Anal. Chem. 2009, 81, 6572–6580
High-Resolution Microspectroscopy of Plasmonic Nanostructures for Miniaturized Biosensing Andreas B. Dahlin,* Si Chen, Magnus P. Jonsson, Linda Gunnarsson, Mikael Ka¨ll, and Fredrik Ho¨o¨k* Department of Applied Physics, Chalmers University of Technology, Gothenburg, Sweden In this article, we demonstrate how to perform microscale spectroscopy of plasmonic nanostructures in order to minimize the noise when determining the resonance peak wavelength. This is accomplished using an experimental setup containing standard optical components mounted on an ordinary light microscope. We present a detailed comparison between extinction spectroscopy in transmission mode and scattering spectroscopy under dark field illumination, which shows that extinction measurements provide higher signal-to-noise in almost all situations. Furthermore, it is shown that rational selection of nanostructure, hardware components, and data analysis algorithms enables tracking of the particle plasmon resonance wavelength from a 10 µm × 50 µm area with a resolution of 10-3 nm in transmission mode. We investigate how the temporal resolution, which can be improved down to 17 ms, affects the noise characteristics. In addition, we show how data can be acquired from an area as small as 2 µm × 10 µm (∼240 particles) at the expense of higher noise on longer time scales. In comparison with previous work on macroscopic sensor designs, this represents a sensor miniaturization of 5 orders of magnitude, without any loss in signal-to-noise performance. As a model system, we illustrate biomolecular detection using gold nanodisks prepared by colloidal lithography. The microextinction measurements of nanodisks described here provide detection of protein surface coverages as low as 40 pg/cm2 ( 0.14 Abs. Interestingly, also when other surfaces with nanoparticles are looked at (spheres,14 rods,13 triangles,11 cubes,7 rings,8 “islands”,9 and “caps”10), all have an extinction significantly higher than the critical value when λ is close to λpeak. It should now be recalled that the plots in Figure 3B are derived under the assumptions that the illumination is as effective in dark field and that there is no contribution from absorption. These assumptions are not realistic, which speaks even more for transmission mode as the preferable method. We emphasize that even if it would be possible to increase the intensity of light arbitrarily, transmission mode is still preferable. The assumption of isotropic scattering from plasmon decay is not strictly valid either, for instance since light is refracted at the backside of the glass support when the immersion medium is not oil (see above). However, a different interval of possible values for β does not change the concluding critical extinction value of 0.14 Abs significantly. For instance, if one assumes that as much as 50% of the scattered light can be collected, which requires an infinite objective lens, the critical extinction value is still well below 0.2 Abs. As shown by this analysis, the only case when scattering mode may be preferable is for samples with extremely low extinction Analytical Chemistry, Vol. 81, No. 16, August 15, 2009
6577
Figure 4. Representative noise data when tracking the centroid of the nanodisk plasmon resonance peak on a 10 µm × 50 µm area (blue) and on a 2 µm × 10 µm area (red) at the fastest temporal resolution of 17 ms (inset). The white graph (superimposed on the blue graph) represents the data for 10 µm × 50 µm after averaging, with a final temporal resolution of ∼4 s. The data are presented as a surface plot in part B, where the standard deviation of the difference between two centroid measurements is shown as a function of the temporal resolution (number of averages) and the correlation time (the time elapsed between the two estimates of the centroid). The red circle illustrates the region with lowest noise for the larger area, and the red arrow indicates the lowest noise when measuring on the smaller area.
(E < 0.1 Abs in Figure 3B). Such low extinction can result from small (40 nm) particle is analyzed and is indeed the method generally used in such experiments.26-29 Details and support to the conclusions of this analysis are presented in the Supporting Information. Motivated by the discussion above, noise estimations of λpeak and sensing experiments were performed using microextinction measurements in transmission mode. The air immersion objectives with 20× (NA 0.40) or 100× (NA 0.75) magnification were chosen since they collect the lowest amount of scattered light, as illustrated by the microextinction peaks in Figure 2A. Our previous study introduced a data analysis algorithm based on calculating the centroid wavelength (C) of the LSPR peak.32 In brief, centroid methods are based on calculating the geometric average of a spectrum. In this way, more data values are utilized in the calculation, which generally reduces noise compared to direct estimations of λpeak that take fewer pixels into account. For asymmetric peaks, there is an offset between C and λpeak but the changes in the parameters induced by biomolecular 6578
Analytical Chemistry, Vol. 81, No. 16, August 15, 2009
binding generally only differ by a few percent.32 Figure 4A shows characteristic noise data when tracking C without spectral averaging at the highest temporal resolution (τres) of 17 ms (limited by software) at 20× (blue) and 100× (red) magnification. Interestingly, as shown in Figure 4, the system displays different noise characteristics depending on magnification. At 100×, additional fluctuations occur when looking at a longer time scale. To investigate this further, we introduce the parameter ∆Cstd, which is defined as the standard deviation of the difference between two measurements of C. Figure 4B shows surface plots of ∆Cstd estimated from noise data, as a function of τres and correlation time (τcorr). Here τcorr is the time elapsed between the two spectral acquisitions used to calculate ∆Cstd while τres equals the number of averages multiplied by 17 ms. The surface plots illustrate the influence from short-term and long-term noise in the system and provide a means to estimate the optimal temporal resolution for lowest short-term noise. (The noise level in C is strictly defined in the Supporting Information.) At 20× (10 µm × 50 µm), ∆Cstd is only weakly dependent on τcorr, i.e., the system is stable over longer time periods. Thus, lower noise can be achieved by averaging spectra at the expense of a higher value of τres. The lowest noise in C is almost exactly 10-3 nm and occurs for τres close to 4 s and reasonably short τcorr (3 nm shift (Figure 5) originating from molecules bound to the glass is negligible.13 The active surface area of a gold disk in this study is 24 500 nm2 (subtracting the bottom which is in contact with the glass). Since the coverage of NeutrAvidin is approximately 120 ng/cm2 and the protein has a molecular weight of 60 kDa, the number of protein molecules on a disk can be estimated to be 295. When the sample surface is in focus, 240 disks are analyzed (Figure 1B), which then corresponds to ∼70 000 molecules (10 amol) in total. As explained above, when measuring in focus at 100× magnification, the long-term noise increases in the system (Figure 4). However, since molecular binding events are virtually instantaneous, it is the short-term noise (τcorr ) τres) which determines if it is possible to resolve one protein molecule binding to (or dissociating from) the surface. Strictly, the molecule must then induce a change in C between two subsequent acquisitions which is higher than what can be expected from the noise level, which is 0.003 nm according to the data in Figure 4B. Therefore, if 70 000 molecules induces a shift of 3 nm, the detection limit in terms of the number of molecules is below 100 and thus comparable to values reported for the single nanoparticle sensing approach.29,45 In addition, the microspectroscopy approach gives the advantage of providing a detection limit in surface coverage (and thus concentration) which is orders of magnitude better than for single particle sensors. In general, the Au disks prepared by colloidal lithography offer several advantages as a nanostructure in LSPR sensing. Their resonance peak is high in magnitude and quite narrow, which reduces the noise in C. The disks are also robust and easy to fabricate.32 The optical properties can be tuned by varying the aspect ratio of the disks, which controls λpeak19 and the absolute size of the disks, which controls the scattering contribution to the extinction (σsca/σext). A higher λpeak is associated with a higher bulk RI sensitivity19 but also results in a field which extends further into the liquid.39 It is therefore of interest to find the aspect ratio of disks which gives the highest surface sensitivity to binding of proteins of analytical interest. Naturally, the recognition layer thickness also plays a critical role in this optimization.9 Possible future improvements of the structure involve directing binding events to the disk edges, where the sensitivity is highest,39 for instance by SiO2 sandwiching40 and material-specific chemistry.44 This could be of interest for single molecule resolution since it increases the signal in λpeak per molecule, although it will not provide a higher signal for a given surface coverage. CONCLUDING REMARKS Using simple instrumentation, we have demonstrated a miniaturized plasmonic biosensor based on microspectroscopy of gold nanodisks. We have shown how extinction spectroscopy in transmission mode outperforms scattering spectroscopy in dark field in almost all cases, the primary exception being single particle spectroscopy. With the microextinction approach, we have shown a resolution in resonance wavelength of 10-3 nm at a 10 µm × 50 (45) Nusz, G. J.; Curry, A. C.; Marinakos, S. M.; Wax, A.; Chilkoti, A. ACS Nano 2009, 3 (4), 795–806.
Analytical Chemistry, Vol. 81, No. 16, August 15, 2009
6579
µm area for a temporal resolution of a few seconds. Similar performance in short-term noise could be maintained on an area as small as 2 µm × 10 µm, corresponding to 240 nanodisks. Furthermore, we have demonstrated biomolecular binding and a detection limit of 40 pg/cm2, in terms of protein coverage, on the microscale. We have also shown that the system is essentially as close to single molecule resolution as sensors based on single plasmonic particles. In order to reach this goal, the noise in λpeak must be reduced or the shift in λpeak per molecule increased. The number of disks on the surface can be changed, which in combination with the analyzed area will be important when aiming for single molecule resolution.45 However, such an optimization is highly complex since a lower number of disks strongly reduces the peak magnitude and introduces more noise in C. Future improvements in performance can also be accomplished by, e.g., active vibration damping, microscope auto focus control, and actuation in feedback loops based on the camera image, although such additional components make the concept lose some of its attractive simplicity. To the best of our knowledge, the performance shown here is comparable to or better than what can be achieved in any commercial instrument, such as imaging SPR, with respect to both
6580
Analytical Chemistry, Vol. 81, No. 16, August 15, 2009
miniaturization and detection limit on the microscale.18 However, it should be noted that our current system cannot perform parallel readout of sensor spots in arrays. In principle, it is possible to perform 1D spectral imaging by reading each pixel row separately in the spectrometer detector.15 A complete sensing array in 2D can be accomplished by using monochromatic light source and a suitable camera,10 although important information may be lost since the full spectrum cannot be measured. We consider parallel readout with maintained LSPR sensor performance the next challenge in the tradition of this work. ACKNOWLEDGMENT This work was financially supported by the Swedish Research Council and the FP7-NMP-ASMENA project. SUPPORTING INFORMATION AVAILABLE Details on data acquisition and analysis, noise characterization, and analysis of the active sensing area. This material is available free of charge via the Internet at http://pubs.acs.org. Received for review May 29, 2009. Accepted July 15, 2009. AC901175K
