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

Biosensors are indispensable tools for public, global, and personalized healthcare as they provide tests that can be used from early disease detection and treatment monitoring to preventing pandemics. We introduce single-wavelength imaging biosensors capable of reconstructing spectral shift information induced by biomarkers dynamically using an advanced data processing technique based on an optimal linear estimator. Our method achieves superior sensitivity without wavelength scanning or spectroscopy instruments. We engineered diatomic dielectric metasurfaces supporting bound states in the continuum that allows high-quality resonances with accessible near-fields by in-plane symmetry breaking. The large-area metasurface chips are configured as microarrays and integrated with microfluidics on an imaging platform for real-time detection of breast cancer extracellular vesicles encompassing exosomes. The optofluidic system has high sensing performance with nearly 70 1/RIU figure-of-merit enabling detection of on average 0.41 nanoparticle/µm2 and real-time measurements of extracellular vesicles binding from down to 204 femtomolar solutions. Our biosensors provide the robustness of spectrometric approaches while substituting complex instrumentation with a single-wavelength light source and a complementary-metal-oxide-semiconductor camera, paving the way toward miniaturized devices for point-of-care diagnostics.

shifting of the resonance due to the analyte binding. a) shows that a higher Q resonance (346) with the full-width at half-maximum (FWHM) of 2.4 nm exhibits a 53% intensity change when the resonance is red-shifted for 1 nm. On the other hand, b) shows that a lower Q resonance (88) with the FWHM of 9.4 nm exhibits only a 14.83% intensity change for the same red-shift of 1 nm. c) and d) show the influence of the Q on the dynamic range of the sensor. Here, the dynamic range is chosen as the part of the resonance when the slope has less than 1% of deviations from the linear fit. c) shows that as the higher Q resonance has smaller bandwidth, it leads to a smaller dynamic range. On the other hand, d) shows that a lower Q resonance offers a larger dynamic range. Supplementary

Supplementary Note 1: Important parameters in designing refractometric biosensors
Multiple parameters affect the performance of a refractometric biosensor; hence, it is essential to consider the relevant parameters simultaneously. Below, we discuss some of them : i) The quality factor (Q) contributes to the performance of a biosensor utilizing optical resonances.
An ultra-high Q resonance resembling a delta function is not suitable for refractometric imagingbased detection because the intensity changes induced by the wavelength shifts cannot be used to resolve continuous variation of intensity (Supplementary Figure 1). On the one hand, it is preferable to have relatively sharp resonances, i.e., with steep slopes of the intensity profile, to provide bigger changes in the output signal for a given resonance shift. Supplementary Figure 2a and S2b show realistic cases with a high Q value of 346 and a lower Q of 88. A bigger output value leads to the easier detection of more dilute samples, resulting in a better limit-of-detection (LOD).
On the other hand, a lower Q offers a wider dynamic range of a biosensor, as shown in the example ( Supplementary Figures 2c and 2d). Thus, there is a trade-off between the LOD and dynamic range that we can achieve for a given biosensor.
ii) Another important consideration in designing metasurfaces for sensing applications is the enhancement of the near-fields and their accessibly to the analytes. It is essential that enhanced near-fields extend out of the resonator and spatially overlap with the samples on the sensor surface (see also Section 3). This also brings a trade-off because maximizing the field's spatial overlap with analytes (e.g., by not confining the mode tightly in a high-index meta-unit) reduces the Q value. Here, it is useful to emphasize that BIC designs that can achieve impressively high Qs with tightly confined field enhancements inside the resonators are less suited for biosensing and more appropriate for nonlinear or lasing applications.
iii) In biosensing applications, it is important to show operation in an aqueous environment because water is essential to preserve the bioactivity and extend its applicability for real-time and in-flow measurements. However, the optical loss of water hampers the high-Q values of the nanoresonators. In our work, we optimized a design for operation in solution (with the integration of a microfluidic chip) while still supporting suitably high Q value as well as accessible high nearfield enhancements at 850 nm wavelength range.

Supplementary Note 2: The general concept of quasi-BICs in asymmetric diatomic metasurfaces
In this section, we develop a general theory of quasi-bound states in the continuum (quasi-BICs) in dielectric diatomic metasurfaces. Contrary to earlier works, here, we intentionally use a complex unit cell (meta-molecule) with two meta-atoms per unit. Compared to single-atom metasurfaces, diatomic structures provide advanced flexibility in the engineering of the in-plane asymmetry while keeping the design easy to fabricate and feasible for biosensing applications.
We consider a silicon metasurface composed of a square lattice of elliptical dimers deposited on a glass substrate. Parameters: disk height is 100 nm, x-period and y-period are 535 and 340 nm, respectively, the distance between disk centers is 270 nm, the radius of the short and long semiaxis of the ellipse is 50 nm and 90 nm, respectively. The structure supports a true BIC with an infinite quality factor at a wavelength of about 855 nm. To achieve a finite Q of the resonance, we transform the BIC to a quasi-BIC by introducing the in-plane asymmetry of the unit cell. Due to the reciprocity, the quasi-BIC can also be observed in response to an external excitation of the structure. For such diatomic metasurface, the asymmetry can be introduced in different ways, including squeezing of one meta-atom, changing its ellipticity, and rotation of one or two metaatoms, as shown in Supplementary Figure 3a. The asymmetry parameter is a linear function of perturbation and has a specific definition for each considered case, as shown in Supplementary   Figure 3d. An important difference of diatomic metasurfaces from monoatomic is a higher number of degrees of freedom, which allows achieving better precision in the control of the quasi-BIC Q.
Depending on the application and the device requirements, one can use the most suitable of the mentioned types of asymmetry to obtain a quasi-BIC with a specific Q and field enhancement. We also note that, unlike squeezing and rotation, to the best of our knowledge, changing the ellipticity was not explored earlier as a way to introduce asymmetry for diatomic metasurfaces.
The eigenmode calculations show that for diatomic metasurfaces, the radiative Q of the quasi-BIC decreases with the increase of asymmetry following the inverse square law rad = 0 −2 , as shown in Supplementary Figure 3b. For simulations of the radiative Q, we use the material dispersion for silicon without absorption losses. The total Q factor is limited by the non-radiative quality factor , which takes into account the material losses, fabrication imperfections, and sample finite −1 = rad −1 + nrad −1 . Assuming the weak dependence of on the asymmetry, we obtain a closed-form expression for the total Q (equation (1) To spectrally characterize the uniformity of the structures, we used our hyperspectral imaging setup 1 . This set-up enables the extraction of resonance maps, which gives the resonance wavelength over the entire sensor area at a single-pixel resolution. Supplementary Figures 5a and 5b show the spectral resonance map and the histogram of resonance wavelength distribution for our metasurface, respectively.
The extracted mean value indicates a negligible difference between the simulation (835 nm) and the experiment (829.05 nm). The measurements also reveal a standard deviation of 0.97 nm, highlighting the uniformity in the optical response over the sensor area and the correspondingly high quality of the optimized nanofabrication process.

Supplementary Note 5: Bioregconitions assay for nanoparticles
To demonstrate the biosensing capability of the platform, we performed real-time in-flow detection of nanoparticles to mimic the bioparticles (i.e., extracellular vesicles and viruses). To target the biotinylated silica nanoparticles, streptavidin was immobilized on the detection sensors while the control sensor was blocked with bovine serum albumin (Supplementary Figure 6).