LETTER pubs.acs.org/NanoLett
Plasmon-Enhanced Colorimetric ELISA with Single Molecule Sensitivity Si Chen,† Mikael Svedendahl,† Richard P. Van Duyne,‡ and Mikael K€all*,† † ‡
Department of Applied Physics, Chalmers University of Technology, 412 96 G€oteborg, Sweden Chemistry Department, Northwestern University, Evanston Illinois 60208-3113, United States
bS Supporting Information ABSTRACT: Robust but ultrasensitive biosensors with a capability of detecting low abundance biomarkers could revolutionize clinical diagnostics and enable early detection of cancer, neurological diseases, and infections. We utilized a combination of localized surface plasmon resonance (LSPR) refractive index sensing and the well-known enzymelinked immunosorbent assay to develop a simple colorimetric biosensing methodology with single molecule sensitivity. The technique is based on spectral imaging of a large number of isolated gold nanoparticles. Each particle binds a variable number of horseradish peroxidase (HRP) enzyme molecules that catalyze a localized precipitation reaction at the particle surface. The enzymatic reaction dramatically amplifies the shift of the LSPR scattering maximum, λmax, and makes it possible to detect the presence of only one or a few HRP molecules per particle. KEYWORDS: Surface plasmon, ELISA, biosensor, single molecules, single particles, spectra imaging
T
he ability to detect protein biomarkers within a limited reaction time is vital for disease monitoring and rapid public health screening. Unfortunately, many proteins of potentially high clinical relevance have such low abundance that detection using established methods, such as conventional enzyme-linked immunosorbent assay (ELISA), surface plasmon resonance (SPR), waveguide or quartz crystal microbalance sensors, becomes problematic. This has driven the development of a range of novel ultrasensitive detection methodologies, including Au nanoparticleDNA biobarcodes,1 whispering-gallery mode,2 and nanomechanical3 resonators. A particularly interesting strategy is to use single molecule fluorescence immunoassays,46 which can be combined with magnetic beads to overcome the slow diffusion process.5 But many diagnostic applications require additional features apart from high sensitivity, such as multiplexing capability, ease of use, low cost, and portability.7 Surface bound metal nanoparticles that exhibit localized surface plasmon resonances (LSPRs) fulfill many of the basic requirements for these functionalities. A single gold or silver particle, a few tens of nanometers in diameter, has a sensing volume in the attoliter range, which is well suited for detection of adsorbed protein molecules.8 It has a scattering cross-section at resonance many orders of magnitude higher than the fluorescence cross-section of the best fluorophores available and it does not suffer from bleaching or blinking,9 thus enabling simple colorimetric readout. Metal nanoparticles can also be printed in large high-density arrays using, for example, nanoimprint lithography,10 nanosphere lithography,11 or hole-mask colloidal lithography,12 thereby offering cost-effective platforms for multiplexed biosensing. r 2011 American Chemical Society
Similar to classical label-free SPR biosensing, the most common signal transduction method in nanoplasmonic sensing is to utilize the shift in the LSPR due to the change in refractive index caused by target binding.13,14 The LSPR wavelength can be tracked with very high accuracy for ensembles of nanostructures, resulting in detection sensitivities as low as 40 pg/cm2 of adsorbed mass.15 However, single or close to single molecule sensitivity is required to detect protein concentrations in the femtomolar range within acceptable reaction times using surface supported nanostructures.16 In practice, this implies that one needs to be able to resolve the peak shift induced by a single protein adsorbing on a single nanoparticle. Moreover, many single particles need to be interrogated simultaneously to yield a reasonable dynamic range.6 Even though single molecule LSPR sensing has been reported for highly irregular nanoparticles,17 most previous studies have only been able to resolve full or close to full monolayer protein coverage.18,19 This is not very surprising because the peak shift induced by a protein monolayer, usually a few hundreds of molecules on a single particle,13,19 is in the range of a few nanometers in aqueous solution. Thus, a single protein molecule can only be expected to induce a sub-0.1 nm peak shift on average,17,20 which is below the detection limit of current single particle LSPR measurement technologies. To circumvent this problem, we developed a multiplexed single particle LSPR assay that utilizes enzymatic signal amplification. The inspiration comes from the ELISA technique and several recent reinvestigations of the signal Received: February 22, 2011 Revised: March 15, 2011 Published: March 23, 2011 1826
dx.doi.org/10.1021/nl2006092 | Nano Lett. 2011, 11, 1826–1830
Nano Letters
LETTER
Figure 1. Schematic representation of the experiment. (a) Dispersed, biotinylated, gold nanoparticles are studied on glass substrates. Each biotin can bind a streptavidin-HRP conjugate that can catalyze a precipitation reaction on the particle surface. The high refractive index of the precipitate in turn leads to a detectable shift in the LSPR λmax. (b) A large number of nanoparticles are interrogated simultaneously using dark-field imaging and a tunable narrow bandpass liquid crystal filter. (c) Images recorded at discrete wavelengths are combined so as to yield scattering spectra for each individual nanoparticle.
enhancement ability of enzymatic proteins in biosensors.10,21 We demonstrate that it is possible to measure the shift of the LSPR λmax caused by the enzymatic activity of one or a few horseradish peroxidase (HRP) molecules per single particle, thus providing a basis for further development of simple and robust colorimetric bioassays with single molecule resolution. The experiment is outlined in Figure 1a. The basic idea is to use individual plasmonic nanoparticles as transducers that report on the enzymatic activity of single HRP molecules bound to the particle surface. We investigate circular gold nanoparticles fabricated by hole mask colloidal lithography (HCL) on a glass substrates.12 The particles are 60 nm in diameter, 60 nm in height, and of slightly conical shape. The LSPR occurs at ∼630 nm and the bulk refractive index sensitivity is ∼100 nm per refractive index unit (RIU), see Supporting Information for further characterization information. The particles are biotin-functionalized through immersion in a 1 mM solution of 1:3 biotinylated tri(ethylene glycol)-undecane-thiol and 1-octanethiol for at least 24 h to allow for the formation of a well-defined self-assembled monolayer (SAM). We then use streptavidin-HRP conjugates (Sigma) diluted in phophate-buffered saline (PBS) buffer to induce a simple biorecognition reaction on the particle surfaces. Samples were incubated with different SA-HRP concentrations for one hour under stagnant conditions, after which unbound HRP where removed through several buffer exchanges. The enzymatic reaction is then initiated by introducing 1 mM 30 3-diaminobenzidine (DAB) together with 1 mM H2O2 into the measurement chamber. The soluble DAB monomers are rapidly oxidized by H2O2 in the presence of HRP and then polymerize to an insoluble precipitate around the catalytic site.22 The precipitation reaction takes only 10 min and is stopped by removing the precursor through additional buffer exchanges. The experimental setup for spectroscopic imaging has been described previously.23 In brief, a substrate with well-separated gold nanoparticles is placed in a flow cell attached to the sample stage of an inverted microscope (Nikon TE2000E) equipped
with a dark-field condenser. Single particle scattering images are then recorded using a 100 objective and a liquid crystal tunable filter (LCTF, VariSpect CRI) placed before the CCD camera (SpectroPro 2500i, Princeton Instruments). The LCTF scans through the visible spectrum in 1 nm increments, thereby producing an image stack that can be used to reconstruct the scattering spectra for each of the ∼50100 nanoparticles in the field of view. The single particle data were complemented by high-resolution extinction spectroscopy measurements, see, for example, ref 13 for experimental details. These measurements were performed on ensembles of particles of identical size and shape as in the single particle studies but with much higher surface coverage; see Supporting Information for details. In the following, it will be shown that single particle LSPR measurements exhibit a measurable response down to the level of a single HRP enzyme molecule per particle. This will be done in several consecutive steps. First, ensemble measurements are used to construct a calibration curve that relates the λmax shift observed after the enzymatic precipitation Δλproduct to the much smaller λmax shift induced by the HRP adsorption itself ΔλHRP. Second, we estimate the number of HRP molecules (#HRP) per particle that corresponds to a certain peak shift ΔλHRP. Together, this allows one to convert from Δλproduct to #HRP at the single particle level even at the lowest HRP concentrations. The results are crosschecked by comparing the spread in #HRP values between different single particles with Poisson statistics and by comparing the average #HRP values with what is expected from a numerical solution of the diffusion equation for the specific HRP concentration used in the experiments. Figure 2a displays the kinetics for a typical assay measured on ensembles of nanoparticles using the extinction method for an intermediate HRP concentration of 4.5 nM. The curve shows resonance wavelength shift versus time while SA-HRP diffuses and binds to the biotinylated particles (between t ≈ 7 and 67 min), during rinsing (t ≈ 6777 min), and finally during the enzymatic 1827
dx.doi.org/10.1021/nl2006092 |Nano Lett. 2011, 11, 1826–1830
Nano Letters
Figure 2. LSPR peak shifts induced by SA-HRP coupling to biotinylated Au particles and subsequent enzymatic enhancement from ensemble averaged extinction measurements. (a) Peak shift versus time during binding of SA-HRP (red curve, multiplied by a factor 50) followed by HRP induced precipitate formation on particle surface. The inset shows the baseline variation before HRP injection. (b) SAHRP induced peak shifts and corresponding enzyme enhanced peak shifts versus HRP concentration. The lines are guides to the eye only. (c) Calibration curve for HRP induced peak shift from the data in (b).
reaction (starting at t ≈ 77 min). Note that the enzymatic reaction enhanced the colorimetric response on the order of 50 times! Note also that the noise level in the peak shift determination (given by the fluctuation in baseline, see inset) is only 0.002 nm or less! This means that SA-HRP can be detected at very low surface coverage in ensemble measurements, as will be shown below. Figure 2b shows SA-HRP induced peak shifts ΔλHRP and the corresponding enzymatic shifts Δλproduct versus HRP concentrations spanning 6 orders of magnitude. The data were obtained from the kinetics curves, such as Figure 2a, 60 min after HRP injection and 10 min after starting the enzymatic reaction, respectively. The peak shifts allow us to define an enhancement
LETTER
factor M = (Δλproduct Δλnegative)/(ΔλHRP), where Δλnegative ≈ 0.7 nm is the shift induced by injection of the enzyme substrate in the absence of HRP. We find that M ranges between ∼90 and ∼16 from the lowest to the highest HRP concentrations (see Figure 2b). This variation can be explained by self-quenching that is caused by newly formed products that react with the enzyme and diminish its catalytic activity. The physical mechanism in the present case is simply clogging of the enzyme from the precipitated DAB polymer,24 an effect that will be most pronounced for high surface densities. To be able to convert a precipitation response back to the amount of HRP on the surface of a nanoparticle, a calibration curve between ΔλHRP and Δλproduct is drawn as in Figure 2c. We find that a linear relationship ΔλHRP = M1(Δλproduct 0.7), with M = 85 ( 1.0 gives a good fit at the lowest HRP concentrations, where selfquenching is minimized. We can then estimate #HRP in two ways, based on the physical size of the molecule and based on the refractive index sensitivity of the nanoparticles, respectively. First note that the two highest SA-HRP concentrations in Figure 2b, 0.1 and 1.7 μM, result in the same peak shift ΔλHRP = 2.76 nm. This shift obviously corresponds to saturation, that is, formation of a protein monolayer. Assuming that the proteins attach to the nanoparticles through random sequential adsorption, we expect that saturation coverage corresponds to ∼55% percent of the total surface area being occupied by adsorbed molecules.25 The dimension of a single SA-HRP conjugate can be estimated to be ∼5 5 10 nm3 based on the crystalline structure of the two separate proteins. From the available surface area of a nanoparticle (∼14 000 nm2), one then finds that the maximum number of molecules per particle should fall in the range #HRP ≈ 150300, corresponding to all molecules either laying down or standing up on the surface, respectively. These numbers compare well with an estimate based on the refractive index sensitivity of the LSPR. As detailed in the Supporting Information, one can utilize the measured bulk refractive index sensitivity and an approximate field decay length of the LSPR to calculate how many SA-HRP molecules that are needed to produce the saturation peak shift ΔλHRP = 2.76 nm.13 The result is #HRP ≈ 190250, again corresponding to molecules laying down or standing up on the particle surface. To not overestimate the sensitivity, it is assumed that the LSPR peak shift at saturation coverage corresponds to the maximum number in the two estimates, that is, to 300 molecules. We now turn to the single particle measurements. It is useful to first make an estimate of the LSPR peak shift corresponding to a single molecule binding event. The lowest concentration investigated in the ensemble measurements, [SA-HRP] = 3.5 pM, produced a peak shift of ΔλHRP = 0.013 ( 0.002 nm, where the error is set by the baseline noise level. Using #HRP = 300(ΔλHRP/2.76), this implies an average of 1.4 proteins per particle, or 0.3% of a full monolayer. The corresponding precipitate induced peak shift is ∼M times larger, or Δλproduct ≈ 2 nm. Thus, if we can resolve peak-shifts of the order of 1 nm for single particles after the enzymatic reaction, we have in principle reached the single molecule detection limit. Figure 3a shows examples of typical single particle LCTF spectra and summarizes the peak shifts for HRP concentrations down to 350 fM. The shifts are sorted in 1 or 2 nm bins and are deduced from Lorentzian fits to spectra obtained before and after the HRP induced precipitation reaction. As for the ensemble measurements, the SA-HRP solution was incubated for one hour with the biotinylated particles and the enzymatic reaction proceeded until saturation occurred. We also show a negative control obtained after DAB injection to particles without HRP. 1828
dx.doi.org/10.1021/nl2006092 |Nano Lett. 2011, 11, 1826–1830
Nano Letters
LETTER
Figure 4. (a) Number of HRP molecules per particle estimated by applying the calibration curve in Figure 2c. A Poisson distribution is plotted based on the estimated average number of molecules per particle. (b) Poisson distributions generated from FEM simulations of SA-HRP diffusion to a single nanoparticle. The only adjustable parameter in the simulations is the SA-HRP concentration in solution.
Figure 3. (a) Peak shifts Δλproduct induced by HRP induced precipitation on single particles for different concentrations of SA-HRP ranging from 3.5 nm down to 350 fM. The column to the right shows typical single particle raw spectra obtained before (black) and after (red) the enzymatic precipitation reaction. (b) Average enzymatic peak shift Δλproduct versus [SA-HRP] from single particle (variance displayed as error bars) and ensemble measurements. The lines in the figure are sigmoidal fits as a guide to the eye.
The average shift is the same as for the ensemble, Δλnegative ≈ 0.7 nm. As shown in Figure 3a, however, the average Δλproduct for the single particles are slightly higher than the corresponding ensemble data (Figure 3a). We interpret this gain in sensitivity as a result of the much lower particle coverage in the single particle measurement, which means that each particle will accumulate a larger number of SA-HRP molecules for a given incubation time and SA-HRP concentration. As outlined above, we now convert the single particle peak shifts Δλproduct to #HRP using the calibration curve in Figure 2c. The results for the three lowest concentrations are displayed in Figure 4a. The data show that the average number of HRP molecules initiating the precipitation reaction is h#HRP = 3, 8, and 21 per particle for [SA-HRP] = 350 fM, 3.5 pM, and 35 pM, respectively. The two lowest concentrations, which yields average shifts well separated from the negative control, thus correspond to only one or a few molecules per particle. To confirm this conclusion, we then generate Poisson probability distributions
P(N) = (#hHRP)Ne#hHRP/N! defining the probability of finding N HRP molecules on a particle given the average h#HRP. These distributions are clearly very similar to the experimental variance in #HRP obtained from the spread in peak shift values, as can be seen from Figure 4a. Measurement uncertainties, in contrast, would be expected to result in a variance independent of h#HRP. As an additional consistency check for the number distributions in Figure 4a, we performed finite element method (FEM) simulations for the diffusion of SA-HRP molecules to a single surface supported hemispherical nanoparticle of the same surface area as in the experiment, see Supporting Information for details. We used the diffusion constant for SA-HRP obtained from the StokeEinstein’s equation, the same diffusion time (one hour) as in the experiments, and streptavidin and biotin adsorption and desorption constants from experiments on planar gold surfaces.26 The only input parameter is thus the SA-HRP bulk concentration. The simulations yielded h#HRP values of 0.2, 2, and 19 molecules per particle for the three lowest concentrations. As can be seen from the corresponding Poisson distributions in Figure 4b, the agreement with the experimental estimates is quite remarkable, in particular when considering that no adjustable parameter is used in the simulation. In conclusion, we have demonstrated a simple and straightforward method for colorimetric detection of a very low number of HRP molecules present on a single plasmonic nanoparticle. Specifically, for the lowest bulk HRP concentrations investigated, we find measurable plasmon peak shifts corresponding to only one or at most a few HRP molecules per particle. This single molecule sensitivity should persist if HRP is linked to a secondary antibody that binds clinically relevant antigenes in a sandwich assay. Thus, we believe that the results will pave the way for the development of nonfluorescence single molecule ELISA and other simple but ultrasensitive detection schemes suitable for clinical environments. 1829
dx.doi.org/10.1021/nl2006092 |Nano Lett. 2011, 11, 1826–1830
Nano Letters
’ ASSOCIATED CONTENT
bS
Supporting Information. Additional information regarding particle characterization, how to calculate the number of molecules on a nanoparticle, and parameters used for the COMSOL simulations of molecular adsorption. This material is available free of charge via the Internet at http://pubs.acs.org.
’ AUTHOR INFORMATION Corresponding Author
*E-mail:
[email protected].
’ ACKNOWLEDGMENT We thank Dr. Julia Bingham for technical support. This work was supported by the Swedish Research Council through the Linnaeus Center Bioinspired Supramolecular Function and Design (SUPRA) and Sweden’s innovation agency VINNOVA through Project 201002762. In addition, the experiments done at Northwestern University were supported by the National Science Foundation (Grants EEC-0647560, CHE-0911145, DMR-0520513, and BES-0507036) as well as the National Cancer Institute (1 U54CA119341-01), ’ REFERENCES (1) Nam, J. M.; Thaxton, C. S.; Mirkin, C. A. Nanoparticle-Based Bio-Bar Codes for the Ultrasensitive Detection of Proteins. Science 2003, 301, 1884–1886. (2) Armani, A. M.; Kulkarni, R. P.; Fraser, S. E.; Flagan, R. C.; Vahala, K. J. Label-Free, Single-Molecule Detection with Optical Microcavities. Science 2007, 317, 783–787. (3) Burg, T. P.; Godin, M.; Knudsen, S. M.; Shen, W.; Carlson, G.; Foster, J. S.; Babcock, K.; Manalis, S. R. Weighing of Biomolecules, Single Cells and Single Nanoparticles in Fluid. Nature 2007, 446, 1066–1069. (4) Rondelez, Y.; Tresset, G.; Tabata, K. V.; Arata, H.; Fujita, H.; Takeuchi, S.; Noji, H. Microfabricated Arrays of Femtoliter Chambers Allow Single Molecule Enzymology. Nat. Biotechnol. 2005, 23, 361–365. (5) Rissin, D. M.; Kan, C. W.; Campbell, T. G.; Howes, S. C.; Fournier, D. R.; Song, L.; Piech, T.; Patel, P. P.; Chang, L.; Rivnak, A. J.; Ferrell, E. P.; Randall, J. D.; Provuncher, G. K.; Walt, D. R.; Duffy, D. C. Single-Molecule Enzyme-Linked Immunosorbent Assay Detects Serum Proteins at Subfemtomolar Concentrations. Nat. Biotechnol. 2010, 28, 595–599. (6) Tessler, L. A.; Reifenberger, J. G.; Mitra, R. D. Protein Quantification in Complex Mixtures by Solid Phase Single-Molecule Counting. Anal. Chem. 2009, 81, 7141–7148. (7) Giljohann, D. A.; Mirkin, C. A. Drivers of Biodiagnostic Development. Nature 2009, 462, 461–464. (8) McFarland, A. D.; Van Duyne, R. P. Single Silver Nanoparticles as Real-Time Optical Sensors with Zeptomole Sensitivity. Nano Lett. 2003, 3, 1057–1062. (9) Anker, J. N.; Hall, W. P.; Lyandres, O.; Shah, N. C.; Zhao, J.; Van Duyne, R. P. Biosensing with Plasmonic Nanosensors. Nat. Mater. 2008, 7, 442–453. (10) Lee, S.-W.; Lee, K.-S.; Ahn, J.; Lee, J.-J.; Kim, M.-G.; Shin, Y.-B. Highly Sensitive Biosensing Using Arrays of Plasmonic Au Nanodisks Realized by Nanoimprint Lithography. ACS Nano 2011, 5 (2), 897–904. (11) Haynes, C. L.; Van Duyne, R. P. Nanosphere Lithography: A Versatile Nanofabrication Tool for Studies of Size-Dependent Nanoparticle Optics. J. Phys. Chem. B 2001, 105, 5599–5611. (12) Fredriksson, H.; Alaverdyan, Y.; Dmitriev, A.; Langhammer, C.; Sutherland, D. S.; Z€ach, M.; Kasemo, B. Hole-Mask Colloidal Lithography. Adv. Mater. 2007, 19, 4297–4302. (13) Chen, S.; Svedendahl, M.; K€all, M.; Gunnarsson, L.; Dmitriev, A. Ultrahigh Sensitivity Made Simple: Nanoplasmonic Label-Free
LETTER
Biosensing with an Extremely Low Limit-of-Detection for Bacterial and Cancer Diagnostics. Nanotechnology 2009, 20, 434015. (14) Svedendahl, M.; Chen, S.; Dmitriev, A.; K€all, M. Refractometric Sensing Using Propagating versus Localized Surface Plasmons: A Direct Comparison. Nano Lett. 2009, 9, 4428–4433. (15) Dahlin, A. B.; Chen, S.; Jonsson, M. P.; Gunnarsson, L.; K€all, M.; H€o€ok, F. High-Resolution Microspectroscopy of Plasmonic Nanostructures for Miniaturized Biosensing. Anal. Chem. 2009, 81, 6572–6580. (16) Sheehan, P. E.; Whitman, L. J. Detection Limits for Nanoscale Biosensors. Nano Lett. 2005, 5, 803–807. (17) Mayer, K. M.; Hao, F.; Lee, S.; Nordlander, P.; Hafner, J. H.; Single, A Molecule Immunoassay by Localized Surface Plasmon Resonance. Nanotechnology 2010, 21, No. 255503. (18) Raschke, G.; Kowarik, S.; Franzl, T.; S€onnichsen, C.; Klar, T. A.; Feldmann, J.; Nichtl, A.; K€urzinger, K. Biomolecular Recognition Based on Single Gold Nanoparticle Light Scattering. Nano Lett. 2003, 3, 935–938. (19) Baciu, C. L.; Becker, J.; Janshoff, A.; S€ onnichsen, C. ProteinMembrane Interaction Probed by Single Plasmonic Nanoparticles. Nano Lett. 2008, 8, 1724–1728. (20) Haes, A. J.; Van Duyne, R. P. A Unified View of Propagating and Localized Surface Plasmon Resonance Biosensors. Anal. Bioanal. Chem. 2004, 379, 920–930. (21) Pita, M.; Cui, L.; Gaikwad, R. M.; Katz, E.; Sokolov, I. High Sensitivity Molecular Detection with Enzyme-Linked Immuno-Sorbent Assay (ELISA)-Type Immunosensing. Nanotechnology 2008, 19, No. 375502. (22) Graham, R. C., JR.; Karnovsky, M. J. The Early Stages of Absorption of Injected Horseradish Peroxidase in the Promixal Tubules of Mouse Kidney: Ultrastructural Cytochemistry by a New Technique. J. Histochem. Cytochem. 1966, 14, 291–302. (23) Bingham, J. M.; Willets, K. A.; Shah, N. C.; Andrews, D. Q.; Van Duyne, R. P. Localized Surface Plasmon Resonance Imaging: Simultaneous Single Nanoparticle Spectroscopy and Diffusional Dynamics. J. Phys. Chem. C 2009, 113, 16839–16842. (24) Klibanov, A. M.; Tu, T.-M.; Scott, K. P. Peroxidase-Catalyzed Removal of Phenols from Coal-Conversion Waste Waters. Science 1983, 221, 259-a. (25) Adamczyk, Z.; Siwek, B.; Zembala, M.; Belouschek, P. Kinetics of localized adsorption of colloid particles; Elsevier Science B.V.: Schlieben, 1994. (26) Tang, Y. J.; Mernaugh, R.; Zeng, X. Q. Nonregeneration Protocol for Surface Plasmon Resonance: Study of High-Affinity Interaction with High-Density Biosensors. Anal. Chem. 2006, 78, 1841–1848.
1830
dx.doi.org/10.1021/nl2006092 |Nano Lett. 2011, 11, 1826–1830
