Ultra-sensitive vibrational spectroscopy of protein monolayers with plasmonic nanoantenna arrays

We engineer nanoplasmonic antenna arrays for enhanced collective response starting with the individual antenna behavior. To obtain individual response with a good signal to noise ratio using a conventional Fourier-transform infrared (FTIR) microscope, we used randomly positioned antennas. The random arrangement of the nanoantenna array cancels out any diffractive behavior (24). As shown in SI Text and Fig. S1, the spectral measurements are independent from the randomization process and clearly reflect the individual antenna response. Fig. 1A shows the infrared reflectance measurements from randomized arrays consisting of varying length and a fixed width (230 nm) of nanorods pointing in the same direction on an area of 100 × 100 μm². The structures are fabricated on silicon substrates using electron beam lithography. A thin titanium adhesion layer is first deposited by electron beam evaporation, followed by ≈70 nm thick gold layer. After lift-off, well-defined rod shaped structures were formed as shown in the scanning electron microscope (SEM) image of an individual nanoantenna (Fig. 1A Inset). For incident light polarized parallel to the long axis of the rods (//-polarized), the nanoantenna resonance follows a linear scaling relation (Fig. 1B). This scaling behavior could be understood from antenna theory concepts. A traditional half-wave dipole antenna of length L has a tiny feed-gap in between two segments of length λ/4 at which the antenna is connected to its load. When the feed-gap impedance is matched to the impedance of the load, induced current is maximized in the antenna and the load circuit. Accordingly, resonance occurs at a wavelength corresponding to the total length of the two λ/4 segments (i.e., L = λ_Res/2). A nanorod antenna is fabricated as single particle with no feed-gap and the impedance matching condition is automatically met. As a result, the resonant excitation wavelength of a nanorod antenna with length L occurs approximately at λ_Res = 2Ln_eff. Here, n_eff is the effective refractive index, accounting for the inhomogeneous dielectric environment (substrate and air) surrounding the antenna. Slight deviations from the ideal half-wave dipole antenna behavior are due to the increasing penetration depth of the incident radiation at IR frequencies and decreasing aspect ratio of the long axis versus the short axis. Complete resonance, which is overlapping of the plasmonic excitations with the vibrational amide-I and amide-II bands, occurs for the nanoantenna length L = 1,100 nm. For incident light that is polarized perpendicular to the rod long axis, plasmon resonances are not excited over the spectral region of interest (Fig. 1A). The cross sectional near-field profile of the individual nanoantenna shown in Fig. 2A is obtained by three-dimensional finite difference time domain (3D FDTD) simulations. Enhanced field intensities with a factor of ≈10²–10³ are observed for parallel polarized (E_//) incident light near the tips of the nanoantenna. Although these enhancement factors are quite impressive, single monolayer measurements require even higher near-field intensities. This cannot be simply achieved by changing the nanoantenna geometry (25).

Collective excitations in well engineered antenna ensembles, however, can be exploited to achieve dramatically enhanced near-field intensities with much narrower far field spectral responses (15–19). In a periodically ensembled antenna array with a period d, long range dipolar interactions among the nanoantennas become dominant over the localized near-field couplings for separations more than several tens of nanometers. At a fixed incident wavelength, the interaction term (given by the second component in Eq. 1) accumulates larger phase delays when propagating between the antennas with increasing lattice spacing. The narrow resonance occurs for d slightly smaller than a critical periodicity d_c where the light fields corresponding to the diffractive grating order (i,j) change from evanescent to radiative:Here, n_Si is the refractive index of the substrate and (i,j) are the diffraction grating orders. Collective excitations are created when the dipolar interactions among the antennas are almost in-phase just before the appearance of the grating order resulting in radiation damping (Fig. 2B). Tuning of the individual antenna LSPs to this incident wavelength strongly enhances the collective response and results in efficient funneling of the incident light into much stronger near-field excitations (15–19). This behavior is well captured by the FDTD simulations presented in Fig. 2A. An order of magnitude stronger near-field intensities than those of individual antennas are observed at slightly shorter wavelengths than λ_Si(1,0). Here, λ_Si(1,0) is the critical wavelength where the (1, 0) grating order of the silicon interface becomes radiative at a given lattice periodicity (λ_Si(1,0) = n_Sid). This collectively enhanced near-field behavior is reflected as a narrowing of the far-field spectral response due to the suppression of the radiation damping with the confinement of the electromagnetic field within the array (16).

To enhance absorption signals, arrays are optimized using far field reflectance measurements by spectrally tuning the narrow resonances to the protein vibrational bands. Fig. 3A shows an SEM image of a fabricated periodic array of rod shaped antennas. The arrays consist of 1,100 nm long rods with varying periodicities ranging from 1.5 μm to 2.0 μm, and a random set that serves as a control to probe the individual antenna response. The resulting reflectance spectra are shown in Fig. 3B. For periodicities smaller than 2 μm, collective resonance peaks are above the (1, 0) grating order cutoff of the silicon interface (dashed line in Fig. 3C) where the field is evanescent (d < d_c). This results in narrower plasmonic far field resonances with respect to individual antenna as shown in Fig. 3C (red triangles). A progressive blue shift is expected for greater separations as the coupling among antennas is reduced, and the optical response converges back to that of an isolated particle. For d > d_c, the grating order field changes from being evanescent to radiative form in nature, which is associated with strong damping of the collective resonance and broadening of the line-width. In fact, at 2 μm periodicity, resonance occurs at a wavelength below the grating cut-off (d > d_c) and results in a broader linewidth. In comparison with the randomized arrays, the arrays with 1.5 μm and 1.6 μm periodicities have significantly narrower resonance line-widths (≈1 μm versus 2.75 μm for the individual particle behavior). As shown in Fig. 3C, the 1.6 μm periodic array offers the best combination of narrow line-width and spectral overlap with the protein amide-I band at 1,660 cm⁻¹.

Direct identification of the vibrational signatures of the protein monolayers is achieved by using this collective excitations. Here, the protein layer is applied to the nanoantenna substrates by uniform spin coating of a thin film of silk fibroin (detailed in SI Text). As mentioned before, a unique advantage of silk is that it allows fine control over the protein film thickness by varying its concentration in solution (22). We coat homogeneous films on the nanorod substrates as thin as 2 nm, corresponding to essentially a single protein monolayer. Atomic force microscopy (AFM) is used to confirm the uniformity and the thickness of the film. Initially, the scanning of the tip is performed in contact mode to scratch away the film down to the silicon substrate. Subsequent measurements at reduced force are then used to determine the film thickness (Fig. 4A). A combination of AFM and profilometer measurements confirmed the linear dependence of the film thickness on solution concentration.

As shown in Fig. 4C, protein absorption bands are clearly noticeable within the optical spectra collected from the protein-coated antenna arrays with 1.6 μm periodicity. Dips in the plasmonic response as a result of the amide I and II absorption bands are indicated in the figure at 1,660 and 1,537 cm⁻¹, respectively. Capacitive loading of the antenna with the protein layer results in slight red shifting of the plasmonic resonances (26, 27) (dashed curve in Fig. 4C). This shift is corrected using a polynomial fitting procedure in difference spectrum measurements (ΔR/R₀ = R_before/R₀ − R_after/R₀, where R₀ is the reflection signal from reference mirror). Fig. 4D shows the difference spectra of the periodic and the randomized arrays. The enhanced absorption signals well above the noise level are observed for the antenna arrays with narrow line-width (1.6 μm periodicity). Control measurements were performed on the bare silicon substrate with protein films of the same thickness in a region far from any fabricated nanoantennas. No absorption features were observed from the control samples. The observed signal (ΔR/R₀) of 6.8% in the periodic structures is nearly an order of magnitude higher than that of the randomized array, where the difference signals were 0.9%. This observation is in agreement with the enhanced near-field intensities predicted by the FDTD simulations (Fig. 2A).

To calculate the near-field enhancement factor, we compared the enhanced signal collected from the 1.6 μm periodic array to the expected reflectance signal from a 2 nm thick silk film on bare silicon substrate. Because the signal from protein films on bare silicon is below the noise level, instead we performed IR reflection absorption spectroscopy (IRRAS) measurements to obtain the expected value of the absorption signal for normally incident light. The frequency dependent dielectric function of the silk film was experimentally determined by fitting the measured IRRAS signal to a Lorentzian oscillator model (28). For a thin film (i.e., t ≪ λ) deposited on a metal substrate, the reflectance signal is approximately given by (10):where θ is the angle of incidence, t is the film thickness and ε₂ = ε_2,Re + iε_2,Im is the dielectric function of the silk layer. IRRAS measurements are performed at a grazing angle (80°) for 2 nm thick protein film coated on a 100 nm thick gold layer deposited on a silicon substrate. The incident light was polarized in the plane of incidence. The resonant frequencies, oscillator strengths and full-width half maxima of the Lorentz oscillator model for ε₂,were varied to obtain a fit with the data using Eq. 3. An estimated difference absorption signal of 4.7 × 10⁻²% is obtained by using three-layer Fresnel equations and ε₂ fitted through the IRRAS measurements. For an accurate estimate of the enhancement factor, we also need to include the following factors (see SI Text for details). The enhanced signal mainly comes from a small quantity of molecules at the close vicinity of the N² nanorod tips (n = 63 is the number of rows and columns). Second, in contrast to the studies conducted with self-assembled monolayers (SAMs) (7–13), here the protein molecules are physisorbed (see SI Text), which results in the following differences. (i) Unlike the SAMs, the transition dipole moments of the physisorbed proteins have no fixed orientation with respect to the metal surface normal. Accordingly, we expect that approximately one third of the transition dipoles of the molecules at the nanorod tips contribute to the absorption signal (see SI Text). (ii) Additionally, unlike in chemisorption, lack of molecular binding to the metal surface rules out any contribution of the chemical effects. (iii) Finally, given that the silk film is only 2 nm thick, it is unlikely that the physical deposition method results in the same degree of uniform coverage over the 70 nm high vertical sidewalls of our nanorods as would be possible with a SAM method. With these in consideration, we estimated the signal enhancement to be within the range of 10⁴–10⁵ (see SI Text). Such large enhancements allow us to detect absorption signals even in the raw spectral measurements from monolayer protein films with a commercial FTIR microscope. The detection volume of the antenna was calculated by considering the lateral area of the nanotips and the thicknesses of the monolayer protein film. Given that the density of the protein is 1.4 g/cm³ (29) and the molecular mass is ≈375 kDa (30), we estimate that the measured absorption signals are obtained from ≈300 zeptomoles for the entire array, corresponding to only 145 silk molecules per antenna (see SI Text). The large signal to noise ratios achieved in our measurements indicate that we should be able to observe vibrational signatures from even smaller quantities of protein molecules down to a few tens of zeptomoles.

In addition to the strong enhancements observed for the collective resonances, we observe variations in the absorption lineshapes as the spectral overlap between the antenna resonances and those of the protein amide bands are varied. We explain this experimental observation using a model based nonequilibrium Green's function formalism, which has been successfully applied in molecular/nanoelectronics (20–21). The strength of the NEGF formalism is that it provides a natural framework for describing the wave nature of the elementary excitations (electrons, phonons, vibrons, etc) in the presence of incoherent and dissipative processes. Conceptually, the system is partitioned into a nanoantenna and a protein load driven by the excitation of the plasmonic resonances as shown in Fig. 5A. Direct absorption of the incident light by the protein load is neglected because this absorption is expected to be very weak. The coupling of the incident light to the nanoantenna is expressed using a self-energy matrix [Σ_inc] whose anti-Hermitian component Γ_inc = [Σ_inc − Σ_inc⁺] is the broadening of the plasmonic antenna resonance due to the radiative coupling. Vibrational absorptions of the molecules are incorporated into the antenna/protein Hamiltonian [H_Ant+Load] using a coupling matrix [κ_Ant,Pro]. This matrix is designed to give unit transmission in spectral regions far from the absorption bands. A combination of Lorentz oscillator dips in the matrix strength are used to account for the absorption bands at the spectrally related regions. Reflected light is calculated through the second self-energy term [Σ_ref] with Γ_ref = [Σ_ref − Σ_ref⁺], which takes into account the open boundary condition in a similar fashion to the perfectly matching layers (PML) used in FDTD simulations. Reflected light intensity is obtained by R = tr(Γ_incG_Ant+LoadΓ_refG_Ant+Load⁺). Here, [G_Ant+Load] = [(ℏω)I − H_Ant+Load − Σ_inc − Σ_ref]⁻¹ is the Green's function of the antenna/protein system coupled to the open boundaries and [I] is the identity matrix. Retarded Green's function [G_Ant+Load] is a frequency dependent non-Hermitian matrix incorporating complex self-energy terms through which the radiative/nonradiative decaying of the plasmonic excitations are taken into account. The model absorption lines are defined as Lorentz oscillators at 1,660 cm⁻¹ and 1,537 cm⁻¹ with a width of 30 cm⁻¹. As shown in Fig. 5B, for the amide-II band, when there is an exact match between the plasmonic antenna resonance and the absorption band, the spectral lineshape is Lorentzian. For slightly detuned absorption resonance, such as for the amide-I, an asymmetric Fano type line-shape is clearly noticeable. The relative strength of the absorption resonances is due to the relative coupling strength of the incident infrared radiation to the protein vibration band. As expected, with the detuning of the antenna resonance the relative absorption strength of the vibration bands can be modulated (Fig. 5C).

The ability to control the thickness of the silk protein films from several nanometers to several micrometers provides a unique opportunity to probe the near-field behavior of the nanorod antenna. Due to the rapid decaying of the strongly enhanced near fields with distance from the nanorod surfaces (as can be seen in the FDTD simulations in Fig. 2A), saturation of the enhancement is expected to occur for films as thin as 40 nm. Although the numerical calculations offer insight into this observation, we use silk films as a near-field probe and experimentally demonstrate the surface nature of the enhancement by varying the film thicknesses (2, 4, 20, 40, and 100 nm) on identically patterned nanoantenna substrates. Fig. 6A shows the difference absorption spectra for the 1.6 μm periodic array with L = 1,100 nm. Measurements are also taken from a periodic structure with 2 μm perodicity and a randomized array for the same rod length. The strength of the absorption signal (amide-I) is plotted as a function of the film thickness in Fig. 6B for all of the structures (signals are normalized to take into account the difference of nanoantenna numbers in each array). Enhancement in the absorption signal strength is significant with increasing protein film thickness from 2 to 20 nm, but it appears to saturate as the thickness reaches to 40 nm. This behavior is in contrast to a steady linear dependence one would expect from Beer-Lambert's law for very thin films (see SI Text for details). Beyond 100 nm film thickness, bulk infrared absorption signals from thick protein films become observable. The other critical observations in Fig. 6 are as follows. Both of the periodic arrays with collective resonance display markedly higher signals than the randomized arrays, in agreement with the discussion presented above. The consistency of the data from periodic samples emphasizes the repeatability of the CEIRA measurements on the lithographically patterned substrates. Finally, experimentally measured enhancement saturation resulting from decaying near-field intensity distribution is confirmed by numerically calculating sampling volumes (for the 1.6 μm periodic array). For a given silk film thickness, t, the volumes are computed from the FDTD by summing over the simulation grids, which are within a distance t from the rod surface and have a near-field intensity >1/e² of the average maximum. The agreement between the calculated mode volumes and the absorption signal saturation strongly emphasizes the surface nature of the CEIRA effect.

Nanoparticle antenna arrays were fabricated on silicon wafers by a single layer liftoff process as described in SI Materials and Methods.

All spectral data were taken on a Fourier transform infrared spectrometer with an IR microscope as described in SI Materials and Methods.

Silk films were prepared by a spin coating procedure described in SI Materials and Methods.

CST Microwave Studio was used for the FDTD simulations as described in SI Materials and Methods.