Quantum relaxometry for detecting biomolecular interactions with single NV centers

Significance

Quantum sensing using nitrogen-vacancy (NV) centers in diamond promisingly offers spin observation window to detect biomolecular interactions at single-molecule level. A critical bottleneck, however, lies in achieving high-density molecular immobilization on the diamond surface without compromising bioactivity—a challenge stemming from inefficient surface functionalization and steric interference. We overcome this limitation by engineering a nanogel-coated interface that enables dense molecular anchoring while preserving native physiological functions. This innovation not only enables observation of single-molecule interactions but also resolves ensemble-averaged interaction behaviors at the microscale. By bridging the gap between single-molecule stochasticity and population-level biophysics, our platform establishes a versatile framework for high-precision analysis of molecular interactions, with broad implications for proteomic studies, diagnostics, and the fundamental understanding of biological processes.

Abstract

The investigation of biomolecular interactions at the single-molecule level has emerged as a pivotal research area in life science, particularly through optical, mechanical, and electrochemical approaches. Spins existing widely in biological systems offer a unique degree of freedom for detecting such interactions. However, most previous studies have been largely confined to ensemble-level detection in the spin degree. Here, we developed a molecular interaction analysis method approaching single-molecule level based on relaxometry using the quantum sensor, nitrogen-vacancy (NV) center in diamond. Experiments utilized an optimized diamond surface functionalized with a polyethylenimine nanogel layer, achieving $\sim$10 nm average protein distance and mitigating interfacial steric hindrance. Then we measured the strong interaction between streptavidin and spin-labeled biotin complexes, as well as the weak interaction between bovine serum albumin and biotin complexes, at both the micrometer scale and nanoscale. For the micrometer-scale measurements using ensemble NV centers, we reexamined the often-neglected fast relaxation component and proposed a relaxation rate evaluation method, substantially enhancing the measurement sensitivity. Furthermore, we achieved nanoscale detection approaching single-molecule level using single NV centers. This methodology holds promise for applications in molecular screening, identification, and kinetic studies at the single-molecule level, offering critical insights into molecular function and activity mechanisms.

Studying molecular interactions and dynamics at the single-molecule level is of vital significance for exploring the intrinsic mechanisms of biological and chemical processes. To date, a variety of single-molecule measurement techniques have been developed, utilizing approaches like optics, mechanics, and electrochemistry. Notable examples include single-molecule fluorescence resonance energy transfer (smFRET) (1, 2), fluorescence correlation spectroscopy (FCS) (3, 4), Raman spectroscopy (5, 6), high-speed atomic force microscopy (AFM) (7–9), optical and magnetic tweezers (10–12), and nanostructured devices based on nanopores (13–15) and nanowires (16–18). These techniques have been extensively applied to single-molecule measurements, including studies of DNA hybridization and melting kinetics (19, 20), protein binding (21), and enzyme activity (16). However, optical-based methods suffer from photobleaching, limiting the long-term observation. Electrochemical methods based on nanopores and nanowires are susceptible to interference from ionic concentrations in the solution (22). Furthermore, an emerging trend in single-molecule research is the integration of multiple complementary physical measurement windows to acquire more comprehensive and precise insights into molecular interactions (23, 24).

Spin-based measurement methods are promising in overcoming the challenges faced by the aforementioned optical and electrochemical approaches. Biological systems inherently possess diverse spin sources, such as nuclear and electron spins in proteins, free radicals involved in metabolic activities, and transition metal ions, which collectively provide a broad signal bandwidth (25). While traditional magnetic resonance (MR) techniques excel at ensemble-level analysis, they lack the sensitivity for single-molecule detection. The nitrogen-vacancy (NV) center has garnered extensive attention in recent years as a quantum sensor, due to its excellent magnetic field sensitivity and outstanding biocompatibility. Notably, individual biomolecule detections based on spin signals have been achieved using NV centers (26–29). This advancement provides both technical infrastructure and conceptual frameworks for detecting molecular interactions at this single-molecule level. NV-based quantum relaxometry, with simple pulse sequences, has been successfully applied to detect magnetic noise of free radicals (30–39), pH (40, 41), and metal-based paramagnetic spins such as Gd³⁺, Mn²⁺ and Cu²⁺ ions (26, 42–49). These metal-based spin labels offer photostability and subnanometer dimensions superior to fluorescent labels, presenting great potential in molecular interaction detection.

Up to now, there is still a lack of effective attempts to measure biomolecular interactions at the single-molecule level in the spin degree. In this work, we presented a biomolecular interaction analysis method that integrates metal-based spin labeling with quantum relaxometry employing NV centers in diamond. The basis of this methodology is the precise engineering of the sensor–biomolecule interface, which optimizes bonding density while preserving molecular activity. By functionalizing the diamond surface with a polyethylenimine (PEI) nanogel, we achieved a protein bonding distance of $\sim$10 nm and minimized steric hindrance. A relaxation rate evaluation method, reconsidering the often-overlooked fast relaxation component, was developed to improve the measurement sensitivity of ensemble NV centers. This enabled sensitive detection of both strong and weak interactions between biomolecules at the micrometer scale. We further extended this approach to single NV centers, achieving measurements at nanoscale and near single-molecule level.

Results

Principle and Conceptual Design.

The basic idea of our experiment is depicted in Fig. 1A. Protein A (yellow) is immobilized on the functionalized diamond surface and then the solution of protein B (blue) labeled with paramagnetic metal spin is introduced onto the interface. If there is an interaction between protein A and B, the capture of B by A results in a close distance between the spin labels and shallow NV centers, thus fluctuating magnetic fields with nonzero root mean square (RMS) of the spin labels will influence NVs’ spin relaxation process. Therefore, quantum relaxometry based on the NV center can be used to detect the magnetic noise of the metal-based spin label. The measurement sequence for the relaxation time $T_1$ is performed as Fig. 1B. The spin state of NV is polarized to $|0⟩$ by a 532 nm laser pulse (or to $|\pm 1⟩$ by a following microwave $\pi$-pulse), and freely decays to the thermal equilibrium state after a dark evolution time $\tau$. The difference of decay curves with initial spin state of $|0⟩$ and $|\pm 1⟩$ is normalized and the relaxation time $T_1$ is obtained (Fig. 1C). The signal of the metal-based spin label is quantified by the acceleration $\mathrm{\Delta }\mathrm{\Gamma }$ of the relaxation rate ${\mathrm{\Gamma }}_1=1/T_1$.

Fig. 1.

Quantum relaxometry for detecting biomolecular interactions using NV centers. (A) Schematic of the experimental setup. Protein A (yellow) is immobilized on the diamond surface to capture spin-labeled protein B (blue). The NV center detects magnetic noise from the spin label via acceleration of its spin relaxation rate. (B) $T_1$ measurement sequence. The spin state of the NV center is initialized and read out using 532 nm laser pulses (green). A microwave $\pi$-pulse (blue) is applied to flip the spin state between $|0⟩\leftrightarrow |\pm 1⟩$. Varying the dark evolution time $\tau$ yields the $T_1$ relaxation profile. (C) $T_1$ curves measured with only protein A (yellow) and with captured spin-labeled protein B (blue). Two relaxation curves were measured using the upper and lower pulse sequences shown in (B), corresponding to initial spin states $|0⟩$, $|\pm 1⟩$. The difference of the curves was normalized to the reference point at $\tau =0$, yielding the normalized intensity defined as $(I_{0,\tau }-I_{\pm 1,\tau })/(I_{0,0\mathrm{\mu }\text{s}}-I_{\pm 1,0\mathrm{\mu }\text{s}})$.

In this work, we used Gd³⁺, Mn²⁺ and Cu²⁺ ions chelated by a ligand as the magnetic labels, which are important labels of traditional electron paramagnetic resonance (ESR). They have a long lifetime under laser irradiation, thus providing a sufficient time window for long-term monitoring. We evaluated the photostability of the three labels with a typical laser condition of 30$\mathrm{\mu }W$ 532 nm on a home-built optically detected magnetic resonance (ODMR) spectroscope (SI Appendix, section 1). The results show that the signals of Gd³⁺ and Mn²⁺ are maintained above 60$\%$ after more than 15 h of continuous laser irradiation (SI Appendix, Fig. S1), while the lifetime of Cu²⁺ is $\sim$11 h (SI Appendix, Fig. S2). Therefore, continuous signal accumulation for $\le$120 h can be performed at room temperature, considering that the laser duty cycle is about 0.1 in most quantum relaxometry experiments. The biotin–ubiquitin (biotin-Ub) complex is a powerful synthetic tool. Combined with the high affinity of biotin to streptavidin (SA), it is widely used in biochemical and cellular research to study ubiquitination—a critical posttranslational modification that regulates protein degradation, trafficking, and signaling (50, 51). Therefore, we chose the interaction between SA and the Mn²⁺ labeled biotin–ubiquitin (biotin-Ub(Mn), details in SI Appendix, Fig. S3) to demonstrate our biomolecular interaction detection approach.

Diamond Surface Functionalization.

As an interface analysis method, the experiments required protein immobilization on the surface of the bulk diamond. Commonly used oxygen terminal diamond surface has functional groups such as carboxyl, hydroxyl, carbonyl, and ether, in which carboxyl groups can be directly utilized for molecular immobilization (52). Strategies based on carboxyl residuals have achieved an average molecular spacing of $\sim$20 nm (26, 28, 29). However, interaction studies demand further surface optimization to enhance bonding density while preserving biomolecular activity. We observed that the function of SA was limited by steric hindrance when it was directly attached onto the oxidized diamond surface. Specifically, SA failed to capture biotin-Ub when anchored to tri-acid-oxidized surfaces (SI Appendix, Figs. S4 and S5). To address this, we used PEI as a buffer layer to increase the distance between SA and the diamond surface. PEI, a polymer rich in amine groups, has shown significant effectiveness for the functionalization of nanodiamonds with small molecules (38, 53, 54). Here, we extended this strategy to the application for protein conjugation on the surface of bulk diamond, demonstrating both high immobilization efficiency and preserved protein biological activity. The bulk diamond surface is carboxylated with a tri-acid solution, and then an amino cross-linked thin layer is formed on the diamond surface with the mixed solution of PEI and 4 arm-PEG-NHS used to form an amino cross-linked thin layer on the diamond surface (details in SI Appendix, section 3.2). The freshly prepared PEI exhibits a thickness of 1 to 6 nm and it can collapse to $\ll$1 nm after being stored under room temperature atmospheric conditions for extended periods exceeding 24 h. Notably, the PEI layer showed no measurable impact on the $T_1$ relaxation of NV centers (SI Appendix, Fig. S6). Additionally, the density and uniformity of the Ub(Mn) bonding on the PEI surface are evaluated by spin signals. Spin signal analysis revealed an average Ub(Mn) bonding distance of $\sim$10 nm (SI Appendix, Fig. S7), with only $\sim$13% signal variation across four regions spaced 500 $\mathrm{\mu }m$ apart, confirming high surface homogeneity (SI Appendix, Fig. S8). Besides, fluorescence analysis also shows a consistent bonding spacing of $\sim$8 nm (SI Appendix, Fig. S9). Subsequent interaction experiments demonstrated that the PEI-functionalized surface effectively mitigates steric hindrance, providing a robust platform for biomolecular interaction studies.

Micrometer-Scale Detection of Biomolecular Interaction with Ensemble NV Centers.

Initially, we employed a bulk diamond containing shallow ensemble NV centers, with an average depth of 6.5 nm below the [100] surface, to demonstrate the reliability of the method in studying the overall behavior of molecular interactions. As illustrated in Fig. 2A, the diamond surface was functionalized with PEI to introduce amino groups, followed by the immobilization of SA or BSA (bovine serum albumin) onto the surface. Subsequently, a solution of biotin-Ub(Mn) was added to the diamond and allowed to react for 40 min. The diamond was then cleaned with deionized water and dried using nitrogen gas. We utilized the atomic force microscope (AFM) to characterize the diamond surface after the bonding process, and the result for SA$+$biotin-Ub(Mn) is presented in Fig. 2B. AFM contact mode was employed to remove the protein from a square area and the surrounding topography was scanned using tapping mode of AFM. The measured height of the protein layer is 5.8 $\pm$ 1.3 nm. Given that the average hydrodynamic diameters of SA and Ub are $\sim$5 nm (55) and $\sim$3 nm (56), the 5.8-nm height suggests the formation of a protein monolayer.

Fig. 2.

Micrometer-scale biomolecular interaction detection using ensemble NV centers. (A) Schematic of biomolecular interaction detection with ensemble NV centers in bulk diamond. The diamond surface is functionalized with a polyethylenimine (PEI) nanogel to achieve surface amination. Then the immobilized proteins, such as streptavidin (SA) or bovine serum albumin (BSA), are sequentially bonded to the surface. Finally, spin-labeled molecules are introduced for detection. (B) Topography image of the diamond surface coated with SA and biotin-Ub(Mn) using atomic force microscope (AFM). Proteins in the square depression area were removed using AFM contact mode. Inset shows a total thickness of 5.8 $\pm$ 1.3 nm. (C–E) Relaxation results of SA$+$biotin-Ub(Mn), as shown in Inset in (C). (C) $T_1$ curves measured in regions with (27 red lines)/without (21 blue lines) proteins shown in (B). The dashed lines are experimental curves and the hollow dots are averaged data. These dashed lines were fitted using a biexponential decay function to obtain weighted relaxation rates (details in Materials and Methods). (D) Comparison of the averaged relaxation rate derived from the dashed lines in (C). The error bar is the SD. The difference in relaxation rates between the two regions is (4.8 $\pm$ 0.5) $\times {10}^3$ s⁻¹. (E) 2D map of the relaxation signal. By fixing the waiting time of the $T_1$ measurement sequence at ${\tau }_{\text{fix}}=350\mathrm{\mu }s$, the fluorescence intensity normalized to $\tau =0\mathrm{\mu }s$ was obtained around the depression area shown in (B). Inset: Profile of the black line in the 2D map. (F–H) Relaxation results of BSA$+$biotin-Ub(Mn), as shown in Inset in (F). (F) $T_1$ curves measured in regions with (29 red lines)/without (31 blue lines) proteins. (G) Comparison of the averaged relaxation rate derived from the dashed lines in (F). The relaxation rate difference between the two regions is (0.8 $\pm$ 0.3) $\times {10}^3$ s⁻¹. (H) 2D map of the relaxation signal at ${\tau }_{\text{fix}}=350\mathrm{\mu }s$. The proteins in the square depression area have been removed.

Fig. 2C illustrates $T_1$ curves of the region in Fig. 2B with (red) and without (blue) protein coating. This comparison highlights a significant relaxation acceleration for the protein-coated region relative to regions without proteins or coated only with SA (as depicted in SI Appendix, Fig. S10). To quantify the signal intensity, we fitted the $T_1$ curves using a biexponential decay function $y=A_{\text{short}}\text{exp}(-t/T_{1,\text{short}})+A_{\text{long}}\text{exp}(-t/T_{1,\text{long}})$, which provided superior fitting compared to the single-exponential decay, consistent with prior studies (33–35, 43). These studies have primarily emphasized the slow relaxation component (considered more sensitive to external spin signals) while neglecting the fast relaxation component from surface-proximal NV centers. Our Monte Carlo simulations reveal that shallow NV centers contribute a higher proportion to the detectable signal (details in Materials and Methods), as the relaxation change $\mathrm{\Delta }\mathrm{\Gamma }$ originating from the coupling strength between NV and spin label exhibits a stronger sixth-power inverse dependence on separation distance $r$ ($\mathrm{\Delta }\mathrm{\Gamma }\propto 1/r^6$, SI Appendix, section 8.2), which outweighs the fourth-power depth dependence of NV relaxation rate ${\mathrm{\Gamma }}_1$ on depth $d$ (${\mathrm{\Gamma }}_1\propto 1/d^4$) (57, 58). To utilize this fast component, the ensemble relaxation rate was quantified using the amplitude-weighted characteristic rate ${\mathrm{\Gamma }}_{1,\text{w}}=(A_{\text{short}}/T_{1,\text{short}}+A_{\text{long}}/T_{1,\text{long}})/(A_{\text{short}}+A_{\text{long}})$, yielding a sensitivity enhancement of up to 4.5-fold (details in Materials and Methods). As shown in Fig. 2D, the mean ${\mathrm{\Gamma }}_{1,\text{w}}$ for protein-coated regions (red) exceeded that of bare regions (blue) by (4.8 ± 0.5)$\times {10}^3$ s⁻¹. This fast relaxation acceleration indicates that SA has a high binding efficiency for the biotin-Ub complex, which is attributed to the high affinity between SA and the biotin component of the complex. Then we fixed the dark evolution time $\tau$ of $T_1$ measurement sequence and scanned around the square region without protein. The relaxation rate distribution was visualized through the normalized fluorescence intensity. The region with protein exhibited faster relaxation, corresponding to lower fluorescence intensity. The 2D map in Fig. 2E aligns with the AFM topography. In addition, we substituted SA with BSA, which is commonly used as a blocking agent and lacks specific biotin-binding sites like SA (Inset in Fig. 2F). However, the surface plasmon resonance (SPR) analysis revealed weaker binding efficiency between BSA and biotin-Ub complex (SI Appendix, Fig. S11). This observation was corroborated by our relaxation measurements. Similarly, the AFM topography indicates a protein monolayer with a thickness of $\sim 8$ nm (SI Appendix, Fig. S13). The relaxation rates in regions with/without proteins show a little difference and the mean relaxation acceleration is (0.8 $\pm$ 0.3) $\times {10}^3$ s⁻¹ (Fig. 2 G and H). In addition, we demonstrated the method’s efficacy under solution conditions (SI Appendix, Fig. S22). The ensemble NV-based results highlight the capability of our quantum relaxation methodology in effective differentiation between strong and weak biomolecular interactions, holding potential applications for molecular recognition and screening.

Nanoscale Detection of Biomolecular Interaction with Single NV Centers.

We next conducted experiments employed single NV centers. The diamond sample was fabricated with nanopillar arrays on the surface and the NV centers were implanted at an average depth of 5.5 nm (details in Materials and Methods, SI Appendix, section 7.2). The chemical treatment of the diamond surface followed the same protocol as for ensemble NV diamonds (Fig. 3A). A total of 195 single NV centers with long background relaxation times ($T_{1,\text{BG}}>$0.5 ms) were selected as quantum sensors. Fig. 3B shows the $T_1$ curves of a typical single NV center under four conditions: 1) bare diamond, 2) surface bonded only with SA, 3) surface bonded with BSA followed by biotin-Ub(Mn) solution, 4) surface bonded with SA followed by biotin-Ub(Mn) solution. The data points were fitted with a single-exponential decay function. The obtained relaxation rate showed minimal variation under the first three conditions, with significant acceleration observed in the presence of SA$+$biotin-Ub(Mn). Statistical analysis was conducted on the relaxation rates of the 195 single NV centers. As shown in Fig. 3C, the average relaxation rates of the four conditions are $(0.92\pm 0.04)\times {10}^3{\text{s}}^{-1}$, $(0.9\pm 0.1)\times {10}^3{\text{s}}^{-1}$, $(1.0\pm 0.2)\times {10}^3{\text{s}}^{-1}$, $(6.5\pm 0.7)\times {10}^3{\text{s}}^{-1}$. The change of average relaxation rate for BSA$+$biotin-Ub relative to the bare diamond ($\sim$0.08$\times {10}^3{\text{s}}^{-1}$) is smaller than that observed in the above ensemble NV experiments, which could potentially be attributed to the selection of relatively deep single NV centers (background rate, i.e., without protein, ${\mathrm{\Gamma }}_{1,\text{BG}}<2,000s^{-1}$) and the variations in surface modifications observed between diamond samples. And the relaxation rate change $\mathrm{\Delta }\mathrm{\Gamma }={\mathrm{\Gamma }}_{1,\text{w/ protein}}-{\mathrm{\Gamma }}_{1,\text{BG}}$ for each NV center is given as a histogram in Fig. 3D. For the SA$+$biotin-Ub condition, the distribution of $\mathrm{\Delta }\mathrm{\Gamma }$ is broad, primarily attributed to the random positioning of proteins and the varying depths of NV centers. It is noted that some NV centers exhibit significantly accelerated relaxation rates induced by protein aggregation (above $1\times {10}^4{s}^{-1}$), thereby increasing the average signal intensity beyond typical measurement ranges. Monte Carlo simulations suggest that the mean spacing between SA proteins is approximately 12 nm (details in SI Appendix, section 10.2).

Fig. 3.

Nanoscale biomolecular interaction detection using single NV centers. (A) Schematic of the interaction detection with single NV centers in nano-pillar. The expanded view shows the side profile of a nano-pillar coated with PEI, preimmobilized proteins (e.g., SA or BSA), and spin-labeled protein (e.g., biotin-Ub(Mn)). (B) $T_1$ curves of a typical single NV center under four conditions. The dots are experimental data points and the solid lines are fitting curves to the single exponential decay function. (C) Relaxation rates of 195 single NV centers under these four surface conditions. The mean rates for each condition are presented as mean $\pm$ SD, which are $(0.92\pm 0.04)\times {10}^3{s}^{-1}$, $(0.9\pm 0.1)\times {10}^3{s}^{-1}$, $(1.0\pm 0.2)\times {10}^3{s}^{-1}$, $(6.5\pm 0.7)\times {10}^3{s}^{-1}$. (D) Histogram distributions of relaxation accelerations for 195 single NV centers relative to the bare diamond. The dashed lines represent Gaussian fits. The fitting results presented as the peak $\pm$ SD are $(-0.2\pm 0.3)\times {10}^3{s}^{-1}$ (green line, SA) and $(-0.2\pm 0.4)\times {10}^3{s}^{-1}$ (purple line, BSA$+$biotin-Ub). The solid red line is the theoretical curve obtained from the Monte Carlo simulation, assuming an average distance of 12 nm between SA proteins (details in SI Appendix, Fig. S21).

Theoretical Simulation for the Magnetic Signal of Metal-Based Spin Label.

From the distribution of relaxation rate acceleration detected by single NV centers, we were able to estimate the number of proteins around NV centers. The theoretical model is employed as below. Generally, the relaxation rate of the NV center is affected by the magnetic noise of its environment and given by

$$\begin{array}{c}\hfill {\mathrm{\Gamma }}_1={\mathrm{\Gamma }}_{1,\text{BG}}+\mathrm{\Delta }\mathrm{\Gamma }=\frac{1}{T_{1,\text{BG}}}+3{\gamma }_e^2⟨B_{\perp }^2⟩\frac{{\tau }_c}{1+{\omega }_0^2{\tau }_c^2},\end{array}$$ [1]

where ${\mathrm{\Gamma }}_{1,\text{BG}}=1/T_{1,\text{BG}}$ is the background relaxation rate (without proteins), which is due to the lattice relaxation and spin bath near the NV center (57, 58). Here, we simplify the background relaxation rate ${\mathrm{\Gamma }}_{1,\text{BG}}={\mathrm{\Gamma }}_{1,\text{surf}}+{\mathrm{\Gamma }}_{1,\text{bulk}}$. ${\mathrm{\Gamma }}_{1,\text{surf}}$ is attributed to the magnetic noise of surface impurities, which varies with the depth of NV centers, while ${\mathrm{\Gamma }}_{1,\text{bulk}}$ is independent of NVs’ depth (details in SI Appendix, section 9.2). ${\gamma }_e$ is the NV gyromagnetic ratio. $B_{\perp }$ is the RMS magnetic field of the metal-based spin labels at the location of the NV center, which is determined by both the protein bonding density and the NV depth. As the width of the spectral density function of Mn²⁺ is $\sim$GHz (43), our experiment is performed at zero magnetic field. Thus the resonant frequency is ${\omega }_0=2\pi D_{\text{gs}}$ ($D_{\text{gs}}=2.87$GHz is the zero-field splitting of the NV center between $|0⟩$ and $|\pm 1⟩$) and ${\tau }_c$ is the autocorrelation time of the magnetic noise of spin labels. $\mathrm{\Delta }\mathrm{\Gamma }$ quantifies the signal intensity, representing the acceleration of the relaxation rate caused by molecular interaction events (details in SI Appendix, section 8.2).

To analyze the distribution of relaxation rates, we implemented a Monte Carlo simulation involving 10,000 single NV centers, where the depth of NV centers approximately follows a Gaussian distribution according to the stopping and range of ions in matter (SRIM) simulation (SI Appendix, Fig. S18C). First, we derived the density of the surface paramagnetic impurities from the statistical distribution of ${\mathrm{\Gamma }}_{1,\text{BG}}$. It is then assumed that SA proteins, which have an approximate cubic shape with a size of 5.8 nm, are randomly distributed on the diamond surface, with each SA capturing four biotin-Ub molecules (Fig. 4A). The relaxation acceleration $\mathrm{\Delta }\mathrm{\Gamma }$ of 10,000 NV centers at a specific SA bonding density was thus simulated and analyzed statistically. By comparing the distribution of the simulated $\mathrm{\Delta }\mathrm{\Gamma }$ with the experimental data, the average SA bonding density consistent with the experimental results was obtained. The solid red line in Fig. 3D represents the simulated statistical distribution that best fits the experiment, corresponding to an average SA bonding distance of 12 nm, i.e., a bonding density ${\sigma }_{\text{SA}}\approx 0.007{\text{nm}}^{-2}$ (SI Appendix, Fig. S21).

Fig. 4.

Simulation of the magnetic noise signal measured with single NV centers. (A) Schematic of the Monte Carlo simulation. Single NV centers are modeled with depths ($d$) following a Gaussian distribution ($\mu =5.5\mathit{nm}$, $\sigma =2.8\mathit{nm}$) derived from SRIM ion implantation simulations. SA is approximated as a cubic structure with 4 biotin-binding sites distributed on the diamond surface at an average spacing of 12 nm. (B) Correlation between the NV center’s background relaxation rate ${\mathrm{\Gamma }}_{1,\text{BG}}$ and the detected magnetic signal intensity $\mathrm{\Delta }\mathrm{\Gamma }$. Yellow circles are the experimental data from 195 single NV centers. The colormap displays the simulated probability density of obtaining $\mathrm{\Delta }\mathrm{\Gamma }$ at given ${\mathrm{\Gamma }}_{1,\text{BG}}$. Outliers (low ${\mathrm{\Gamma }}_{1,\text{BG}}$, high $\mathrm{\Delta }\mathrm{\Gamma }$) indicate localized protein aggregates or depth estimation deviations. (C) Single-molecule detection probability. The background illustrates the simulated single-molecule detection probability $P_{\text{single}}$, depicted by the color scale. $P_{\text{single}}$ quantifies the probability that the observed $\mathrm{\Delta }\mathrm{\Gamma }$ originates from individual SA-biotin-Ub(Mn) complexes at the given ${\mathrm{\Gamma }}_{1,\text{BG}}$. The blue curve marks the contour of $P_{\text{single}}=0.5$.

We further analyzed the probability of single SA proteins interacting with biotin-Ub(Mn) as detected by single NV centers. Fig. 4B shows the relationship between background relaxation rates of NV centers and signal intensity $\mathrm{\Delta }\mathrm{\Gamma }={\mathrm{\Gamma }}_{1,\text{SA+biotin-Ub}}-{\mathrm{\Gamma }}_{1,\text{BG}}$ at ${\sigma }_{\text{SA}}\approx 0.007{\mathit{nm}}^{-2}$. The color map depicts the probability density derived from Monte Carlo simulations at specified background relaxation rate ${\mathrm{\Gamma }}_{1,\text{BG}}$ and signal intensity $\mathrm{\Delta }\mathrm{\Gamma }$. Simulations indicate that NV centers with larger background relaxation rates are more likely to measure higher signals. Experimental data points from 195 single NV centers (yellow dots) align closely with the simulated distribution, with the majority of data points clustering in high-probability regions. A small subset detects anomalously high signals (scattered data points in low-probability regions). These outliers can be attributed to two factors: surface protein aggregates generating strong localized magnetic noise or discrepancies between theoretical and actual NV depths. Specifically for the second, NV centers shallower than predicted—due to sparse local surface electron distribution—exhibit enhanced coupling with spin labels on proteins, causing their signals to deviate from statistical expectations. In our analysis, we defined a single-molecule detection event as one where over 70$\%$ of the signal originates from a single SA-biotin-Ub(Mn) complex, based on Monte Carlo simulations. The probability of detecting such events was mapped in a two-dimensional parameter space of background relaxation rate (${\mathrm{\Gamma }}_{1,\text{BG}}$) and relaxation acceleration ($\mathrm{\Delta }\mathrm{\Gamma }$), as shown in Fig. 4C. The results indicate that for most of the 195 NV centers, the probability of detecting a single SA-biotin-Ub(Mn) complex is below 0.5, with only 5 NV centers falling within the range of 0.5 to 0.7. This low probability is attributed to the selection of relatively deep NV centers. To improve detection efficiency, selecting NV centers with shallower depths (2.5 to 5 nm) is recommended, as their closer proximity to the surface enhances sensitivity to interactions at the single-molecule level.

Discussion

In conclusion, we have developed a quantum relaxation-based molecular interaction analysis method utilizing NV centers as quantum sensors and paramagnetic metal spins as labels, enabling multiscale analysis of biomolecular interactions from micrometer to nanoscale levels. For ensemble NV measurements at the micrometer scale, they achieved superior sensitivity owing to exceptional fluorescence counting rates and large effective sensing areas. Moreover, we introduced a refined relaxation rate evaluation protocol for ensemble NV centers, achieving an up to 4.5-fold sensitivity enhancement. This improvement facilitate studies of transient biological phenomena with NV relaxometry, particularly in vivo free radical dynamics—a critical frontier in biomedical research (33–39). However, the spatial resolution of ensemble measurements remains constrained by the diffraction limit ($\sim$600 nm spot diameter), where signals averaged over hundreds of NV centers obscure nanoscale heterogeneity. In contrast, single NV centers provide nanoscale resolution ($<$10nm), enabling single-molecule detection, which is critical for resolving intrinsic variations in molecular binding kinetics, conformation, and local chemical environments. This nanoscale capability nevertheless faces challenges of experimental efficiency, in which high-throughput pillar arrays and improved surface modification can help. The PEI nanogel was proven to be effective to provide a universal surface modification strategy with $\sim$0.01nm⁻² protein bonding density and reduced steric hindrance. Moreover, the amide bond linking PEI to the carboxylated diamond surface can be hydrolyzed using alkaline solutions, which permits controlled surface regeneration (59). Combined with microfluidics, this recyclable interface could enable in situ cleaning and refunctionalization, allowing rapid sensor reconfiguration for diverse analytes in diagnostic arrays and enhancing cost-efficiency. Further optimizations include NV depth control ($<$5 nm) to increase single-molecule detection probability and readout methods selection like spin-to-charge conversion readout schemes (60) to improve experimental efficiency and temporal resolution.

Quantum relaxometry with metal ion spin labels provides a photostable platform for detecting biomolecular interactions, overcoming the photobleaching limitations of conventional optical techniques and enabling long-term monitoring. While Gd³⁺/Mn²⁺ are detected at zero magnetic field due to GHz-scale magnetic noise, Cu²⁺ requires external magnetic field alignment ($\sim$450 Gs) due to hundreds of MHz-scale spectral linewidth, allowing magnetic “bicolor” labeling in complex biological environments. Strategic integration with emerging microdroplet or microfluidic architectures could further transform this platform into a high-sensitivity, high-throughput, and low-volume biodetection system (61, 62), with promising applications in drug screening. Moreover, the demonstrated single-molecule sensitivity opens opportunities in clinical diagnostics, particularly for liquid biopsy applications requiring ultralow-abundance biomarker detection (63, 64). Additionally, combining our relaxometry approach with nanoscale biointerfaces such as nanopillar arrays, offers unique capabilities for in situ cellular activity studies (65). This includes probing plasma membrane protein organization (66, 67), monitoring nuclear deformation (68), and tracking ion channel dynamics (69). The minimally invasive nature of NV centers further permits intracellular biosensing without disrupting cellular integrity. In short, this work not only provides a fundamental analysis of molecular interactions but also establishes a versatile platform based on large-scale quantum biosensor arrays, promoting a crucial step toward the construction of a high-throughput and high-sensitivity platform and the engineering applications of quantum biotechnology.

Materials and Methods

Biexponential Shape of T₁ Curve of Ensemble NV Centers.

The $T_1$ curve of the shallow ensemble NV center exhibits a biexponential decay, as shown in Fig. 5A, which has been reported (33–35, 43). The fast relaxation component is usually attributed to the effect of cross-relaxation of NV pairs, or NV centers close to diamond surface. Given that the implanted dose of the ensemble diamond used in this work is ${10}^{-13}{\text{cm}}^{-2}$ (with an estimated average distance of $\sim$30 nm between NV centers), the effect of cross-relaxation can be neglected. We excluded the impact of charge relaxation using a 594 nm laser for readout and observed that the $T_1$ curve of the ensemble NV center still exhibited a biexponential shape (SI Appendix, Fig. S16). According to the following simulations, the biexponential behavior is demonstrated to arise from the heterogeneous relaxation rates of individual NV centers within the ensemble.

Fig. 5.

New evaluation method for relaxation rate of ensemble NV center. (A) $T_1$ curve of an ensemble NV center. The solid (dashed) line is the (bi)exponential decay fitting curve. (B) $T_1$ curves of three single NV centers. Solid lines are fitting curves to the single exponential decay function. (C) Histogram of background relaxation rates of 471 single NV centers. The solid line is a fitting curve based on the surface magnetic noise model (SI Appendix section 9.2 and Fig. S20). (D) Averaged $T_1$ data of 471 single NV centers in (C). The pink (green) line is a fitting curve to the (bi-)exponential decay function. (E) Schematic of surface magnetic noise model for shallow NV center. Bulk magnetic noise and surface paramagnetic impurities jointly influence background relaxation rates. (F) Simulated distribution of ${\mathrm{\Gamma }}_{1,\text{BG}}$ for shallow single NV centers as ${\sigma }_{\text{surf}}=0.40{\mathit{nm}}^{-2}$ (SI Appendix, section 9.3). (G) Comparison of the simulated $T_1$ curve (purple line) and the experimental $T_1$ data (black dots) for the ensemble NV center. The simulated $T_1$ curve is the average data of single exponential decay curves from the distribution in (F). (H) Comparison of simulated (solid lines) and experimental $T_1$ data (hollow dots). The diamond surface is only bonded with Ub(Mn) and the derived bonding density is ${\sigma }_{\text{Ub}}={(1/9)}^2{\mathit{nm}}^{-2}$. (I) Dependence of the ensemble NV relaxation rate change $\mathrm{\Delta }\mathrm{\Gamma }$ on the Ub(Mn) bonding density ${\sigma }_{\text{Ub}}$ with relaxation rate evaluation methods: 1) characteristic rate $1/T_{1,\text{stre}}$ from stretched exponential fits, 2) slower characteristic rate $1/T_{1,\text{long}}$ from biexponential fits, 3) weighted rate ${\mathrm{\Gamma }}_{1,\text{w}}$ (our method). Black points are the arithmetic average of relaxation rate accelerations for single NV centers constituting an ensemble NV center. The solid lines are linear fitting curves of data corresponding to the bonding spacing from 20 to 7 nm.

Single shallow NV centers typically display single-exponential $T_1$ decay (Fig. 5B and SI Appendix, Fig. S17), but their relaxation rates vary widely (100 to 40,000 $s^{-1}$, Fig. 5C). Superimposing these 471 single-exponential $T_1$ curves reproduces the ensemble’s biexponential profile (Fig. 5D), confirming the attribution of the biexponential shape to the relaxation rate broadening of single NVs. As described in the main text, the depth-dependent surface magnetic noise broadens the relaxation rates of single NV centers (Fig. 5E and SI Appendix, section 9.2). According to the SRIM simulation, NV depths follow approximatively a Gaussian distribution (truncated at $\ge$2 nm from the surface considering the shallow NVs’ instability; SI Appendix, Fig. S18C). Averaging $T_1$ curves of 40,000 single NV centers with surface spin density of $\sim$0.40 ${\mathit{nm}}^{-2}$ simulates the experimental $T_1$ curve of ensemble NV center (Fig. 5 F and G and SI Appendix, Fig. S19), the consistency demonstrating the reliability of the surface magnetic noise model.

Evaluation for Relaxation Rates of Ensemble NV Centers.

The biexponential decay reflects contributions from shallower (fast-relaxing) and deeper (slow-relaxing) NV centers. To enhance the sensitivity, we introduce a weighted relaxation rate: ${\mathrm{\Gamma }}_{1,\text{w}}=(A_{\text{short}}/T_{1,\text{short}}+A_{\text{long}}/T_{1,\text{long}})/(A_{\text{short}}+A_{\text{long}})$, where $A_{\text{short}}$, $A_{\text{long}}$ and $T_{1,\text{short}}$, $T_{1,\text{long}}$ are amplitudes and relaxation times from biexponential fitting. This evaluation method is compared to two conventional approaches: (1) using the slower component $T_{1,\text{long}}$ from biexponential fits (33–35, 43), and (2) extracting $T_{1,\text{stre}}$ from stretched-exponential fits (45, 49, 70–72). We presented a comparison of the three methods by numerical simulations as follows: first, simulated background $T_1$ curve of ensemble NV center were generated by averaging 40,000 individual decays $y_i=\text{exp}(-t·({\mathrm{\Gamma }}_{1,i}+\delta {\mathrm{\Gamma }}_{1,i})$. Then local relaxation accelerations $\delta {\mathrm{\Gamma }}_{1,i}$ for each single NVs were introduced due to the randomly distributed Ub(Mn) on the surface at density ${\sigma }_{\text{Ub}}$. These modified decays $y_i=\text{exp}(-t·({\mathrm{\Gamma }}_{1,i}+\delta {\mathrm{\Gamma }}_{1,i})$ were averaged to produce the $T_1$ curve with Ub protein. Fig. 5H shows a typical comparison of simulated (solid lines) and experimental $T_1$ data (yellow dots) at ${\sigma }_{\text{Ub}}={(1/9)}^2{\mathit{nm}}^{-2}$. The relaxation rate difference between with and without Ub $\mathrm{\Delta }\mathrm{\Gamma }$ reflects the simulated measured signal intensity. All three methods exhibit linear responses to Ub(Mn) density (Fig. 5I). However, ${\mathrm{\Gamma }}_{1,\text{w}}$ captures 2.0$\times$ and 5.5$\times$ greater sensitivity to protein bonding than $1/T_{1,\text{long}}$ and $1/T_{1,\text{stre}}$, respectively, as quantified by linear-fit slopes around the experimental bonding density. While all methods underestimate the true single-NV average signal (black points, $\sum _i\delta {\mathrm{\Gamma }}_{1,i}/40,000$), ${\mathrm{\Gamma }}_{1,\text{w}}$ aligns most closely with the average signal intensity. Despite nonlinearity at low densities of ${10}^{-4}$ to ${10}^{-3}{\mathit{nm}}^{-2}$, ${\mathrm{\Gamma }}_{1,\text{w}}$ nearly equals the true average signal.

Diamond Membrane.

Single crystalline diamond membranes (Element Six, Electronic Grade) were used to create NV centers via ion implantation. Ensemble NV centers were generated by implanting 8 keV 14${\text{N}}_2^{+}$ ions at a dose of $1\times {10}^{13}{\mathit{cm}}^{-2}$ into the [100] cut bulk diamond (2 mm$\times$2 mm$\times$0.5 mm), followed by vacuum annealing (800 ^°C, 2 h). SRIM simulations indicated an average NV depth of $6.5\pm 2.8$nm and a density of $\sim 1,000\mathrm{\mu }{m}^{-2}$. The uniformity of fluorescence and spin properties is demonstrated in SI Appendix, Fig. S14.

For single NV center experiments, an ultrathin diamond membrane (2mm$\times$2 mm$\times$0.05 mm) was used. Four regions of the diamond were implanted with 4 keV 15${\text{N}}^{+}$ at doses of $1\times {10}^{11}{\mathit{cm}}^{-2}$, $3\times {10}^{11}{\mathit{cm}}^{-2}$, $6\times {10}^{11}{\mathit{cm}}^{-2}$, $1\times {10}^{12}{\mathit{cm}}^{-2}$. Cylindrical nano-pillars were fabricated by a self-aligned patterning technique (73). The process began with spin-coating the diamond surface with double layers: a lower polydimethylglutarimide (PMGI) layer ($\sim 270$nm) and an upper polymethyl methacrylate (PMMA) layer ($\sim 210$nm). Electron-beam lithography (EBL) patterned smaller holes in the PMMA, while wet etching with 2.38$\%$ tetramethylammonium hydroxide created larger undercuts in the PMGI layer, forming a dual-mask structure. Nitrogen ions were then implanted through these masks, confining NV center formation to regions within a 15 nm radius. Then the smaller hole array of PMMA on the top was fabricated by electron beam lithography (EBL) and the bigger hole array of PMGI below was generated by the wet etching with 2.38$\%$ tetramethylammonium hydroxide. Next, the N⁺ ions were implanted into the diamond through the PMMA and PMGI holes. The well-designed double-hole layers constrain the implantation region to within a 15 nm radius from the center. After removing PMMA using acetone, depositing the titanium (Ti) layer by electron beam evaporation, and removing PMGI using N-methyl pyrrolidone, a circular Ti mask array was formed on the diamond surface. The nano-pillars were fabricated by reactive ion etching (RIE) with the mixed gas of CHF₃ and O₂. The height of the conical cylinder ($\sim$350 nm) was controlled by the etching time. Finally, the Ti layer was removed with a buffered oxide etch. The diamond was annealed at 1,000 ^°C and subjected to a boiling tri-acid solution (98$\%$ H₂SO₄, 70$\%$ HNO₃, 60$\%$ HClO₄ at a 1:1:1 volume ratio) for 8 h to eliminate surface contaminants. A final air-annealing step at 580 °C for 20 min further optimized NV proximity to the surface, yielding an average depth of $\sim$5.5 nm.

Fluorescence measurements of NV centers within the nano-pillars revealed counting rates of approximately 1.9 Mcps under 532 nm laser saturation (SI Appendix, Fig. S15). We selected the single NV centers with background relaxation times exceeding 500 $\mathrm{\mu }$s as sensors for biomolecular interaction detection to ensure high sensitivity and signal-to-noise ratios.

Metal Spin-Labeled Proteins.

In this work, ubiquitins were used to label Gd³⁺, Mn²⁺, or Cu²⁺. The cysteine mutation sites were introduced into Ub and probe molecules 4PhSO2-3Br-PyMTA were linked to these sites (SI Appendix, Fig. S3). Following the chelation reaction with Gd³⁺, Mn²⁺ or Cu²⁺ solutions, the Ub protein can be labeled and Ub(Gd/Mn/Cu) molecules were obtained. The biotin-Ub(Gd/Mn/Cu) complex was formed with the linkage between biotin and Ub(Gd/Mn/Cu) via a short peptide.

Diamond Surface Functionalization with PEI.

The detailed protocol for diamond surface functionalization with PEI is mainly referred to ref. 54 (as shown in SI Appendix, Fig. S6). 1) Carboxylation of the diamond surface. The diamond surface was carboxylated by immersing it in a tri-acid mixture (H₂SO₄/HNO₃/HClO₄) at 180 ^°C for 3 h. This step was preceded by sequential cleaning in concentrated nitric acid, 1 M NaOH solution and 1 M HCl solution at 100 ^°C for 1 h each, followed by thorough rinsing with deionized water (ddH₂O washing). 2) PEI bonding. The carboxyl groups on the diamond surface were activated with a mixture of 2 mM 1-ethyl-3-(-3-dimethylaminopropyl) (EDC, Thermo Fisher Scientific, 22980) and 5 mM N-hydroxysuccinimide (NHS, Thermo Fisher Scientific, 24510) for 15 min. After activation, the diamond was washed to remove the excess EDC/NHS mixture. The activated diamond was then incubated in 66$\mathrm{\mu }$L of a 15mg/mL PEI solution (molecular weight$\sim$25 kDa, dissolved in PBS buffer, Sigma-Aldrich, 408727) and allowed to react for 30 min at room temperature. 3) PEI cross-linking. To cross-link the PEI layer, 18 $\mathrm{\mu }$L of a 40 mg/mL 4-arm-PEG-NHS solution (molecular weight 20 kDa, dissolved in PBS buffer, Axispharm, AP14802) was added. The mixture was gently shaken for 90 min to ensure complete cross-linking. Finally, the diamond was rinsed with ddh₂O to remove any unreacted reagents.

Protein Immobilization on the Diamond Surface.

The streptavidin (SA, Thermo Fisher Scientific, 434301) was prepared at a concentration of 0.6mg/mL in PBS buffer (pH 7.2), while biotin-Ub(Mn) was dissolved in MES buffer (0.1M, pH $=$ 6.0) at a concentration of 0.6 mM. The PEI-coated diamond was incubated in a freshly prepared mixture containing EDC (20 mg/mL, 5 $\mathrm{\mu }$L), SA (0.6 mg/mL, 5 $\mathrm{\mu }$L), and MES buffer (0.1 M, pH $=$ 6.0, 8$\mathrm{\mu }$L) for 2 h at room temperature. After the reaction, the diamond was rinsed thoroughly with MES buffer. Following the SA immobilization, 5 $\mathrm{\mu }$L of 0.6 mM biotin-Ub(Mn) was applied to the diamond surface and allowed to react for 40 min. The surface was then rinsed repeatedly with MES buffer to remove unbound molecules. The immobilization protocol for bovine serum albumin (BSA) followed the same procedure.

Experimental Setup.

The experiments were performed on a home-built confocal microscopy platform comprising the optical, microwave, and control systems. The optical system utilized 532 nm and 594 nm laser beams for spin state initialization and readout. The 532 nm laser (Changchun New Industries, MGL-III-532-150 mW) passed twice through an acousto-optic modulator (AOM, Gooch&Housego, 3200-121), while the 594 nm laser was modulated by a fiber-coupled AOM (Gooch&Housego, Fiber-Q 633 nm). Both laser beams were combined and delivered through a fiber bundler (Thorlabs, RGB46HF). Then the laser beam was focused on the diamond membrane by an objective [Olympus, PlanApo 60$\times$, numerical aperture (NA), 1.42]. The diamond sample was mounted on a high-precision nanopositioner (PI, P562) for scanning. NV center fluorescence was collected through the same objective, filtered by a 561-nm long-pass dichroic mirror (Semrock, Di03-R594-t1-25x36), and detected by avalanche photodiodes (Excelitas, SPCM-AQRH). As for the microwave system, microwaves were generated by a microwave source (Anapico, APSIN6010), and routed through a switch (Mini-Circuits, ZASWA-2-50DRA+). After amplification (Mini-Circuits, ZHL-16-43-S+), the microwaves were delivered to the NV centers via a customized Omega-shaped microwave antenna to manipulate their spin states. An external magnetic field can be applied using a permanent magnet. The system is controlled by an arbitrary sequence generator (Guoyi Quantum (Hefei) Technology, ASG8000) for synchronizing laser and microwave switching, photon data acquisition, and other experimental sequences. Data acquisition and analysis were automated through a dedicated computer, ensuring reproducible experimental workflows.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

The AFM image, relaxation data related to the measurements of molecular interactions, and simulation data for ensemble NV centers in this study have been deposited in Zenodo database: https://doi.org/10.5281/zenodo.16780745 (74). All other data and materials are included in the article and/or SI Appendix.