of the GNRs.

FIG. 2.

Correlated light and electron microscopy (CLEM) images of single gold nanorods. (a) Scanning electron microscopy image of single GNRs dispersed on an ITO-coated glass substrate. (b) Two-photon image of the same GNRs as in (a). (c) Orientation of the rods in two-photon microscopy is measured by rotating the polarization of the linear excitation light. (d) GNR close-ups from the electron microscopy image and comparison of the orientation of the GNRs.

Close modal

Similar to obtaining polarization spectra, we measured excitation TPL spectra of single GNRs in wide-field by scanning the excitation wavelength from 730 to 900 nm in successive images. Figure 3(a) shows the TPL image of a typical field of view of GNRs deposited on a glass coverslip and immersed in water. The background-corrected signal intensity of the GNRs varied between 9 and 140 kHz. The intensity at a specific excitation wavelength can, however, not be used to identify single GNRs as significant differences in luminescence intensity can not only arise from small size differences of the GNRs, but also from small differences in aspect ratio, or small differences in the refraction index of their environment. Both factors shift the plasmon resonance and lead to a different excitation efficiency when exciting at a fixed wavelength. This is especially important for two-photon excitation, which has a much narrower excitation spectrum than one-photon excitation. The spot size is also not suitable for identifying single GNRs, as clusters of multiple GNRs may result in a single diffraction limited spot that cannot be differentiated from a single GNR. The excitation spectra, however, were more informative. The TPL excitation spectra of six GNRs are plotted in Fig. 3(b). Spectra were fitted to a squared Lorentzian [Eq. (1)], which is characteristic for two-photon excitation. The spectrum of area 5, however, showed two distinctive peaks and better matched the sum of two squared Lorentzians, suggesting two GNRs in a single spot. Figures 3(c) and 3(d) show the results of fitting TPL excitation (TPE) spectra of hundreds of GNRs measured in three field-of-views. All fits to Eq. (1) with an r-squared value larger than 0.6 were included in the analysis. The distribution of SPRs in this population of GNRs varied in the range of 730–850 nm, largely resembling the bulk absorption spectrum, though SPR dropped more rapidly after 825 nm in single GNR measurements. This probably reflects the reduced intensity of our laser at larger wavelength. The power spectrum of the laser is plotted for reference in the supplementary material, see Fig. S1. Figure 3(d) shows the width of the TPE spectra of 478 single GNRs, yielded a FWHM of 31 ± 1 nm (mean ± sd). The very narrow distribution not only reflects the very homogeneous optical properties of the batch of GNRs, but also demonstrates the accuracy of TPL excitation spectroscopy using scanning multi-focal microscopy.

FIG. 3.

Multiplexed two-photon excitation spectroscopy images of individual GNRs. (a) The spectra were reconstructed from a series of images, similar to the one shown here. Scale bar = 10 µm. (b) Spectra of single rods were fitted to a squared Lorentzian. The spectrum of GNR 5 was fitted to two squared Lorentzians. (c) The distribution of SPRs measured in multiple field-of-views compared to the bulk spectrum. (d) The FWHM of the two-photon spectra shows an average width of 31 nm. Parameters were acquired from fitting Eq. (1) to the raw data.

FIG. 3.

Multiplexed two-photon excitation spectroscopy images of individual GNRs. (a) The spectra were reconstructed from a series of images, similar to the one shown here. Scale bar = 10 µm. (b) Spectra of single rods were fitted to a squared Lorentzian. The spectrum of GNR 5 was fitted to two squared Lorentzians. (c) The distribution of SPRs measured in multiple field-of-views compared to the bulk spectrum. (d) The FWHM of the two-photon spectra shows an average width of 31 nm. Parameters were acquired from fitting Eq. (1) to the raw data.

Close modal

For bio-sensing applications, it is important that the SPR of single GNRs is stable over time. Absorption of intense femtosecond pulses can lead to heat-induced reshaping of the GNR, resulting in a blue-shift of the spectrum. We mapped the stability of GNRs during continuous spectral measurements by sweeping the excitation wavelength between the range of 730 and 850 nm every 25 s. Figure 4(a) shows the changes in excitation spectrum of a GNR with a laser power of 4.1 mW. We observed reshaping of the GNR from 780.7 to 755.0 nm at a rate of 3 nm per sweep. At a power of 1.6 mW (2.56 µW/focus), however, the spectrum was quite stable, as shown in Fig. 4(b). At the start of the measurement, we fitted the SPR at 766.4 nm. The last spectrum, after 216 measurements, yielded an SPR at 766.8 nm, indicating the absence of reshaping. We estimate this laser power to be equivalent to ∼31 fJ per pulse. Indeed, this laser power is below the reported damage threshold of 60 fJ,1 demonstrating that, at sufficiently low excitation power, two-photon excitation does not affect the stability of a GNR.

FIG. 4.

Gold nanorods remain stable during nearly 30 min of two-photon excitation. (a) Spectrogram of one GNR as the excitation wavelength is continuously swept back and forth at full power. The bottom graph shows the spectrum at the start of the measurement (red) and at the end (blue). (b) Spectrograph and corresponding graph of a single GNR when the laser power is reduced by 40% to ±31 fJ. The SPR remains within 1 nm of the SPR measured at the start of the measurement. (c) The SPRs of multiple rods with the laser power at ±75 fJ/pulse. Single rods are depicted in red and the average in black. (d) SPRs of multiple rods with lower excitation power.

FIG. 4.

Gold nanorods remain stable during nearly 30 min of two-photon excitation. (a) Spectrogram of one GNR as the excitation wavelength is continuously swept back and forth at full power. The bottom graph shows the spectrum at the start of the measurement (red) and at the end (blue). (b) Spectrograph and corresponding graph of a single GNR when the laser power is reduced by 40% to ±31 fJ. The SPR remains within 1 nm of the SPR measured at the start of the measurement. (c) The SPRs of multiple rods with the laser power at ±75 fJ/pulse. Single rods are depicted in red and the average in black. (d) SPRs of multiple rods with lower excitation power.

Close modal

The trends observed in these two GNRs were representative of most GNRs. Figure 4(c) shows the spectral drift of 4 other GNRs as well as the average trend for 49 GNRs. Differences in the reshaping rate probably reflect differences in the GNR size, as larger GNRs will heat faster due to a larger absorption cross section. The SPR of all GNRs stabilized around 725 nm. This probably reflects the reduced laser power at smaller wavelengths. At 1.6 mW, all spectra remained largely stable, as shown in Fig. 4(d). Only in the first wavelength-sweep, we did observe a small shift for some GNRs, which we tentatively attribute to laser induced changes in the immediate surrounding of the GNRs, rather than reshaping of the GNR itself. Note that the signal to noise ratio (SNR) did not decrease significantly upon decreasing the laser power. This may be due to the reduced heating, which may stabilize the optical response of the GNR. It should be noted though that the measured luminescence in our setup is not shot-noise limited, as discussed below. In any case, the spectrum of individual GNRs was stable for tens of minutes, opening the way for GNR based label-free sensing applications using TPL.

To demonstrate the sensitivity of TPL spectroscopy on single GNRs, we measured their spectral response when changing the refractive index of the medium from that of water (n = 1.333) to two different sucrose solutions. The bulk spectrum, as plotted in Fig. 5(a), red-shifted by 12 and 14 nm when the refractive index increased from 1.333 to 1.376 and subsequently to 1.432. This spectral shift corresponds to a sensitivity of 255 nm/refractive index unit (RUI) for this batch of GNRs. The TPL spectra of single GNRs showed a similar trend to those plotted in Figs. 5(b) and 5(c). By comparing the shift relative to the first measurement, we obtained a narrow distribution of changes in SPR for a large population of GNRs. A shift of 8 ± 3 nm was observed when going from n = 1.333 to n = 1.376, 12 ± 7 nm when changing from n = 1.376 to n = 1.432, and 21 ± 8 nm when the medium was changed from n = 1.333 to n = 1.432. The change in SPR was reversible as changing the medium back from n = 1.432 to n = 1.333 reduced the SPR by −26 ± 9 nm. Despite that the SPR varied over more than 80 nm between GNRs, the shift of individual GNRs was very reproducible. In Fig. 5(e), the SPRs for a number of GNRs are plotted as a function of the refractive index of the medium. The majority of the GNRs showed a linear increase in their SPR, though some deviated slightly from the linear trend. The slope is independent of the location of the SPR. The average change in SPR was proportional to Δn, as shown in Fig. 5(f), and yielded a slope of 237 ± 9 nm per refractive index unit, in fair agreement with the bulk value.

FIG. 5.

The LSPR shift of individual gold nanorods upon changes in the refraction index of the medium. (a) UV–vis spectra of bulk solutions of GNRs in: water (black), 30% sucrose (red), and 70% sucrose (blue) solutions. (b) and (c) The spectrum of individual GNRs in water and the two different concentrations of sucrose. Both rods show a red-shift when the refractive index is increased. (d) The shift of the SPR of single GNRs when changing the refractive index. (e) The SPRs of multiple GNRs upon changes in refractive index. (f) The average change in SPR follows a linear increase.

FIG. 5.

The LSPR shift of individual gold nanorods upon changes in the refraction index of the medium. (a) UV–vis spectra of bulk solutions of GNRs in: water (black), 30% sucrose (red), and 70% sucrose (blue) solutions. (b) and (c) The spectrum of individual GNRs in water and the two different concentrations of sucrose. Both rods show a red-shift when the refractive index is increased. (d) The shift of the SPR of single GNRs when changing the refractive index. (e) The SPRs of multiple GNRs upon changes in refractive index. (f) The average change in SPR follows a linear increase.

Close modal

Next, we did sensing experiments. Prior to the measurement, GNRs were functionalized with two types of oligo’s. One oligo sequence (OS) for immobilization of the GNR on the glass and another for specific and long-lasting binding to a third complementary OS, see Fig. 6(a). After GNRs were immobilized on the glass surface, the excitation spectrum was measured by sweeping the laser from 730 nm to 900 nm while taking consecutive images. SPR peaks of the rods were fitted to a 2D Gaussian, and the corresponding amplitudes were used to produce an excitation spectrum for every rod in the FOV. The laser power was increased to 3.0 mW to improve SNR at a trade-off of rod stability. We found that most of the rods were stable and were suitable for sensing experiments. Once the excitation spectra were acquired, the wavelength was fixed at 800 nm and the luminescence of each GNR was measured continuously. A solution containing the 25 nM neutravidin–oligo complex was flushed in the flow cell after 120 s of imaging. The complementary sequences of OS3 and OS4 allowed for specific binding of the neutravidin to the GNRs. After 380 s, a second excitation spectrum was taken and fitted to a squared Lorentzian [Eq. (1)]. All spectra with a FWHM larger than 35 nm were discarded from further processing to exclude GNR clusters from the data. After this selection, the measurement yielded 82 traces. A control measurement, where only buffer was flushed in the flow cells, yielded 73 traces.

FIG. 6.

Specific binding of DNA oligo–neutravidin complexes induce a shift in the SPR. (a) Schematic of the experiment. Inside a flow cell, GNRs are immobilized on a glass surface by binding of two complementary oligo strands (pink). A second oligo pair (green) mediates binding of neutravidin to the GNR. (b) and (d) Excitation spectra of two GNRs before (black) and after (red) addition of neutravidin to the medium. In both instances, the SPR is red-shifted, indicating an increase in the refractive index of the medium surrounding the rod. The data are fitted to a squared Lorentzian (solid lines) according to Eq. (1). The derivative of the fitted curve (dotted line) indicates the expected signal change upon a shift in the SPR. Both rods were excited at 800 nm (vertical red line). (c) and (e) Time traces of the GNRs corresponding to the spectra in the column of b. After 120 s of imaging (dotted black line), 25 nM neutravidin was flushed in the flowcell, resulting in signal intensity changes. (f) and (g) The excitation spectra and time trace of a GNR where buffer, instead of neutravidin, was flushed in after 120 s of imaging. No SPR shift was observed and the time trace did not feature significant changes in signal intensity. (h) Normalized and weighted average signal intensity, according to Eq. (3), of the ensemble of GNRs when flushing in neutravidin (red) or only buffer (black) at 120 s (dotted black line). The red trace is offset by 0.15 for clarity. (i) The SPR shift of the ensemble of GNRs. When neutravidin was added, the SPR features an average red shift of 5 ± 1 nm.

FIG. 6.

Specific binding of DNA oligo–neutravidin complexes induce a shift in the SPR. (a) Schematic of the experiment. Inside a flow cell, GNRs are immobilized on a glass surface by binding of two complementary oligo strands (pink). A second oligo pair (green) mediates binding of neutravidin to the GNR. (b) and (d) Excitation spectra of two GNRs before (black) and after (red) addition of neutravidin to the medium. In both instances, the SPR is red-shifted, indicating an increase in the refractive index of the medium surrounding the rod. The data are fitted to a squared Lorentzian (solid lines) according to Eq. (1). The derivative of the fitted curve (dotted line) indicates the expected signal change upon a shift in the SPR. Both rods were excited at 800 nm (vertical red line). (c) and (e) Time traces of the GNRs corresponding to the spectra in the column of b. After 120 s of imaging (dotted black line), 25 nM neutravidin was flushed in the flowcell, resulting in signal intensity changes. (f) and (g) The excitation spectra and time trace of a GNR where buffer, instead of neutravidin, was flushed in after 120 s of imaging. No SPR shift was observed and the time trace did not feature significant changes in signal intensity. (h) Normalized and weighted average signal intensity, according to Eq. (3), of the ensemble of GNRs when flushing in neutravidin (red) or only buffer (black) at 120 s (dotted black line). The red trace is offset by 0.15 for clarity. (i) The SPR shift of the ensemble of GNRs. When neutravidin was added, the SPR features an average red shift of 5 ± 1 nm.

Close modal

Figure 6(b) shows the TPL spectra of a rod before and after neutravidin was added to the medium. The SPR was red-shifted by 15 ± 1 nm, indicating an increase in the local refractive index. The initial sensitivity of the GNR is around 25 (photons/s)/nm based on the derivative of the fitted Lorentzian squared. Note that the sensitivity increases to 50 (photons/s)/nm when the SPR red-shifts, as the rod is excited at a steeper part of its SPR spectrum. In Fig. 6(c), the signal intensity remained relatively stable during the first 120 s of the measurement. Upon flushing-in oligo–neutravidin solution, the signal intensity increased, which we attribute to a red-shift of the SPR. Discrete steps of ∼212 photons/s can be discerned in the time trace. Based on the sensitivity, a 15 nm red shift of the SPR is possible. For such a large change, it is most plausible that multiple neutravidin molecules bind to the GNR. The signal intensity decreased after 320 s, suggesting that the rod reshaped when excited at a higher absorption cross section. However, the spectrum that was measured afterward showed a red-shifted peak, relative to the original spectrum, suggesting that desorption may have happened.

A similar response was found for the rod in Figs. 6(d) and 6(e). The initial signal intensity of the time trace is 1019 ± 227 photons/s. It decreased instantly to 357 ± 164 photons/s after addition of neutravidin, suggesting multiple binding events. The decrease in signal intensity is consistent with the location of the SPR peak relative to the excitation wavelength. More examples of spectra and time traces resembling the ones shown here are plotted in Fig. S2 in the supplementary material.

We flushed-in buffer without oligo–neutravidin to verify that the SPR perturbations are not caused by measurement artifacts like microscope instabilities, thermal drift, or buffer impurities. The spectra of one GNR from this control measurement, plotted in Fig. 6(f), show that the SPR shifted marginally, by 0.4 ± 0.5 nm, upon flushing in buffer. The time trace of the corresponding GNR, shown in Fig. 6(g), continues as a constant signal intensity. The spectra and time traces plotted in Fig. 3 illustrate similar behavior of other GNRs measured in this control experiment.

To quantify the response of all rods, we averaged and weighted the time signals according to Eq. (3) and plotted the results in Fig. 6(h). The average signal intensity changed 12 ± 4% with the addition of neutravidin. Flushing-in buffer absent of neutravidin resulted in a shift of 4 ± 2%. The larger average intensity change when adding neutravidin indicates a specific response. The standard deviation (SD), defined in Eq. (4), of both averages also increased after 120 s. This could be caused by mechanical instabilities, impurities in the solution, or reshaping of the GNRs, which would also explain the relatively wide distribution of the SPR shift in Fig. 6(i). By fitting a Gaussian function to the distribution of SPR shifts, we find that the SPR shifts during the control measurement on average 0.1 ± 0.8 nm, with a FWHM = 16 ± 3 nm. Note that the SPR is both red- and blue-shifted without a specific direction, which is also seen in the averaged time trace and depends on the initial position of the SPR relative to the excitation wavelength. The average SPR of the GNRs with oligo–neutravidin was red-shifted by 5 ± 1 nm, which matches the trend we saw in the average time trace in Fig. 6(h), and confirms a specific interaction of oligo–neutravidin with the GNR.

We did not observe single neutravidin binding events, despite the high sensitivity of TPE. Considering a GNR with a sensitivity of 100 (photons/s)/nm, a single binding event would change the signal by 50 photons/s. With a signal with an intensity of 1019 ± 227 photons/s, as plotted in Fig. 6(e), we would expect a shot-noise level of 32 photons/s. The background noise, surrounding the GNR peaks, however, appeared to be 99 ± 5 photons/s (data not shown), which is excessive compared to the dark- and readout-noise of an air-cooled sCMOS camera. Mechanical instabilities of the microscope or temporal fluctuations in the spiral-scanning pattern may cause this excess noise, which may be relieved by a faster and/or better distributed scanning pattern. Based on the expected SPR shift and the theoretical shot-noise limit, we expect that the setup should be capable of imaging single binding events once such measures have been implemented. For now, however, the diverse response of individual GNRs to flushing in neutravidin solution and the excess noise precludes single-molecule sensing at this stage.

We measured the TPE spectra of single GNRs in parallel by scanning the excitation wavelength of a multi-focal two-photon microscope. The narrow peaks of the plasmon resonances follow a squared Lorentzian, which confirms two-photon excitation. Irradiation of rods with an energy estimated at 31 fJ per pulse allowed for stable spectra measurements for nearly 30 min, whereas increasing the excitation power above the threshold level resulted in reshaping accompanied by a blue-shift of the plasmon resonance. We demonstrated that by changing the refractive index of the medium, the TPE spectra shifted accordingly. We observed specific interactions of oligo-functionalized GNRs with complementary oligo–neutravidin molecules in solution.

To identify TPL from single GNRs, we correlated SEM with two-photon images. Besides SEM, single GNRs were also identified by making use of light polarization and the excitation spectra. Considering the spectra measurements, as shown in Fig. 3(b), clusters of GNRs resulted in a broader SPR because the spectrum is the sum of multiple individual rods combined. Previous studies also concluded that the spectrum of a cluster of rods is a combination of multiple single GNRs.24,37,38 Discriminating single GNRs from clusters by using light polarization was used as an alternative method. The cos4(φ) dependence on polarization shown in Fig. 1 for a single rod agrees with earlier findings of Wang et al. and Imura et al., which furthermore confirms that the collected signal originates from a two-photon excitation process.39 

The plasmon resonance for single GNRs was determined with sub-nanometer accuracy. The average FWHM we found is ±35% smaller compared to the reported one-photon spectra, which lies approximately between 45 and 60 nm.17,42–42 Moreover, the width of 24 nm for the narrowest peaks is nearly a factor of two smaller than any reported one-photon spectra we could find. Narrow peaks are preferred for sensing applications as narrower peaks translate to a larger derivative around the SPR and, therefore, higher sensitivity. We attribute differences in FWHM to inhomogeneities in the size of the GNRs as the width of the plasmon resonance depends on the volume of the rod.43 

TPE spectra from previous studies, although more limited in resolution, reported comparable reduction in the FWHM when compared to one-photon excitation. Zijlstra et al. measured a reduction in the FWHM of ∼60%.27 Molinaro et al. showed that their measured TPE spectra overlap with the model of absorption calculated using Gans-Mie theory, which describes the scattering and absorbing of small ellipsoid shaped nanocrystals.28 Considering the differences in the SPR width between one- and two-photon excitation, the Gans-Mie model results in reduced peak widths of a factor 2221.29. Both the findings from Zijlstra and Molinaro fit with our own data, where the narrowest peaks are as narrow as 20 nm and the broadest are around 40 nm.

The location of the SPRs correspond to the absorption spectrum of the bulk solution. A sharp cut-off below 740 nm can be explained by the lower limit of the excitation wavelength at 730 nm, which impairs imaging of the full plasmon peak. For wavelengths larger than 850 nm, we also do not observe SPR maxima. This may originate from limitations in the DOE pattern, i.e., by loss of focus caused by chromatic aberrations and a reduced lower laser power at extreme wavelengths (see the supplementary material, Fig. S1). Still, the remaining bandwidth of 120 nm, in which we can image GNRs, is sufficient to measure the larger part of the distribution of rods considering the bulk spectrum in a highly parallel and accurate fashion.

The EM-CCD camera readout for single rods ranged between 15 × 103 and 20 × 103 kcounts/s. Considering the EM-gain of 40 and the quantum efficiency of the camera, the readout signal translates to a photon count of ∼340–450 photons/s. The luminescence of GNRs is known to reach intensities of thousands of photons per second. However, reshaping of rods due to high laser intensities limits the excitation power and thereby lowering the photon emission from the rods. Nevertheless, SNR is excellent as the non-linear TPE results in an almost negligible background, dominated by readout noise of the camera.

By lowering the excitation power under the damage threshold, we measured stable TPE spectra. The majority of GNRs blue-shifted their SPR by a few nanometers upon initial irradiation. For these small shifts in wavelength, reshaping does not seem to be the underlying mechanism as a change in aspect ratio would induce shifts of an order of magnitude larger.42 Moreover, Zijlstra et al. showed that for rods with a plasmon resonance between 750 and 800 nm, which roughly corresponds to our distribution of measured rods, reshaping does not occur until 150 fJ of laser power is absorbed by a GNR. Nevertheless, significant reshaping below the melting temperature of gold was observed by Taylor et al.33 They showed that their observations fitted well with a surface diffusion model from Mullins. Diffusion around the tips of the rod changes the aspect ratio causing a shift in the plasmon resonance. However, for rods with an aspect ratio between 3.8 and 4.1—similar to the rods we use—Taylor measured a thermal threshold of ∼60 and 40 fJ, respectively. From these results, we expect that we do not see reshaping but rather that upon initial exposure to the laser beam, sedimented particles and remaining CTAB coating (n = 1.435)44 residing around the rods are ablated from the surface. Increasing the excitation power to above the damage threshold resulted in a linear decrease in the average SPR shift per wavelength sweep between a SPR of 800 and 740 nm. Based on Ekici et al., we expect that the GNRs almost instantaneously reach thermal equilibrium when illuminated in-resonance, as there is enough time for the energy to disperse in 12.5 ns between each femtosecond pulse of the 80 MHz pulsed laser.28 Therefore, illuminating the rods in-resonance for longer periods of time results in a linear response of the SPR shift. As the SPR is linearly dependent on the aspect ratio of the GNR, the linear decrease indicates an also linear decrease in the aspect ratio of the rod. In the oligo–neutravidin measurements, we increased the laser power to slightly above the damage threshold of the GNRs, which may have resulted in a fraction of unstable rods. The difference in robustness of the GNRs to the laser illumination may arise from residual variations in the spatial distribution of laser intensity across the FOV or from a slightly tilted excitation plane, which would reduce the absorption of a rod.

Our reference measurement for the sensitivity of the GNRs in bulk [Fig. 5(a)] yielded a sensitivity of 255 nm/RUI. This result is higher than the sensitivity of 237 ± 9 nm/RUI we measured for single GNRs in the two-photon microscope. However, differences in SPR shift between individual rods and the bulk solution are to be expected, as sensitivity to refractive index changes is highly dependent on the aspect ratio and the rod-to-substrate interface.15,45 A lower sensitivity for individual rods when compared to bulk was also found by others and also attributed to the presence of the substrate near immobilized GNRs. Martinsson et al. studied the influence of a negatively charged glass substrate on the SPR of several differently shaped metallic nanoparticles. They measured a reduction in sensitivity of 34% when comparing GNRs in bulk and distributed on glass.46 Conveniently, the multiplexed capabilities of the setup can be used to select the more sensitive rods from the ensemble.

A bulk sensitivity of 237 ± 9 nm/RUI for individual rods on a substrate resembles the findings of other studies. Piliarik et al. measured a bulk sensitivity of 140 nm/RUI of GNRs with a dimension of 35 × 75 nm2 (AR = 2.14), Zijlstra et al. measured a bulk sensitivity of 202 ± 15 nm/RUI with an average GNR size of 37 × 9 nm2 (AR = 4.1), and Martinsson et al. found 255 nm/RUI for rods with an average size of 67 × 19 nm2 (AR = 3.5).15,46,47 Differences between sensitivities are attributed to the differences in the shapes of the GNRs. The AR, absolute length, and diameter of a nanorod define the sensitivity of a GNR for changes in the bulk refractive index, where larger rods with a larger AR tend to have a higher sensitivity.12,48

Spectral studies on gold nanoparticles have been limited to scattering and photo-thermal microscopy. In terms of in vivo environments, TPE is better suited for biological applications. Besides deeper penetration depth, the large anti-Stokes shift results in low background, higher sensitivity, and better selectivity would favor TPE over scattering and possibly photo-thermal approaches. By utilizing the deep-tissue imaging capabilities of two-photon microscopy, one could even envision spectral studies in live animals like zebra fish embryos and mice.

Sensitivity of the measurements can also be increased by making use of bipyramid-shaped nanoparticles instead of rods. Although thermally more unstable compared to rods, the more confined electrical field around the sharper tips of bipyramids do result in a higher sensitivity to refractive index changes.13 For addressing the thermal stability of gold nanoparticles, Chen et al. found that by coating GNRs with a silica layer, they were considerably more stable to pulsed laser light than when coated with CTAB or PEG.49 

In our bio-sensing experiments, we saw a relatively large shift of the SPR when oligo–neutravidin was added to the buffer solution, which we tentatively attributed to multiple binding events of neutravidin. To estimate how many neutravidin molecules can bind at the sensitive ends of the rods, take an average size of the rod of 40 × 40 × 10 nm3. Approximately 1/4 of the area would be blocked by the substrate. Thus, a rod should have around 235 nm2 exposed area. The surface area of neutravidin is ∼22.4 nm2 and, thus, around ten neutravidin molecules could bind to a rod.50 Based on streptavidin (53 kDa) experiments by Zijlstra et al., who use similar GNRs to us, we expect a SPR shift of a little over 0.5 nm per neutravidin (60 kDa) binding event.51 Note though that we used a DNA linker to mediate neutravidin–GNR interactions. This yields a larger distance to the GNR and hence a lower sensitivity. Ignoring this difference, a maximum number of ten neutravidin molecules per rod would, therefore, result in a SPR shift of 5 nm. However, some of our reported spectra feature a spectral shift of up to 20 nm, which could indicate that we measure a larger SPR shift per molecule, or that unspecific binding of impurities in the medium induces an additional red-shift. The average plasmon shift of 4 ± 1 nm of the ensemble of GNRs better matches the shift we expect. We did not observe steps in the time traces, see Figs. 6(e) and S2(d) for examples. One strategy to improve sensitivity would entail decreasing the linker-size between the target molecule and the rod. For example, by using smaller molecules instead of oligo’s, neutravidin could move closer to the surface of the rod and experience a stronger field intensity and cause a larger shift in the plasmon resonance.52 

In conclusion, we demonstrate a multi-focal scanning two-photon microscope that makes it possible to measure TPL response and excitation spectra of hundreds of GNRs measured in parallel. The multi-focal scanning presents a niche in terms of excitation power that allows for a sufficient signal intensity for high quality imaging, while maintaining the heat sufficiently low to prevent thermal reshaping. We observed a decrease in peak width by almost 60% upon moving from one- to two-photon excitation. The narrow TPL spectra suggest an increase in the sensitivity of sensing applications based on plasmonic nanoparticles. Combining high sensitivity with multi-focal microscopy enables the measurement of GNRs in wide-field—allowing for acquiring statistical significant data within minutes. We demonstrated the ability of the setup to measure the change in the TPL spectra when changing the refractive index of the medium and take the first steps in single-molecule bio-sensing. Given the TPL response on the refractive index and the ability to measure GNRs without reshaping, we paved the path for TPL plasmonic single-molecule sensors.

See the supplementary material for the spectral response of our microscope and additional data on GNR sensing using TPL.

This work was supported by The Netherlands Organisation for Scientific Research (NWO/OCW), as part of the Frontiers of Nanoscience program, by NWO-VICI Research Program Project No. 680-47-616.

The authors have no conflicts to disclose.

The data that support the findings of this study are available from the corresponding author upon reasonable request.

1.
A.
Mooradian
,
Phys. Rev. Lett.
22
,
185
(
1969
).
2.
M. B.
Mohamed
,
V.
Volkov
,
S.
Link
, and
M. A.
El-Sayed
,
Chem. Phys. Lett.
317
,
517
(
2000
).
3.
H.
Wang
,
T. B.
Huff
,
D. A.
Zweifel
,
W.
He
,
P. S.
Low
,
A.
Wei
, and
J.-X.
Cheng
,
Proc. Natl. Acad. Sci. U. S. A.
102
,
15752
(
2005
).
4.
B.
Van den Broek
,
B.
Ashcroft
,
T. H.
Oosterkamp
, and
J.
van Noort
,
Nano Lett.
13
,
980
(
2013
).
5.
R.
Zhang
,
J.
Erler
, and
J.
Langowski
,
Biophys. J.
112
,
450
(
2017
).
6.
C.
Yu
and
J.
Irudayaraj
,
Anal. Chem.
79
,
572
(
2007
).
7.
D.
Shajari
,
A.
Bahari
, and
P.
Gill
,
Colloids Surf., A
543
,
118
(
2018
).
8.
M.
Eghtedari
,
A.
Oraevsky
,
J. A.
Copland
,
N. A.
Kotov
,
A.
Conjusteau
, and
M.
Motamedi
,
Nano Lett.
7
,
1914
(
2007
).
9.
A.
de la Zerda
,
S.
Prabhulkar
,
V. L.
Perez
,
M.
Ruggeri
,
A. S.
Paranjape
,
F.
Habte
,
S. S.
Gambhir
, and
R. M.
Awdeh
,
Clin. Exp. Ophthalmol.
43
,
358
(
2015
).
10.
M. R. K.
Ali
,
H. R.
Ali
,
C. R.
Rankin
, and
M. A.
El-Sayed
,
Biomaterials
102
,
1
(
2016
).
11.
R. S.
Riley
and
E. S.
Day
,
Wiley Interdiscip. Rev.: Nanomed. Nanobiotechnol.
9
,
e1449
(
2017
).
12.
G. J.
Nusz
,
A. C.
Curry
,
S. M.
Marinakos
,
A.
Wax
, and
A.
Chilkoti
,
ACS Nano
3
,
795
(
2009
).
13.
S. M.
Marinakos
,
S.
Chen
, and
A.
Chilkoti
,
Anal. Chem.
79
,
5278
(
2007
).
14.
W.
Ni
,
X.
Kou
,
Z.
Yang
, and
J.
Wang
,
ACS Nano
2
,
677
(
2008
).
15.
P.
Zijlstra
,
P. M. R.
Paulo
, and
M.
Orrit
,
Nat. Nanotechnol.
7
,
379
(
2012
).
16.
I.
Ament
,
J.
Prasad
,
A.
Henkel
,
S.
Schmachtel
, and
C.
Sönnichsen
,
Nano Lett.
12
,
1092
(
2012
).
17.
G. J.
Nusz
,
S. M.
Marinakos
,
A. C.
Curry
,
A.
Dahlin
,
F.
Höök
,
A.
Wax
, and
A.
Chilkoti
,
Anal. Chem.
80
,
984
(
2008
).
18.
A. B.
Taylor
and
P.
Zijlstra
,
ACS Sens.
2
,
1103
(
2017
).
19.
M. A.
Beuwer
,
M. W. J.
Prins
, and
P.
Zijlstra
,
Nano Lett.
15
,
3507
(
2015
).
20.
J.
Olesiak-Banska
,
M.
Waszkielewicz
,
P.
Obstarczyk
, and
M.
Samoc
,
Chem. Soc. Rev.
48
,
4087
(
2019
).
21.
D.-S.
Wang
,
F.-Y.
Hsu
, and
C.-W.
Lin
,
Opt. Express
17
,
11350
(
2009
).
22.
A. M.
Funston
,
T. J.
Davis
,
C.
Novo
, and
P.
Mulvaney
,
Philos. Trans. R. Soc., A
369
,
3472
(
2011
).
23.
F.
Wackenhut
,
A. V.
Failla
, and
A. J.
Meixner
,
J. Phys. Chem. C
117
,
17870
(
2013
).
24.
A.
Tcherniak
,
S.
Dominguez-Medina
,
W.-S.
Chang
,
P.
Swanglap
,
L. S.
Slaughter
,
C. F.
Landes
, and
S.
Link
,
J. Phys. Chem. C
115
,
15938
(
2011
).
25.
C.
Molinaro
,
Y.
El Harfouch
,
E.
Palleau
,
F.
Eloi
,
S.
Marguet
,
L.
Douillard
,
F.
Charra
, and
C.
Fiorini-Debuisschert
,
J. Phys. Chem. C
120
,
23136
(
2016
).
26.
W.
Zhang
,
M.
Caldarola
,
X.
Lu
, and
M.
Orrit
,
ACS Photonics
5
,
2960
(
2018
).
27.
P.
Zijlstra
,
J. W. M.
Chon
, and
M.
Gu
,
Nature
459
,
410
(
2009
).
28.
O.
Ekici
,
R. K.
Harrison
,
N. J.
Durr
,
D. S.
Eversole
,
M.
Lee
, and
A.
Ben-Yakar
,
J. Phys. D: Appl. Phys.
41
,
185501
(
2008
).
29.
S.
Link
and
M. A.
El-Sayed
,
J. Chem. Phys.
114
,
2362
(
2001
).
30.
M. A.
El-Sayed
,
Acc. Chem. Res.
34
,
257
(
2001
).
31.
P.
Zijlstra
,
J. W. M.
Chon
, and
M.
Gu
,
Phys. Chem. Chem. Phys.
11
,
5915
(
2009
).
32.
G.
González-Rubio
,
P.
Díaz-Núñez
,
A.
Rivera
,
A.
Prada
,
G.
Tardajos
,
J.
González-Izquierdo
,
L.
Bañares
,
P.
Llombart
,
L. G.
Macdowell
,
M. A.
Palafox
,
L. M.
Liz-Marzán
,
O.
Peña-Rodríguez
, and
A.
Guerrero-Martínez
,
Science
358
,
640
(
2017
).
33.
A. B.
Taylor
,
A. M.
Siddiquee
, and
J. W. M.
Chon
,
ACS Nano
8
,
12071
(
2014
).
34.
Y.
Gidi
,
S.
Bayram
,
C. J.
Ablenas
,
A. S.
Blum
, and
G.
Cosa
,
ACS Appl. Mater. Interfaces
10
,
39505
(
2018
).
35.
J.
Li
,
B.
Zhu
,
Z.
Zhu
,
Y.
Zhang
,
X.
Yao
,
S.
Tu
,
R.
Liu
,
S.
Jia
, and
C. J.
Yang
,
ACS Appl. Mater. Interfaces
19
,
6800
(
2014
).
36.
T.
Zuo
,
H. J.
Goldwyn
,
B. P.
Isaacoff
,
D. J.
Masiello
, and
J. S.
Biteen
,
J. Phys. Chem. Lett.
10
,
5047
(
2019
).
37.
M.
Garai
,
T.
Zhang
,
N.
Gao
,
H.
Zhu
, and
Q.-H.
Xu
,
J. Phys. Chem. C
120
,
11621
(
2016
).
38.
J.
Wu
,
X.
Lu
,
Q.
Zhu
,
J.
Zhao
,
Q.
Shen
,
L.
Zhan
, and
W.
Ni
,
Nano-Micro Lett.
6
,
372
(
2014
).
39.
K.
Imura
,
T.
Nagahara
, and
H.
Okamoto
,
J. Phys. Chem. B
109
,
13214
(
2005
).
40.
P. V.
Ruijgrok
,
N. R.
Verhart
,
P.
Zijlstra
,
A. L.
Tchebotareva
, and
M.
Orrit
,
Phys. Rev. Lett.
107
,
037401
(
2011
).
41.
K.-S.
Lee
and
M. A.
El-sayed
,
J. Phys. Chem. B
109
,
20331
(
2005
).
42.
W.
Zhang
,
M.
Caldarola
,
B.
Pradhan
, and
M.
Orrit
,
Angew. Chem.
129
,
3620
(
2017
).
43.
A.
Henkel
,
W.
Ye
,
Y.
Khalavka
,
A.
Neiser
,
C.
Lambertz
,
S.
Schmachtel
,
R.
Ahijado-Guzmán
, and
C.
Sönnichsen
,
J. Phys. Chem. C
122
,
10133
(
2018
).
44.
P.
Kekicheff
and
O.
Spalla
,
Langmuir
10
,
1584
(
1994
).
45.
J.
Becker
,
A.
Trügler
,
A.
Jakab
,
U.
Hohenester
, and
C.
Sönnichsen
,
Plasmonics
5
,
161
(
2010
).
46.
E.
Martinsson
,
M. M.
Shahjamali
,
N.
Large
,
N.
Zaraee
,
Y.
Zhou
,
G. C.
Schatz
,
C. A.
Mirkin
, and
D.
Aili
,
Small
12
,
330
(
2016
).
47.
M.
Piliarik
,
P.
Kvasnička
,
N.
Galler
,
J. R.
Krenn
, and
J.
Homola
,
Opt. Express
19
,
9213
(
2011
).
48.
H.
Chen
,
X.
Kou
,
Z.
Yang
,
W.
Ni
, and
J.
Wang
,
Langmuir
10
,
5233
(
2008
).
49.
Y.-S.
Chen
,
W.
Frey
,
S.
Kim
,
K.
Homan
,
P.
Kruizinga
,
K.
Sokolov
, and
S.
Emelianov
,
Opt. Express
18
,
8867
(
2010
).
50.
C.
Rosano
,
P.
Arosio
, and
M.
Bolognesi
,
Biomol. Eng.
16
,
5
(
1999
).
51.
P.
Zijlstra
and
M.
Orrit
,
Rep. Prog. Phys.
74
,
106401
(
2011
).
52.
Z.
Omair
and
M. A.
Talukder
,
Plasmonics
14
,
1611
(
2019
).

Supplementary Material