key: cord-0454541-4dvhjo1a
authors: Hu, Jack; Safir, Fareeha; Abendroth, John M.; Dixon, Jefferson; Pinsky, Benjamin A.; Jeffrey, Stefanie S.; Lawrence, Mark; Dionne, Jennifer A.
title: Rapid genetic screening with high quality factor metasurfaces
date: 2021-10-15
journal: nan
DOI: nan
sha: e6250d8f5ff8b8eaa9b0eb9a9cec9523e1858385
doc_id: 454541
cord_uid: 4dvhjo1a

Genetic analysis methods are foundational to advancing personalized and preventative medicine, accelerating disease diagnostics, and monitoring the health of organisms and ecosystems. Current nucleic acid technologies such as polymerase chain reaction (PCR), next-generation sequencing (NGS), and DNA microarrays rely on fluorescence and absorbance, necessitating sample amplification or replication and leading to increased processing time and cost. Here, we introduce a label-free genetic screening platform based on high quality (high-Q) factor silicon nanoantennas functionalized with monolayers of nucleic acid fragments. Each nanoantenna exhibits substantial electromagnetic field enhancements with sufficiently localized fields to ensure isolation from neighboring resonators, enabling dense biosensor integration. Quantitative detection of complementary target sequences via hybridization occurs simultaneously for arrays of sensing elements patterned at densities of 160,000 pixels per cm$^2$. In physiological buffer, our nanoantennas exhibit average resonant quality factors of 2,200, allowing detection of purified SARS-CoV-2 envelope (E) and open reading frame 1b (ORF1b) gene fragments with high sensitivity and specificity (up to 94$%$ and 96$%$) within 5 minutes of nucleic acid introduction. Combined with advances in nucleic acid extraction from complex samples (eg, mucus, blood, or wastewater), our work provides a foundation for rapid, compact, and high throughput multiplexed genetic screening assays spanning medical diagnostics to environmental monitoring.

Genetic screening methods have enabled significant advances in the prediction, detection, treatment, and monitoring of organism and ecosystem health. For example, respiratory panels identify pathogen nucleic acids indicative of infectious diseases like influenza and Coronavirus disease 2019 [1, 2] ; tissue and liquid biopsies detect cancerous genetic mutations, likelihood of recurrence, and are used to guide treatment [3, 4] ; and emerging environmental DNA sensors monitor the health of oceans, freshwater, livestock, soil and air [5, 6] . Current genetic screening methods include polymerase chain reaction (PCR), next-generation sequencing (NGS), Sanger sequencing, and DNA microarrays. Each utilizes oligonucleotide amplification followed by optical tagging to sensitively detect target sequences. Despite their tremendous utility in laboratory settings, translation of these screening methods to clinical and point-of-care applications is ultimately limited by their reliance on "traditional" optical signal transduction (absorption and fluorescence). Even with the best optical tags, sensitive and specific readouts are generally only achieved with time consuming thermal cycling and/or costly reagents for nucleic acid amplification.

Rather than amplifying the concentration of the biomolecule, we postulated that light could be resonantly amplified to help enable compact, point-of-care biomarker screening methods. Photonic devices strongly confine and scatter light; when decorated with molecular probes, target analyte binding alters the optical signal due to subtle changes in the polarizability or refractive in-dex of the resonator environment. Plasmonic sensors are among the most common affinity-based biosensors [7, 8, 9, 10, 11] , but have larger limits of detection set by the metals' intrinsic absorption; the resulting low quality factor (Q) resonances (Q∼10) give rise to poor differentiation of small binding signals; the resonator's sensing figure of merit (FOM), defined as sensitivity [resonant wavelength shift per refractive index unit (RIU) change] divided by the full width at half maximum (FWHM) of the mode, is ca. 1-10 RIU −1 for plasmonic sensors. More recently, dielectric nanoantennas and metasurface based sensors have been designed with Q factors of 10's-100's, with similar improvements in the FOM [12, 13, 14, 15, 16, 17, 18, 19] . Unlike high-Q whispering gallery mode resonators [20, 21, 22] and photonic crystal microcavity devices [23, 24] , these metasurfaces can be illuminated from free space and far field scattering can be readily controlled, an advantage in the scalability and integration of sensors in imaging based devices. [25] However, these systems typically rely on delocalized resonant modes formed from extended two-dimensional arrays to improve Q factors and the resultant large modal volumes reduce responses to binding of small amounts of target molecules. Additionally, the larger footprint of these arrays limits the dense incorporation of sensing elements for multiplexed analyte detection and data driven analyses.

In this work, we report a new genetic analysis platform based on our lab's development of high quality factor metasurfaces [26] . These metasurfaces consist of subwavelength nanoantennas that strongly confine light in the near field while affording precise control over far-field scattering. We design resonators that exhibit high average Q's in buffered biological media of 2,200, with strong field penetration into the surrounding environment for sensitive biomarker detection. We show that the FOM of our sensors is 400 RIU −1 , in good agreement with our computational model and significantly larger than existing nanophotonic sensors. We functionalize our resonators with self-assembled monolayers of DNA probes complementary to the SARS-CoV-2 E and ORF1b gene sequences. Hybridization of target nucleic acid fragments to the surface probes results in a rapid (<5 minute) change in the resonant wavelength, with sensitivities and specificities up to 94% and 96%, respectively. Due to the spatially localized nature of the high-Q resonances, individual sensing pixels can be patterned at densities of 160,000+ features per square cm, promising analyte parallelizability across a multitude of biomarkers.

Individually addressable high-Q resonator sensing platform Figure 1a illustrates our sensor design, which consists of rows of silicon nanoblocks illuminated with near-infrared light. Each row constitutes a one-dimensional guided-mode resonant (GMR) metasurface; the periodic modulation of block widths within each row, characterized by ∆d, allows for finite, but suppressed dipolar radiation and free space coupling to otherwise bound waveguide modes (Supplementary Note 1 and Supplementary Fig. 3 ) [26, 27, 28, 29] . The resulting long resonant lifetime translates to strong electric near-field enhancements (Fig. 1b) . Notably, electric fields at the surface of Si blocks are enhanced by 80x. Due to the gaps between discrete silicon blocks within the resonator, 29% of the electric field energy is exposed to the surrounding medium compared with 8% in a continuous or partially notched waveguide (Supplementary Note 2 and Supplementary Fig. 4 ). This field concentration in the gaps leads to greater sensitivity to surfacebound analytes. Additionally, these silicon resonators exhibit sharp scattering responses in the far-field. As seen in Figure 1c , calculated transmission spectra Q-factors exceed 5,000 for ∆d=50 nm, and can be further increased with decreased ∆d (vide infra).

We fabricate silicon resonators atop a sapphire substrate (Fig. 1d ) (see Methods). Utilizing a near-infrared supercontinuum laser and spectrometer equipped transmission microscope (Supplementary Fig. 1) , we illuminate the metasurfaces at normal incidence and simultaneously measure the transmitted spectra from multiple resonators (Fig. 1e) . By modulating the block lengths in adjacent nanostructures by ± 5 nm, we intentionally vary the spectral position of the resonant mode, highlighting that each waveguide structure can be individually addressed and tuned as a distinct resonator ( Fig. 1e & 1f) ; in other words, our high-Q resonances do not rely on inter-chain coupling or an extended 2-D array effect. This spatial localization of the optical modes makes our platform ideally suited for the integration of densely distributed and multiplexed sensor arrays.

Our metasurfaces are sealed in a 3-D printed fluid cell (Fig. 2a) and characterized in phosphatebuffered saline (PBS) solution (1x concentration) to represent physiological conditions for biomolecule detection. In Fig. 2b , we vary the perturbation ∆d along the block chain from ∆d = 100 nm to ∆d = 30 nm and observe a decrease in the resonant linewidth for 25-30 individual resonators at each condition ( Fig. 2b & 2c) . Importantly, in our high-Q metasurface design, the coupling strength between free space radiation and the GMR is dictated by the degree of asymmetry along the waveguide. Since silicon is lossless in the near infrared, radiative loss dominates the GMR resonant lifetime and Q factor. Thus, shrinking ∆d we observe scattering responses with average Q factors of 800 (at ∆d = 100 nm) increasing to 2,200 at ∆d= 30 nm and even observing Q's above 3,000 for individual resonators (Fig. 2c) . These Q factors represent a two to three order of magnitude increase compared to reported plasmonic biosensors, and a significant (>5-10x) increase compared to other metasurface biosensors [16, 17, 19, 30, 31] , yielding a FOM of ∼400 (Supplementary Fig. 7 ). Our experimental Q factors are slightly lower than numerically predicted and are likely limited due to scattering losses caused by fabrication imperfections. We also note that water has non-negligible absorption in the 1,500 nm wavelength range that may limit our attainable experimental Q factors (Supplementary Note 3 and Supplementary Fig. 5 ). Designing future metasurface resonances in an optical transparency window of biological media (such as 1,300 nm) and optimizing fabrication processes may further improve performance, with Q factors in the millions potentially attainable; future iterations of our metasurface could offer the single particle sensitivity of high-Q microcavities, [20, 22] but with the ease of integration and compactness afforded by free-space coupling.

Due to the localization of the mode along each individual chain, resonators can be spaced laterally at least as close as 3 µm without affecting the GMR (Fig 2d) . Based on our fabricated waveguide length of 200 µm, our devices feature sensor arrays with densities of over 160,000 sensors per cm 2 . Due to the slow group velocities of the GMR's, losses due to finite size effects can be suppressed [29, 32] , and 50 µm waveguides can be fabricated with comparable Q ( Supplementary  Fig. 6 ), yielding feature densities over 600,000 sensors per cm 2 . These large sensor densities offer an avenue for robust statistical analysis in diagnostic studies as well as a platform for multiplexed detection of many distinct biomarkers in parallel.

To utilize our sensor arrays for gene detection, we modified the silicon surface with DNA monolayers, where complementary nucleic acid sequences serve as capture molecules for a specified target. Self-assembled monolayers (SAMs) are deposited in a three-step process to covalently link 26 base pair single-stranded DNA (ssDNA) probes over the entire metasurface chip surface. The silicon surface is first functionalized with an amine-terminated silane (11-aminoundecyltriethoxysilane, AUTES), and then cross-linked via a heterobifunctional molecule (3-maleimidobenzoic acid Nhydroxysuccinimide ester, MBS) to thiolated ssDNA probes (Methods and Supplementary Fig. 2 ). In this study, we considered nucleic acid fragment targets of the envelope (E) and open reading frame 1b (ORF1b) genes of the SARS-CoV-2 virus (GenBank accession: MT123293.2 positions 26326→ 26351 and 18843→ 18866, respectively, also see Supplementary Table 1 ) [33, 34] (Fig.  3a) . As a proof of principle, we use synthetic DNA targets, but note that viral RNA will analogously hybridize to complementary DNA probes [35, 36] .

In Fig. 3b , measured spectra show clear resonant wavelength shifts as consecutive molecular monolayers of AUTES, MBS, and the probe DNA are grafted to the resonator surface. Monolayers were modeled as thin dielectric shells surrounding the silicon blocks and simulated responses show close agreement with the experimental resonance shifts (Fig. 3c )(Supplementary Note 6 and Supplementary Fig. 8 ). Upon adding the target SARS-CoV-2 gene, a clear, 0.4nm resonant shift is observed (Fig 3d) . Data was collected from N=75 individual resonators or chains of silicon blocks, and we note that the high density of sensing elements on our chips can enable significant increases in measurement throughput compared to typical photonic sensors where signals are averaged over larger 2-D arrays. The deviation between experimental and simulated wavelength shifts for the AUTES and MBS layers is likely due to the tendency for aminosilane molecules to form multilayer structures; differences in the attachment of DNA probes and subsequent target hybridization are likely due to a strong influence of steric hindrance and electrostatic repulsion effects on the packing density and hybridization efficiency of the DNA strands [37, 38, 39, 40] .

Pairing our resonators with specific probe DNA sequences offers specificity in target gene detection. To confirm specificity, we modify target DNA strands with ATTO590 fluorescent labels and incubate sensors functionalized with probes that are only complementary to the nCoV.E sequence. Fluorescence imaging of sensors exposed to 10 µM solutions of target nCoV.E and HKU.ORF1 show significant binding only for the complementary E gene target and minimal signal for the noncomplementary ORF1 strands ( Supplementary Fig. 10 ). This target specificity is also measured in the resonator scattering spectra, where resonance wavelength shifts are significant for complementary target-probe conditions and suppressed for non-specific binding (Fig. 4a) .

Our sensors exhibit concentration dependent responses from 1 µM to 10 nM (Fig. 4b) . Measurements are taken for N=50-75 individual resonators at each target and concentration condition. The large variability in resonant wavelength shifts at each concentration are likely due to the stochastic nature of surface binding; notably, the signal from any particular resonator will depend on the local concentration and spatial position of bound targets (and hence binding at sites with the greatest electric field concentration will produce larger resonant shifts). Additionally, signal variation is also likely introduced through the hydrolytic degradation of silane layers in aqueous solutions during functionalization and hybridization experiments [37, 38] (as seen in blue-shifted data points in Fig. 4b ). We expect optimization of the surface functionalization homogeneity and stability to dramatically improve the performance of our sensors. Importantly, we are confident that this background signal, which is currently the dominant factor limiting our resolution, comes entirely from the instability of the sensing environment and not the photonic resonators themselves. We ultimately expect our detection threshold to be limited by the resonant linewidth, with shifts <0.1*FWHM easily measurable. We note that on our metasurface chips the silane linking chemistry will also non-specifically functionalize DNA probes to the surface oxide of the sapphire substrate. We estimate that only 0.0003% of surface bound target molecules contribute to the resonance shift of each resonator. Thus, even with the current background signal, our limit of detection could be reduced from 10 nM down to 10 fM with the introduction of microfluidic channels where only resonator regions are exposed to target molecules, utilization of a silicon specific surface functionalization process, or incorporation of additional nanostructures to isolate resonators from one another and increase sensor densities further [41] . Additionally, the concentration dependent range of our device can potentially be tuned to different values of analyte concentration through modification of surface probe densities [42] .

Two-way analysis of variance (ANOVA) and post-hoc Tukey's range test indicates that differences in scattered shift signals were statistically different for complementary vs. non-complementary targets at all tested concentrations (Fig. 4b inset) (Supplementary Tables 2 and 3 ). The increased measurement throughput and larger sample sizing of our platform can be used to significantly improve the accuracy of diagnostic studies, where multiple measurement redundancies allow for improved quantification and classification of sample populations. For example, we can classify "positive" complementary target detection against "negative" non-complementary target detection at each concentration based on thresholding resonant wavelength shift signals. Varying the threshold signal produces a receiver operating characteristic (ROC) curve, and the positive and negative signal discrimination is quantified as the area under the ROC curve (AUC). From this analysis, our sensors exhibit AUC values up to 0.98 (where AUC = 1 indicates perfect signal discrimination and 0.5 represents no discrimination) and high sensitivity and specificity of 94 and 96% respectively ( Supplementary Fig. 11 ). This increased digitization of target gene binding may also be paired with machine learning based analysis for further improved accuracy or to allow for discrimination of small signals due to genetic variants and point mutations [43] .

Real-time measurement of resonators shows rapid target binding responses for a 100 nM solution of nCoV.E complementary targets measured across six representative resonators (Fig. 4c) . Changes in the resonant wavelength greater than the measurement noise are detected within seconds and the binding signal plateaus within 5 minutes of sample introduction. The signal response shows excellent agreement with the Langmuir adsorption model (dashed line Fig. 4c ) with an observed hybridization rate constant of 7x10 −3 s −1 , comparable to other hybridization capture assays [44, 45] . These fast binding kinetics highlight a key advantage of chip-based approaches over conventional detection techniques that require time-intensive sequence amplification cycles.

Our nanophotonic device offers a new platform for high throughput molecular analysis. We have demonstrated free space illuminated resonators with high-Q resonances in physiological media (2,200+ ) that can be patterned, tuned, and measured at densities exceeding 160,000 pixels per cm 2 . Even larger Q's and greater feature densities are attainable in our platform with improved fabrication processes to reduce scattering losses from structural inhomogeneities, reduced absorption losses from biological media, and inclusion of photonic mirror elements to suppress light leakage as resonator chains are truncated below 50 µm. Interfaced with DNA probes, our metasurface design enables rapid, label-free, and highly digitized genetic screening that can bridge many of the challenges faced by conventional genetic analysis techniques. Paired with bioprinting procedures where different gene sequence probes are spotted across distinct sensing pixels, our high-Q metasurface chips can provide the foundation for rapid, label-free, and massively multiplexed photonic DNA microarrays. Furthermore, our nanophotonic chips are amenable to intensity imaging and/or hyperspectral imaging techniques that provide signal binding information without the need for a spectrometer [15, 30] , further reducing complexity and costs towards point of care genetic screening. Our platform promises unique possibilities for widely scaled and frequently administered genetic screening for the future of precision medicine, sustainable agriculture, and environmental resilience. 

The metasurfaces were fabricated using standard lithographic procedures. First, 500 nm, single crystal silicon-on-sapphire (MTI Corp.) substrates were cleaned via sonication in acetone and isopropyl alcohol. The substrates were baked at 180 • C before spin coating with hydrogen silsesquioxane (HSQ) negative tone resist (XR-1541-06, Corning). The resist was baked for 40 min at 80 • C. To reduce charging, a charge dissipation layer (e-spacer, Showa Denko) was spin coated over the HSQ resist and baked again for 5 min at 80 • C. The metasurface patterns were defined by a 100 keV electron beam in a JEOL JBX-6300FS EBL system. Patterns were developed for 120 seconds in a 25% solution of tetramethylammonium hydroxide. Reactive ion etching with Cl2, HBr, and O2 chemistries were utilized to transfer the pattern to the silicon layer (Lam TCP 9400). The HSQ resist was removed using 2% hydrofluoric acid in water and the samples were then cleaned using a Piranha solution (9:1 H 2 SO 4 :H 2 O 2 ) heated to 120 • C. The silicon nanostructures were passivated by heating for 30 min at 800 • C in a furnace to grow a ∼ 4 nm oxide layer.

Resonator spectra were measured in a home-built near-infrared microscope shown in Supplementary Figure 1 . Samples were illuminated via a broadband NKT supercontinuum laser with a collimated fiber output. A polarizer P1 was set to create linearly polarized incident illumination at a 45 • angle with respect to the metasurface structures. The beam is weakly focused onto the sample through the sapphire substrate at normal incidence with a lens L2 (f= 50 mm) to an approximate spot size of 200 µm. Additionally, all optical measurements in this work were taken with sample chips sealed in a fluid cell and immersed in PBS 1X. The scattered light is collected through a 50X objective lens (Olympus LCPLN50XIR) and directed through a cross-polarized polarizer P2 at -45 • to reduce the substrate Fabry-Perot signal. The scattered light is then focused via a lens L3 (f=75 mm) into a SPR-2300 spectrometer (Princeton Instruments). The broadband signal is diffracted via a diffraction grating (600 g/mm, blase wavelength 600 nm, Princeton Instruments) and focused onto an air-cooled InGaAs detector (NiRvana, Princeton Instruments). All spectral measurements are collected as the average of three successive 200 millisecond acquisitions. Spectral features were analyzed by fitting the data with the function:

where T is the scattered intensity from a superposition between a constant complex background, a r + a i i, and a Lorentzian oscillator with resonant frequency f 0 and full-width at halfmaximum of 2γ. The quality factor is then calculated as Q = f 0 /2γ.

Self-assembled monolayers of single stranded probe DNA was interfaced to the silicon metasurfaces through a multi-step chemical functionalization process summarized in Supplementary Figure 2 . To activate the silicon surface for functionalization, the samples were immersed in a Piranha solution (9:1 H 2 SO 4 :H 2 O 2 ) heated to 120 • C for 20 min to hydroxylate the surfaces. Next, samples were immersed in a 0.1 mM solution of 11-aminoundecyltriethoxysilane (Gelest Inc.) in ethanol, sealed, and left for overnight for 18-24 hrs. The samples were rinsed in fresh ethanol for 5 min (3X) and then baked for 1 hr at 150 • C to form a stable silane layer. A hetero-bifunctional cross linking molecule was attached to the silane layer through immersion in a 1mM solution of 3-maleimidobenzoic acid N-hydroxysuccinimide ester (Millipore Sigma) dissolved in a 1:9 (v/v) mixture of dimethyl sulfoxide and PBS for 1 hr. Samples were then rinsed thoroughly with deionized water and blown dry with N 2 gas. Single stranded DNA probes were obtained from Integrated DNA Technologies (Coralville, IA) modified with a disulfide tether on the 5' ends. The as received DNA probes were disperesed in 50 µL of tris-EDTA buffer, pH 8.0, and mixed with 30 mg of DL-dithiothreitol for at least 1 hr to reduce the disulfide moieties to thiols. The probes were then purified via gravity-flow size exclusion chromatography using illustra NAP-5 columns. The concentration of the eluted DNA solutions were determined using UV absorption signatures (Varian Cary 500 UV-Vis Spectrophotometer). For the functionalization reaction, portion of the stock solution were then diluted to 20 µM in PBS 1x with added divalent cations of 100 mM MgCl 2 . The DNA probe solution was pipetted onto each sample and incubated overnight (∼18-24 hrs) in a dark and humid environment. Samples were rinsed with PBS 1X and then soaked in a PBS solution with added salt to a concentration of 1M NaCl for 4 hours to remove any loosely bound or physiosorbed oligonucleotides. Samples were then rinsed with PBS 1X and deionized water and dried with N 2 gas. Samples corresponding to optical measurements in main text Fig. 3 were measured before and after each functionalization step with additional deionized water rinsing and N 2 drying before the next chemical processing step. Samples corresponding to main text Fig. 4 were optically characterized only before and after target DNA hybridization.

For static DNA hybridization measurements (all presented data in main text excluding Fig. 4c) , a baseline spectroscopic measurement was taken on metasurfaces that had been functionalized with a probe DNA monolayer. Probes with sequences corresponding to the E gene of the SARS-CoV-2 virus were used in all experiments. Following baseline measurements, samples were rinsed with DI water and dried. A target DNA solution corresponding to either complementary E gene or noncomplementary ORF1b gene fragments (Supplementary Table 1 ) was produced by diluting a 100 µM stock solution to the desired concentration in 1X PBS. Additional divalent cations corresponding to 100 mM MgCl 2 were added to the solution to increase hybridization efficiency and speed. A 100 µL droplet of the target solution is then pipetted onto each sample chip and incubated for 30 min in a dark environment. Samples are rinsed in PBS 1X and deionized water before final optical characterization.

For dynamic DNA hybridization measurements presented in Fig. 4c of the main text, samples functionalized with DNA probes were placed in a fluid cell and mounted in the optical transmission set up described above. Spectral acquisitions were collected at 10 second intervals, and baseline measurements of the metasurfaces immersed in a pure hybridization solution with no nucleic acids were taken for 4 minutes. Next, excess volume of the target solution containing nucleic acids was flowed into the fluid cell from a syringe for 10 seconds via inlet tubing to displace the pure hybridization solution and completely fill the cell with target solution. Spectra were monitored for an additional 20 minutes and wavelength shifts were calculated based on changes compared to the average resonance wavelength obtained from the initial 4 minute baseline measurement. Considering an unperturbed silicon waveguide made from a 1-D array of subwavelength silicon blocks, we calculate the waveguide dispersion for the lowest order mode (Supplementary Fig.  3a) . The waveguide mode possesses larger momentum than free-space radiation and is "bound" or does not couple to free-space illumination. Upon introducing periodic perturbations in the length of every other silicon block along the waveguide, our unit cell spacing, a, effectively doubles, folding the first Brillouin zone in half. Now modes that were previously inaccessible to freespace illumination lie above the light line ( Supplementary Fig. 3b ). This band structure is maintained when the magnitude of the perturbation is changed and only the coupling strength and Q factor are modulated as discussed in the main text.

The sensitivity of a resonant mode to minute changes in the local refractive index can be estimated by the fraction of electric field energy residing outside the resonator. We calculate the exposure of the mode utilized in our sensors with the following equation:

where out and in are the permittivity of the medium containing the analyte and the permittivity of the resonator and substrate, respectively. V out and V in represent the volumetric regions of the analyte containing medium and the portions inside the resonator or substrate that do not overlap with any bound materials or molecules. Performing this analysis on the sensor design described in the main text as well as guided mode resonant structures previously described in reference [?] composed of notched silicon waveguides, we find that our silicon block chains significantly increase field penetration into the surrounding environment. Field profiles of the two structures are plotted in Supplementary Fig. 4 showing similar transverse electric waveguide modes. Due to the subwavelength spacing of the discrete silicon blocks in our sensors, we still excite the localized waveguide modes along the periodic direction that are seen in continuous silicon wire waveguides. However, the grating-like structure exposes regions of the mode to the surroundings while also reducing the effective mode index of the waveguide, leading to further extension of the fields out of the resonator. This design results in the fraction of the mode energy in the surroundings to increase to f U E = 0.29 compared to only f U E = 0.08 for notched or continuous waveguide structures.

As discussed in the main text, introducing an asymmetry along a silicon waveguide allows for the excitation of previously bound modes. Reduction of the asymmetry, ∆d in the case of our metasurfaces, decreases the coupling strength of the mode to free-space radiation thereby increasing the Q factor. For a material that exhibits no intrinsic absorption losses, such as silicon in the near infrared, the Q factor can be arbitrarily increased as the perturbation strength approaches zero. This dependence of the Q factor on subtle structural deviations have been previously described through temporal coupled-mode theory and perturbation theory[?, ?, ?]:

where B is a constant that depends on the resonator geometry and α is a unit-less asymmetry parameter represented by ∆d/d 0 in our metasurface. This relationship is shown in Supplementary  Fig. 5 , where theory (solid line) and numerical simulations (stars) indicate diverging Q factors as ∆d is decreased. We also observe that experimentally observed Q factors are lower than predicted values (experimental data from Main text Fig. 2 ). One significant factor limiting our experimental quality factors is the absorption coefficient of water at telecommunication wavelengths. Since all our optical measurements are performed in aqueous solutions, dissipative losses are expected to decrease our measured Q factors as shown by the dashed line in Supplementary Fig. 5 , which represents numerical calculations including water absorption. The effects of absorption losses are particularly strong as ∆d is decreased, as longer resonance lifetimes lead to greater interaction between the resonant mode and the absorptive background medium. Future iterations of our sensor can be designed in the water absorption window around 1300 nm to maximize performance of the resonators. Additionally, fabrication imperfections such as surface roughness or non-uniformity in the metasurface structures will introduce scattering losses and reduce the observed Q factor.

While the resonators shown in the main text exhibit high-Q modes in longer 1-D arrays (200 µm), we show that the resonators can be scaled down significantly while maintaining sharp spectral features. Our metasurface design features low scattering losses out the ends of the waveguides, and hence are relatively robust to resonator finite size effects due to the high index contrast between separated silicon blocks and gaps containing the background medium. In Supplementary Fig. 6a , we show calculated dispersion diagrams for three different resonators consisting of a solid silicon waveguide with increasing depths of notch corrugations. The waveguide has width of 600 nm and from top to bottom, the bands correspond to notches added on both sides of the waveguide with depths of 50, 150, and 300 nm. We observe flattening of the bands as the notch depth is increased all until 300 nm, where the waveguide is now separated into distinct silicon blocks. The flatter bands indicate a much smaller group velocity due to strong in-plane Bragg scattering, which reduces the propagation of the mode out the waveguide ends and reduces effects of shrinking the resonator on the Q factor. We experimentally verify that we can maintain high quality factors while shortening the overall length of each resonator. In Supplementary Fig. 6b ,c we show SEM images of multiple resonators with varying lengths from 300 µm down to 50 µm and representative spectra. Fitting N=6 resonators for each condition, Supplementary Fig. 6d shows little change in the Q factor with varying waveguide length. Resonators with ∆d = 50 and 30 nm maintain high Q factors exceeding 1000 even in resonators down to 50 µm. Each resonator could potentially be further scaled down with added dielectric mirrors patterned on the waveguide ends to reduce scattering losses. Thus, it is possible to envision individual free space coupled high Q resonators on the order of a few µm. where ∆λ/∆n s the resonant wavelength shift induced by a change in the background medium refractive index or bulk refractive index sensitivity, Q is the resonator quality factor, and λ 0 is the resonant wavelength. To determine the bulk refractive index sensitivity of our devices, we take a series of optical measurements in various saline solutions with differing concentrations of NaCl dissolved in deionized water. Increasing NaCl concentrations have been shown to increase the refractive index of water.

[?] In Supplementary Fig.7 , the change in resonant wavelength indicates our sensors have a sensitivity of ∆λ/∆n = 270 nm/RIU. The sensitivity of the resonators does not vary significantly with ∆d, as the block asymmetry alters the Q factor, but the modal overlap with the surrounding medium does not change. Given our resonator Q factors of ∼2200 (Main text Fig.  2) , we obtain a sensing FOM of around 400. This value is larger than previous demonstrations in plasmonic or dielectric metasurface based sensors.[?, ?, ?, ?, ?] We also note that modification of our metasurfaces, such as introducing subwavelength gaps or slots along the waveguide that further expose the resonant mode to target analytes, could dramatically improve the sensitivity and FOM of future iterations of the devices.

To estimate the resonant wavelength shifts corresponding to successive molecular layers deposited on our sensors (Main text Fig. 3c) , we model each surface step with a thin dielectric shell and numerically calculate the response with FDTD as described above. The dielectric shells extend from all exposed silicon faces of our metasurface nanoblocks as shown in Supplementary Fig. 8 . The initial bare sensor before surface functionalization is calculated with a 4 nm silicon dioxide layer due to the thermal passivation step performed after nanofabrication of our sensors.

[?] The AUTES layer is assumed to bind as a uniform monolayer of thickness 1.8 nm and refractive indices ranging from n = 1.40-1.45, based on reported literature values. The refractive index is calculated for a range that corresponds to typical estimated optical properties of biomolecular layers. The MBS layer is estimated as a 0.7 nm thick layer based on the reported spacer arm length of the molecule and refractive index of n = 1.40-1.45. Due to stable secondary structures that are likely to have reduced our probe density by up to 50% the ssDNA probes are calculated with optical properties n = 1.37-1.382.[?, ?, ?] We estimate the structure of our single-stranded DNA probes using the open source software "mfold" to predict the secondary structure in a 1X PBS solution and the most stable conformation is shown in Supplementary Fig. 9 . The probe layer thickness is estimated by the wormlike chain model[?]:

where the monomer spacing a = 0.6 nm, persistence length L p = 1 nm, and contour length l = Na (N = number of nucleotides) give a radius of gyration, Rg, and layer thickness of ∼ 4 nm.

[?] The dsDNA probe refractive index is approximated as n = 1.4-1.423 as estimated from the densification and increased polarizability of duplexed DNA compared to single stranded DNA. [?] The thickness is similarly estimated by the wormlike chain model, but with a = 0.3 nm and Lp = 30 nm, which returns a similar radius of gyration of 4 nm. Based on the short fragment length of our probe and targets at 26 nt, we do not expect a significant monolayer thickness change upon DNA hybridization.

Fluorescence experiments were performed after DNA hybridization experiments with target nucleic acids tagged with ATTO590 dye on the 5' end. Dried samples were placed in a Zeiss Ax-ioImager system and imaged with a 20x objective. Fluorescence images were acquired with 1000 ms exposures on a Zeiss Axiocam 506 mono camera. Fluorescence intensity values were averaged over a 80 x 40 µm area and were normalized to the maximum intensity values from chips hybridized with complementary E gene targets as seen in Supplementary Fig. 10 . of 1 signifies perfect separability of specific and non-specific DNA binding. Supplementary Fig.  11 shows the ROC curves at each tested concentration. The dashed line represents an AUC value of 0.5, corresponding to random classification or no discrimination of signals. The AUC value of all curves is above 0.7, representing good classification of specific to non-specific binding. We also observe sensitivity (true positives) and specificity (1-false positives) up to 94% and 96%, respectively.

Supplementary Fig. 3 . a, Simulated waveguide dispersion in an unperturbed chain of subwavelength silicon blocks. b, Brillouin zone folding introduced via symmetry breaking in biperiodic guided mode resonator.

Rates of co-infection between sarscov-2 and other respiratory pathogens

Detection of respiratory viruses by molecular methods

Liquid biopsy enters the clinic -implementation issues and future challenges

Current and future perspectives of liquid biopsies in genomics-driven oncology

A system of coordinated autonomous robots for lagrangian studies of microbes in the oceanic deep chlorophyll maximum

The ecology of environmental DNA and implications for conservation genetics

Biosensing with plasmonic nanosensors

Highly specific plasmonic biosensors for ultrasensitive microrna detection in plasma from pancreatic cancer patients

Dual-Functional plasmonic photothermal biosensors for highly accurate severe acute respiratory syndrome coronavirus 2 detection

Label-free detection and molecular profiling of exosomes with a nano-plasmonic sensor

Symmetry breaking in plasmonic nanocavities: Subradiant lspr sensing and a tunable fano resonance

Imaging-based spectrometer-less optofluidic biosensors based on dielectric metasurfaces for detecting extracellular vesicles

Highly sensitive biosensors based on all-dielectric nanoresonators

Unravelling the role of electric and magnetic dipoles in biosensing with si nanoresonators

Imaging-based molecular barcoding with pixelated dielectric metasurfaces

Dielectric nanohole array metasurface for high-resolution near-field sensing and imaging

All-dielectric metasurface analogue of electromagnetically induced transparency

Spectrally selective chiral silicon metasurfaces based on infrared fano resonances

All-dielectric crescent metasurface sensor driven by bound states in the continuum

On-chip single nanoparticle detection and sizing by mode splitting in an ultrahigh-q microresonator

Porous silicon ring resonator for compact, high sensitivity biosensing applications

Single virus detection from the reactive shift of a whispering-gallery mode

Photonic-crystal waveguide biosensor

A multiplexed optofluidic biomolecular sensor for low mass detection

High-q nanophotonics: sculpting wavefronts with slow light

High quality factor phase gradient metasurfaces

Theory and applications of guided-mode resonance filters

Analysis of guided resonances in photonic crystal slabs

Dimerized high contrast gratings

Ultrasensitive hyperspectral imaging and biodetection enabled by dielectric metasurfaces

Chirped guided-mode resonance biosensor

Wave localization of doubly periodic guided-mode resonant grating filters

Detection of 2019 novel coronavirus (2019-nCoV) by real-time RT-PCR

Molecular diagnosis of a novel coronavirus (2019-nCoV) causing an outbreak of pneumonia

Kinetics and thermodynamics of DNA, RNA, and hybrid duplex formation

Detecting DNA and RNA and differentiating Single-Nucleotide variations via Field-Effect transistors

Control and stability of self-assembled monolayers under biosensing conditions

How to prepare reproducible, homogeneous, and hydrolytically stable aminosilane-derived layers on silica

A unified view of polymer, dumbbell, and oligonucleotide DNA nearestneighbor thermodynamics

Kinetic study of DNA hybridization on DNA-modified gold nanoparticles with engineered nano-interfaces

Controlling evanescent waves using silicon photonic all-dielectric metamaterials for dense integration

Aptamer-field-effect transistors overcome debye length limitations for small-molecule sensing

Advancing biosensors with machine learning

Real-time reliable determination of binding kinetics of DNA hybridization using a multi-channel graphene biosensor

Supplementary Fig. 1. Schematic of near-infrared microscope set up utilized to collect spectra from metasurface samples

Supplementary Fig. 2. Multi-step surface functionalization of a DNA probe monolayer on silicon metasurfaces

Upper panel shows an x-y cut through the center of the structure and the lower panel is an x-z cut through the center of the notch perturbation where fields are most strongly concentrated. b, Electric field profile for asymmetric chain of silicon blocks on a sapphire substrate

Resonator quality factor as a function of difference in neighboring silicon block length

Simulated waveguide dispersion in waveguides with varying notch depth. b,Representative optical spectra from waveguides with length of 300 µm and 50 µm. c, SEM image of resonators fabricated with different waveguide lengths. d, Quality factors as a function of waveguide length

Resonant wavelength measurements as a function of background medium refractive index. Circles and error bars represent experimental measurements for N=25-30 resonators at each condition

Schematic of calculated structure with dielectric shell around silicon nanostructures representing the molecular monolayers. b, Estimated layer thicknesses and refractive indices

Schematic of fluorescently tagged target DNA sequences. b, Fluorescence images and integrated intensities for sensors exposed to complementary nCoV.E sequences (top) and noncomplementary HKU.ORF1 sequences (bottom). Fluorescence imaging confirms the specificity of immobilized DNA probe molecules to complementary nucleic acid sequences. All metasurface sensors were functionalized

The dashed line represents AUC = 0.5, representing a line in which there is no separation of specific and non-specific DNA binding signals. E gene Probe 5

The authors thank Sahil Dagli, Elissa Klopfer, Dr. Harsha Reddy, Dr. Loza Tadesse, Dr. David Barton, Dr. Halleh Balch, Dr. Lisa Poulikakos, Chris Siefe, and Baba Ogunlade for insightful discussions. The authors acknowledge funding from a NSF Waterman Award (Grant number

Concentration dependent resonant wavelength shift responses in main text Fig. 4 were fit to the Hill equation (formally equivalent to the Langmuir isotherm):is the saturated maximum binding signal at high target concentrations, X is the target concentration, h is the Hill coefficient which describes the slope of the curve, and K d is the concentration value that corresponds to half-maximum binding signals. Furthermore, the Langmuir adsorption model is also used to fit the time varying response in main text Fig. 4D . The wavelength shift response as a function of time is described as [?] :where θ eq is the saturated equilibrium binding signal and k is a "observed" rate constant that accounts for both target hybridization and reversible dehybridization rates.

The resonant shift responses of the sensor to complementary and non-complementary targets in main text Fig. 4 are analyzed by two-way analysis of variance (ANOVA). Omnibus statistics are reported in Supplementary Table 2 . Comparison of nCoV.E and HKU.ORF1 targets at each concentration were performed using a post hoc Tukey's honest significant difference test and pvalues for each concentration are shown in Supplementary Table 3 . In all cases, the low p-values were considered statistically significant.

Additionally, data sets from resonant wavelength shift signals for nCoV.E and HKU.ORF1 targets can be further analyzed with a thresholding method to determine if binding responses are due to specific complementary probe-target hybridization. For nCoV.E and HKU.ORF1 data sets at each concentration condition, a response threshold is applied where resonant shifts above this value are considered as "positive" samples. Data points corresponding to nCoV.E targets are considered true positives while HKU.ORF1 targets are labelled as false positives. The threshold value is varied to obtain different true and false positive rates and to produce a receiver operator characteristic (ROC) curve. ROC curves are quantified by the value of the area under the curve (AUC), where a value Concentration (M) p-value 1 * 10 −5 2 * 10 −7 1 * 10 −6 2 * 10 −7 5 * 10 −7 2 * 10 −7 1 * 10 −7 2 * 10 −7 5 * 10 −8 9 * 10 −6 1 * 10 −8 0.0012 Supplementary Table 3 . Concentration dependent p-values for comparisons between nCoV.E and HKU.ORF1 targets.