## Introduction

Photonic biosensors are indispensable tools for life sciences and medical diagnostics. For example, commercial surface plasmon resonance (SPR)1,2 instruments are widely used to measure the binding kinetics and affinities of receptor–ligand interactions via the evanescent waves of surface plasmon polaritons (SPP) in a gold film. The real-time, label-free sensing capability of this modality has been critical for the functional and quantitative characterization of antibodies3, drugs, aptamers, target-drug interactions4, and virus-host-cell receptor interactions5. SPR generates invaluable time-resolved kinetics data that cannot be obtained with conventional end-point binding assays that employ target labelings, such as enzyme-linked immunosorbent assays (ELISA) or radioimmunoassays. State-of-the-art SPR instruments can achieve limits of detection (LOD) that are comparable to ELISA and they are increasingly being applied toward the clinical analysis of patient biofluids to detect proteins, microRNAs, drugs, and small molecules associated with various disease conditions6. The refractive index (RI)-sensing transduction mechanism of SPR eliminates the need for labeling and washing steps, provides real-time kinetic information and is fast, which can be advantageous for clinical applications. Furthermore, SPR is useful for characterizing low-affinity analytes, which in equilibrium ELISA assays would require larger sample volumes and cause unwanted dissociation of weakly-bound analytes during wash steps6.

To miniaturize and further improve the performance of SPR sensors, researchers have leveraged both top-down lithography and bottom-up synthesis to build “nanoplasmonic” sensors by engineering the flat gold films of conventional SPR into nanoparticles, nanoholes, or collections of sub-wavelength unit cells called “metasurfaces” of various shapes. Such nanostructures and metasurfaces can extend resonances to broader frequency ranges (i.e., from visible to near- or mid-IR), and exhibit optical phenomena such as localized surface plasmon resonance (LSPR), radiative coupling, and extraordinary optical transmission7,8,9. While nanoplasmonic structures add fabrication complexity and costs, they can improve multiplexing capacity, miniaturization, and sensitivity while also offering considerable design flexibility. Early demonstrations focused on enhancing the performance of conventional SPR by reducing the footprint of the sensing area to provide portability for point-of-care (POC) applications and/or increase parallel detection for high-throughput screening. For example, the widely used gold nanohole arrays10,11 can be excited with simple collinear optics compatible with a standard microscope, while simultaneously enabling array-based sensing via parallel imaging techniques12. The challenge of producing large-scale patterns with ease and low cost has been addressed using interference lithography13, replication from templates14, and photolithography15.

Nanoplasmonic sensors can perform tasks inaccessible to conventional SPR (Fig. 1). The main advantage of SPR compared to other diffraction-based optical techniques is the use of the tightly-confined evanescent field of SPP waves beyond the diffraction limit that is present at the interface of metal and medium. By shaping the flat metal films into nanoparticles, even tighter 3D field confinement is possible, and this eventually led to the ultimate feat in label-free sensing—detection of a single protein molecule (Fig. 1a, b)16,17,18. By taking advantage of the nanohole geometry, which can selectively localize receptor molecules or biological nanoparticles, researchers have shown unique sensing functions utilizing nanovesicles11 (Fig. 1c) and virus-like particles19. Nanohole SPR sensors integrated with sophisticated microfluidics have also been used to enhance molecular binding kinetics20,21 (Fig. 1d) or measure secreted molecules from a live single cell on a chip (Fig. 1e–h)15.

More recently, a rapidly expanding family of van der Waals (vdW) materials22,23 with extraordinary optical, electrical, and mechanical properties have revolutionized the fields of electronics and optics. The reduced dimensionality of these materials enhances plasmonic field confinement, and their much-reduced dielectric screening confers sensitive electrostatic tunability and enables the excitation of different polariton modes such as plasmons, excitons, and phonons (Fig. 2a–c) for new sensing modalities24,25,26. While still a nascent technology relative to ELISA or SPR, two of the most extensively studied low-dimensional vdW structures—one-dimensional (1D) single-walled carbon nanotubes (SWNTs) and two-dimensional (2D) graphene—have already demonstrated novel sensing capabilities that are inaccessible to metal/dielectric nanophotonic sensors. For example, SWNTs have been used for single-molecule detection27 (Fig. 2d) and in vivo detection28 (Fig. 2e) via excitonic effects. 2D vdW materials such as graphene have been shown to increase plasmonic field confinement much tighter than metallic nanostructures. Graphene plasmonics has also demonstrated the unique potential for dynamically tunable infrared absorption spectroscopy for probing structural changes in molecules and vibrational-mode fingerprinting26. vdW materials could also enable the on-chip integration of electrical readouts, nanopore sensing29 (Fig. 2f), molecule trapping mechanisms (Fig. 2g), on-chip photodetectors30 (Fig. 2h), and nanofluidics. Hence, rather than merely competing for the same applications as existing modalities (e.g., refractometric sensing of receptor–ligand binding kinetics with metal-based SPR), we envision vdW nanophotonic sensors ushering in new capabilities for biosensing based on novel physical principles and materials properties31 (Fig. 2i).

In this review, we explore the potentials of vdW materials for nanophotonic biosensing (Fig. 2). We start by investigating performance metrics for nanophotonic sensors and then discuss how vdW materials can push their limits of performance, overcome tradeoffs, or enable new functionalities. We next discuss technological challenges awaiting researchers, including efficient excitation of high-momentum polaritons in vdW materials, nano-patterning, device architecture, surface chemistry, in vivo sensing, and toxicity. Finally, we highlight recent progress in harnessing vdW materials to demonstrate new modalities that are difficult to perform with conventional metal/dielectric nanophotonic sensors.

## Performance metrics for nanophotonic biosensors

How can one fairly evaluate the performance of various nanophotonic biosensors featuring different substrate materials, structures, transduction mechanisms, operating frequencies, and detection schemes? Consider SPR-based refractometric biosensing, in which signal transduction occurs via interfacial RI changes caused by surface-bound analytes. Ideally, such devices should provide resonance peaks or dips that undergo large spectral shifts upon analyte binding, are narrow as measured by the quality (Q) factor, and are of high contrast. When these conditions are met, analyte molecules can result in large intensity changes in the photodetector (Box 1).

As illustrated in Box 1, a single performance metric (e.g., spectral shift, Q factor, FOM, or resolution) may not be enough to characterize the LOD of disparate biosensors. A more fundamental quantity for nanophotonic structures is the evanescent field decay length of the resonant field into the sensing medium (Box 1f). The decay length is important not only for refractometric sensing but also for surface-enhanced Raman scattering (SERS) and surface-enhanced infrared absorption (SEIRA). Indeed, a marquee advantage of nanophotonics is the ability to beat the diffraction limit and confine optical energy into sub-wavelength scales. For instance, the decay length of the evanescent field into the dielectric medium of an SPP wave in a semi-infinite metal film is ~200 nm for water/metal interfaces at visible wavelengths. Decay lengths associated with LSPR in nano-patterned structures can be much shorter (i.e., tens of nanometers), allowing them to overcome the lower bulk sensitivity and detect thin films, or even an unlabeled single protein molecule. Near-field confinements in 2D vdW materials can be extreme: ~106 smaller than the diffraction limit.32 In graphene nanoribbons, the plasmon field decays exponentially as $${\exp }\left(-\frac{\pi \alpha }{W}z\right)$$, with $$W$$ being the width of the ribbon, where $$\alpha \, \space {\rm{of}} \sim \!1$$ is generally assumed but can approach $$\sim \! 100$$ at the edges of the graphene nanoribbon33. In a typical nanoribbon of $$W= \, \sim \!100\text{nm}$$, more than 50% of the plasmon intensity can be confined to a ~5 nm distance from the graphene surface. The practical LOD will depend on a combination of the field decay length and the sizes of the receptors and analytes34,35. Besides these intrinsic performance metrics, the real-world performance, and utility of nanophotonic biosensors depend on other factors such as surface modification and blocking strategies, multiplexing capacity, pre-concentration techniques, and signal-enhancing schemes.

## VdW materials for enhancing conventional prism-coupled SPR biosensors

Building on the mature capabilities of prism-coupled gold-film SPR instruments, researchers have explored the benefits of coating the gold film with various 2D materials. Groups have reported improved SPR sensitivity from graphene-coated gold surfaces, which was attributed to enhancement of the surface electric fields provided by an added charge transferred from graphene to gold36,37. While the surface of gold SPR sensors is commonly functionalized via a self-assembled monolayer of alkanethiols, the surface of graphene or other vdW materials requires different strategies for functionalization, which presents challenges as well as new opportunities. Graphene surfaces can be modified by attaching some molecules via π-stacking or other functional groups (e.g., carboxyl). However, the ease of attachment via π-stacking also implies increased nonspecific binding of interferents from biological samples, necessitating a proper blocking procedure. Xue et al. used antimonene-modified gold SPR chips to detect microRNA hybridization events31. Density functional theory energetic calculations predicted that antimonene surfaces have a higher affinity for single-stranded (ss) DNA than double-stranded (ds) DNA. By using gold nanorods to further boost SPR signals, they were able to detect ssDNA-microRNA hybridization events. vdW materials can also protect the surfaces of reactive metals—especially copper and silver—which is necessary for SPR biosensing applications38. Wu et al. characterized the potential of layer materials (e.g., graphene, boron nitride, and carbon nanomembranes) to protect the chemical and optical properties of plasmonic metals39 At a wavelength of ~1 µm, they achieved bulk RI sensitivity of ~10,000 nm/RIU, which is comparable to the sensitivity of conventional spectroscopic techniques.

While these studies demonstrate impressive LOD for molecules in buffer, the response of hybrid vdW-metal surfaces in biofluids for practical biosensing applications (e.g., plasma) needs to be characterized to ensure whether the same low LOD can be maintained. Also, instead of single-channel measurements, reference-channel measurements, as well as temperature stabilization, will be essential for ultrasensitive detection.

As shown in these examples, the applications of vdW materials for conventional SPR or other nanophotonic RI sensors are rapidly expanding40. However, the large-area growth and transfer of vdW materials atop metal films add complexity, labor, and costs. As such, in order to justify their use in practical biosensing applications, the benefits need to go beyond incremental gains in sensitivity and enhanced surface adsorption of analytes. Ideally, vdW nanophotonic biosensors should be used in tandem with other sensing modalities, such as electrical or electrochemical detection, or to enable novel functionalities such as vibrational fingerprinting with spectroscopy.

## VdW materials for reconfigurable vibrational spectroscopy

Refractometric sensors made with metals, dielectrics, and metal-vdW hybrids have shown impressive performance metrics, but cannot identify bound analyte molecules. Vibrational spectroscopy techniques like infrared (IR) absorption spectroscopy and Raman scattering complement SPR and enable such “fingerprinting” functions for molecules, viruses, extracellular vesicles, and cells. With their electrically tunable doping and ultra confined mid-IR and terahertz (THz) plasmons, graphene and other 2D vdW materials are especially promising for these applications.

IR absorption spectroscopy is a powerful analytical tool due to its ability to reveal molecular and structural information of samples without using external labels. Samples in solid, liquid, and gas phases have distinct vibrational modes associated with their molecular bonds, particularly within the mid-IR spectrum of ~3–20 μm (3000–600 cm−1). IR spectroscopy taps into this so-called fingerprint region and is used in numerous scientific fields (e.g., biology, chemistry, material science) and in the industry (e.g., food safety, pharmacology, environmental monitoring, and forensics). However, the large mismatch between micrometer IR wavelengths and nanometer-sized molecules leads to low sensitivity, limiting the use of IR spectroscopy for measuring trace amounts of samples. Nanophotonics can bridge this gap by producing strong, tightly localized near-fields in the vicinity of resonant optical nanostructures. This approach is called SEIRA (Box 2)41,42,43,44, and is closely related to SERS. VdW materials have been used to enhance SERS in the absence of plasmon excitation. The effect is not direct but occurs via chemical enhancement of the polarizability of graphene-bound molecules45,46. We focus here on mid-IR SEIRA enabled by the plasmons of vdW materials.

Progress in nanophotonics has fueled research into increasing low IR signals for SEIRA and expanding its application space. Resonantly excited plasmons in SEIRA substrates (e.g., nanorods or nanogaps) typically exhibit lower enhancement factors (EF) of ~103–107 but provide a wider probing range extending a few hundred nanometers from the metal surface, which is suitable for sensing adsorption of large protein molecules, binding of surface-linked receptors and target molecules, biomembranes, or nanovesicles. These benefits are illustrated in recent experiments (Fig. 3a, b), which detected the kinetics of multiple biological analytes in aqueous solutions using resonant gold nanorods coated with a thin SiO2 layer to facilitate the formation of biomembranes47. Limaj et al. characterized the distance dependence of SEIRA EF using SiO2 overlayers of different thicknesses (Fig. 3c). They showed that the absorption signal from molecules was detectable for SiO2 overlayer thicknesses—and hence, metal-to-molecule separation distance—of up to ~100 nm in aqueous solutions48. This distance dependence is consistent with the dry characterization of gold antennas for SEIRA49. Although SERS has much shorter probing ranges of just a few nm, the longer range of SEIRA is advantageous (Box 2) by allowing the use of receptor molecules for specificity as well as to detect larger analytes (e.g., biomacromolecules, extracellular vesicles, and viruses). Khatip et al. utilized graphene to trap a thin layer of sample solution for tip-enhanced IR spectroscopy in an aqueous environment50 (Fig. 3d).

To date, SEIRA has mostly employed metal-based plasmonic resonators with engineered nanostructures in various shapes and array configurations. However, this landscape is changing rapidly, with novel nanomaterials that can address the intrinsic shortcomings of metals51. For instance, while metals can support strong plasmonic responses due to their high electron concentration, they suffer from high loss and provide resonators with low Q-factors (typically < 10). In contrast, Mie resonances in low-loss and high-index dielectric metasurfaces can achieve several orders of magnitude higher Q-factors. Tittl et al. used CMOS-compatible high-Q-factor (>100) silicon resonators in a sensing approach that can convert mid-IR molecular fingerprints into chemical-specific barcode-like images while eliminating the need for bulky IR spectroscopy instruments52. Leitis et al. used germanium metasurfaces to expand the spectral coverage of resonant SEIRA over 1000 cm−1 wavenumber and extracted vibrational signatures of different molecules in a multi-step assay53.

Graphene and other 2D vdW materials are also bringing exciting prospects for SEIRA26,46,54. The unique optoelectronic properties of graphene enable external control over plasmonic resonance frequency by changing its carrier density (i.e., Fermi level) with electrostatic biasing55,56,57. Such active tuning is used to realize dynamically configurable IR sensors tuned to specific molecular vibrational modes. Rodrigo et al. showed the first graphene-based tunable mid-IR biosensor and its potential for quantitative chemical-specific detection26. They measured plasmon resonance spectral shifts accompanied by narrow dips corresponding to the surface-enhanced molecular vibration bands of a protein analyte over a broad spectrum (Fig. 4a–e). The plasmonic resonance is electrostatically tuned to sweep continuously over the two protein vibrational bands, amide-I and amide-II. Based on this dynamic tunability, they combined both SEIRA and LSPR sensing in a single device and showed extraction of complex refractive indices of nanometer-thick samples on the sensor surface. Since then, graphene-based SEIRA has been used for sensing biomolecules, polymers, self-assembled monolayers, ion gels, metal ions, and gases54,58,59,60,61.

Another unique feature of 2D vdW materials is their extreme field confinement. Given the semi-metallic nature of these 2D materials and the fact that their thickness can be close to atomic-layer scale, they can confine near-fields at extreme (i.e., ~106-fold smaller than the diffraction limit). Graphene nanoribbons can provide up to two orders of magnitude higher field confinement than metallic plasmonic dipole antennas, and thus enable unprecedentedly strong overlap of light with nanometric biosamples26 (Fig. 4d, e). This feature is attractive for achieving superior sensing performance, especially for small molecule detection. Consequently, graphene nanoribbons can also be exploited to detect thin layers of adsorbed gas molecules, as recently demonstrated in a pressure-controlled chamber for label-free identification of various gases including SO2.62. Even tighter field confinement is possible by coupling graphene plasmons with their image charges in a metal mirror and forming a hybridized mode called acoustic graphene plasmons54,63,64,65,66 (Box 3). In this mode, the field energy is confined in a nanometric gap between graphene and mirror, thus sample insertion into the gap remains a practical challenge.

Graphene can achieve high Q-factors, which can lead to increased sensitivity due to longer interaction time between the probing IR light and the analytes. In theory, Q-factors well over 100 can be expected with exfoliated graphene32. These values favorably compare to metal-based resonators, which exhibit Q-factors of ~10 at similar frequencies41. Experimental Q-factors of graphene plasmons are far lower, however—typically <10. This is because chemical vapor deposition (CVD) is typically used to synthesize large-area graphene, and the resulting grain boundaries, impurities and defects result in lower quality graphene, as shown by its reduced carrier mobility (~1000 cm2 V−1 s−1). Furthermore, nanopatterning of graphene with small structures (e.g., nanoribbons with widths ~50 nm) is often required to excite mid-IR plasmons, but the rough edges induced by the nanofabrication process lead to additional loss and further deteriorate the quality of the resonators. Limited plasmon lifetimes (~50 fs) and inelastic scattering with optical phonons have also been reported in graphene nanoribbons67. However, further effort is needed towards the development of high-quality 2D vdW materials that can be scalably manufactured for real-world applications. Observation of low-loss graphene from the encapsulation of exfoliated samples with hexagonal boron nitride (hBN)68 and recent advances in CVD graphene transfer involving proper cleaning and encapsulation, which result in room-temperature mobility in excess of 10,000 cm2 V−1 s−1, demonstrate some of the important ongoing progress in materials development69.

Dynamic tuning of graphene plasmons (see Box 3) is achieved mainly by applying an external voltage (hence controlling the Fermi energy Ef), but this requires some care in terms of fabrication and design. In the past, devices typically relied on a thick dielectric spacer between the graphene and the gating layer. This leads to low capacitance values, and hence requires the application of high bias voltages (tens of volts). However, recent research shows that incorporation of a field-effect transistor (FET) device architecture and suitable spacer films as a substrate can greatly reduce the bias voltage while enabling high-speed operation and incorporation of imaging70. A related issue stems from the dependence of the graphene response on the optical properties of the spacer layer. Graphene plasmons can strongly couple with the phonon modes of the substrate, and this hybridization can produce complex spectral features and limit the operating spectral window. Hu et al. incorporated CaF2 nanofilms as spacers; since CaF2 has no phonons in the relevant range, it eliminates this problem and allows the use of SEIRA over a broader spectral range59.

Although extreme light confinement with 2D materials is an attractive feature for sensing, it is characterized by weak coupling efficiency of the external light to graphene plasmons. The extinction values are typically quite low (below 5%), which is not practical for device applications. The coupling efficiency is further hampered by the low carrier density and the weak oscillator strength of graphene plasmons. In order to excite stronger plasmonic responses, recent works have used multi-layer stacks71, integration with photonic cavities (i.e., Fabry-Perot)72, and hybrid substrates incorporating plasmonic nanostructures73 (Fig. 4f–h). Strong field confinement also requires the presence of samples directly at the sensing surface, and in this context, graphene’s high adsorptivity for hydrophobic species can be leveraged to concentrate analytes. Functional polymers and active analyte transport schemes can also be utilized to enrich samples at the sensing layer to boost both signal strength and device response time. Active trapping methods, e.g., dielectrophoresis (DEP; Box 4) can also be implemented with vdW materials.

To benefit from these additional functions, graphene-based SEIRA devices need to be compatible with in-solution measurements and microfluidics. To this date, most SEIRA experiments have been conducted in a dry medium due to the high absorption of water and polymers used for microfluidics in the IR spectrum. These constraints are circumvented for metal-based SEIRA by utilizing strong plasmonic confinement and performing plasmon excitation and signal collection in the reflection configuration with an IR-transparent substrate74. Because the field intensity decays exponentially from the plasmonic substrate surface with a decay length of ~100 nm, one can selectively probe surface-adsorbed molecules and minimize interference from water molecules in the sample. In contrast, graphene resonances are extracted from either the extinction spectra in the transmission configuration or the absorption spectra obtained in the reflection configuration with an optically thick metal layer, which make in situ measurements not possible. Recent work has combined graphene with attenuated total reflection (ATR) prisms for in situ studies, but these approaches involve inherently bulky instrumentation and are not suitable to exploit tunable and strongly confined plasmonic resonances75,76. Graphene and other liquid-impermeable vdW materials offer alternative routes to avoid water absorption by trapping water in nanometric volumes. This approach has been successfully demonstrated with tip-enhanced IR absorption spectroscopy using a graphene liquid cell50, and in graphene-based nanofluidics77.

Besides graphene plasmons, phonon-polaritons in hBN have been utilized for confining mid-IR light78,79,80,81. The resonances of hBN, however, are spectrally limited in biosensing applications due to the fixed Reststrahlen bands—where the energy band between the longitudinal and transverse optic phonon frequencies Re(ε) is negative and surface phonon polaritons can be excited—which are not aligned with the amide absorption bands of proteins. However, recent work has shown that the Reststrahlen band of MoO3 can be tuned82, which may be useful for chemical sensing applications, although such tuning is not suitable for biomolecules such as proteins and nucleic acids.

## Carbon nanotubes: 1D vdW for biophotonic sensors and single-molecule imaging

While 2D vdW sensor substrates enable powerful in vitro sensing functions, 1D vdW SWNTs95,96 can address an entirely new set of problems—such as in vivo nanophotonic sensing and bioimaging. Shortly after the discovery of SWNTs’ near-infrared (NIR) fluorescence, this phenomenon was recognized as ideal for optical imaging and sensing97. The unique electronic properties of SWNTs that result in their bandgap fluorescence have been reviewed by Weisman98. Briefly, SWNTs can be thought of as single sheets of graphene rolled into cylindrical geometries. Their electronic—and thus photophysical—properties depend on the direction of rolling, which can be specified by lattice vectors n and m (Fig. 6a). Since specific lattice directions in graphene conduct, the rolling direction determines the SWNT bandgap and whether a specific (n, m) chiral species is metallic or semiconducting. The nanoscale-cylindrical shape of semi-conducting SWNTs confines electrons in one dimension along its circumference, leading to what is known as van Hove singularities in the electron density (Fig. 6b). The density of states in SWNTs scales as E−1/2 and diverges at the minima of electronic bands in 1D materials. Photoexcitation of SWNTs generates excitons, where the electron and hole are bound. The limited charge screening in quasi-1D structures allows strong electron-hole Coulomb interaction99 and the formation of stable excitons at room temperature. The exciton binding energy reaches 0.6 eV in 1-nm diameter SWNTs as determined by theoretical100 and experimental work101. The binding energy decreases as SWNT diameter increases. When the exciton is generated, it diffuses along the nanotube; for example, a diffusion length of 120 nm in 20 ps was reported for (6,5) SWNTs102. The diffusion length is limited by the presence of defects in the nanotube or the supporting substrate. Fluorescence occurs when the exciton relaxes back down to the band edge and emits a photon with energy equivalent to the exciton energy. SWNTs are commonly interrogated by exciting through a range of visible light in the range of the E22 transition, and their NIR E11 emission is then collected. Different semiconducting chirality species with different bandgaps/exciton energies can then be visualized using a 2D excitation-emission profile (Fig. 6c)97.

Due to intrinsically strong vdW interactions, SWNTs tend to stick together as bundles and do not naturally disperse in an aqueous solution or biological media. Covalent and non-covalent surface modifications of SWNTs are used to promote solubility. Covalent modification is achieved by chemical reactions on the nanotube surface to introduce carboxyl, amine, sulfhydryl, or other functional groups (Box 5)103. While offering a more stable suspension, covalent modification often alters the intrinsic mechanical strength and conductivity of SWNTs and increases exciton quenching. Thus, non-covalent modification is preferred to preserve their inherent electrical and optical properties. Non-covalent functionalization104,105,106 can be realized by entropy-driven interactions, such as hydrophobic interactions using surfactants such as sodium dodecyl sulfate (SDS)107, sodium dodecylbenzene sulfonate (SDBS)108,109,110, and sodium cholate (SC)111,112, and/or enthalpy-driven interactions such as π−π bonding, hydrogen bonding (CH−π, NH−π), or ionic interactions between the SWNT surface and dispersants such as aromatic polymers, cationic polymers, and block polymers.

A new technique exploiting non-covalent interaction between SWNTs and polymers called corona phase molecular recognition (CoPhMoRe)28 has been developed to achieve selective recognition of specific molecules (Fig. 7). In CoPhMoRe, SWNTs are colloidally stabilized by amphiphilic polymers such as block copolymers or ssDNA, which form a 3D configuration called a “corona” in conjunction with the nanotube surface that can act as a synthetic molecular recognition site. Changes in the fluorescence of SWNTs upon molecular binding to the corona allow for the detection of such recognition events. Molecular recognition can influence SWNT fluorescence in several ways, including emission wavelength shifts due to solvatochromism and emission intensity changes caused by charge-transfer transition bleaching and exciton quenching. In solvatochromism, the introduction of analyte molecules changes the local dielectric environment of the SWNT by replacing solvent molecules, which modifies the optical transition energies113. In the electron transfer mechanism, the frequency at which the E11 fluorescence transition occurs is modulated by the interaction between the SWNT and the target molecule. The electrons that are otherwise allowed to release or prohibited from releasing photons through the E11 transition instead interact with the molecular orbitals of the analytes, which in turn changes the frequency of the transition114,115. Mechanistic understanding of SWNT fluorescence wavelength shift and intensity changes have enabled the development of SWNT-based fluorescent biosensors for the selective detection of a large array of analytes. To date, CoPhMoRe sensors have been fabricated to detect nitric oxide115,116, hydrogen peroxide117, vitamins28, neurotransmitters118, carbohydrates119, proteins120,121, and small-molecule drugs and steroids28. These measurements have been performed in a variety of environments, including buffer, cellular media, and living plant systems27. The design of corona phases for CoPhMoRe sensors draws upon a huge range of amphiphilic polymers, ranging from DNA sequences to phospholipids, to synthetic polymers. Lambert et al. used directed evolution of DNA sequences as a high-throughput strategy to quickly screen and optimize an SWNT wrapping that maximizes sensor fluorescent response122. Different approaches to interface biomolecules with metals and vdW materials are summarized in Box 5.

Several groups have worked on improving the sensitivity of the CoPhMoRe method to image a single molecule. This is of great importance, as it enables detection of stochastic and dynamic interactions between biomolecules in bulk samples with high heterogeneity. Such measurements can be achieved by monitoring excitons that are induced by laser excitation and move along the sidewall of SWNTs with a characteristic diffusion length. When molecular quenchers bind to the corona/SWNT surface, the SWNT electronic structure is perturbed and excitons within a diffusion length of the adsorption site are quenched and undergo nonradiative recombination. Due to the excitonic nature of SWNT luminescence, the nanotubes do not photobleach, unlike traditional organic fluorophores. This allows for longer exposure times and higher laser fluences that can ultimately enable imaging of single molecules123. Cognet et al. have observed single proton and diazonium salt binding and release from SDBS-functionalized SWNTs by detecting stepwise changes in fluorescence124. This approach was extended by Zhang et al.115 to detect nitric oxide with SWNTs deposited onto functionalized glass sides, and by Jin et al.125 to detect H2O2 with SWNTs embedded in type 1 collagen thin films. During measurements, each SWNT serves as an individual optical probe, and individual stepwise changes in sensor states correspond to single-molecule binding or release events. The observed bulk SWNT fluorescence is a combination of these single-molecule interactions, and the bulk concentration is obtained through hidden Markov modeling125. The collagen-based SWNT sensors have also been applied to detect real-time H2O2 generation with high spatial resolution from individual live A431 human epidermal carcinoma cells stimulated with epidermal growth factor117 (Fig. 8). This approach can also be applied to understand small analyte efflux from microorganisms at the cellular level with fine spatial and temporal resolution. For example, Landry et al. immobilized RAP1 aptamer-anchored SWNT sensors in a microfluidic chamber and detected unlabeled RAP1 GTPase and HIV integrase selectively from various cell lines126. Similarly, Kruss et al. deposited a layer of SWNT sensors on a glass slide, cultured PC12 cells over them, and tracked dopamine efflux from single cells, correlating the curvature of the cell to neurotransmitter release127. At the tissue scale, SWNT sensors have been used to detect neurotransmitters in acute cultured brain slices128.

## Towards SWNT-based POC diagnostics and in vivo applications

SWNT-based sensors have also been developed to detect disease biomarkers at low concentrations, with potential application to POC diagnostics, enabling timely treatment at a reduced cost129,130. Label-free optical SWNT sensors are preferable to traditional POC detection methods such as ELISA in that they do not require additional processing steps after sample collection and allow transient data collection131. One example is a fluorescent SWNT-based microarray for detecting real-time protein-protein interactions132. Here, chitosan-wrapped SWNTs are conjugated to nitrilotriacetic acid (NTA) groups that can chelate Ni2+. The NTA-Ni2+ complex then binds to a His-tag on the target proteins, and the binding event is detected by the reduction of SWNT fluorescence. This platform has been further engineered to monitor 1156 capture protein-protein analyte interactions in a signaling network important in apoptosis. Using a different molecular chelator, Dong et al.133 conjugated immunoglobulin-binding proteins to SWNT and detected IgG, IgM, IgG2a, and IgD. Capture protein-modified sensors can also be printed onto a microarray that examines cross-reactivity, competitive, and nonspecific binding of analyte mixtures134. Such optical nanosensors have great potential to advance real-time, multiplexed biomarker detection for disease diagnostics.

The unique optoelectronic properties of SWNTs are also ideal for in vivo biosensing applications. SWNTs fluoresce within the NIR window (900–1500 nm), where scattering is reduced and tissue absorption is minimal, allowing SWNT fluorescence to penetrate through deeper layers of tissue relative to visible light135,136. A central question that limits biological applications relates to the biocompatibility of SWNTs. Their toxicity has been evaluated in vitro, with a variety of factors including chirality, length137, synthesis method138, SWNT wrapping139,140 and cell type125,140,141, all affecting biocompatibility. SWNTs are a broad class of materials with varying dimensions, functionalization, and formulation, and since many aspects of a given preparation can affect biological responses142, it is important to evaluate toxicity on a case-by-case basis. In addition, response in cell culture might not always reflect in vivo tissue effects, where cell populations are far more heterogeneous, and materials might also be modified once introduced into the body. Thus, it is crucial to test SWNT sensors in vivo, under conditions reflecting the intended route of administration, site of action, and dose.

SWNTs have been widely applied in animal models. Iverson et al. delivered ss(AAAT)7-wrapped SWNTs via tail vein injection to detect transient inflammation, using nitric oxide as a signaling marker in mouse liver. The authors demonstrated long-term nitric oxide monitoring by subcutaneous implantation of alginate hydrogel-encapsulated SWNT sensors. The sensor fluorescence signal remained stable for over 400 days, and histological analysis of the implantation sites showed negligible adverse response116. Harvey et al. loaded DNA-wrapped SWNT sensors into a semipermeable membrane and implanted these into the intraperitoneal (IP) space in mice. The sensors successfully detected oligonucleotide hybridization in vivo, and the mice exhibited no obvious behavior changes143. SWNT sensors have also been used to detect lipid concentrations in vivo in the mouse liver144. Galassi et al. injected DNA-wrapped SWNTs through the tail vein in mice, which localized to the liver144. These DNA SWNTs showed a wavelength shift in response to changing lipid concentrations, and once the mice were administered a high-calorie diet, the resulting fat accumulation within the liver could be detected and used to determine the onset of fatty liver disease. SWNTs implanted in the liver maintained the ability to shift wavelength in response to their analyte over a period of three months, showing the stability of such sensors.

Williams et al. encapsulated antibody-conjugated/DNA-wrapped SWNTs into a dialysis bag and implanted them into the IP space of mice145. These sensors detected the ovarian cancer biomarker human epididymis protein 4 (HE4) from both exogenous administration and endogenous production. DNA-wrapped SWNTs encapsulated in poly(ethylene glycol) diacrylate hydrogels have also been used in various marine organisms for biologging purposes, with minimal changes in tissue architecture for catfish, shark, and eel as detected by high-resolution ultrasound146. These examples show that SWNT sensors of various designs have the potential to function safely in vivo.

## Conclusions and future perspectives

We have explored nanophotonic biosensing technologies enabled by low-dimensional vdW materials. With continually improving large-scale synthesis methods for vdW materials, high-resolution lithography, and ultrasensitive detection schemes, it may soon be possible to harness these versatile materials for practical POC biosensing applications. Further in the future, this technology could grow to enable previously inaccessible applications such as label-free monitoring of protein conformational dynamics, flexible and wearable biosensors, and in vivo nanophotonic biosensors. Such real-time in vivo sensors could potentially provide quantitative measurements of biological networks, and in humans, could allow prolonged data collection for early detection of disease, or even be coupled to drug delivery devices to enable closed-loop disease control—analogous to the ‘artificial pancreas’ systems that have been developed for continuous glucose control in diabetes. We also expect vdW materials to benefit other functions beyond conventional sensing that will still contribute to better biosensors, such as harnessing vdW materials for on-chip liquid manipulation (e.g., liquid sealing, or nanofluidics), high-gradient-field particle trapping, or building waveguide-integrated light sources and photodetectors.

Research into graphene and 2D quantum materials for SEIRA sensing is still in its infancy. In the near future, we will likely witness innovations combining high-quality 2D materials with smart choices of substrates and device configurations. These efforts will undoubtedly improve the performance and expand the versatility of SEIRA-based sensors for a wide range of applications.

Another exciting strategy for accessing physical and chemical information entails the physical confinement of individual analyte molecules in a configuration with lower degrees of freedom. One such detection modality is based on the confinement of molecules into structures fabricated in the matrix of the 2D material, such as nanoscale apertures or nanopores (Supplementary Note 1).

The library of vdW materials and heterostructures is rapidly expanding to offer customizable choices for fabricating unique nanophotonic biosensors. For SEIRA and other biosensing applications, we envision that hybrid substrates integrating graphene, polar materials, vdW heterostructures, metal antennas, high-Q dielectric metasurfaces52,147, and low-loss dielectric waveguides will present unique opportunities for future innovation. Two-dimensional TMDCs have demonstrated strong exciton resonances in optical spectra24,148, and recent work has shown that the presence of molecules can transform dark states into bright excitons, potentially enabling unambiguous detection of the adsorbed molecules. Excitons in 2D TMDCs can also be tuned by various approaches149. In addition, integration with birefringent and chiral metasurfaces e.g., chiral plasmons or twisted graphene layers150,151 or valley-dependent chiral coupling in 2D TMDCs152 for measuring chiro-optic response could advance enantioselective biochemical sensing applications153.

While this review focused on nanophotonic sensors, electronic and electrochemical sensors based on vdW materials have also been extensively investigated154,155. Ultimately, vdW materials may empower researchers to realize one of the overarching ambitions of nanophotonics—seamless integration of photonics and electronics in the field of biosensing.