Introduction

Improvements to current magnetic resonance imaging (MRI) techniques will allow chemical analysis of samples with high resolution down to sub-cellular volumes. However, conventional magnetic imaging is hindered by inductive detection requiring a sufficient magnetic flux from the sample to induce a detectable current in the pickup coil. Increases in spatial resolution reduce the number, n, of detected spins until the statistical polarization, whose magnitude is of order 1, exceeds the magnetization due to Boltzmann polarization. The resultant vanishing mean field strength and its stochastic nature render common detection schemes ineffective. Precision magnetometers provide exceptional sensitivities reaching a few fT/√Hz2,3,4,5, yet their relatively large size and/or operational conditions prohibit sensing of small numbers of spin magnetic moments at ambient conditions and sub-micron resolution. Sensors based on nitrogen-vacancy (NV) centres in diamond are promising magnetometers because of their atomic size. This allows placement of the sensor with few nanometre proximity to the sample while retaining superior volume-to-sensitivity scaling, room temperature operation and optical readout6,7,8. Precision metrology has already been demonstrated for DC and AC fields9,10,11,12,13,14, spins within the diamond lattice15,16,17,18 and surface spins19. Here, we demonstrate sensing and imaging of stochastic magnetic fluctuations originating from freely diffusing electron spins such as paramagnetic oxygen (O2, S=1), MnCl2 (S=5/2) and Gadolinium ions (Gd3+, S=7/2) in liquids, immobilized in polymers and linked specifically to cellular structures.

Results

NV relaxometry

In low external magnetic fields, freely diffusing ions exhibit a zero-mean field, but magnetic field fluctuations due to statistical spin polarization cause a non-zero RMS field with random phase. Such stochastic fields are difficult to detect, but the NV offers a striking avenue to measuring such random fields by monitoring its quantum relaxation20,21. To benefit from ensemble sensing sensitivity22,23, we employ an array of atomic sized NV sensors (~1,000 μm2) with a calculated mean depth of h=6.7 nm (ref. 24), as illustrated in Fig. 1. After preparation of a distinct NV spin state, interaction with the environment leads to NV relaxation with rate Γ=Γintenv. The contributions intrinsic to the diamond, Γint, include spin impurities and vibrational lattice dynamics, whereas Γenv depends on the external environment to which the system can deliberately be exposed. Relaxation may occur via the transverse and longitudinal relaxation channels characterized by their respective decay times T2 and T1. The transverse dephasing rate Γ2=1/T2 is increased by fluctuations at low frequencies (kHz to MHz) and can be monitored using spin echoes or higher order dynamical decoupling microwave control protocols25,26,27. The longitudinal relaxation rate Γ1=1/T1 describes the population decay of a polarized spin into thermal equilibrium. It is susceptible to frequencies at the NV Larmor precession ω0=D±γB0, where γ is the gyromagnetic ratio and B0 is a magnetic offset field. Owing to the large NV zero-field splitting of D=2.87 GHz this is in the GHz range at low fields.

Figure 1: Widefield magnetometry with microfluidic detection.
figure 1

(a,b) Widefield excitation (green) of the NV spin array (red arrows) and spatially resolved CCD detection of NV fluorescence. Homogeneous spin manipulation of the NV array is achieved via a lithographic Ω-microwave structure (yellow). Gd3+ ions (blue spheres in a,b in aqueous solution are introduced using a microfluidic channel placed directly on top of the sensor proximal to the NV array with h=6.7 nm. (c) Energy level scheme of NV centre illustrating the high fluorescent signal extending from 637–800 nm. The |0› spin sublevel has a stronger fluorescence than the |±1› states as they emit less photons owing to a higher probability to enter the long-living singlet states. NV relaxation is thus readout optically. (d) Brightfield image of the microfluidic channel with channel boundaries (dotted lines) and 30 × 30 μm2 detection region of interest (ROI, red rectangle). Scale bar, 20 μm.

Spin noise sensing

First, we demonstrate magnetic sensing of various chemical environments by probing both relaxation channels along one preferred NV axis (B0=5 mT) in the presence of air, water and a solution of 1 M Gd3+ (see Table 1). Particularly the large magnetic moment of Gd3+ has made the Gd3+ chelate a prime candidate as relaxation contrast agent in MRI28. In this system, probing Γ2 by a Hahn echo sequence shows only marginal changes in T2 for Gd3+, while multi-pulse CPMG81 is mildly responsive with a change in T2 of 13.5%. The longitudinal T1, however, exhibits a prominent reduction of 94% in the presence of Gd3+ and was similarly responsive to dissolved MnCl2 and with O2 saturated water (Fig. 2a and Supplementary Fig. S1). Focusing on the dominant Gd3+, the relaxation rate ΓGd induced by freely diffusing Gd3+ depends on the corresponding RMS magnetic field and its spectral density SGd(ω). For a Gaussian process, the latter is given by , where ω0 is the Larmor frequency of Gd3+. Boltzmann polarization and ω0 can be neglected for the low B0 fields applied. SGd(ω) is instead dominated by statistical polarization and substantial broadening effects of zero-mean fluctuations fGd=fdipole+fvib+ftrans+frot. The contribution from the concentration-dependent dipole coupling between the Gd3+ is given by fdipole=cGd·77 GHz M−1 (Supplementary Note 1), where cGd is the spin concentration in mol l−1. Intrinsic vibrational spin relaxation of the complexed Gd3+ ion yields a constant fluctuation of fvib~50 GHz29, while rotational motion frot and translational diffusion ftrans cause fluctuations of ~140 MHz (Supplementary Fig. S2). The resulting broadening of the spectral density SGd(ω) is therefore effectively constant up to a few tens of GHz (Fig. 2b), where resonant Gd3+-induced fluctuations Bx and By (NV axis defines z) cause Γ1 relaxation of the NV. The magnetic field variance is derived from the dipolar NV-Gd3+ coupling and is given by

Table 1 Transverse (T2) and longitudinal (T1) NV spin relaxation in various environments.
Figure 2: Microfluidic spin sensing.
figure 2

(a) T1 relaxation curves of the NV ensemble in the presence of water (black), oxygenated water (~1.5 mM, blue), 1 M MnCl2 (green) and 1 M Gd3+ (red). Solid curves are bi-exponential fits to the data. Optimized spin detection via direct fluorescent readout is achieved at a discrete interrogation period τ (orange). The inset shows the measurement pulse sequence and normalization. (b) Spectral density SGd(ω) of Gd3+ for two distinct concentrations (red curves) illustrating the broadening effect due to Gd3+ coupling (fdipole) at higher concentrations. Although the sensitivity windows of T2-decoherence (F2, blue) is limited to low MHz fluctuations, T1-relaxometry allows to probe a wide frequency range up to GHz with two sensitivity windows (F1 and F1+, black). can be Zeeman-shifted via B0 enabling experimental detection of SGd(ω). (c) Experimental relaxation rate Γ1,Gd as a function of the Gd3+ concentration (black squares with 1 σ s.e. and three independent measurements) and analytical predictions for specific mean implantation depths h (red curves). (d) Dynamic microfluidic single-τ detection (τ=100 μs) of varying Gd3+ concentrations with a single data point acquisition time of tm~2 s.

where NA is the Avogadro constant, μ0 the vacuum permeability, h the mean depth of the NV centres and γNVγGd (Supplementary Fig. S4). The effective RMS field strength is then given by . The time-dependent probability of finding the NV in the |0› state (Supplementary Note 2) is given by

where are the individual decay rates of each sensitivity window (Fig. 2b) integrated over SGd(ω). The overall decay rate Γ1,Gd of the longitudinal NV magnetization is then given by

This theoretical prediction was experimentally verified by varying cGd (Fig. 2c), showing excellent agreement for the non-trivial dependence on cGd. At nanometre length scales, the sample-sensor distance is a key parameter as and thus represents a viable tool to access the average implantation depth of nitrogen ions (Fig. 2c). The relaxometric SNR=ΓGdint of T1 is much higher as Γ2,int»Γ1,int due to the low spin-orbit coupling accompanied by a low phonon density in diamond30. Hence, in this system longitudinal T1 outperforms T2-relaxometry by two orders of magnitude (Supplementary Fig. S3) and is employed from here on.

Microfluidic single-τ detection

Next, we optimized the microfluidic detection by converting the T1 signal directly into a measureable fluorescence. Instead of detecting the full relaxation curve, the fluorescent NV response at a single-τ point is utilized as direct concentration readout (Fig. 2a). Maximum sensitivity is achieved at τ~T1,Gd/2 for cGd<10 mM and approaches T1,Gd for higher cGd (Supplementary Note 4 and Supplementary Figs S6–S8). Therefore, varying Gd3+ concentrations can be monitored dynamically with temporal resolutions in the order of a second (Fig. 2d). For maximal sensitivity all four crystallographic NV orientations can be employed, as exhibits a random magnetic field orientation. This is achieved by measuring in near-zero field (50 μT earth field) and probing all NV orientations at D=2.87 GHz, which effectively improves the fluorescent single-τ contrast and increases the number of sensing NV spins by a factor of four (Supplementary Fig. S9). We applied a statistical t-test (2 σ) to the fluorescence traces to identify the lowest concentration still significantly different to pure water. After a total measurement time of tm=20 s and an optimized τ=400 μs, we resolved =250 μM and 80 μM for single and all four NV axes, respectively. In addition, the multiplexed charge-coupled device (CCD) detection allows varying the detection voxel equivalent to an effective change in spatial resolution of magnetic sensing. For fixed tm and CCD detection, depends on the shot noise of detected photons and thus on the applied pixel binning (Supplementary Fig. S3). Although single pixel analysis is feasible, the lower limit of the spatial resolution for the widefield detection is Δrxy~430 nm (Supplementary Fig. S10). In our experimental setup this translates into 4 × 4 pixels (equivalent to Δrxy~460 nm). For such a small voxel, we detected =500 μM, which approaches typical ion concentrations encountered in oxidative bursts31 or senile placques32. As demonstrated previously, lower spin concentrations can conveniently be sensed by spatially averaging over larger voxels. To determine the sensitivity in terms of number of spins per detection voxel we emphasize that Γ1,Gd integrates the signal from the spins above the surface while proximal ions (r=h) contribute most due to the h−3 scaling. We determined the height dependence by employing polymer spacers between NV and Gd3+, which yielded a decay of Γ1,Gd(r) to 1/e at Δrz=15 nm (Supplementary Fig. S11). Taking the experimental sensitivity to =500 μM detected in tm=20 s and a spatial resolution of Δrxy=460 nm (4 × 4 pixels), this corresponds to an experimental detection of 1,000 statistically polarized spins of which only 32 ions contribute to an effective net magnetization. This spin sensitivity is an improvement by more than four orders of magnitude compared with other state-of-the-art magnetic sensing techniques operating at ambient conditions33,34,35. Single spin detection has been demonstrated using magnetic resonance force microscopy3,8, but at the expense of cryogenic temperatures and significantly longer acquisition times. The combination of spatio-temporal resolution, spin sensitivity and the operability under ambient conditions using widefield NV relaxometry paves the way towards minimally invasive real-time observation of chemical and biological processes involving magnetic spins on the sub-cellular level.

Relaxometric imaging

Finally, this unprecedented spatial-temporal resolution is used to image magnetic fluctuations originating from samples on the sensor array (Fig. 3a). A periodic grid of lithographically patterned Gd3+ is easily resolvable in the reconstructed T1 image (Fig. 3b). However, tm can be reduced by orders of magnitude applying the optimized single-τ detection, yielding yet higher contrast of magnetic imaging (Fig. 3c). For high-resolution magnetic imaging in biological samples, we have specifically labelled the plasma membrane of HeLa cells with caged Gd3+ ions and an Alexa532 fluorophore via biotinylated poly-L-lysine. As the NV sensor is most sensitive to proximal spins, we placed 150 nm thin ultramicrotome sections of labelled cells onto the diamond sensor. The control fluorophore indicates a successful label of the magnetic marker to HeLa cells (Fig. 3d), while the boundary of the cell is clearly present in the magnetic image (Fig. 3e). Furthermore, a line scan through the magnetically labelled plasma membrane verifies a spatial resolution of Δrxy=472 nm approaching the inherent optical limit of the conventional widefield technique. A complete simulation of the single-τ detection based on shot noise limited photon detection agrees with the experimental sensitivities (Fig. 4, Supplementary Fig. S8). The key advantage of magnetic spin labels is the potential for chemically selective spin contrast imaging as each spin label is expected to have a distinct S(ω), for example, by binding to a certain biological complex. Using the NV, the spectral density can be measured with a high bandwidth as Γ1 provides a narrow and via B0 tunable sensitivity window (F1 in Fig. 2b, experimental realization in Supplementary Note 3 and Supplementary Fig. S5).

Figure 3: Spin contrast imaging.
figure 3

(a) Schematic of magnetic spin imaging. (b) T1 weighted image of lithographically patterned Gd3+ grid (blue rectangular regions with low T1) on top of diamond sensor. Three data sets, each containing the full T1-decay information with varying τ, were acquired within tm=45 min, subsequently averaged and fitted to obtain T1 for each pixel. (c) Single-τ imaging (τ=150 μs) directly yields dark areas where Gd3+ is present due to the increased NV relaxation. Although a single image (tm=2 s) is sufficient to identify the pattern, the image shown was averaged for 10 min to enhance the contrast. (d) Fluorescent control image of an ultramicrotome sectioned HeLa cell (150 nm), where the plasma membrane was labelled with biotin-poly-L-lysine-Gd3+-DTPA-Alexa532 (Alexa532 fluorescence spectrally filtered from 550–575 nm). (e) Magnetic imaging via a single-τ measurement (tm=15 min, τ=440 μs, B0~50 μT) evidencing the presence of magnetic Gd3+ at the cell membrane (dark structures). (f) Line scan through plasma membrane shown in (e) demonstrating spatial resolution of 472 nm. Errors bars, 1 σ s.e. of six independent line scans. Scale bars, 5 μm.

Figure 4: Sensitivities of optimized single-τ detection.
figure 4

The minimum number of detectable spins nGd was simulated for tm=20 s as a function of sample-sensor distance h and spatial resolution Δrxy. Experimental results with h=6.7 nm and two distinct pixel binnings (black dots, WF: widefield detection) demonstrate good agreement. Fewer spins per voxel can be detected by decreasing h and/or the detection voxel equivalent to an increase in spatial resolution. Note that T1,int remains to be experimentally determined for shallow NV depths with h<5 nm and is assumed to be constant for the simulation.

Discussion

The results presented here demonstrate highly sensitive minimally invasive spin sensing and imaging of unperturbed electron spins at room temperature. We have shown the NV-based relaxometric technique offers several advantages over other techniques capable of sensing statistical polarization3,33,35,36 as it is operable under ambient conditions with no requirements of strong magnetic fields or radiofrequency pulses. Quantum relaxation of the NV sensor has the potential to emerge as a novel technique of high-throughput analytical sciences and contrast-enhanced optical-MRI at the nanoscale. A considerable improvement in sensitivity is possible by decreasing the NV depth h and the voxel size Δrxy equivalent to enhancing the spatial resolution (Fig. 4 and Supplementary Fig. S8). With h=2.5 nm and typical resolutions of Δrxy=50 nm for structured widefield illumination37, a sensitivity in the order of ten Gd3+ spins is expected. At the expense of longer integration, scanning techniques such as stimulated emission depletion would boost Δrxy down to 8 nm38 reaching even single spin sensitivities. Alternatively to our ensemble sensor, local spin densities could be monitored even inside living cells by employing single NVs embedded in nanodiamonds39,40. The high temporal resolution of widefield magnetometry favors sub-cellular visualization also of label-free dynamic processes, for instance the production of free radicals in cell death, the regulation of homoeostasis through ion channels41 or haemoglobin trafficking by imaging paramagnetic oxygen.

Methods

Experimental setup

15N2 at a fluence of 1013 cm−2 with an energy of 4 keV per atom was homogeneously implanted into a 80-μm thin ultrapure type-IIa diamond (Element6) yielding a density of sensing NV spins of ~1,000 μm2. According to a SRIM simulation (stopping and range of ions in matter)24, the mean depth of the NV sensor is h=6.7±2.8 nm. A 500 MHz PulseBlaster card (ESR-Pro-II, Spincore) was employed for timing the triggering of microwave pulses, CCD integration and the laser excitation. Laser (1.5 W of a 532- nm cw; Verdi, Coherent) was directed through an AOM (Crystal Technology) and focused onto the back-focal plane of a 60 × oil objective, 1.49 NA (Olympus). An 800 ns laser pulse allows optimal NV readout with subsequent re-polarization into the |0› ground state enabling CCD integration over multiple repetitions (N~104 per image). The fluorescent NV response was spectrally filtered (LP650, Omega) and projected onto a 512 × 512 cooled EM-CCD (CascadeII, Roper Scientific) yielding an effective pixel size of 115 nm. Homogeneous and resonant microwave manipulation of the NV sensor with an optical field of view of ~60 × 60 μm2 was achieved using a lithographically grown broadband Ω-shaped structure. The flow rate in the microfluidic channel made of transparent polydimethylsiloxan (PDMS, Sylgard) was adjusted to 1 μl s−1. All experiments were conducted with commercially available Gd3+ solution (Gadovist, Bayer Schering Pharma), except for the spin contrast imaging of cellular structures (see cell preparation and synthesis of magnetic/fluorescent spin label in detailed method section). Alexa532 was imaged using a separate fluorescence filter (Semrock, HC 565/24).

Cell culture, labelling and preparation

In all, 2.5 × 106 HeLa ATCC cells were harvested and suspended in PBS buffer at a concentration of 25 × 106 cells per ml. Ten microlitre containing 20 mM solution of EZ-Link NHS Biotin (Pierce, 20217) in dimethylsulphoxide were added for 20 min to the cells for biotinylation. The cells were then fixed with 4% paraformaldehyde, extensively washed and resuspended in PBS buffer. After 5 min incubation with 100 μg streptavidin, cells were washed in PBS buffer, resuspended and incubated for 5 min with ~1 nmol of biotin-poly-L-lysine Gd-DTPA (see synthesis below). After extensive washing with PBS Puffer, cells were resuspended in 1 ml of 1.25 M sucrose solution and harvested by centrifugation. The cell pellets were mounted on an aluminium stub and vitrified by submersion in liquid nitrogen. Then, thin cell sections with a thickness of ~150 nm were produced with an Ultramicrotome (Leica UCT) equipped with a LKB Cryokit and freshly prepared glass knives (Leica KMR2). Frozen cell cuts were transferred onto the diamond under ambient conditions and embedded in PDMS to minimize changes in the refractive index.

Synthesis of Biotin-poly-L-lysine-Gd3+-DTPA-Alexa532

Twenty-five microgram poly-L-lysine (Sigma-Aldrich, P2636) was dissolved in 0.1 M sodium hydrogencarbonate buffer (pH8.3). Then, 600 μmol diethylenetriaminepentaacetic-dianhydride (Sigma-Aldrich, 284025) was added slowly while stirring and regulating to pH8. The solution was stirred for 2 h at 4 °C and was extensively dialyzed against 0.1 M sodium hydrogencarbonate using a ZelluTrans dialysis membrane (Carl Roth, E671.1). The obtained poly-L-lysine-DTPA was biotinylated by addition of 3 μmol EZ-Link NHS-Biotin. After 2 h of stirring at 4 °C the solution was dialyzed against 1 M citrate buffer (pH6.5). Six hundred micromol gadolinium chloride hexahydrate (Sigma-Aldrich, G7532) was dissolved in 0.1 M sodium acetate (pH6) and slowly added to the biotin-poly-L-lysine-DTPA solution. After 20 h of stirring at 4 °C, the solution was first dialyzed against 1 M citrate buffer (pH6.5) to remove free Gd3+ ions within the solution followed by a dialysis against 0.1 M sodium hydrogencarbonat buffer. Then, 1 mg Alexa532 carboxylicacid-succinimidylester (Invitrogen, A-2010) was dissolved in 200 μl of dimethylsulphoxide and slowly added to the biotin-poly-L-lysine-Gd3+-DTPA solution. After stirring at 4 °C for 30 min, the free dye was removed while simultaneously concentrating the labelled probe using a Centriprep Ultracel (Millipore, YM.10).

T1-decay fitting

Recorded T1 curves were fit with a bi-exponential function , as the decay exhibited two components with T1,short <<T1,long and Along>Ashort. We attribute the shorter component (T1,short <15 μs) to a few clustered and thus very proximal NV pairs subjected to cross-relaxation. The relaxation rates Γ1 and the corresponding relaxation times T1 given in the text are determined by the main decay of the second and significantly longer T1 component.

Additional information

How to cite this article: Steinert, S. et al. Magnetic spin imaging under ambient conditions with sub-cellular resolution. Nat. Commun. 4:1607 doi: 10.1038/ncomms2588 (2013).