WorksPhysics and Astronomy
Multi-day single-cell magnetic imaging of living cancer cells using Quantum Cell Monitoring
Abstract
Live-cell imaging of cellular and tissue samples relies predominantly on optical modalities, such as fluorescence, confocal, and multiphoton microscopy, in which cellular information is encoded in emitted photons. Photons scatter and attenuate in biological samples, restricting depth resolution, while the excitation light required to produce them induces photodamage that perturbs the very biology under observation. Magnetic fields, by contrast, traverse biological samples without attenuation at cellular length scales, but the imaging modality that extracts the encoded information, magnetic resonance imaging, cannot resolve individual cells. Here we present Quantum Cell Monitoring (QCM), a widefield nitrogen-vacancy magnetometry platform that encodes single-cell information in stray magnetic fields rather than emitted photons and reads it out at subcellular resolution. Integrated with a specialized live-cell incubator, QCM tracks magnetically labeled living cancer for up to 72 h through their individual magnetic field signatures. It captures single-cell mitosis and morphological remodeling, as well as cell cluster formation, splitting, and migration even within dense regions where brightfield contrast collapses. By shifting the carrier of single-cell information from propagating optical photons to static magnetic stray fields, QCM decouples cell from cellular optical accessibility and auto-fluorescence, enabling sustained live-cell observation in regimes inaccessible to optical microscopy.
main.pdf
Introduction
Fluorescence microscopy, the prevailing method for live-cell imaging, faces three coupled limitations that together impede multi-day single-cell observation: (i) durability of the label, (ii) strong excitation light directly incident on the cell, and (iii) scattering of the fluorescence emission restricting depth resolution. While photobleaching of small-molecule dyes under extended imaging durations can be avoided by genetically encoded labels such as photostable fluorescent proteins [1, 2], the excitation light needed to drive these labels affects cell behavior well below the threshold of cell damage. As a result, mitotic delay and altered migration speeds are observed at doses routinely incurred during multi-day imaging [3, 4]. Taken together, these limitations preclude diffraction-limited fluorescence imaging at biologically relevant depths, even with multiphoton excitation [5]. These constraints are not independent symptoms but consequences of using photons as the carrier of cellular information. Mitigating one typically increases the others [6], imposing fundamental limitations on the applicability of this technique for extended single-cell observation. Magnetic resonance imaging (MRI), by contrast, replaces photons with magnetic fields as the information carrier, exploiting their ability to penetrate biological tissue without attenuation or interaction with the sample at cellular length scales, but lacks the spatial resolution required to resolve individual cells [7, 8, 9]. As a result, events at the single-cell length scale, such as division, morphological remodeling, and short-range migration inside live tissue remain inaccessible to existing imaging techniques over multi-day timescales. Closing this gap requires both a magnetic label that encodes single-cell information into magnetic stray fields and a sensor that detects those fields at the cellular length scale under live-cell conditions.
Such a label is provided by the selective labeling of cells with superparamagnetic iron oxide nanoparticles (SPIONs) [10], which provide a stable and biocompatible [11, 12] magnetic contrast whose faint stray fields can in principle be detected to identify and track individual cells. These labels can be selectively functionalized for specific cellular targets, based on the desired application through antibody targeting. To date, however, no imaging modality has been demonstrated that resolves these stray fields at the single-cell level under long-term live-cell conditions. Such measurements require a detection method that simultaneously delivers high magnetic field sensitivity, subcellular spatial resolution and full compatibility with live-cell culture conditions. Nitrogen-vacancy (NV) centers in diamond meet these requirements, combining high magnetic field sensitivity with diffraction-limited spatial resolution and an inert, biocompatible sensing substrate [13]. This combination, in principle, unites the magnetic-field information carrier with the single-cell resolution of optical microscopy, closing the carrier-versus-resolution tradeoff that has so far separated the two modalities.
Widefield NV magnetometry has been demonstrated to image magnetically labeled single cells [14], to resolve internalized SPIONs in fixed samples [15], and to map excised tumor tissue [16]. While these prior pioneering approaches successfully resolved static biomarker expression in short proof-of-principle demonstrations with single time point [17] or short-term imaging [18] of living cell samples, sustained longitudinal imaging of living cells has remained fundamentally elusive. The critical bottleneck is that multi-day live-cell diamond magnetometry requires the complex orchestration of non-invasive continuous microwave delivery, high sensitivity spin readout under strictly minimized photodamage, and a specialized incubation environment. Consequently, monitoring single-cell dynamics like proliferation or mitosis through purely magnetic signatures over biologically relevant timescales has never been achieved.
Here, we present Quantum Cell Monitoring (QCM), a self-contained quantum-biological platform built around a custom-engineered live-cell incubation architecture [19]. Unlike conventional commercial chambers, this proprietary hardware is co-optimized for fluidic and optical performance to solve a dual challenge: it suppresses parasitic stray light to ensure efficient light coupling at the sensing interface, while simultaneously maintaining long-term cell viability by providing a stable nutrient supply and strict homeostatic regulation. This innovation breaks the long-standing barrier of thermal and phototoxic cell degradation during continuous diamond magnetometry, transforming widefield NV sensors from a static endpoint tool into an autonomous, longitudinal live-cell imaging modality.
We resolve single-cell migration, proliferation and magnetic-signal dynamics in a monolayer culture over up to 72 h. Together, these results establish QCM as a live-cell imaging technique in which single-cell identity is carried by stray magnetic fields rather than emitted photons, decoupling cell monitoring from the optical accessibility of the sample and resolving single-cell dynamics on living samples over multi-day timescales.
Results
Widefield nitrogen-vacancy center magnetometry for live-cell imaging
We use commercially available 20 nm core size SPIONs with a polymer shell that is functionalized with streptavidin, allowing for antibody-based targeting of the particles to cell surfaces (Figure 1a). We target the epithelial cell adhesion molecule (EpCAM), a surface marker commonly expressed in epithelial cancers, including the HT29 colorectal cancer cell line used in this study. Optimizing the labeling protocol, we achieve saturated binding of SPIONs to the cell surface, resulting in a uniform distribution of magnetic labels across the cell membrane. Previous work [20] has shown, that fully saturated HT29 cells can bind up to 2.5 pg of SPIONs, which corresponds to an approximate (using the mean density of maghemite/magnetite and assuming single-particle linkage) number of SPIONs on the cell surface. The SPION-labeled cells are seeded directly onto the diamond sensor, hosting a thick ensemble layer with a NV center density of 1 ppm (Figure 1b). The cells adhere to the diamond surface, mediated by a Poly-L-Lysine (PLL) coating that improves cell adhesion. An external magnetic bias field magnetizes the SPIONs on the cell surface (Figure 1a,Figure 1b) and subsequently induces a cumulative magnetic dipole field for each individual cell approximately following a sum of magnetic dipoles:

Figure 1. Widefield NV center magnetic field microscope for long-term live-cell imaging. a, Illustration of the magnetic labeling protocol. SPIONs (black) are coated in a polymer shell (green) and functionalized with Streptavidin. The primary antibody (yellow) attaches to the EpCAM surface marker (blue) of the cell membrane (brown). The biotinylated secondary antibody (pink) attaches to the primary antibody and provides a binding site for the streptavidin on the SPION surface, resulting in the attachment of SPIONs to available EpCAM surface markers. Magnetization of the sample by a pair of permanent magnets causes the superparamagnetic particles to generate a magnetic dipole field (black field lines). b, Render of the sample assembly. The magnetically labeled cells (brown with SPIONs illustrated in green) are seeded on the diamond’s NV layer surface (red). A pair of permanent ring magnets (gray) create a magnetizing bias field aligned with a [111] axis of the diamond. The magnetic dipole signature of the cells is detected by the NV centers through widefield ODMR measurements (color map). c, NV Center spin resonance shifts induced by local magnetic dipole fields from a magnetically labeled cell (brightfield image left in black and white contrast). The surface-bound SPIONs cumulatively produce a magnetic field signature that shifts the NV center’s resonance frequencies from the magnetic bias field resonance. This shift (blue: on the cell, orange: off the cell) allows reconstruction of the local absolute magnetic field value affecting the NV centers of the respective pixels. Background subtraction of the magnetic bias gradient by a rolling ball algorithm reveals the magnetic dipole signatures of the individual cells. The selected pixels observe a shift of the resonance lines by ≈134 kHz, resulting in an absolute magnetic field difference between the two pixels of 4.79 µT. After background subtraction, the magnetic field signature of the proximal cell (magnetic field image on the right) reveals a 3.49 µT magnetic field strength between the two pixels. d, Render of the specialized incubation system integrated with the widefield NV microscope. Green, incubation chamber enclosing the cell culture atmosphere; red, the mm diamond sample emitting into the objective below; green line, excitation laser; orange, microwave antenna in proximity to the diamond sample, which is mounted on a mm glass cube for total internal reflection (TIR) excitation of the NV layer.
where is the vector from the -th SPION at to position , is the magnetic field vector at position , is the number of SPIONs on the cell surface, is the vacuum permeability, is the magnetic moment of the -th SPION. The resulting magnetic field encodes information about the position and shape of the cell in the form of shifting and subsequently the magnetic field signature of the cell relative to the NV center layer [20]. This magnetic dipole field of each cell causes local deviations of the much larger magnetic bias field, which in turn locally shifts the NV center’s spin resonance frequencies based on the electron Zeeman effect. The resulting local shift from the magnetic bias field at location is , where kHz/T and is the local magnetic field generated by the cell (Eq. (1)) projected on the interrogated NV center axis [111]. For a cell surface saturated with SPIONs, this results in a magnetic dipole field in the T range, which in turn causes shifts of the NV center’s spin resonance frequencies on the order of a few 100 kHz (Figure 1c). In order to detect these faint signals, laser excitation optically cycles the NV centers continuously, and a dual-tone microwave signal with a hyperfine splitting governed separation of 3.05 MHz is swept in frequency across both the NV center’s spin transitions. This oscillating magnetic field induces resonance-conditioned spin state driving. The convolution of this dual-tone microwave signal with the electron spin transitions results in three distinct resonance features in the ODMR signal. From these, the center feature representing the resonance where both microwave tones are resonant with their respective hyperfine transitions is sampled and evaluated (Figure 1c). This approach improves sensitivity by a factor of 1.4 due to the higher contrast achieved by simultaneously driving both hyperfine transitions of the N-coupled electron spin. Interleaved frames, with and without the microwave signal present, are captured by the camera to compute the spin contrast governed by the spin-state-dependent emission intensity of the NV centers. Evaluating this contrast on a pixel-by-pixel basis with a spatial resolution of per pixel for our binned sensor allows sampling of the resonance features of the spin transitions in this widefield continuous-wave optically detected magnetic resonance (ODMR) measurement. Thus, we capture the resulting electron Zeeman shift caused by both the magnetizing bias field and the magnetic dipole fields locally resolved across the field-of-view (FoV) (Figure 1b,Figure 1c). Our hyperfine resolved ODMR measurement achieves a linewidth for the dual-tone driving of MHz. Extracting the electron Zeeman shifts for each pixel of the captured ODMR data, results in a map of absolute magnetic field values , where is the magnetic bias field along the NV [111] axis. Finally, we subtract the background magnetic field gradient using the rolling ball method [21]. Due to their localized size, the magnetic field signatures of individual cells are preserved by the background subtraction method, resulting in clear signals on the imaged sensor surface [14, 17, 20]. Cells were allowed to adhere for 8 h to the diamond surface before the first ODMR measurement. Due to the small size of the incubation volume, the culture medium was refreshed in 8 h intervals, to maintain nutrient supply. To sustain cell viability over the multi-day experiment while retaining ODMR capability, we constructed a specialized incubation system (Supplementary Information Sec. S1) around the diamond sample inside our widefield microscope (Figure 1d) [19]. The incubator is placed between the bias magnets above the microscope objective and encloses the diamond sample, providing a closed-off atmosphere for the living samples. The system stabilizes temperature, humidity and CO at the diamond surface (Supplementary Information Sec. S1) and provides the optical and microwave access required for ODMR.
By establishing these precise environmental conditions directly on the sensing substrate, this platform serves as an essential enabling technology that transcends the static, fixed cell imaging paradigm of prior diamond magnetometry [17] into a continuous, multi-day live-cell observation window. Consequently, individual cells can now be resolved and tracked in the magnetic field channel, allowing observation of their dynamic behavior over the course of the experiment. The ODMR measurement protocol was optimized to minimize photodamage (Supplementary Information Sec. S2) through evanescent wave coupling and stray light by optimizing the laser power in a TIR configuration to 25 mW over 5 min acquisition within our FoV. At this operating point, our per-pixel magnetic field sensitivity of remains sufficient to resolve the stray fields of individually labeled cells. This operating regime enables repeated ODMR acquisition over multiple days, exploiting the inherent photostability of magnetic labels to track individual cells on timescales inaccessible to fluorescence-based methods [10]. The combination of the specialized incubation system [19] and optimized ODMR protocol thus establishes a platform for long-term live-cell magnetometry, enabling subsequent experiments on monitoring cell dynamics.
Long-term monitoring of cancer cells in monolayer
To demonstrate sustained cell monitoring, we observed SPION-labeled HT29 cells across multiple long-term experiments spanning up to 72 h. Taking advantage of the unique permanence of magnetic labels which, unlike fluorophores do not photobleach, the QCM sampling interval was set to 8 h. This spacing was strategically chosen to match the slow diffusion kinetics of adherent single cells [22] and their 24 h to 72 h doubling time [23], thereby minimizing unnecessary laser induced grid interrogation while fully capturing cellular functions like proliferation and migration.
Concurrently, strictly to establish an independent, high-resolution ground truth for the initial algorithmic verification of this prototype, low-intensity brightfield reference frames were captured at 2 min intervals. As mathematically validated later by our mean-squared displacement analysis (Figure 2a), this high-frequency optical tracking is an overdetermined baseline, confirming that the 8-hour magnetic channel is entirely self-sufficient for identity retention. These optical co-registrations were processed via Cellpose [24] and Bayesian tracking [25] purely to validate the analytical accuracy of our magnetic field maps, proving that QCM maintains absolute identity retention across days without relying on continuous optical intervention (Figure 2b).

Figure 2. Multi-day correlated brightfield and magnetic field imaging of SPION-labeled HT29 cells over a total of 48 h timespan. a, Full FoV brightfield image showing adherent cells with an average diameter of 20 µm on the diamond surface after 8 h seeding. Color overlays indicate segmented cell boundaries with assigned cell identities. b, Time series of correlated brightfield (top) and magnetic field (bottom) images at the ODMR measurement time points spanning the 48 h timeframe. Colored overlays remain consistent. Cell identities are omitted for visual clarity. Individual magnetic dipole signatures of cells are visible, and cell outlines are overlaid based on the brightfield segmentation mask. All scale bars: 20 µm.
Using the same camera for both brightfield imaging (first and third row in Figure 2b) and corresponding ODMR measurements (second and fourth row in Figure 2b), allows direct comparison between the two imaging methods, by assigning cell identities across both modalities and outlining the exact cell position within the magnetic field maps. Performing ODMR measurements and subsequent reconstruction of the magnetic field across the FoV provides a magnetic field map of the camera’s FoV. Magnetically labeled cells appear as characteristic dipole signatures in the magnetic field channel (Figure 2b), arising from the projection of each stray field onto the interrogated NV axis. This unique magnetic footprint allows locating individual cells across the FoV. As fluorescence microscopy and subsequently standard biological viability assays are not possible in our experimental setup, we instead opt for a behavioral observation assessment of cell viability. Over 72 h, cells migrated, proliferated and underwent morphological changes, confirming that the measurement protocol sustains normal cell culture conditions.
These results demonstrate that QCM achieves robust single-cell monitoring within dynamic cellular environments without compromising cell viability. Notably, this represents the first instance in quantum sensing where living biological specimens have been continuously interrogated via diamond magnetometry over multi-day timescales while remaining fully viable. Since the early demonstrations of widefield magnetic imaging on fixed cells [20], achieving long-term sample survival under sustained quantum control sequences has remained a major barrier. By demonstrating that normal proliferation, and migration can be tracked continuously over 72 h, these findings validate QCM not only as a non-invasive monitoring tool for oncology but as a fundamental breakthrough in quantum-biological sensing.
Magnetic imaging of single-cell dynamics
The subcellular resolution of the magnetic field imaging modality allows direct observation of single-cell dynamics over the course of a 48 h experiment. For example cell mitosis is directly observable in the splitting of magnetic dipole signatures of individual cells into two daughter cells (Figure 3a). Specifically, the mother cell #10 separates into two distinct cells clearly visible in the brightfield imaging, as well as the magnetic field map at the 24 h time point. The daughter cells #1023 and #1024 show distinct individual magnetic field signatures, consistent with the division of the cell’s surface area in two during mitosis. The resulting change in magnetic field strength from ≈ 3.9 µT peak amplitude of the mother cell is in line with the reduction of SPION density on the daughter cell surface due to the mitosis, as during this process the mother cell grows and separates into two equal daughter cells. This results in a theoretical increase in surface area of 26 %, causing a subsequent reduction of SPION density of ≈ 21 %, which can be observed in the magnetic field image with the resulting magnetic field peak amplitudes of the daughter cells of 3.1 µT (#1023) and 2.2 µT (#1024) respectively. We attribute the discrepancy to the deformed dipole signature of cell #1024 due to movement during the acquisition time. Cell #1023 shows a well-structured, symmetric dipole that better reflects the absolute labeling strength. This demonstrates clearly how the information of the occurring processes is encoded in the magnetic field information. Furthermore, evidence of either tumor cell cannibalism [26] or entosis [27, 28] is observed, as cells #46 and #135 merge into cell #411 between the 40 h and 48 h time points (Figure 3b). We attribute the faint magnetic field signature of the resulting merged cell to a larger size, as well as a change in the distance of the cell’s center of mass and thus a change in the mean distance of SPIONs on the surface of both involved cells away from the sensor surface. Additionally, due to the coupling between the SPION position and the cell surface, morphological changes of the cell are also observable in the magnetic field channel (Figure 3b, Figure 3c). Cell #333 shows filopodia outgrowth, characteristic for cancer cells, and Cell #186 shows vertical stretching of its cell membrane, which are both visible in the magnetic field image as a change in the shape of the magnetic dipole signature, as the cell’s shape is encoded in the magnetic field information. This washing out of the magnetic dipole field is attributed to the elongation of the cell as well as movement during the 5 min integration window of the ODMR measurement. This subsequently causes partial compensation of the magnetic dipole field, as the positive and negative lobes of the dipole field cancel out over a larger area. Additionally, the movement of the cell during this process washes out the magnetic field signature, as the magnetic field affecting the NV centers is averaged over the integration time of the measurement.

Figure 3. Selected ROIs showing observable single cell behavior instances spanning mitosis, entosis and morphological changes. a, Cell #10 undergoing mitosis resulting in daughter cells #1023 and #1024. The magnetic dipole is clearly split up into two distinct magnetic field signatures. b, Cell #46 and #135 undergo either tumor cell cannibalism or entosis. Observable in the merging of magnetic field signatures to cell #411. c, Cell #186 undergoes morphological changes by stretching along the vertical direction. This is reflected in an elongation and partial compensation of the magnetic dipole signature, as the positive and negative lobes of the dipole field cancel out over a larger area due to a change in shape of the cell surface and thus of the individual positions of the contributing SPIONs (Eq. (1)). All scale bars: .
These demonstrations illustrate the capability of the magnetic field imaging modality to resolve single-cell dynamics based on the change of magnetic field signatures induced by magnetically labeled cells. As the resulting magnetic field of a cell is composed of the cumulative contribution of individual SPIONs situated on the cell surface (Eq. (1)), changes in the cell’s position and shape are reflected in the magnetic field signatures. QCM thus provides a unique method to observe fundamental cellular processes irrespective of optical accessibility or necessary excitation doses incident on the sample, which would otherwise degrade or interfere with the sample viability.
Magnetic imaging of cell population dynamics
In addition to single-cell behavior, a larger perspective is resolvable in the magnetic field imaging in the form of cell-cell interactions and cluster formation. Over the course of three experiments with identical conditions, but different time scales (48 h for Run 1, 66 h for Run 2, and 72 h for Run 3, not including the initial seeding time), the HT29 cells proliferate, migrate and form dense monolayer clusters as they move within the FoV. To evaluate the migration of cells over the course of the experiment, we analyzed the trajectories of individual cells whose tracks exceeded 6 h of continuous observation (Supplementary Sec. S3) in the spider plot (Figure 4a). It is evident that cells show limited migration across the FoV, with most cells showing net-zero migration and only a few outliers within 2-3 cell diameters ( µm). To further evaluate the migratory behavior, we evaluated the mean-squared displacement (MSD) of the tracked cells (Figure 4b), where (Supplementary Sec. S3) is the diffusion coefficient, is the anomalous diffusion exponent and the evaluated time delay between time points. Evaluation of the MSD was limited to a delay due to centroid noise dominating for low caused by the segmentation, tracking errors, and loss of cell tracking due to cluster formation. The upper limit of is in line with the commonly applied criterion for limiting where is the minimum trajectory length [29, 30]. The resulting MSD curve shows a slight sub-diffusive behavior, with values between 0.88 and 0.98 across the experiments. This quantified sub-diffusive exponent () serves as a fundamental physical justification for our discrete sampling approach. Because the structural constraints of the monolayer restrict the maximum cellular displacement over 8 h to a fraction of the average intercellular distance, cell identity can be unambiguously tracked and reconstructed purely from consecutive magnetic dipole maps. This directly demonstrates that the magnetic channel is fully self-sufficient, rendering the high-frequency optical co-registrations an overdetermined baseline used strictly for independent ground truth verification.

Figure 4. Selected ROIs showing observable cell culture dynamics captured in correlated brightfield and magnetic field imaging of a 66 h timespan experiment. a, Spider plot showing the trajectories of cells that could be tracked for >6 h continuously. The trajectories are colored by experiment and illustrate that migration is limited to a few tens of µm over the course of the experiment. b The mean-squared displacement (MSD) of the tracked cells shows a slight sub-diffusive behavior (), as expected for adherent single cells in a monolayer culture [22]. Low brightfield image quality, segmentation, and tracking errors in the dense cell clusters cause significant centroid noise, which dominates for low values (shaded region). The MSD curve is thus evaluated for min, where the MSD curve shows sub-diffusive behavior. c, A group of cells undergoing cluster formation, subsequent splitting and migration out of the FoV. Merging and separation of the cell cluster is clearly visible in magnetic field images. Cell IDs are omitted for visual clarity, as well as faulty segmentations due to the incoherent cluster structure posing a challenge for the segmentation algorithm. d, Complex cell interactions imaged with correlated imaging modalities. After formation of larger cell clusters individual cells remain partly visible in the magnetic field map, in contrast to the brightfield image. All scale bars: 20 µm. e, Cell localization predicted from the magnetic field map, without any input from brightfield data. From left to right, the experimentally captured magnetic field map, the modeled field obtained by sparse dipole deconvolution and peak localization (cells of identical 20 µm size, white circles), and the brightfield image overlaid with the magnetically recovered cell positions (white circles). Because the magnetic stray fields traverse biological matter without attenuation, individual cell identities are recovered from their dipole signatures alone, even within dense clusters where brightfield contrast collapses.
These processes are clearly visible in the magnetic field channel, allowing us to track the dynamics of individual cells as they form, and interact with, clusters of cells. Magnetic field imaging successfully resolves the grouping and splitting of cell clusters, as well as their migration across the FoV during the 66 h experiment duration, based on their magnetic dipole signatures. Figure 4c illustrates this as a group of cells (right side) splits at the 38 h time point, and one of the resulting subgroups subsequently merges into another at the 63 h time point. Additionally, the magnetic field channel allows us to track individual cells within clusters, while the brightfield image shows immediate loss of contrast, as well as merging of cell boundaries at the 38 h and 46 h time points (Figure 4d). Because the cells are all labeled with SPIONs, we still resolve some degree of individuality in the magnetic field channel, while the brightfield image shows a merged cluster of indistinguishable cells. Figure 4c (left) and (d) (center) show the formation of a dense cell cluster, whose details remain unresolvable in the brightfield modality, but the respective magnetic field maps of the captured time points resolve details of individual cells within the cluster.
Building on this, the position and iron mass of individual cells can be recovered directly from the magnetic field map alone, without any optical input (Figure 4e). We demonstrate this by applying a lightweight localization algorithm to the magnetic field map of Run 2 at the 38 h time point, where cells assembled to larger clusters. As a leading order approximation that keeps the reconstruction numerically tractable, each cell is modeled as a single point dipole of magnetic moment magnetized along the interrogated NV axis, which defines a kernel that is invariant under translation (Eq. (1)). We recover the underlying source distribution through a sparse, nonnegative, regularized deconvolution of the field map with the dipole kernel. This collapses each cell’s extended dipole signature into a single localized peak and suppresses sidelobe artifacts, so that the cell count emerges directly from the deconvolution. A fixed minimum separation between cells extracts the candidate centers, and a joint nonnegative least squares fit of the superposed dipole fields eliminates residual false detections while reading out the peak field amplitude of each cell. For this first order approximation, holding the cell size fixed at its nominal value, this amplitude is converted into a per cell iron mass through the known specific magnetization of the SPIONs [20], so that the procedure returns an estimation of both position and iron mass across the ROI. We overlay the recovered positions onto the brightfield image as circles of the chosen fixed size of 20 µm (Figure 4e), and direct comparison with the brightfield image shows that the magnetic localization recovers the positions of individual cells within the cluster. This proof of concept demonstrates that the magnetic field map alone contains sufficient information to recover individual cell identities, even within dense clusters where brightfield contrast collapses. This is a key advantage of magnetic-field imaging over brightfield: cell identity is preserved through cluster formation, a capability usually associated with fluorescence labels.
This performance highlight uncovers the core advantage of QCM over state-of-the-art optical modalities: when cellular boundaries completely collapse in the brightfield channel due to local tissue density rendering classical segmentation algorithms blind, the magnetic channel successfully maintains individual cell identities. This sustained multi-day observation is directly enabled by our specialized custom incubation system: while its optimized optics isolate the excitation plane to minimize background noise, its integrated micro-environmental control ensures a continuous, optimal supply of nutrients and metabolic stability directly on the diamond substrate. Consequently, QCM transcends the physical limits of photon scattering without compromising long-term cell kinetics, proving its capability to operate as an independent imaging modality where optical microscopy fundamentally fails. Unlike fluorescence, however, magnetic-field imaging is unaffected by photobleaching or scattering, enabling sustained monitoring over multi-day timescales regardless of microenvironmental conditions of the sample [5, 6].
Discussion
We demonstrate that single-cell information from living cells can be encoded in stray magnetic fields and read out at subcellular resolution over multi-day timescales. SPION based magnetic labeling combined with widefield NV magnetometry inside a specialized incubator allows QCM to resolve proliferation, mitosis, entosis, morphological remodeling, and cluster dynamics in HT29 colorectal cancer cells over 72 h, including inside dense clusters of cells where brightfield contrast collapses. This capability addresses a long-standing gap left by existing live cell imaging modalities. Optical methods are bound by their information carrier and trade photon budget against duration and depth resolution, meaning that multi-day fluorescence imaging inevitably induces phototoxic cell stress and suffers from optical scattering [5, 6]. QCM fundamentally circumvents these constraints by replacing optical photons with static magnetic stray fields as the primary carrier of cellular information.
While widefield NV center magnetometry has previously been introduced for single-cell evaluation, prior state-of-the-art approaches remained strictly limited to static, single time endpoint measurements of fixed samples [14–18]. This historical limitation stems from an unresolved biophysical dilemma in requiring continuous, high-power microwave and laser delivery, which inherently introduces severe phototoxic and thermal stress, rendering multi-day live observation impossible [3, 4]. Our platform overcomes this technological bottleneck by introducing a self-contained architecture. By combining a specialized incubation system directly with a carefully designed microscope architecture, we decouple sustained spin readout from environmental and phototoxic degradation of the sample. This architectural synergy transforms diamond magnetometry into a fully functional, longitudinal live cell imaging modality capable of operating under true physiological conditions.
Consequently, the physical parameters mapped by QCM provide a robust framework for monitoring long-term cell kinetics, specifically cell migration, without the need for optical accessibility. Adherent single cells within dense biological assemblies typically exhibit complex, sub-diffusive migratory trajectories that are severely obscured or entirely lost in brightfield imaging once clusters form [31]. Our tracking platform successfully addresses this challenge by preserving absolute cell identity through continuous cluster formation, fusion, and mitotic splitting based purely on shifting magnetic dipole signatures. Crucially, these multi-hour reconstructions serve as a quantitative validation of this spatial framework.
Because static magnetic stray fields traverse biological matter without loss at cellular length scales, QCM bypasses both optical scattering and attenuation within magnetically translucent environments, such as soft tissue [7, 8, 9]. At current SPION labeling strengths, individual cell signatures remain resolvable several tens of µm above the sensor surface based on extrapolation and an SNR of 1 (Supplementary Sec. S4). Advancements in the magnetic moment and size tuning of SPION labels offer the potential to at least double the saturation magnetization of these nanoparticles [32]. Additional tuning of SPIONs to form clusters further amplifies the per-cell signal strength, due to more iron per EpCAM binding site, which directly scales the depth at which single cells remain trackable via QCM, extending the depth range to and beyond 100 µm (Supplementary Sec. S4) [20].
It is important to emphasize that the 8 h sampling interval for magnetic field image capture used in this study does not represent a fundamental technical lower limit of the QCM hardware. Rather, it was strategically chosen as an operational optimum tailored to the slow baseline kinetics of HT29 cells while strictly minimizing cumulative laser exposure, which makes QCM a well-suited choice for imaging slow cell dynamics. For future applications in highly dynamic environments or dense 3D tissue models where optical co-registration is entirely unavailable, the magnetic sampling frequency can be scaled up significantly. This scalability is governed by the underlying sensor physics, where the optical NV-readout is shot-noise limited and the signal-to-noise ratio scales as , with being the local cell magnetic field strength and the ODMR integration time. Implementing high-moment SPION clusters addresses this in two distinct regimes. For dynamic cellular events occurring within the first dozens of micrometers away from the sensor, the amplified magnetic field drastically reduces the required integration window. Since the integration time scales quadratically with the field strength for a constant target SNR (), a conservative 10-to-20-fold increase in local magnetization reduces the necessary integration time from the current 5 min down to less than a second. Due to this exponential drop in laser exposure per frame, the actual measurement interval between individual scans can be compressed from 8 h down to minutes while maintaining a lower cumulative phototoxic dose, enabling high-frequency, optics-free monitoring. Second, for cells situated deeper within dense 3D architectures ( µm) where migration is naturally slowed down by microenvironmental constraints, the enhanced magnetic moment ensures that individual signatures remain resolvable above the sensor’s noise floor (SNR ), sustaining robust longitudinal monitoring over multi-day timescales even under the sparse sampling regime.
Beyond extending the imaging range and improving the imaging rate, the reconstruction itself offers room for refinement. Conceivable improvements include a more extensive biologically-informed forward model that represents cells as extended sources to capture their 3D shape and expresses the magnetic field through a spherical harmonic (multipole) expansion rather than point dipoles, more advanced sparse deconvolution schemes for this underdetermined inverse problem, and machine learning approaches that directly learn the mapping from magnetic field maps to cell position, shape, and iron mass information [16, 20].
Building on this capability, the presented monitoring approach establishes a versatile technological baseline, demonstrating that single-cell kinetic pathways can be continuously resolved over dozens of hours under physiological conditions forming the foundation for adaptation into complex 3D tissue architectures, which is especially relevant with rising interest in 3D culture models such as organoids and tumor spheroids [33–35] (Supplementary Sec. S5). This technological baseline directly opens up avenues for advanced oncology research that previously had to rely on retrospective inferences from fixed histology or employ imaging modalities that ultimately trade phototoxic dose, label photostability and penetration depth against observation time. Where previous experimental research relies on multiphoton or confocal imaging to investigate the process of cancer cell invasion, QCM would resolve the individual cells invading and migrating deep inside dense stroma tissue over multi-day timescales, capturing the invasion dynamics that current methods only infer from snapshots [36, 37]. Especially for imaging of cancer cell dynamics in 3D environments, where confocal microscopy is insufficient to resolve the full depth of the sample, QCM would allow continuous monitoring of the sample in contrast to light sheet microscopy, which requires imaging of independent samples for each time point due to terminal mounting procedures [38]. As single cell models do not capture the full complexity of the tumor microenvironment, experiments using spheroids or organoids are increasingly used to study cancer cell invasion. Recent studies investigate the development of such micro-tumors using light sheet microscopy, an imaging modality that necessitates imaging of independent systems for subsequent time points [39]. Furthermore, current research in cell invasion based on spheroids reduces the observation to a 2D equatorial plane of the organoid relying on local-maximum detection instead of cell segmentation, which is insufficient to capture the full 3D dynamics of the system [40]. QCM would allow continuous monitoring of the same organoid over multiple days, resolving individual labeled cells deep within the sample, further improving the quality of fundamental biological research.
Ultimately, these milestones bridge a long-standing gap at the intersection of quantum physics and the life sciences. For biomedical imaging, QCM introduces a paradigm where subcellular spatial resolution and multi-day continuous observation no longer come at the cost of phototoxic cell stress, remaining fully functional even within optically dense cellular environments. For quantum sensing, this work demonstrates that non-invasive magnetic stray fields can be successfully translated into quantitative biophysical models of geometry and monitoring kinetics. QCM firmly establishes magnetic field-encoded imaging as a powerful, self-contained modality for the future of translational cancer research.
Methods
Widefield NV-center magnetic field microscope
The widefield magnetic field microscope is built around a [100] bulk diamond with a thick, 1 ppm NV-doped surface layer with isotopic enrichment [41]. The diamond is mounted on a polished glass cube using optical adhesive (NOA61, Norland) to enable TIR excitation of the NV layer, preventing the 532 nm laser (Verdi G Series, Coherent) from interacting with the biological sample [19]. The laser was focused to illuminate the FoV without clipping on the sample edges. The power was then tuned to 25 mW over the 376 μm diameter FoV, which was found to be a good compromise between magnetic field sensitivity and phototoxicity over the 5 min integration duration of a single ODMR measurement. Two samarium cobalt ring magnets provide a static bias field of up to 40 mT along the selected NV axis to magnetize SPIONs present in the sample [20]. NV fluorescence is collected by a long working distance objective (MY50X-805, Mitutoyo), filtered by a 635 nm long-pass filter (EdgeBasic) and imaged onto a binned electron multiplying charge-coupled device camera sensor with a pixel sensor (Evolve13, Teledyne). A dual-tone microwave signal, generated by IQ mixing a dual-tone intermediate frequency from an arbitrary wave-form generator (OPX+, Quantum Machines) with a swept local oscillator (SynthNV Pro, Windfreak), drives both hyperfine transitions of the NV ground state [42]. The center frequency of the microwave signal is swept over 4 MHz intervals across the and transitions. The camera was subsequently exposed for 200 ms with and without the microwave signal to compute the spin contrast. The resulting ODMR contrast is evaluated per pixel to extract the electron Zeeman splitting and subsequently the local magnetic field per pixel, yielding spatially resolved magnetic field map over a field of view. Background subtraction using a rolling-ball algorithm removes low-frequency spatial noise, isolating the dipole signatures of individual SPION-labeled cells [17].
Incubation system for live-cell imaging
To maintain cell viability during extended imaging sessions, a specialized incubation platform compatible with the NV widefield microscope geometry was employed [19]. The incubation chamber consists of an acrylic cylinder bonded to the glass cube with optical adhesive (NOA61, Norland), sealed to prevent leakage of culture medium. Temperature control is provided by regulated aluminum heating elements (MP825-20.0-1%, Caddock) with negative temperature coefficient thermistor feedback, calibrated to maintain an incubation temperature of 37 °C at the diamond surface. The incubator lid is heated 2 °C to 3 °C above the incubation temperature to suppress condensation. A premixed gas atmosphere (5 % , 95 % air; 2GF Mixer, Okolabs) is sensed via a sensor and subsequently humidified (CO2-500ml, Bioscience Tools) before entering the 3D printed chamber (BioMed White Resin, Formlabs). An internal water reservoir and overflow system maintain stable humidity to prevent medium evaporation and pH drift. exchange with the culture medium is facilitated through grooves on the incubation volume lid, while a sealed cultivation compartment minimizes evaporative loss. All hardware components are positioned outside the incubation atmosphere to avoid contamination and interference with the ODMR measurement. Further details of the incubator design are provided in [19]. Brightfield imaging of the cells is performed by a light emitting diode illuminating the sample from above and collecting the transmitted light through the imaging objective on the camera.
Cell segmentation and tracking
Brightfield time-series images were segmented using Cellpose [24], a deep-learning-based generalist algorithm for cellular segmentation. The cell diameter was fixed to 70 px, flow threshold and cell probability threshold were set to 0.6 and respectively based on empirical testing. The resulting instance masks were linked across frames using btrack [25], which employs a Bayesian multi-object tracking algorithm to reconstruct cell trajectories and detect proliferation events. A branching of 50, prospected segmentation miss rate of 10 %, search radius of 50 px, optimization tolerance of 0.5 % was used for the tracking. Together, the pipeline provides per-cell position, lineage, and morphological information from the brightfield channel, which is spatially correlated with the corresponding magnetic field maps. Due to the limited contrast and resolution of the brightfield images acquired through the NV microscope objective, segmentation accuracy is insufficient for quantitative morphological analysis. Cell outlines are therefore displayed as visual guides to facilitate spatial correlation between the brightfield and magnetic field channels.
Magnetic field cell localization
To recover cell positions from a magnetic field map after background subtraction, without optical input, each cell is approximated as a point magnetic dipole oriented along the interrogated NV [111] axis. The NV projected stray field of a unit dipole ( Eq. (1)) defines a single kernel that is invariant under translation, so the map is modeled as with , a sparse map of dipole amplitudes per pixel. We estimate by minimizing subject to with a fast iterative shrinkage thresholding algorithm (FISTA) [43]. The penalty enforces sparsity, collapsing each cell’s bipolar signature into a single peak and suppressing the sidelobe artifacts that a plain matched filter () produces. Cell centers are extracted from by nonmaximum suppression with a fixed minimum separation and an edge margin, and a single joint nonnegative least squares fit of the superposed dipole fields removes residual spurious candidates and yields the peak field amplitude of each cell. Holding the cell size fixed at its nominal value, this amplitude is mapped to a per cell iron mass through the known specific magnetization of the SPIONs, calibrated by the peak field of a single dipole that carries a reference iron mass at the nominal cell depth [20]. An optional subpixel refinement of the position of the brightest sources further sharpens the localization. The detection noise floor is estimated from the median absolute deviation of the discrete Laplacian of the map. On synthetic ground truth data the recovered positions reach an accuracy of root mean square.
Cell culture and SPION labeling
Prior to cell seeding, the polished diamond surface was functionalized with PLL to promote cell adhesion [17]. The incubation volume above the diamond was filled with 100 L of PLL solution and incubated for 1 h at 37 °C. The PLL solution was subsequently aspirated, and the surface was allowed to dry under sterile conditions before cell seeding.
The mycoplasma-free human colorectal cancer cell line HT29 was cultivated at 37 °C and 5 % in RPMI-1640 medium (Gibco Life Technologies, #P04-18500) supplemented with 10 % fetal bovine serum (PAN Biotech, #P30-3031), penicillin (100 L/mL) and streptomycin (100 L/mL) (PAN Biotech, #P06-07100). Cells were used at 70 % to 80 % confluence, between passages 3 and 13.
For labeling, cells were detached from the plate by trypsinization. cells were resuspended in 1 mL of cultivation medium. All subsequent incubation steps were performed at 37 °C on a shaker at 300 rpm. Between incubation steps, the cell solution was centrifuged (300 for 5 min) to remove excess antibody or nanoparticles and washed with Dulbecco’s phosphate buffered saline (DPBS, PAN Biotech, #P04-36500). Cells were first incubated with anti-EpCAM antibody (1:250 dilution in cultivation medium; Cell Signaling Technology, #2929) for 45 min, followed by a biotinylated secondary goat-anti-mouse antibody IgG H&L (1:500 dilution in cultivation medium; Abcam, #ab6788) and streptavidin-coated superparamagnetic iron oxide nanoparticles (1:2 dilution in DPBS; Ocean NanoTech, #SHS-20) for 30 min each.
Monolayer imaging protocol
To image cell migration and proliferation in a monolayer culture, HT29 cells were seeded at a concentration of 40 000 cells per culture medium directly onto the PLL-coated diamond surface. After allowing h for cell adhesion, the first ODMR measurement was acquired. Subsequent ODMR measurements were performed at 8 h intervals, while brightfield images were acquired every 2 min to 5 min to track cell migration and proliferation dynamics. The culture medium was refreshed with prewarmed medium after each ODMR acquisition to maintain nutrient supply and optimal culture conditions.
Data availability
The presented datasets generated during the current study are available from the corresponding authors upon reasonable request.
Code availability
The custom code used for data acquisition and analysis in this study is available from the corresponding authors upon reasonable request.
Supplementary information
Acknowledgements
We thank Lukas Heinen and the team of the Section of Experimental Oncology and Nanomedicine (SEON) at the University Hospital Erlangen for fruitful discussions and support on the physical and biochemical aspects of superparamagnetic iron-oxide nanoparticles. We are grateful to Daniela Thalheim and Jessica Knittel of the Chair of Experimental Tumor Pathology for their support with cell and tissue culture, labeling, and sample preparation. We further thank Prof. Dr. Jörg Wrachtrup, Prof. Dr. Ronald Walsworth, Dr. Durga B. R. Dasari, Dr. Amit Finkler and Dr. Yaron Antebi for fruitful discussions and editorial input.
Funding
A.P. and R.N. disclose support for the research of this work from the European Union under the Key Digital Technologies Joint Undertaking (KDT JU), now Chips Joint Undertaking (Chips JU), in the projects ARCHIMEDES (grant nos. 16MEE0329 and 101112295) and MOSAIC (grant nos. 16MEE0494 and 101081238), and from the German Federal Ministry of Education and Research (BMBF, now BMFTR) in the projects INNOBAT (grant no. 03XP0492D) and QuaLiProM (grant no. 03XP0573C). Regine Schneider-Stock was supported by the Deutsche Forschungsgemeinschaft (DFG) SCHN477/19-1 and BMBF (QUBIS).
Author contributions
A.P. designed and built the experimental setup, performed measurements, data analysis and prepared the figures. A.P. and R.N. conceived the study. S.B. and D.T. prepared cell cultures and SPION labeling protocols. R.S.-S. supervised the biological experiments and contributed to data interpretation. A.P. and R.N. wrote the manuscript with input from all authors.
Competing interests
A.P., R.S.-S., R.N. are inventors on a patent application related to the optical excitation and collection scheme described in this work (application number EP 24198554.8 and EP 10 2025 109 562.0).
supp_inf.pdf
NV magnetometry compatible incubation platform
Long-term cultivation directly on the diamond sensor surface requires an incubation environment compatible with the inverted total internal reflection (TIR) illumination geometry, microwave access and permanent magnets capable of reaching , which precludes the use of available commercial stage-top or whole-microscope incubators [1] (Figure S1). The incubation chamber is constructed from an acrylic cylinder bonded to the polished glass cube via optical adhesive (NOA61, Norland) [2], defining a small cultivation volume of directly above the diamond surface. The cylinder is enclosed by a 3D-printed atmospheric chamber, manufactured from a biocompatible resin (BioMed White Resin, Formlabs) on an SLA printer (Form 4, Formlabs). The chamber is printed as a single body in which the gas-inlet and condensate-overflow hose connectors are integrated directly into the resin part, minimizing the amount of metal introduced near the sensing region and the magnetizing bias field and yielding a drop-in replaceable component that requires no post-processing. Its internal volume is partitioned into an outer reservoir, which holds the water that sustains the humidified atmosphere and acts as a thermal buffer, and an inner incubation volume that seals over the open-top of the acrylic cylinder (Figure S1). The full assembly fits between the samarium-cobalt bias magnets and around the microwave antenna without compromising ODMR sensitivity, and all active hardware, namely the heaters, thermistors, and gas and humidity supply, is kept outside the incubation atmosphere to avoid contamination of the culture and interference with the ODMR measurement [1].

Figure S1. Integrated incubation system on the NV widefield microscope. a, Render of the incubation assembly, showing the atmospheric chamber (green) and lid (gray) enclosing the incubation volume between the two samarium-cobalt bias magnets (purple). The heated sample mount stages the glass cube required for total internal reflection (TIR), on which the diamond sample is mounted. The acrylic culture cylinder is bonded to the polished glass cube, same as the diamond sample and microwave antenna. The excitation laser is incident from below and reflects symmetrically out of the assembly. The optical access from below allows collection of NV emission onto the camera optics. The gas inlet, condensate overflow and water reservoir are integrated in the 3D printed atmospheric chamber. b, Cross-section of the incubation chamber enclosing the cultivation volume within the acrylic cylinder (gray). The water reservoir provides thermal and humid buffering, while the gas inflow supplies a constant stream of water and the premixed incubation gas, resulting in a stable incubation volume (turquoise shading). Two pairs of resistive heaters heat the lid and the sample stage base to maintain the incubation temperature at . The excitation laser is cleanly reflected at the diamond–medium interface, while the NV emission is collected through the optical path from below.
A premixed 5% / 95% air atmosphere is supplied to the chamber, with the fraction set by metering the mass flow into a gas mixer (2GF Mixer, Okolabs) against the ambient air delivered by an air pump. The gas mixture is humidified in a temperature-controlled humidifier (CO2-500ml, Bioscience Tools) before entering the chamber, where the internal water reservoir and a condensate overflow maintain saturated humidity to suppress evaporation of the culture medium. A diffusion lid with peripheral grooves separates the cultivation volume from the surrounding humidified atmosphere, permitting exchange while suppressing medium evaporation and the associated pH drift. The concentration is monitored by a non-dispersive infrared (NDIR) sensor (ExplorIR-20W) placed upstream of the humidifier, since elevated humidity degrades the sensor, and is logged at 2 Hz over a USB-TTL serial interface alongside all thermistor readings. Slow drift of the concentration, arising from variations in ambient pressure, temperature, and supply-gas pressure, is corrected by periodically readjusting the flow at intervals no longer than the 8 h spacing between ODMR acquisitions, at which point the culture medium is also replenished [1].
Temperature is regulated by two heated aluminum elements driven in independent parallel PID loops by a microcontroller (Arduino Nano) using thick-film resistors (MP825-20.0-1%, Caddock) and NTC thermistors as temperature sensors. The base heater is the primary element: it mounts the glued glass-cube, diamond, and microwave-antenna stack together with the incubation chamber via a nylon set screw, provides optical access from below for fluorescence collection, and leaves lateral gaps for the TIR excitation, heating the incubation volume from the substrate side for a faster and more direct thermal response than heating the lid alone. The heated lid carries an identical pair of resistive heaters and thermistor and is closed by a transparent cover that permits brightfield illumination and visual monitoring of the culture-medium level. It is operated 2 °C to 3 °C above the incubation temperature to suppress condensation on the optical surfaces. To avoid placing a sensor inside the cultivation volume, where it would perturb the ODMR measurement, the gradient between the external base-heater sensor and the diamond surface was calibrated against an in-medium reference sensor, yielding a stable offset of 3 °C. The heater set point of 34 °C therefore maintains the diamond surface at 37 °C while bounding the worst-case temperature excursion below physiological limits in the event of thermistor beta-value drift. The temperature is monitored at four points, namely the base heater, the lid, the incubation atmosphere, and the ambient laboratory environment, and a control measurement confirmed that the applied microwave power produces no measurable heating of either deionized water or culture medium, so the microwave drive does not perturb the incubation temperature [1]. Surface functionalization of the diamond with Poly-L-Lysine (PLL) ensures rapid cell adhesion prior to the start of multi-day ODMR acquisition.
Starting from room temperature, the Arduino-based PID-controlled heaters reach their set points within approximately 30 min and subsequently hold the surface at the target temperature, with residual fluctuations of = 43 mK at the base and 56 mK at the lid, i.e. below 100 mK and within the control range of the loop. Over a 72 h period the CO₂ concentration varies by only ≈ 0.12%, confirming that the system maintains a stable physiological environment across multi-day experiments (Figure S2). To verify that this environmental control, rather than the medium or surface functionalization, is responsible for the observed cell viability, a stress test was performed in which HT29 cells were cultured for 48 h with the temperature and CO₂ regulation disabled [1]. Under these conditions the cells showed no proliferation, adhesion, or migration and exhibited membrane blebbing characteristic of early apoptosis, demonstrating that the controlled incubation environment is essential for sustaining the cultures imaged in this work.

Figure S2. Long-term temperature and CO₂ stability of the incubation system. Base- and lid-heater temperatures and CO₂ concentration recorded over a 72 h incubation period of Run 3 from the main manuscript. The temperature traces show a steady state for the majority of the experiment ( = 43 mK base, 56 mK lid) and the ≈ 0.12% CO₂ drift. The temperature traces show semi-periodic drops in temperature, which are attributed to the periodic refreshment of the culture medium, requiring opening the incubator lid.
Cell damage due to high laser flux
Repeated ODMR acquisition over multi-day experiments imposes a cumulative optical dose on the biological sample that, if not carefully constrained, leads to severe phototoxic damage. To characterize the operating regime within which cell viability is preserved, we first monitored a HT29 monolayer over 48 h under high-flux ODMR acquisition every 30 min, imaging the full FoV at magnification (Figure S3).

Figure S3. Cell damage due to high laser flux. Timeseries of a 48 h experiment with periodic ODMR measurements every 30 min. The full FoV of is imaged with a magnification. The green shaded area illustrates the area of the Gaussian excitation beam that is totally reflected by the diamond surface. Significant cell damage due to the excitation laser is observable within the area of the beam spot, while cells at the edges of the field of view show healthy morphology and proliferation behavior.
During the initial 8 h settling period, cells adhered uniformly to the PLL-functionalized diamond surface across the full FoV. Beyond h, cells within the totally reflected Gaussian beam spot (green overlay) exhibited progressive loss of morphological integrity and accumulation of cellular debris. In contrast, cells at the periphery, where the local laser flux is significantly lower, retained a healthy state and continued to proliferate and migrate. By h, no viable cells remained within the central beam region. The spatial coincidence between the damaged region and the beam footprint immediately implies excessive optical flux as the dominant damage source.
Three mechanisms could in principle account for this damage: laser-induced heating of the diamond surface, microwave-induced heating of the culture medium, or direct phototoxicity from the excitation laser. Laser-induced heating of the diamond is excluded by the absence of any measurable temperature rise at the surface under comparable conditions [1]. The diamond’s high thermal conductivity would in any case dissipate heat quickly across the sample, rather than produce the beam-localized damage observed here. Microwave-induced heating of the medium is excluded by the antenna geometry, the wire surrounds the field of view, so medium heating would be strongest at the FoV periphery, yet cells closer to the wire remain adherent and viable while damage occurs in the FoV center, far from the antenna. Only phototoxicity, which acts locally where the laser intensity is highest, matches the observed spatial distribution [3, 4].
Although the total internal reflection (TIR) geometry confines the 532 nm excitation beam within the diamond and is intended to protect cells from direct exposure [2], a residual evanescent field penetrates into the culture medium with an exponential decay characterized by a sub-wavelength penetration depth [5]. Simulation of evanescent-wave coupling into water shows that, within the first 100 nm of the surface, cells are exposed to a significant fraction of the excitation laser intensity (Figure S4). This near-surface exposure, compounded by scattering from residual surface imperfections, accounts for the localized damage observed along the beam path. Reflection of the laser away from the sample volume, even for a perfectly flat interface, is therefore insufficient to prevent phototoxic effects. The effect of stray scattered light from the sample holder causes additional optical dose incident on the cells.

Figure S4. Simulated evanescent-field amplitude at the diamond–water interface. Simulation of the evanescent wave coupling into water for a 532 nm excitation wavelength from diamond (n=2.426) into water (n=1.334) for different incidence angles . The evanescent wave penetrating from the total internal reflection barrier into the culture medium or cell surface is governed by an exponential decay with distance from the interface [5]. However, this still corresponds to a significant exposure to the excitation laser for the cells in close proximity to the diamond surface.
To mitigate photodamage, we increased the interval between ODMR measurements to 8 h to allow for cellular recovery. Even under these conditions, cumulative photodamage is not fully eliminated over multi-day acquisition. Over a 40 h observation period, the majority of cells along the excitation beam path progressively detach from the diamond surface, while cells lateral to the beam remain adherent and proliferate (Figure S5). This residual cell loss persists despite an 8 h recovery interval between ODMR measurements, demonstrating that the recovery aids in cellular survival, but is not enough to ensure long-term viability under the current acquisition parameters.

Figure S5. Phototoxicity-induced cell detachment during multi-day ODMR acquisition. a, Correlated brightfield (top) and magnetic field (bottom) images of SPION-labeled HT29 cells at 8 h (left) and 24 h (right) showing characteristic magnetic dipole signatures of individual cells. Colored overlays in brightfield and white outlines in magnetic field images indicate cell boundaries segmented with Cellpose [6]. White arrows mark cells that persist across both time points. b, Full FoV brightfield image at 40 h. Black square encloses the ODMR evaluation region; cell tracking failed due to the detachment of the majority of cells from photodamage; black arrow and green shaded area illustrate the direction of the excitation laser beam path; white boxes enclose regions of cell adhesion lateral to the beam and cell detachment along the beam path. Inset: three-dimensional rendering of the total internal reflection (TIR) excitation geometry, illustrating the laser beam entering from below and reflecting at the diamond–medium interface. The light green overlay indicates the Gaussian beam spot at the TIR plane.
The operating point was not selected arbitrarily or through a purely empirical sweep, but fixed by a sensitivity requirement set by the expected signal. Saturated SPION labeling produces a per-cell stray-field signature of to , consistent with both the labeling density and the dipole amplitudes observed in our magnetic field maps. The acquisition time and laser power must therefore yield a per-pixel sensitivity whose noise floor lies well below this signal, so that individual cells are resolved at a practical signal-to-noise ratio (SNR). For our 5 min acquisition, in which the dual-tone microwave is stepped across 102 frequency points each imaged with and without microwave driving (204 frames of 100 ms exposure) and repeated over 14 successive sweeps, the total signal-integration time is . With a per-pixel sensitivity of , this gives a theoretical noise floor of . This noise floor lies roughly an order of magnitude below the expected cell signal, corresponding to an SNR of to 14, and thus leaves sufficient leeway to reliably resolve the magnetic field signatures of individual cells. This balance defines the operating point: lowering the laser power would further benefit cell viability but raise the noise floor and degrade imaging quality, whereas raising it would improve the SNR only marginally, since the cells are already detected reliably, at the cost of increased phototoxic load. We therefore operate at 25 mW over the 5 min acquisition within our FoV, the minimum optical dose that still guarantees robust single-cell detection. The laser intensity, measurement duration, and recovery interval are thus not free parameters but jointly constrained by the interplay of magnetic field sensitivity and cell viability.
Cell migration tracking
Evaluating the migration behavior of the tracked cells is crucial for understanding the dynamics of the cell population and the underlying biological processes. By analyzing the trajectories of the tracked cells, we can extract information about their migration patterns, such as their speed, directionality, and persistence. The histogram of observation durations for the tracked cells in the long-term tracking dataset shows a steep decline in the number of tracks with increasing duration, which is why we clipped the trajectories above 6 h continuous observation (Figure S6). This can be attributed to cell clustering, segmentation errors, and tracking issues during the experiment. The average trajectory length of the three experiment runs are 19.2 h (Run 1), 12.8 h (Run 2), and 13.2 h (Run 3) respectively. The histogram of average cell speeds for the tracked cells in the long-term tracking dataset shows a mean around for all three runs (Figure S6). The mean-square displacement (MSD) analysis of the cell trajectories shows a sub-diffusive behavior, which is characteristic of cells migrating in a crowded environment. The resulting MSD fits return and for Run 1, and for Run 2, and and for Run 3 respectively. The distribution of the cell speed should be interpreted with caution, as highly mobile cells are more likely to not be part of a larger cluster and thus are more likely to be tracked, while less mobile cells are more likely to be part of a larger cluster and thus are more difficult to track.

Figure S6. Observation duration and speed distribution of tracked HT29 cells. a, Histogram of observation durations for cells in the long-term tracking dataset. Evaluated cell trajectories are clipped above 6 h to filter out short, non-persistent tracks in order to provide a clearer picture about cell migration during the experiment. The average trajectory lengths of the three experiment runs are 19.2 h (Run 1), 12.8 h (Run 2), and 13.2 h (Run 3) respectively. The steep decline in the number of tracks with increasing duration can be attributed to tracking errors and segmentation faults, that make segmenting and tracking of individual adjacent cells difficult. b, Histogram of average cell speeds for the tracked cells in the long-term tracking dataset. The average cell speeds across the three experiment runs are (Run 1), (Run 2), and (Run 3) respectively. The distribution of cell speeds shows a mean around for all three runs.
SNR-based depth limit of single-cell monitoring
To determine the standoff over which individual SPION-labeled cells remain resolvable in Quantum Cell Monitoring (QCM), we performed a forward simulation of the cellular magnetic stray field and compared its peak amplitude with the per-pixel noise floor of the ODMR readout, following the framework introduced in our previous SPION-labeling study [7]. A cell is approximated as a sphere of diameter whose surface carries the SPION load, in line with the homogeneous distribution of SPIONs on the cell surface observed by fluorescence microscopy [7]. The cell-surface labeling is approximated by points uniformly distributed on the sphere, with each point acting as a perfect magnetic dipole carrying the average moment of the SPIONs assigned to it. For a given iron mass per cell , the number of bound particles follows from the mass-to-moment relation of the streptavidin-coated Ocean NanoTech SHS-20 SPIONs (single-particle saturated moment at 40 mT [7]), and the cluster moment scales linearly with the number of particles, because the bias field used here drives the particles deep into the Langevin saturation region ([7]). All dipole moments are aligned along the external bias field, which lies along the chosen NV axis at the magic angle of to the diamond surface normal. The cumulative magnetic field in the sensing plane at standoff below the cell is computed as the sum of the individual dipole projections onto this NV axis, , on a grid covering the cell footprint [7].
The peak amplitude of the cellular signature is then read off the simulated field map as . At the magic angle, the dipole moments project onto the NV axis as the characteristic positive/negative double-lobe pattern. The maximum of this pattern approximately follows the amplitude returned by a 2D Gaussian-derivative fit, the analytic form commonly used to extract per-cell stray-field amplitudes from experimental -maps [8]. The peak-pick is therefore the simulativ[e] equivalent of the experimental fit and avoids introducing a parametric model on a noiseless map. A monotonic decay curve is generated at a reference iron mass and linearly rescaled by to cover the full sweep, since the cell stray field at fixed diameter and distance is proportional to the total magnetic moment of the surface-bound SPIONs [7]. For each combination of iron mass and integration time the maximum detectable distance is then defined by the crossing , located on the rescaled curve by linear interpolation between adjacent standoffs.
Iron loading was swept across to 10 , spanning the labeling regime reported in our SPION optimization study (saturating around 2.7 pg per cell at 500 SPION concentration [7]), and integration times spanned min to 10 min, covering both the 5 min dwell used here and longer settings accessible with reduced phototoxic load. Two per-pixel sensitivities were considered: , representative of the present widefield setup under live-imaging laser power, and , the sensitivity attainable on the same hardware when the phototoxic dose on the cells is resolved, e.g., by a reflective coating that fully shields the culture from stray light (Figure S7). The resulting curves are monotonic in and in , so that improvements in either SPION moment or integration time translate directly into increased trackable depth [9].

Figure S7. Simulated detection horizon for individual SPION-labeled cells. Maximum standoff at which the peak axial stray field of a single labeled HT29 cell crosses the per-pixel noise floor , as a function of iron mass per cell. Colors denote the ODMR integration time and line styles denote the per-pixel sensitivity, for the present live-cell widefield setup (solid) and for an ideal shielded configuration (dashed). The cell is modeled as a 20 sphere with uniformly distributed SPIONs on its surface; is extracted as the maximum of the simulated axial field map across a sensing plane at standoff below the cell.
Spheroid labeling for cell population dynamics
Beyond the single-cell level, QCM can be applied to track the dynamics of cell populations. A promising approach is the labeling of spheroids, which are three-dimensional aggregates of cells that represent tumor relevant microenvironments. By magnetically labeling the seed cell population of the spheroid or alternatively by continuously re-labeling the outer cells during growth, a strong magnetic signal can be achieved due to the high cell density of the spheroid. Imaging the spheroid under live-sustaining conditions allows for monitoring its growth and migration over time. Within the scope of this work, we have successfully labeled and imaged HT29 spheroids (Figure S8), but have not yet performed a detailed analysis of the spheroid dynamics. However, the preliminary results demonstrate that the approach shows clear magnetic field signatures that open up the possibility to track the growth and migration of the spheroid over time within living tissue.

Figure S8. Brightfield and magnetic field images of magnetically labeled HT29 spheroids. a, Correlated images of a HT29 spheroid. The seed cell population of the spheroid was magnetically labeled with the same protocol as described in the main manuscript. The brightfield image shows discrete edges and clear structure of the cell spheroid. The magnetic field image reveals the underlying individual cells that remained magnetically labeled, even after extended proliferation periods. b, Correlated images of a HT29 spheroid with continuous re-labeling of the outer cells during growth. A significant jump in labeling strength of the outer cells is observable in the magnetic field image. The cumulative magnetic field strength of the spheroid cells closest to the diamond layer diminish the ODMR signal quality, due to a strong electron Zeeman shift outside of the sampled frequency range. Stray individual cells and cell residue is clearly visible outside of the spheroid, which can be attributed to dead cells or cell debris, that still carry some SPIONs.
References
- [1]A. Pointner, D. Thalheim, S. Belasi, L. Heinen, L. R. Carnell, C. Janko, R. Tietze, C. Alexiou, R. Schneider-Stock, and R. Nagy. A compact incubation platform for long-term cultivation of biological samples for nitrogen-vacancy center widefield microscopy. Advanced Materials Technologies, page e70949, 2026.arxiv.org/abs/2512.14482
- [2]D. Le Sage, K. Arai, D. R. Glenn, S. J. DeVience, L. M. Pham, L. Rahn-Lee, M. D. Lukin, A. Yacoby, A. Komeili, and R. L. Walsworth. Optical magnetic imaging of living cells. Nature, 496(7446):486–489, 2013.DOI
- [3]Noriko Matsumoto, Kozo Yoshikawa, Mitsuo Shimada, Nobuhiro Kurita, Hirohiko Sato, Takashi Iwata, Jun Higashijima, Motoya Chikakiyo, Masaaki Nishi, Hideya Kashihara, Chie Takasu, Shohei Eto, Akira Takahashi, Masatake Akutagawa, and Takahiro Emoto. Effect of light irradiation by light emitting diode on colon cancer cells. Anticancer Research, 34(9):4709–4716, 2014.pubmed.ncbi.nlm.nih.gov/25202048
- [4]Sina Wäldchen, Julian Lehmann, Teresa Klein, Sebastian van de Linde, and Markus Sauer. Light-induced cell damage in live-cell super-resolution microscopy. Scientific Reports, 5:15348, 2015.DOI
- [5]D. Axelrod. Cell-substrate contacts illuminated by total internal reflection fluorescence. J. Cell Biol., 89(1):141–145, 1981.pubmed.ncbi.nlm.nih.gov/7014571
- [6]C. Stringer, T. Wang, M. Michaelos, and M. Pachitariu. Cellpose: a generalist algorithm for cellular segmentation. Nature Methods, 18:100–106, 2021.DOI
- [7]A. Pointner, D. Thalheim, S. Belasi, L. Heinen, C. Bonato, T. Luehmann, J. Meijer, R. Tietze, C. Alexiou, R. Schneider-Stock, and R. Nagy. Optimizing SPION labeling for single-cell magnetic microscopy. The Journal of Physical Chemistry Letters, 16(30):7584–7590, 2025.DOI
- [8]D. R. Glenn, K. Lee, H. Park, R. Weissleder, A. Yacoby, M. D. Lukin, H. Lee, R. L. Walsworth, and C. B. Connolly. Single-cell magnetic imaging using a quantum diamond microscope. Nature Methods, 12(8):736–738, 2015.DOI
- [9]Lukas Heinen, Pascal-Raphael Blersch, Constantin Schnell, Karsten Meyer, Roland Nagy, Marcus Halik, Christina Janko, Stefan Lyer, Christoph Alexiou, and Rainer Tietze. Ultrasound-assisted size tuning of polyacrylic acid coated magnetic nanoparticle clusters for biomedical applications. Ultrasonics Sonochemistry, 130:107876, 2026.DOI