Whispering gallery mode sensing through the lens of quantum optics, artificial intelligence, and nanoscale catalysis Open Access

Accessing the length and time scales of single-molecule interactions and their transformations requires innovative techniques to trap and control light at the nanoscale. Light can be trapped inside dielectric resonators in the form of optical whispering gallery modes (WGMs) through near-total internal reflection along the inside boundary of the resonator. A small proportion of the WGM volume extends into the surrounding medium, forming a sensing region where changes in the local environment can perturb the WGM resonance.^1–4 

Plasmonic nanoparticles can be used to focus the light onto the molecules by attaching them to the WGM microresonator, thereby enabling the detection of molecules and their conformational changes.^5–8 These hot spots amplify the signals against the background and allow single molecules to be distinguished from other effects in the bulk environment. The resonance position and linewidth of the WGM optoplasmonic resonance are highly sensitive to changes in refractive index or to the polarizability of molecules in these hot spots. This sensing mechanism allows the label-free detection and study of single molecules, with a range of applications in biosensing and biochemistry.^9,10

The field enhancements of plasmonic nanorods have, to date, provided the highest sensitivity in WGM optoplasmonic sensing, down to single atomic ions interacting with the tips of the nanorod. Transient interactions of Zn²⁺ ions have shown spike signals, while the binding of Hg²⁺ by amalgamation yielded distinct step signals in the traces of the WGM optoplasmonic resonance.¹¹ This experiment sets the stage for investigating other atomistic effects on optoplasmonic WGM sensors, such as the catalytic action of single atoms deposited or synthesized at the tip of Au nanorods, including catalytic Pd, Pt, and other atomic species.^12,13

Single-atom catalysts have been attracting attention due to their ability to boost catalytic activity and selectivity by efficiently utilizing active sites and minimizing material waste, distinguishing them from conventional catalysts. Advanced single-molecule tools like WGM optoplasmonic sensors allow researchers to observe catalytic processes in real-time and study molecular interactions that influence catalytic kinetics. Furthermore, quantum-mechanical methods, such as density functional theory (DFT), allow for calculating the complexation energies of small molecules with the gold surface of the nanorod tip. For example, if small molecular species with different complexation energies compete for binding at the nanorod tip, WGM optoplasmonic sensing can reveal such competition by resolving the switch from transient (spike) interaction to permanent (step) interactions, depending on the molecular species in the experiment or buffer. This could be a useful tool for understanding the energy of molecular interactions at the surface of nanoparticles in complex solvent environments and their effects on catalytic reactions.

Another prospect for WGM optoplasmonic sensors is to utilize the light fields and their hotspots not only for sensing but also to drive a chemically, biologically, or plasmonically powered process. In this respect, artificial photosynthesis is of interest as a promising field that aims to replicate nature's process of photosynthesis for sustainable energy production.¹⁴ 

Despite the exceptional sensitivity of WGM sensors, which has reached single atomic ion sensitivity,¹¹ more could be achieved if the noise level could be reduced further. Noise sources include thermorefractive noise and, at a fundamental level, laser shot noise. Quantum sensing is an aspect of quantum technologies that allows measurements with precision beating classical noise limits such as shot noise. Quantum sensing techniques are increasingly being used in biosensing experiments, and there are a number of reviews in this area.^15–17 

Moving toward quantum optical sensing requires more accurate signal analysis tools. In principle, WGM optoplasmonic signals encode information about light–matter interactions between single molecules and the biosensing fields, assuming the desired sensitivity and time resolution are available. Essentially, the resonance shift is related to the change in polarization energy of the molecule as it moves through nanoscale field gradients.¹⁸ Different types of single-molecule interactions can be identified from the sensor signals, including transient events, binding events and conformational changes, as well as absorption events in thermo-optoplasmonic (TOP) sensing.¹⁹ The analysis of WGM signals is supported by computational models for such nanoscale interactions, giving information about the quantum properties and dynamics of molecules. These models reveal changes to molecule-state transitions affected by quantum effects.

In this perspective, we discuss three future research directions in whispering gallery mode sensing with a focus on single-molecule studies: quantum optical sensing and quantum biology; advanced artificial intelligence (AI) and digital twin signal analysis; and studying catalysis at the nanoscale. In summary, our perspective is structured as follows:

In Sec. II, we discuss quantum optical biosensing in terms of the fundamental limits to the precision of optical measurements. We discuss the prospects for future quantum-enhanced precision in WGM sensing and single-molecule detection. We identify the challenges in realizing this experimentally and the extent to which future improvements in WGM sensors are expected to come from reductions in classical and quantum noise.

In Sec. III, we highlight advanced computational methods that can help to decode WGM signals and identify patterns in the spectroscopic and sensing data. We discuss the prospects of computational models that leverage the precision of electronic structure calculations and the predictive capabilities of machine learning models.

In Sec. IV, we explore future applications of WGM single-molecule sensing in studies of catalysis. These experiments could benefit greatly from improved sensor performance and advanced computational techniques, especially DFT methods introduced earlier. Studies of the catalytic activity and efficiency of catalytic centers on bimetallic nanoparticles down to single catalytic atoms on WGM optoplasmonic sensors would have important applications in biotechnology and environmental sustainability, for example, in the reduction of CO₂.

Finally, in Sec. V, we highlight the future prospects for WGM single-molecule sensors, covering emerging topics, such as quantum optical WGM sensing, applications to studying quantum biology, and digital twins of WGM sensors.

Squeezed states are states with the field quadrature uncertainties $Δ X 1 , Δ X 2$ redistributed so that the uncertainty in one quadrature can be decreased below the QNL, while the other is increased. Squeezed light is typically generated by nonlinear optical processes in crystals, often enclosed in optical cavities to form an optical parametric oscillator (OPO).^23,24

Successful applications of squeezed light have improved measurement sensitivity from macroscopic to microscopic scales. Examples include gravitational wave detection: reducing noise in the LIGO (Laser Interferometer Gravitational-Wave Observatory) experiment by a factor of up to 1.9²⁵ and in biosensing experiments such as quantum-enhanced Raman microscopy.²⁶ 

From Eq. (1), we see that one way to decrease uncertainty is to increase N, i.e., use more optical power. This is possible until increasing the power further introduces new dominant noise sources, or begins to cause damage or undesired photosensitive effects in the sample being studied. In optomechanics, this limit could be due to measurement back-action.²⁹ In the specific case of optical biosensing, there is a limit to optical power due to photodamage, particularly when using focused beams to illuminate cells, such as in optical tweezer studies.^30,31 A strong motivation for quantum optical biosensing then is to increase the signal-to-noise ratio (SNR) of biosensing measurements beyond that allowed by the QNL by reducing the noise level, subject to the constraint on optical power due to photodamage. Figure 1(a) illustrates this schematically: the highest SNR can in principle be achieved by going beyond the QNL at the maximum possible power (represented by photon number per measurement N).

Recent quantum optical biosensing experiments have demonstrated the potential for practical enhancements to the SNR of several relevant measurements on biological systems. Much effort in this area is toward using individual spins such as diamond NV-centers to measure magnetic fields, temperature, and other properties, for example, using optically detected magnetic resonance (ODMR).^32,33 In this perspective, we will focus on quantum sensing experiments using entanglement or squeezing of optical probe states to make measurements on biological samples with enhanced precision.

An early demonstration of quantum optical biosensing was using entangled photon pairs to measure protein concentration with an uncertainty below the QNL (after accounting for photon losses).²⁷ This experiment used a Mach–Zehnder interferometer (MZI) as shown in Figs. 1(b) and 1(c) and shows how integrated photonic circuits can be combined with microfluidics to make small and robust biosensors using quantum optical probe states. More recent photon pair experiments include plasmonic sensing using heralded single photons,³⁴ tapered fiber sensors using photon pairs and path-entangled photons (N00N states),^35,36 and entangled two-photon absorption (ETPA) microscopy, which is proposed as a method to dramatically reduce the optical intensity required for two-photon microscopy.³⁷ 

Squeezed light optical probes offer the possibility of quantum biosensing experiments at optical power similar to that of their classical counterparts. This is essential to realize practical improvements in sensing precision. As an example of this, Raman microscopy using squeezed light has been demonstrated using optical powers above and below the photodamage threshold for yeast cells.^26,38 This illustrates exactly the argument put forward in Fig. 1(a) by showing a 13% SNR enhancement above the QNL at the photodamage threshold. Furthermore, squeezed light experiments include quantum-enhanced particle tracking,^39,40 plasmonic sensing using twin-beam squeezed light,^41,42 squeezed light dark-field microscopy,⁴³ and squeezed light Brillouin microscopy.⁴⁴ 

One study particularly relevant to this article shows a WGM toroid magnetometer with a squeezed light probe state.²⁸ This is shown in Fig. 1(c). A narrowband phase-squeezed field at 1064 nm produced by seeded parametric downconversion in a linear optical cavity is coupled to the microtoroid by a tapered nanofibre.⁴⁵ The input wavelength is matched to a WGM resonance by temperature tuning the microtoroid. Magnetic field changes could be read out via the acoustic modes of the WGM resonator. By using a squeezed probe state, the magnetic field sensitivity of this optomechanical sensor was improved by 20%.

In addition to discussing measurements beating the QNL, we can also ask where these experiments lie in relation to fundamental noise limits, i.e., the Heisenberg limit. Optical phase measurements have been demonstrated with Heisenberg-limited scaling⁴⁶ or close to the absolute Heisenberg limit;⁴⁷ however, these experiments use relatively small numbers of photons, i.e., we find them on the left-hand side of Fig. 1(a). In comparison, photodamage thresholds can be estimated from optical tweezer studies at typically ∼100 mW power in tightly focused beams.^30,31 The squeezed light experiments discussed above that reach or aim to reach the regime where photodamage becomes important are found on the right-hand side of Fig. 1(a) and so far are just beyond the QNL. The main challenge in pushing further toward fundamental noise limits is mitigating photon losses, since losses throughout an experimental setup will degrade the level of squeezing in the optical probe state.

Optoplasmonic WGM sensors are capable of ultra-high sensitivity when used to detect single molecules. However, there is a great motivation to increase the sensitivity of these sensors yet further to detect signals hidden in the noise, such as small motions of enzymes or state changes of proteins, enzymes, or photoreceptors associated with small energy changes that are affected by quantum effects. Single-molecule sensing at the quantum noise limit has previously been achieved using a heterodyne dark-field scattering technique to detect bovine serum albumin (BSA) molecules in the evanescent field of a tapered nanofibre.⁴⁸ Single-molecule detection with quantum-enhanced precision has not yet been achieved. Quantum optical biosensing is exciting as one approach to improve the sensitivity of WGM optoplasmonic sensors and explore how to make single-molecule measurements beyond the QNL.

To propose quantum sensing strategies using WGM sensors, we must consider the noise sources involved—and whether technical noise sources can be mitigated to reach the QNL in the first place—before going beyond QNL sensitivity.

At sufficiently low optical power, e.g., in the photon counting regime, photon shot noise will dominate the sensor signal and a sensing enhancement could feasibly be demonstrated using entangled photon states. A common strategy from quantum sensing would be to use N00N states in a MZI (as in Ref. 27) to make an enhanced phase measurement. This scheme might be used to measure the phase shift near a WGM resonance, as proposed by Ref. 49. WGM resonators could also enable some novel measurement schemes: an add-drop filter configuration⁵⁰ allows photons otherwise lost on resonance to be detected; a type of Hong–Ou–Mandel effect using this scheme has been proposed theoretically.⁵¹ Also making use of polarization modes, polarization entangled states might take advantage of coupling separately to TE and TM modes in a WGM resonator, which are typically detuned from each other. This may enable a polarization interferometer using a single waveguide, as opposed to the MZI.

Restricting ourselves to the photon counting regime is rather limiting since reaching sensitivities similar to current WGM single-molecule sensors requires significantly higher optical power. This is where we might look to squeezed states of light to enhance optoplasmonic WGM sensors, as was shown for optomechanical WGM magnetometers.²⁸ At higher optical power, shot noise is typically far below other noise sources in WGM sensors: laser frequency jitter,⁵² mechanical vibrations at the evanescent coupling point, and thermorefractive noise.^53–55 These noise sources must be dealt with first before being able to demonstrate enhancements beyond the QNL using squeezed light. Methods that show promise for reducing the noise include active locking of the laser to the WGM resonance,^8,56 heterodyne detection methods to move the signal to a frequency band with lower noise,⁴⁸ or measuring the separation of WGM split modes.⁷ Cavity ring-up spectroscopy (CRUS) with WGM sensors has been suggested as a self-heterodyning technique that could be shot noise limited, which also enables high time resolution measurements (16 ns demonstrated).⁵⁷ If the WGM signal in some frequency band can reach the QNL, squeezed states of light can further reduce noise below this limit and enable practical enhancements in WGM quantum sensing. By also using the techniques for optoplasmonic sensing using plasmonic nanoparticles, this is a promising route to quantum-enhanced sensing of single molecules.

It remains to be seen which WGM measurement schemes and squeezed input states can offer the greatest sensitivity advantage. It would be useful to theoretically investigate different squeezed quadrature states, such phase squeezing²⁸ or intensity squeezing.³⁹ Other possibilities include mixing vacuum squeezed states into an interferometer;^58,59 or coupling bright squeezed states to the WGM.²⁸ Additionally, twin-mode squeezed states should also be considered.⁴² Very important to this work is the recent theoretical study showing that for a WGM absorption or refractive index measurement with a single waveguide coupled to the resonator, the quantum Fisher information of the measurement is maximized with a coherent state.⁶⁰ At the optimal measurement conditions, it was shown that there is no advantage to using a squeezed state. Further investigations will identify how this result relates to other setups, such as those using two-mode quantum optical states or two waveguides coupled to the WGM resonator (e.g., add-drop filter configuration), and whether under the constraints applying to current WGM optoplasmonic sensing experiments there can be an advantage to using squeezed light. This very interesting property of a WGM quantum sensor suggests that for some setups the fundamental noise limit for sensor performance could be reached even using coherent states.

Now, to turn to applications for WGM sensing with enhanced sensitivity, we will give some examples of small single-molecule signals, which would be challenging to detect with current sensors. One such application would be detection of single enzyme activity. The optoplasmonic WGM wavelength shift signal due to the conformational change in an enzyme is proportional to the polarizability change due to the changing configuration of the enzyme.¹⁰ Therefore, reducing the noise floor of WGM sensors can enable yet smaller conformational changes of single enzymes to be resolved. At first, this noise reduction can be achieved by reducing classical noise sources, but to improve noise beyond the QNL toward fundamental noise limits will require quantum sensing techniques.

We have shown that quantum optics can enable measurements on biological systems with enhanced sensitivity, but biological systems themselves are increasingly being shown to exhibit “non-trivial” quantum mechanical phenomena.⁶¹ Ultra-sensitive methods for detecting small polarizability changes in single proteins could provide another method for investigating biomolecules involved in quantum biological processes. For example, proteins that would be interesting for these studies include cryptochrome (involved in the quantum mechanical avian compass)^62,63 and bacteriorhodopsin (BR).⁶⁴ We address these possibilities in the outlook Sec. V B.

Advanced computational models and artificial intelligence (AI) can help to decode spectroscopic WGM signals. The sensor detects both global and local changes in the environment within the volume encompassed by the WGM evanescent fields. Global changes have a uniform effect on the sensor signals, resulting in a direct mapping between the sensing parameter and the WGM signals. This facilitates precise calibration strategies, which can be optimized with AI. In contrast, local perturbations (e.g., due to single molecules) lead to heterogeneous signals, since the strength of the light–matter interactions are position-dependent. These signals encode information about the properties of analyte molecules and their location in the near fields. Deciphering these multivariate signals requires additional contextual knowledge about the behavior and dynamics of molecules in optoplasmonic fields. This can be achieved through multiscale computational models,^65,66 which often require substantial computing power. This section provides an overview of decoding WGM signals, highlighting the interplay between scientific computing, supercomputers, and AI.

WGM transmission spectra often contain higher-order or nonlinear spectral features that emerge due to various factors, such as mode-splitting induced by symmetry breaking,⁶⁷ inelastic scattering processes, or photo-thermal effects, among other nonlinear effects.⁶⁸ The transmission spectra can be thought of as “optical barcodes” that represent a unique state of the surrounding environment.⁶⁹ For example, each barcode could correspond to a highly precise global temperature measurement. The one-to-one correspondence between temperature and the optical barcodes can be used to build a calibrated database as a reference for analyzing WGM signals. This approach can similarly be used to measure other global sensing parameters, provided that the measurements and barcodes can be accurately calibrated.

Developing this idea further, several studies apply data-driven approaches to automate and accelerate the calibration and analysis of WGM signals.^70–76 In particular, machine learning algorithms excel at recognizing patterns within high-dimensional spectroscopic data. Such algorithms can analyze the shift of multiple spectral features, learn nuanced correlations between global sensing parameters and optical barcodes, and make predictions about bulk changes in the surrounding environment. This has led to the development of “intelligent” WGM sensors for highly precise temperature,^70,71 pressure,⁷² and bulk refractive index^73–76 measurements. It is important to note that machine learning algorithms must be re-calibrated for different experimental setups, due to variations in the geometry and optical properties of WGM resonators.

While data-driven methods provide a way to analyze bulk signals, it is not immediately clear how to extend these ideas to single-molecule sensing. Molecules are dynamic and have many degrees of freedom. The strength of the light–matter interactions can fluctuate as the molecule moves through the non-uniform optoplasmonic fields. Physics-based computational models can help to make predictions about the behavior of single molecules in optoplasmonic fields. These models can provide information about the dynamics of molecules in plasmonic hot spots, which may help to interpret patterns in the spectroscopic data. Since the intensity of the fields over the molecular volume is not accurately known, it is often more conducive to compare relative changes in the sensor signals, rather than absolute values.

Superficially, it may seem as though no computational modeling problem is out of reach, since the world's fastest supercomputer can perform 10¹⁸ floating point operations per second.⁷⁷ However, the scalability of scientific code is often limited by multiple factors, such as the complexity of the underlying algorithms, communication overheads, and memory requirements. In essence, there is an art to computational modeling that goes beyond brute force number crunching. It requires simplifying the problem with carefully considered approximations, without compromising an adequate representation of the system.

Modeling WGM optoplasmonic sensors is particularly challenging due to the multiscale and multivariate nature of the problem. There are two main types of interactions to consider: the first is between plasmonic nanoparticles and the evanescent WGM fields;⁷⁸ the second is between analyte molecules and the highly confined plasmonic fields. Different approximations can be used to model these interactions, as will be discussed in the following. Furthermore, the solvent environment can affect the field distribution of the sensor and the electronic response of analyte molecules.

In classical electromagnetism, the wave-like nature of light is described by Maxwell's equations.⁷⁹ There are different ways to solve Maxwell's equations, including both analytical and numerical methods. The boundary element method⁸⁰ (BEM) is a numerical grid-based method that is well-suited for simulating WGM resonators⁸¹ and plasmonic nanoparticles,^82,83 since these components are typically homogeneous, isotropic materials enclosed by a boundary surface. BEM can be used to fine-tune optical and plasmonic resonances, as well as visualize the mode profiles of optoplasmonic sensors. Additionally, BEM is generally much faster than other numerical methods, such as the finite difference time domain (FDTD) method and the finite element method (FEM), since only surfaces, rather than volumes, must be discretized in the simulation.

While BEM is an excellent method for simulating continuous structures, it cannot capture the intricate structural details of analyte molecules that may be much smaller than the wavelength of light. In BEM simulations, the complex structure of a biomolecule or protein is often represented as a generic polarizable sphere.⁸⁴ This simplification makes it impossible to study the conformational dynamics of molecules in optoplasmonic fields and their effect on WGM signals. Although an enzyme can be represented by a more realistic hinge-bending shape,⁵ there is still a lack of information about its conformational landscape, preferred binding orientation, and electronic response. Coupling BEM simulations with discrete atomistic models⁸⁵ or electronic structure calculations^86,87 can lead to a more accurate description of the molecule and its behavior in the optoplasmonic fields.

Equation (4) is particularly useful when an analyte molecule has electronic excitations at frequencies within the scanning range (or frequency) of the WGM sensor. For example, chromophores absorb visible light and may resonantly interact with optical WGM sensors. Such experiments could explore the mechanisms behind photo-switchable molecules. Another type of experiment could explore the resonant interactions of aromatic molecules with π (pi) orbitals that absorb ultraviolet (UV) radiation,⁹⁷ using WGM sensors that operate in the UV regime.^98,99

In many experiments, the probe laser is decoupled from molecular excitations, such that we can apply the electrostatic approximation,¹⁰⁰ where $α ( ω WGM ) ≈ Re [ α ( ω ↦ 0 ) ] ≡ α$⁠. Calculating α in the electrostatic limit using the finite field method is significantly faster than performing TDDFT simulations and avoids convergence issues with respect to the cutoff value for the number of states. Additionally, there are additive atomistic models for α that scale well for molecules with tens of thousands of atoms.^101,102

So far, our discussion of the molecular polarizability α did not consider the effects of an aqueous environment. It is important to consider these effects since WGM sensing experiments are often performed in an aqueous environment. In the context of WGM sensing, the polarizability of a solvated molecule is often referred to as the excess polarizability¹⁰³ of the molecule, $α ex ∗$ (i.e., the polarizability of the molecule in excess of the displaced solvent). Our research group recently developed a hybrid quantum-classical polarizability model to calculate $α ex ∗$⁠. The approach combines an implicit solvent model, a three-layer cavity model, and quantum-mechanical polarizability model to predict $α ex ∗$ in the electrostatic limit.¹⁰⁴ Within the point dipole approximation, $α ex ∗$ is proportional to the fractional shift in the WGM resonance.^9,18

Data-driven approaches are becoming more commonplace in computational chemistry.^105–108 This is partly due to the availability of highly accurate quantum-mechanical datasets⁸⁹ that can be used to train machine learning (ML) models. An interesting example of an ML model is AlphaML,^90,91 which is able to predict the gasphase polarizability α of small organic molecules. Once the model is trained and validated, it can rapidly predict α.

The schematic in Fig. 2 describes this process, using the analogy of states and operators in quantum mechanics (except for taking the square of the matrix elements, which is not necessary here). The supercomputer can be thought of as an operator that maps between molecular structures and their polarizability α via electronic structure calculations. The results for α are stored in quantum-mechanical datasets. These datasets provide synthetic data to train ML algorithms. The resulting ML model can rapidly predict α or even the shift $Δ λ$ directly, if the effective volume of the sensor is known.

While there is some computational effort involved to train the ML model, the advantage is that it can later predict α for structures that it has not seen before, as long as the initial structure conforms to certain limitations of the original training data. For example, AlphaML was trained on electrically neutral gas phase molecules and therefore cannot accurately predict the effect of charge in a solvent environment.¹⁰⁴ 

Another interesting example is a deep learning framework called FieldSchNet,¹⁰⁹ which can predict single-molecule interactions with an arbitrary environment. The model can also operate as a polarizable continuum model and predict the effect of an implicit solvent on molecular properties.

Integrating WGM resonators with engineered gold nanoparticles presents a promising approach for the real-time tracking of molecular reactions at the single-molecule level.^11,110 By incorporating a plasmon enhancer into the microcavity, ultra-sensitive nanoscale detection volumes are created within plasmonic sensing hotspots, enabling the detection of single molecules as they enter the plasmon-enhanced sensing volume.¹¹⁰ This advancement in WGM sensing methodology allows for monitoring reactions between molecules weighing less than 1 kDa without the need for complex fluorescent labeling, facilitating real-time molecular analysis.⁶ For example, WGM sensors have already been utilized to track in real-time the disulfide-exchange reaction between individual thiol sites introduced at the surface of the plasmon enhancer and sub-kD thiolated molecules such as the amino acid cysteine. WGM resonance shifts reveal the step-like sensing signals associated with the formation and reformation of disulfide bonds through redox cycling of attomolar reactants.⁶ By adding catalytic sites or centers, such as single catalytic atoms, within plasmonic sensing hotspots of the bimetallic Au nanoparticle, WGM sensors could track the interaction of substrate and product molecules at or near those catalytic sites in real-time.

The WGM sensor is a sophisticated tool for studying catalytic reactions on engineered bimetallic AuNR catalysts at the single-molecule level. It can identify and quantify contributions from synergistic effects such as electron transfer, lattice strain, and bifunctional properties resulting from incorporating secondary metals (Co, Cu, Ag, Ni, Pt, Pd) into AuNRs.^111–113 The presence of two different metals in these nanoparticles facilitates electron transfer and charge redistribution, leading to changes in the d-band center and density of states. This precise modulation of the d-band center allows for adjusting the adsorption strength of reactant molecules on the catalyst surface, influencing reaction kinetics, and selectivity could be evaluated by optoplasmonic sensing signals.^112,114,115 The composition and structure of bimetallic nanoparticles can be tailored for a variety of reactions, including hydrogenation, oxidation, and CO₂ reduction.¹¹⁶ For example, in hydrogenation reactions, adding Pt or Pd can shift the d-band center of Au nanoparticles to enhance the binding and activation of hydrogen molecules, resulting in faster hydrogenation rates and improved chemoselectivity.^117–119 Similarly, in oxidation reactions, the modified electronic structure can enhance oxygen species adsorption, promote the reactive intermediate formation, and improve overall catalytic performance. These examples demonstrate the diverse applications of WGM sensors in chemical and environmental analysis.

To effectively visualize and monitor catalytic conversions using the WGM sensor, one could strategically position secondary metals at the hotspot region of Au nanoparticles, for example, at the tips of the gold nanorods. This deliberate arrangement enables synergistic interactions between the secondary metal and gold, facilitating electron transfer processes and promoting the activation of specific chemical bonds.¹² Moreover, the choice of secondary metal could be customized based on its intrinsic electronic properties and catalytic functionalities, allowing for the design of tailored catalysts optimized for a particular reaction.

We could further apply this approach by incorporating enzymes such as hydrogen-dependent carbon dioxide reductase (HDCR) onto these nanorods. HDCR is an enzyme that plays a crucial role in the reduction of CO₂ to formate by using hydrogen as a reducing agent. This enzyme is of particular interest due to its potential application in the field of biotechnology and environmental sustainability.^120–122 Understanding the mechanisms and functions of HDCR on WGM optoplasmonic sensors could provide valuable insights into developing novel technologies for carbon capture and utilization. The WGM sensor could provide a detailed understanding of the catalytic mechanisms involved in CO₂ reduction at a molecular level by tracking their single-molecule turnover events on these enzymes. There are currently two plausible pathways involved in the electron transfer from a plasmonic particle to an enzyme for the reduction of CO₂. In the first pathway, hot electrons from the plasmonic particle are directly injected into the enzyme for the reduction of CO₂. In the second pathway, the hot electrons are first harvested to reduce protons to hydrogen, which is then utilized by the enzyme for the reduction of CO₂ to formate. A multidisciplinary approach involving WGM sensing in conjunction with other analytical tools will not only shed light on the fundamental processes governing CO₂ conversion but also offer insights into the electron transfer process and developing sustainable catalytic systems for mitigating greenhouse gas emissions and advancing the field of carbon capture and utilization. A visual representation of these prospects and insights for controlling and monitoring catalysis on WGM optoplasmonic sensors is provided in Fig. 3.

Understanding how reactant molecules interact with nanoparticle surfaces is essential for predicting and controlling molecular reactions, with buffer molecules further influencing these interactions. DFT provides a robust computational approach to studying these interactions¹²³ seen in the WGM sensing signals and their impact on molecular reactions. It is noteworthy that reactant molecules could only interact with the nanoparticle surface once they surpass the interaction energies of the buffer molecules (Fig. 4). By constructing a comprehensive DFT lookup table that maps complexation energies to specific molecular interactions, it becomes possible to interpret signals in WGM sensing more easily, and researchers could deepen their understanding of the factors governing reaction rates at the single-molecule level. Furthermore, WGM sensing could be utilized to study the effects of solvents on single-molecule bio-sensing for the first time. Through the examination of how distinct solvent environments influence molecular interactions and reaction kinetics, researchers can explore novel avenues for enhancing the sensitivity and selectivity of biosensors.

We have put forward that pushing toward fundamental noise limits with quantum-enhanced WGM single-molecule biosensors would enable new and important sensing capabilities; however, there remain challenges to realizing this goal.

Initially, proof-of-principle experiments in the current shot noise limited regime could use photon pair states and entanglement to demonstrate a sensing advantage.⁴⁹ The challenge here is to develop narrowband and wavelength-tunable photon pair sources, which can be coupled to WGM resonators. This is possible using cavity-assisted SPDC sources, which have been used to couple to atomic transitions.^124–126 

Demonstrations in the photon counting regime are also interesting in exploring the physics of entangled photons coupled to WGM resonators. One theoretical prediction that could be tested using the same WGM-coupled photon pairs is the HOM effect in a WGM resonator in an add-drop filter setup, as proposed by Alsing et al.⁵¹ Such experiments could have applications beyond sensing in quantum information more generally.

Finally, to make practical improvements to WGM single-molecule sensing requires setups working at an equivalent optical power to current classical WGM sensors. Future work should determine whether squeezed states can lower the noise level of WGM sensors or whether the fundamental noise limit in the case of a mode coupled to a WGM resonator is reached with a coherent state,⁶⁰ and how this generalizes to different sensing schemes. In any case, the challenge in pushing toward fundamental noise limits is to first overcome technical noise sources present in WGM signals. Any method that mitigates noise, classical or quantum, can enable new capabilities for WGM sensors applied in single-molecule biosensing and beyond.

Sensitive single-molecule studies of proteins, such as enzymes, as well as photoresponsive (photoreceptor) proteins, such as cryptochrome and BR, could be an important tool in studying the emerging field of quantum biology. What is required to facilitate a rapid expansion of this potent field are sensors capable of directly probing the biological quantum phenomena at the single-molecule level.

One potential application of quantum-enhanced WGM sensing is the detection of single enzyme activity. Light-activated enzymes provide an ideal system for understanding how dynamic transitions facilitate catalysis via quantum mechanical tunneling, as the reaction chemistry can be initiated using laser pulses, allowing the investigation of the catalytic cycle across multiple timescales. Protochlorophyllide oxidoreductase (POR) serves as a natural, light-activated enzyme that is experimentally accessible. It catalyzes the activation and reduction of protochlorophyllide using NADPH through sequential proton (H⁺) and “hydride” (H⁻) transfers, reactions fundamental to both biology and chemistry. POR plays a crucial role in chlorophyll biosynthesis, a reaction essential for all life on Earth.¹²⁷ 

Addressing a critical question, which has been challenging to explore experimentally due to its requirement for a single-molecule approach, involves understanding the role of protein and solvent motions in coupling to the H-tunneling steps, as well as the dynamics associated with the formation and decay of the photoreceptor excited states.

Bacteriorhodopsin (BR) is another light-activated enzyme, a membrane-based proton pump. BR contains retinal, a biological chromophore ubiquitous in visual receptors of higher life forms, serving also as an antenna in light energy transformation and phototaxis of bacteria. BR is a particularly robust protein that remains active when dried in air and when probed at low temperatures; the retinal photoisomerization dynamics have been studied down to 20 K.¹²⁸ At room temperature, proton tunneling contributions to BR activity have been investigated, and studies have shown the possibility of quantum superposition in BR.⁶⁴ This indicates the relevance of quantum interference between vibrational states in the retinal moiety of BR. The low-temperature long-lived BR components (milliseconds to microseconds) are important targets for single-molecule studies because they can reveal reaction barriers and transition rates affected by quantum effects. In addition, a recent theory proposal¹²⁹ suggests that photoreceptors (such as BR) may be capable of violating the second law of photochemistry when coupled to optical cavities such as optoplasmonic sensors.

We can also consider the response of enzymes to quantum optical states of light. Entangled two-photon absorption (ETPA) experiments aim to demonstrate two-photon microscopy at reduced optical power by using quantum correlations between entangled photons.^130,131 It would be interesting to investigate ETPA in a biomolecule with a strong optical response. BR undergoes a conformational change in response to green illumination,¹³² and this change can be detected with high sensitivity using WGM sensors.^133–135 Potentially, here is a method to measure entangled two-photon absorption cross sections in a biomolecule at very low optical powers using a highly sensitive WGM sensor readout.

We discussed how reducing noise in WGM sensors could enable very small conformational changes of proteins to be resolved. This could help shed light on mechanisms of magnetoreception using cryptochrome by resolving its photocycle with high precision. Cryptochrome (Cry) belongs to a class of highly conserved (flavo)proteins that often function as photoreceptors and/or have a role in the circadian clock. In addition, their magnetic field sensitivity following photoactivation is thought to underlie the avian magnetic compass and similar magnetosensitive phenotypes.⁶³ An outstanding challenge is to observe the magnetic field effects in the structural rearrangement of a single cryptochrome protein as direct proof of a molecular compass. Another challenge is the observation of singlet–triplet interconversion as part of the spin chemistry, thought to lead to observable quantum beats. As the implicated transitions of cryptochrome states alter the molecule's polarizability, these changes can be detected on optoplasmonic sensors. Radiofrequency fields are thought to be able to influence the dynamics of the protein, yet this possibility of quantum control has not been directly shown.

The interpretation of WGM experiments may be facilitated by the use of digital twins. A digital twin is a virtual replica of a real system that can be used to investigate complex multivariate problems. Digital twins use information from both simulations and experiments to reconstruct the behavior of a real system. For instance, digital twins have been used to decode the signals from chiral molecules in cavity-enhanced spectrometry.¹³⁶ In the future, digital twins could be used to decode the spectroscopic signals from WGM sensors; they could help to reconstruct a molecular movie by predicting and matching the sensor signals to the most probable interactions, orientations, and conformational states that a molecule such as protein can access while monitored on the sensor. This is a rather ambitious goal, due to the intricacy of nanoscale interactions. However, with continuous advancements in computational modeling and WGM sensing, this goal is within practical reach.

A future goal is the creation of a user-friendly toolbox that predicts optical biosensor responses by simulating atomistically resolved molecules such as proteins placed in the optical near fields of the sensor. Most sensitive optical biosensors such as the WGM optoplasmonic sensors discussed here detect molecules from shifts of optical, plasmonic, or optoplasmonic resonances. By combining classical electromagnetism simulations and additive atomistic polarizability models, the tool could predict the sensor signals, i.e., resonant frequency shift signals for different molecules perturbing the biosensor's electromagnetic fields. The toolbox could be validated on the optoplasmonic single-molecule sensors discussed here and could allow an automated and fully quantitative signal analysis in single-molecule biosensing. The tool could generate large datasets by combining available structural and simulated molecular dynamics data with sensing signal predictions that are needed for training AI models and systems.

Applied to the actual sensing signals, the tool could extract information about the nature of molecular events (binding or transient interactions), the type of molecule, its biomolecular structure, and its orientation on the sensor, which will transform the use of a large variety of optical biosensors into a much wider range of new applications. For example, the toolbox could provide information about protein conformational changes, which are crucial in drug screening, protein engineering, and understanding molecular mechanisms of diseases such as Alzheimer's. As the capabilities of single-molecule sensors such as the WGM optoplasmonic sensors discussed here advance, for example, by incorporating TOP sensing data on absorption cross section or Raman-like signals that identify chemical groups and molecular vibrational signatures, multiparameter AI prediction models will become available, which could transform optical molecular science and chemical analysis methods. Crucially, the toolbox could apply to a wide range of optical biosensors that operate based on probing molecules by overlapping with the optical field with field gradients on the length scale of the molecules. These biosensors are widely used in academia and industry, examples are single-molecule localized surface plasmon resonance-based sensors,¹³⁷ waveguide and optical fiber-based sensors,⁴⁸ interferometric Mach–Zehnder and ring resonator sensors. One could implement the toolbox as an app for use on a variety of handheld phones and tablets by end users from academia and industry.

The authors acknowledge support from the Engineering and Physical Sciences Research Council (Grant Nos. EP/R031428/1, EP/T002875/1, and EP/X018822/1). E.Z. and M.W. acknowledge support from the German Research Foundation (Grant No. WA 1687/10-1).

The authors have no conflicts to disclose.

Ekaterina Zossimova and Callum Jones contributed equally to this work.

Ekaterina Zossimova: Investigation (equal); Writing – original draft (equal); Writing – review & editing (equal). Callum Jones: Investigation (equal); Writing – original draft (equal); Writing – review & editing (equal). Kulathunga Mudalige Kalani Perera: Investigation (supporting); Writing – original draft (supporting); Writing – review & editing (supporting). Srikanth Pedireddy: Investigation (supporting); Writing – original draft (supporting); Writing – review & editing (supporting). Michael Walter: Supervision (equal); Writing – original draft (supporting); Writing – review & editing (equal). Frank Vollmer: Conceptualization (lead); Project administration (lead); Supervision (equal); Writing – original draft (supporting); Writing – review & editing (equal).

Data sharing is not applicable to this article as no new data were created or analyzed in this study.