The nanohole array (Fig. 1a) acts as a planar waveguide that supports 2D guided mode resonances. It supports two optical modes with different profiles in the wavelength region of interest (Fig. 1b). The structure is realised in oxygen-enriched hydrogenated amorphous silicon (a-SiOx:H), which is transparent down to wavelengths as short as 650 nm (see Methods). The modes are excited with a normally incident collimated beam from an unpolarised halogen light source filtered through a monochromator; all spectra are normalised to a silver mirror. Both modes present a clear Fano lineshape of high Q-factor and high dynamic range (Fig. 1, Supplementary Informations 1).
We have designed the nanohole array in order to optimize the performance of both resonance modes concurrently, in particular aiming at improving the Q-factor, the sharpness of the Fano resonances and the dynamic range (See Supplementary Informations 2)
We have calculated the field distribution of both resonant modes (Fig. 1), assuming an input field polarized along the z-axis, noticing that the reflection spectrum does not change by rotating the polarisation of the input field by 90 degrees, i.e. to the x-axis. Therefore, we define the TE or TM mode according to the mode distribution and not according to the input polarisation.
The TM mode is resonant at λ = 667 nm with a Q = 450 (the Q-factor for a Fano resonance is defined as the spectral distance between the dip and the peak [22]) with a reflectance Rmax_TM = 0.6. We refer to this mode as TM because its dominant E-field is perpendicular to the surface. The second mode at λ = 735 nm exhibits a Q = 300 and has a higher dynamic range with a reflectance Rmax_TE ~ 0.9. We refer to this as the TE mode because its dominant E-field is in the plane of the waveguide.
A remarkable advantage obtained with the dielectirc nanohole array is a high value of SNR related to both modes. We define SNR = DR/σspectrum with DR the dynamic range of the resonant mode and σspectrum is the standard deviation of the signal noise for both modes. Because of a high DR and low noise values, we have verified SNRTM = 78 and SNRTE = 160, with an evident improvement respect to other plasmonic configurations and even compared to related dielectric metasurfaces (See Supplementary Informations 3)
We have verified that both modes are suitable for imaging and sensing, however, their distinctive properties favour each mode for one of those applications. The field of the TM mode is extended in the lateral dimension, more akin to a typical GMR mode, while it is closely confined to the surface in the out-of-plane direction (Fig. 1c and Fig. 1e), which makes it more suitable for sensing. Conversely, the TE mode exhibits a high dynamic range and a very strong confinement to the holes, corresponding to an unusually strong localisation for a Bragg resonance, which makes it suitable for high-resolution imaging applications (Fig. 1d and Fig. 1f). We note that the existence of two different types of modes of distinct field distribution is a unique feature of a dielectric nanohole array, compared to a plasmonic nanohole array [23, 24] which only supports modes of a single polarisation [23]. We first consider the advantageous sensing properties of the TM mode before moving on to the imaging properties of the highly confined TE mode.
Sensing
The commonly used figure of merit for biosensing combines the sharpness of the resonance via its Q-factor with the sensitivity S of the resonance. The sensitivity is expressed as a wavelength change vs refractive index change Δλ/Δn in nm/RIU. So the figure of merit is SQ [25]. The sensitivity is commonly understood with respect to the bulk refractive index change, i.e. the response of the sensor to refractive index changes in the half-space above the sensor surface.
Two-dimensional plasmonic nanohole arrays, for example, can achieve bulk sensitivities above 700 nm/RIU [23, 26] and dielectric nanohole arrays operating at 1550 nm can be even better with a demonstrated sensitivity of S ~ 800 nm/RIU [27] and theoretically up to 4000 nm/RIU in the visible range [19], while 1-D arrays typically achieve between 100–300 nm/RIU [28]. While the bulk sensitivity is easy to measure, it is not the most relevant parameter for a surface-affinity sensor; instead, the surface sensitivity should be used, which describes the response of the sensor to refractive index changes at the very surface of the sensor [25, 29] which is much more representative of surface-bound proteins or DNA. Since there is no agreed thickness for the surface sensitivity, we here chose a thickness of 10 nm and a layer of SiO2 (n = 1.45) to represent the thickness of a typical protein bound to a surface via an antibody [30]. By using amorphous silicon with a relatively high refractive index of n = 2.45 and exciting the TM mode, we are able to confine the mode closely to the surface. We calculate the surface sensitivity as usual with S = Δλ/Δn, where in this case Δλ represents the resonance shift between the bare nanohole array and the same structure with 10 nm SiO2 deposited on the structure, as described in Fig. 2a, while Δn is the refractive index change between SiO2 (with the layer) and water (bare configuration). In both cases, water is assumed as the background medium. Using this method, we observe a surface sensitivity of 20 nm/RIU experimentally (Fig. 2b).
This value is comparable to that of plasmonic structures, which typically exhibit values of order 30 nm/ RIU for a 10 nm layer [23]. While the difference between the two types of structures is rather large in terms of bulk sensitivity, it is surprisingly close in terms of the parameter that actually matters for surface biosensing, i.e. the surface sensitivity (Supplementary Informations 3). In terms of the figure of merit SQ, we note that the Q-factor of the dielectric resonance is an order of magnitude (Q ~ 450, Fig. 1) higher than that of a plasmonic array (Q ~ 40, [23]), resulting in a significantly higher SQ figure of merit when S is the surface sensitivity rather than the bulk sensitivity. Moreover, our nanohole array also compares favourably to related structures, such as the recently introduced structures based on bound states in the continuum (BIC) (Q = 90 and Ss ~ 40 [9]. A higher value of SQ has been obtained in [31] with a dielectric nanohole array exploiting the BIC modes. However, we note that our nanohole array provides a significant enhancement of the SNR for both modes, which plays an important role in the optimization of the imaging and sensing resolution, as described in detail below.
For an overview of the different types of structures reported thus far, please refer to the table provided in the Supplementary Informations (Table S3).
We recognise that the Q-factor of any of these GMR-like structures is much lower than that of waveguide-based resonances such as microring resonators, but we note that waveguide-based resonances require excitation by end-fire coupling or with grating couplers, which require high precision coupling arrangements that are not compatible with the low-cost healthcare diagnostics approach that this work is aimed at.
In order to validate the advantageous properties of our approach, we conducted biological measurements using immunoglobulin G (IgG) as the target protein.
Nanohole array for biosensing application
For the biological measurements, we adopted the chirped configuration [32, 33] to the nanohole array for ease of readout. This configuration is obtained by tapering the period from Λ = 470 nm to Λ = 490 nm over a distance of 500 µm. Accordingly, for single wavelength illumination, the resonance will appear as a bright line at the output of the chirped array, spatially located where the period multiplied by the effective index matches the wavelength and is able to excite a resonance (Fig. 3). Any binding of the target biomarker to the sensor surface then causes a shift of the position of the line due to the change in effective index, thereby translating spectral into spatial information (Fig. 3a). This information is then easily read-out by a CMOS camera. In order to make the system immune to temperature variations of the environment, we include a second channel as a reference (Fig. 3b). The bulk sensitivity of the sensor is 140 nm/RIU (we include here the bulk sensitivity for ease of comparison with literature values) corresponding to a sensitivity for the chirped array of 3960 µm/RIU, defined as the spatial shift of the resonance position per unit change of the bulk refractive index.
The high dynamic range for both modes represents a further advantage of the dielectric nanohole array compared to its plasmonic counterpart, which has a much lower dynamic range with a typical transmission of T < 0.2. The high dynamic range is comparable to that observed with 1-D GMRs [28, 34], but we note that we achieve a significantly higher Q-factor (Q ≈ 450 compared to Q ≈ 150–250 for a 1-D GMR. The high Q-factor, together with the high dynamic range, produces a sharp and bright resonance line as shown in Fig. 3.
From the binding assay, we extract a noise limit of 3σ = 0.183 µm over 30 minutes, corresponding to approximately the time it takes to reach saturation when detecting low protein concentrations.
Together with the sensitivity of 3960 µm/RIU mentioned above, this noise limit translates into a limit of detection of 4.6 × 10− 5 RIU (Supplementary Informations 4).
For the surface functionalisation, we use a spacer layer of SM(PEG)6 between the sensor surface and the antibodies to decrease non-specific binding [35] (Methods).
We use immunoglobulin G (IgG), a generic marker for the human immune response, to quantify the protein sensing performance of the nanohole array. The binding assay (Fig. 4) includes the surface functionalization with anti-IgG antibodies immobilized on the PEG layer, followed by casein as an additional blocker against non-specific binding to optimise the sensor specificity, as previously verified in [35] (Methods).
Following functionalisation, we added different IgG concentrations (1 pg/mL, 10 pg/mL, 100 pg/mL, all in Phosphate Buffer Saline (PBS)) to the channel. In particular, for a concentration of 1 pg/mL, we observe a shift of 1.26 µm (= 0.9 pixels), corresponding to a wavelength shift of about Δλ ~ 30 pm, while the noise level in this specific measurement is 3σ = 0.78 µm (= 0.56 pixels), which confirms that we have achieved a limit of detection better than 1 pg/mL. This high performance is in part due to the use of the SM(PEG)6 spacer layer which we first introduced to GMR-based sensing in [35], where we also showed very low non-specific binding and high-sensitivity detection in human urine.
Remarkably, the demonstrated performance with LOD < 1 pg/mL is comparable to or better than the laboratory standard, i.e. fluorescence-based enzyme-linked immunosorbent assay (ELISA), which typically achieves sensitivities of the 3–5 pg/mL [36], yet our label-free and very simple approach is more suited for point-of-care applications. It also represents an improvement by over two orders of magnitude compared to plasmonic nanohole array [16] and is even better than a sandwich assay using metal nanoparticles based on a similar structure [37, 38] or based on other dielectric metasurfaces [9, 39]. We explain this improvement of performance with the high surface sensitivity, the high values of Q-factor and SNR, together with a sharp Fano resonance, which facilitates easy tracking of the resonance.
We have recently also demonstrated an interferometric approach based on guided mode resonances [40] where we have shown the detection of 1 pg/ml of procalcitonin (PCT) with very high SNR. We note that the main advantage of the nanohole array configuration shown here is its ability to obtain a similar sensing performance together with the imaging capability.
It is interesting to note that the TE mode also exhibits high sensitivity, i.e. its detection limit for IgG is comparable to that of the TM mode (Supplementary Informations 5), which is relevant for the following imaging section. We note, however, that the TM mode performs better in terms of reproducibility and signal-to-noise, which we attribute to the mode distribution being more suitable for detecting surface-bound molecules.
Single bacteria detection with resonant hyperspectral imaging
We now address the spatial confinement and refer to the TE mode (Fig. 1d). We note that the high dynamic range obtained with this mode should translate into high contrast imaging and thus improves the resolution imaging [41]. Furthermore, the fact that the optical field is strongly confined in the holes and less distributed on the surface improves the spatial resolution, which suggests the suitability of the TE mode for imaging. In order to verify this hypothesis, we use hyperspectral imaging applied to a standard configuration of dielectric nanohole array with fixed period providing the same resonance condition in the entire sensor area, whereby the wavelength of the light source is scanned and images are taken at every wavelength step with a simple CMOS camera (Methods). The peak wavelength of the resonance is subsequently extracted for each pixel as the wavelength that maximises reflectance [42, 9]. The spatial resolution of this method, for a dielectric 1-D GMR is typically of the order 2–6 µm [43, 34], which is usually limited by the penetration depth of the GMR into the grating, as discussed above. We note that a higher resolution as low as 0.5 µm has already been quoted with dielectric configurations [42]. However, such values refer to the identification of point sources, not the separation of two features, which instead is the commonly accepted method underpinning the Rayleigh criterion.
In order to test the resolution of the nanohole array, we first (as in [34]), deposit a high-resolution pattern and test for the separation between closely spaced features using hyperspectral imaging. We are able to observe an imaging resolution of better than 1 µm, which confirms the strong spatial resolution of the TE mode (Supplementary Informations 6).
We have also evaluated the imaging performance of the TM mode and obtained a spatial resolution of about 3 µm. As expected, the lower spatial resolution is mainly due to the lower SNR and the weaker localisation of the mode (Supplementary Informations 7).
Encouraged by these results, we turn to applying this high resolution to biological studies. Figure 5 shows that we can clearly resolve the shape and orientation of individual Escherichia coli bacteria (typical size of about 2 × 1 µm) with high accuracy, i.e. comparable to brightfield images obtainable with a microscope.
We demonstrate that our technique provides quantitative information with high spatial resolution by imaging and spatially localizing individual bacteria (Fig. 5 and Supplementary Information 8), enabling the detection of the refractive index changes caused by the presence of bacteria. We use the refractive index calibration determined for sensing (Figs S4, S5) to achieve quantification. We note the ability of the structure to provide imaging information with a refractive index resolution of 4.6 × 10− 5 RIU, which is not possible to reach with conventional imaging techniques. These results demonstrate that the many attractive features of resonant hyperspectral imaging for studies involving e.g. adhesion, secretion and cell viability previously demonstrated with mammalian cells [21, 22, 44] can now also be applied to bacteria.
Individual bacteria have already been studied with traditional techniques based on the use of fluorescent dyes [45]. However, the labelling process associated with these techniques complicates the procedure and may distort the result. In contrast, the nanohole array is based on a completely label-free approach and, in particular, because the structure is defect-free, the sensing area can be very large (up to few mm2, only limited by the field of view of the camera). This allows for the real-time monitoring of bacterial growth in a large area (See Supplementary Informations 9) while preserving a resolution sufficiently high to image individual bacteria. This capability is particularly important e.g. for studying the formation of biofilms and for testing antimicrobial susceptibility, a major problem in the quest to control antimicrobial resistance (AMR).