Paramagnetic fluorinated nanoemulsions for sensitive cellular fluorine-19 magnetic resonance imaging

Abstract

Fluorine-19 magnetic resonance imaging (¹⁹F MRI) probes enable quantitative in vivo detection of cell therapies and inflammatory cells. Here, we describe the formulation of perfluorocarbon-based nanoemulsions with improved sensitivity for cellular MRI. Reduction of the ¹⁹F spin-lattice relaxation time (T₁) enables rapid imaging and an improved signal-to-noise ratio, thereby improving cell detection sensitivity. We synthesized metal-binding β-diketones conjugated to linear perfluoropolyether (PFPE), formulated these fluorinated ligands as aqueous nanoemulsions, and then metalated them with various transition and lanthanide ions in the fluorous phase. Iron(III) tris-β-diketonate ('FETRIS') nanoemulsions with PFPE have low cytotoxicity (<20%) and superior MRI properties. Moreover, the ¹⁹F T₁ can readily be reduced by an order of magnitude and tuned by stoichiometric modulation of the iron concentration. The resulting ¹⁹F MRI detection sensitivity is enhanced by 3-to-5 fold over previously used tracers at 11.7 T, and is predicted to increase by at least 8-fold at clinical field strength of 3 T.

Magnetic resonance imaging (MRI) is becoming a clinical tool for visualizing specific cell populations in the body¹. MRI cell detection using exogenous agents can be used to visualize the in vivo trafficking and behaviour of immune or stem cells used to treat a host of diseases. Fluorine-19 (¹⁹F) ‘tracer’ agents are an emerging approach to intracellularly label cells of interest, either ex vivo or in situ, to enable cell detection via ¹⁹F MRI^{1, 2}. The ¹⁹F label yields positive-signal ‘hot-spot’ images, with no background signal due to negligible fluorine concentration in tissues. Images can be quantified to measure apparent cell numbers at sites of accumulation^{2, 3}, thereby enabling ‘in vivo cytometry’⁴. Tracer agent compositions have mostly focused on nontoxic perfluorocarbons (PFC). Clinical translation of ¹⁹F cell detection has recently been realized in patients⁵ using PFC nanoemulsion to label a dendritic cell cancer vaccine; in these experiments, the cell detection limit was conservatively estimated to be of order 10⁵ cells per voxel⁵.

Improving the sensitivity of ¹⁹F cell detection could lower the barriers for using this technology in a wider range of biomedical applications. One approach for boosting sensitivity is by decreasing the intrinsically-high ¹⁹F spin-lattice relaxation time (T₁) of PFC molecules^6–8. The T₁ ultimately limits the rate of ¹⁹F MRI data acquisitions. Often, ¹⁹F images require summation of multiple acquisitions (i.e., signal averaging) to generate a sufficient signal-to-noise ratio (SNR) for confident interpretation. High ¹⁹F T₁ values require a long repetition time (TR) to allow for longitudinal signal recovery, thus limiting the number of signal acquisitions attainable during a fixed total imaging time (t_i). As t_i is constrained when scanning patients, the key parameter to maximize is SNR/t_i. Shortening T₁ can increase SNR/t_i, sensitivity, and decrease the minimum number of detectable cells per voxel. In practice, reducing T₁ by molecular design can also lead to a reduction in the spin-spin relaxation time (T₂) and line broadening of the resonance; this effect may degrade the SNR if T₂ becomes comparable to the data acquisition sampling time along the frequency encoding direction⁹. The creation of stable and cytocompatible ¹⁹F agents with ‘ultra-fast’ T₁ is an open challenge that can greatly impact the MRI field, enabling accelerated MRI acquisitions and the detection of sparser cell populations in vivo.

The relaxation times T₁ and T₂ can be profoundly altered by high-spin paramagnetic metal ions (e.g., Mn²⁺, Fe³⁺, Gd³⁺). Prior studies⁶ have attached Gd³⁺ to the outer surface of the PFC nanoemulsion droplet resulting in modest reductions in T₁. With increasing distance (r), the steep fall-off (~r⁻⁶) of paramagnetic relaxation rate enhancement from paramagnetic centers limit the efficacy of relaxation agents bound to the surface of PFC nanoparticles.^{8, 10} Thus, effective relaxation enhancement necessitates introduction of metal ions into the fluorous phase, i.e., within the nanoemulsion droplets, to achieve a short T₁ using a minimum amount of a paramagnetic additive.

We describe the scalable synthesis and properties of a family of paramagnetic PFC nanoemulsions with excellent ¹⁹F MRI and biological properties. We show that fluorinated materials incorporating suitable ligands can tightly bind and retain sufficient amounts of metal ions in the fluorous phase of the nanoemulsion to yield ¹⁹F agents with greatly enhanced sensitivity. These new nanoemulsion materials contain metal-binding β-diketones conjugated to linear perfluoropolyether (PFPE). Using these agents, we describe preliminary assessments of the biocompatibility, cell labeling stability, and in vivo MRI studies in mice. Sensitivity enhancement of these materials will potentially accelerate the use of ¹⁹F cell detection in a host of clinical cell therapy trials and for diagnostic inflammation imaging.

Modeling of paramagnetic relaxation enhancement

In the initial design of ¹⁹F probes, we conducted magnetic resonance relaxation time modeling of the impact of dissolving metals ions into PFC. Solomon–Bloembergen–Morgan (SBM) theory^{11, 12} describes paramagnetic relaxation enhancement (PRE) of R₁=1/T₁ and R₂=1/T₂ of surrounding media at a given magnetic field strength, molecular mobility, and metal concentration (See Supporting Information). Using SBM theory, we found optimal parameters for enhancement of R₁ while minimizing linewidth broadening, i.e., R₂. The modeling results (Fig. S1) show that Fe³⁺ uniformly dispersed in PFC will provide the most robust enhancement of ¹⁹F R₁. Mn²⁺ and Gd³⁺ are likely to cause severe line broadening due to a large increase in R₂, especially at high magnetic field strengths. This line broadening originates from very slow electronic relaxation in Mn²⁺ and Gd³⁺ (Fig. S1).

Design and preparation of metal-binding perfluorocarbons

Design of a cytocompatible fluorous-soluble metal chelate requires careful consideration. The steep fall-off of PRE with increasing distance (~r⁻⁶) necessitates solubilisation of individual metal ions, as opposed to incorporating metal-bearing oligomeric clusters or nanoparticles. The metal must not efflux from the fluorous phase during cell labelling and after in vivo administration. The high electronegativity of fluorine imparts very low cohesive energy density¹³ and Lewis basicity¹⁴ to heavily fluorinated compounds, making them extremely poor solvents and ligands. The choice of ligands compatible with fluorous phase is therefore limited to the most hydrophobic scaffolds, with as few intermolecular interactions as possible. To maximize solubility in the fluorous phase, the resulting metal complex should be uncharged and coordinatively saturated. These criteria can be satisfied by using bidentate, monoionic ligands (L) that form high-spin, charge-neutral tris-complexes with trivalent metals (FeL₃, GdL₃) and bis-complexes with divalent metals (MnL₂). Of these, only FeL₃ are coordinatively saturated, due to the small size of the parent Fe³⁺ ion. Coordinatively unsaturated complexes of larger Mn²⁺ and Gd³⁺ tend to be unstable with respect to the formation of oligomeric¹⁵, charged, or ternary complexes¹⁶ (e.g., [GdL₃]_n, [GdL₄]⁻, [GdL₃•(H₂O)_x]). Although gadolinium chelates are widely used contrast agents in clinical ¹H MRI because Gd³⁺ has the highest magnetic moment, we predict that Fe³⁺ is better suited for ¹⁹F applications.

Initially, we tested the results of the PRE modeling [see Supplementary Information (SI) Fig. S1] using small molecules. Fluorinated β-diketone H-fod (Fig. 1) was chosen as starting point. Addition of 2.8 mM H-fod to the aqueous phase of a premade PFPE nanoemulsion displays apparent dissolution of the diketone and appearance of heptafluoropropyl groups in ¹⁹F NMR spectra featuring three broad singlets. Addition of 0.7 mM FeCl₃ led to the slow formation of orange-colored Fe(fod)₃ and a commensurate increase in R₁ from 2.3 to 27.0 s⁻¹ and in R₂ from 4.0 to 85.6 s⁻¹ (at 9.4 T) of the major PFPE resonance (−91.4 ppm) by 24 hours (Fig. S3). With GdCl₃, lower R₁ (12.8 s⁻¹) and higher R₂ (285 s⁻¹) values were obtained. The corresponding gadolinium complex displayed modest R₁ values accompanied by strong line broadening, a likely consequence of both high electronic relaxation time (T_1e) of Gd³⁺ and high rotational correlation time (τ_F) of the oligomeric gadolinium chelate (see Eqs. S1–S4 and Fig. S1 and S2 in SI). However, broad NMR signals of the ligand in the absence of metals and slow metalation kinetics suggested insufficient solubility of H-fod in PFPE.

Figure 1. Comparison of iron and gadolinium diketonates (H-fod) as ¹⁹F relaxation agents for PFPE.

The relaxometry results (9.4 T) are displayed for PFPE emulsions (120 g/L PFPE) containing H-fod (2.8 mM) 24 hours after the addition of 0.7 mM metal ions. R₁ and R₂ values are reported for the main PFPE peak at −91.4 ppm. The results show that Fe³⁺ is a more effective R₁ agent than Gd³⁺.

To improve solubility, we investigated fluorinated β-diketones (FDK) that have a greater fluorine content. We prepared the PFPE-based ligand pAn-FDK using Claisen condensation¹⁷ between PFPE-OMe and excess p-methoxyacetophenone, yielding highly pure pAn-FDK product at >10 g scale upon simple extractive workup (Fig. 2a). ¹H NMR analysis revealed new peaks at 6.46 and 15.35 ppm, characteristic of a diketone in enol form (Fig. S4). This ligand was used for subsequent studies.

Figure 2. Preparation and characterization of metal-binding nanoemulsions for ¹⁹F MRI.

a, Synthesis of metal-binding fluorinated diketones (FDK) from PFPE-OMe (denoted as R_FCO₂Me). b, Structures of fluorocarbons used for ¹⁹F MRI. c, Composition and preparation of various metal-binding (A, B, D, F, G) and control (C, E) fluorocarbon nanoemulsions. d, ¹⁹F NMR spectra (11.7 T) of emulsions A–C (4.5 g/L ¹⁹F, 90% D₂O). Signals from terminal CF₂ of diketone ligands are well separated from other peaks and are used to determine ligand concentration. The peak at −76 ppm is the reference (CF₃CO₂Na, TFA). e, Addition of aqueous metal chlorides to FDK emulsions yields metalated emulsions; Ar = pAn. f, Absorption spectra of metal-binding emulsion B (70 µM diketone, 0.09 g/L ¹⁹F) (color) and control emulsion C (0.09 g/L) (---) in the presence of Fe³⁺. Increasing [Fe³⁺] causes the appearance of ferric tris-diketonate charge transfer bands at 395 and 500 nm that grow linearly in intensity until the ca. 3:1 ligand:Fe ratio is reached at 25 µM Fe³⁺.

To evaluate ¹⁹F MRI properties, we blended pAn-FDK with a variety of perfluorocarbon derivatives and formulated these blended oils into aqueous nanoemulsions using microfluidization. Nanoemulsions (Fig. 2c) included pAn-FDK alone (emulsion A), or as a blend with PFPE diethylamide (DEA) (B), PFPE (D), perfluorooctyl bromide (PFOB) (F), or a short PFPE oligomer perfluorotetraglyme (PF4G) (G). Emulsions C (pure PFPE-DEA) and E (pure PFPE) are controls that cannot bind metals (Fig. 2c). We note that PFOB was tested because of its rapid clearance from the body and prior use clinically, but is not preferred for MRI cell detection due to its multiple ¹⁹F resonances that diminish image quality¹⁸. In all formulations A–G, stable nanoemulsions were formed, with similar physical characteristics. Dynamic light scattering (DLS) measurements in A–G displayed monodisperse nanoemulsions with a polydispersity index (PDI) of <0.2 and average droplet diameters ranging from 140–200 nm and negative ζ-potentials of −27 to −56 mV (Fig. S5). No change in DLS measurements were noted for up to 8 months of storage at 4 °C. Nanoemulsion composition was confirmed by ¹⁹F NMR (Fig. 2d). Terminal CF₂ atoms of PFPE derivatives have resonances between −70 and −85 ppm. Presence of only one major peak in this spectral range in single-component emulsions A and C confirmed high purities of the starting oils; emulsion B shows peaks from both components in the expected 1:1 ratio. Core CF₂CF₂O units resonating at −91 ppm comprise ~90% of the total ¹⁹F spectral weight and is typically the only signal detectable by ¹⁹F MRI, which generally has a much lower SNR compared to conventional ¹H images.

Properties of metalated nanoemulsions

The Pluronic surfactant used in the nanoemulsion formulation is permeable to ions enabling direct metalation of FDK nanoemulsions by the addition of metal chloride into the aqueous phase (Fig. 2e). Optical changes due to the formation of metal complexes were readily observed (Fig. 2f and S6). Among these, europium chelates are notable for their bright photoluminescence which may be useful for studying intracellular localization and trafficking of the PFC droplets (Fig. S11e). Importantly, addition of FeCl₃ caused rapid (k_obs = 0.69±0.10 min⁻¹) appearance (Fig. S7) of characteristic charge transfer bands of ferric diketonates¹⁹ (ε₃₉₀ =23, ε₅₀₀=4.9 mM⁻¹cm⁻¹) that linearly increased in intensity with increasing [Fe³⁺] until the Fe:FDK ratio of ca. 1:3 was reached, consistent with ferric tris-diketonate (Fig. 2f). Henceforth, the term ‘FETRIS’ (FErric TRIS-diketonate) refers to pAn-FDK blended with PFPE and metalated with Fe³⁺.

Relaxometric evaluation of nanoemulsions in the presence of different metals (Fig. 3a) revealed that binding of Fe³⁺ resulted in modest line broadening of all ¹⁹F NMR resonances, including the main PFPE peak at −91 ppm. The highest R₁ observed (158.2±2.5 s⁻¹ at 11.7 T), with a linewidth of 4 kHz, was with FETRIS saturated with Fe³⁺. Despite the largest number of unpaired electrons, Gd³⁺ showed a two-fold lower R₁ compared to Fe³⁺, with severe line broadening. Mn²⁺ gave moderately broad signals with the lowest R₁ of the triad. To confirm that the linewidth of metalated nanoemulsions is dominated by paramagnetism and not by metal binding per se, diamagnetic Ga³⁺, with a similar ionic radius to Fe³⁺, was included in the analysis and was found to have R₁ and R₂ equal to 2.08±0.01 s⁻¹ and 20.9 ± 0.3 s⁻¹, respectively. Also, the small change in relaxation rates relative to the unmetalated emulsion, with R₁ = 2.37±0.01 s⁻¹, and R₂ = 15.1±0.2 s⁻¹ (9.4 T), is attributable to an increase in the effective molecular weight upon formation of the metal complex.

Figure 3. Fluorine-19 relaxometry of metalated PFPE emulsions.

a, R₁ and ¹⁹F NMR spectra of FETRIS nanoemulsion (4.5 g/L ¹⁹F, 3.5 mM diketone) in the presence of 0.5 mM metal ions, 15 mM HEPES, and at pH 7.4. The peaks from different ¹⁹F spectra are scaled to the same absolute intensity. b, Relaxometric analysis of Fe³⁺ and Gd³⁺ binding capacity. Shown are measurements of R₁ for both PFPE (fluorous phase) and trifluoroacetate reference (TFA) added to the aqueous phase. c, Magnetic field dependence at T = 295 K and d, temperature (B₀ = 9.4 T) dependence of observed relaxation rates R₁ (●) and R₂ (x) in FETRIS nanoemulsion (22.5 g/L ¹⁹F, 17.5 mM diketone, 2.8 mM Fe³⁺) and predicted R₁ (—) values using Eqs. S1–S4. Predicted R₁ values represent best fit to SBM equations, with r = 1.19 nm, τ_F (295 K) = 0.80 ns, τ_v (295 K) = 3.59 ps, the Arrhenius temperature dependence with activation energies of 3.6 kcal/mol for τ_F and 4.5 kcal/mol for τ_v. The diamagnetic contributions to R₁ are presumed to be negligible and Δ fixed at 0.2 cm⁻¹. R₁ values increase, while R₂ values decrease, at lower magnetic field strengths, suggesting that there will be no degradation of SNR at clinical fields due to line broadening.

We determined the phase distribution of the paramagnetic ions and the metal binding capacity of FDK nanoemulsions (Fig. 3b). Measurement of R₁ at 11.7 T for both PFPE (fluorous phase) and trifluoroacetate reference (TFA) added to the aqueous phase revealed that nanoemulsions efficiently extracted Gd³⁺ and Fe³⁺ from water into the fluorous phase. R₁ of PFPE reached a plateau at ligand-to-metal ratio of ca. 2.5; increasing metal concentration further affected the R₁ of TFA. Notably, an increase in R₁ of TFA was observed even at the lowest Gd³⁺ concentration in pure PFPE nanoemulsion, confirming that the paramagnetic ion stays in the aqueous phase. We speculate that the modest (~2-fold) increase in R₁ of PFPE in this case was likely due to binding of Gd³⁺ ions to the nanoemulsion surface^{8, 10}. We observed a divergent field and temperature dependence of R₁ and R₂ in FETRIS nanoemulsions (Fig 3c,d, Fig. S8). Further control over relaxation parameters was achieved by tuning molecular weight and viscosity of the emulsion components (Fig. S9). Other rare earths had only a minor effect on R₁, consistent with fast electronic relaxation in these metal ions²⁰ (Fig. S10).

We also evaluated the stability of metal–FDK complexes. Using metal-loaded nanoemulsion, we monitored changes in photoluminescence (Eu³⁺) and absorbance (Fe³⁺) in the presence of excess competing ligands to study potential leakage of metal from the fluorous phase. Ethylenediaminetetraacetate (EDTA), a strong metal chelator²¹, rapidly (<5 min) abolished the photoluminescence of europium-loaded emulsion due to complete sequestration of Eu³⁺ to the aqueous phase to form a non-photoluminescent EDTA complex. In contrast, FETRIS nanoemulsion showed no decrease in characteristic absorbance of the Fe³⁺ chelate, even with prolonged exposure to EDTA. To estimate long-term stability of FETRIS nanoemulsions, relaxation rates were measured in the presence of EDTA (Fig. 4). PFPE-based nanoemulsion showed <20% decrease in R₁ over 2 weeks of incubation at 37 °C with EDTA.

Figure 4. Relaxometry stability of FETRIS nanoemulsions in the presence of competing aqueous ligand.

Nanoemulsion B and F, both metalated with 0.7 mM Fe³⁺, were treated at 37 °C with 75 mM EDTA dissolved in aqueous phase. Shown are R₁ values of PFPE () in nanoemulsion B, and values for blend nanoemulsion F, including PFPE components () and the CF₃ signal of PFOB (). A slight decrease over time is observed, as slow Fe³⁺ efflux occurs from the fluorous phase and irreversibly binds to EDTA. Error bars are standard deviations from three independent replicates.

Next, we examined FETRIS nanoemulsion properties in labeled cells. Ex vivo labeling of a rodent glioma cell line (GL261) with FETRIS showed good viability post-labeling (Fig. 5a), with loadings on the order of ~10^{12 19}F atoms/cell (Fig. 5b). Uptake of FETRIS was evident by the orange color of cell pellets, and optical absorbance in the lysate correlated with the ¹⁹F content determined by NMR (Fig. 5c). Fluorine-19 relaxometry of labeled cells (Fig. S11) showed that FETRIS nanoemulsion did not appear to lose Fe³⁺ to the intracellular milieu over time; moreover, in the same nanoemulsion formulated without added Fe³⁺, it did not appear to sequester endogenous Fe³⁺ from the cell’s labile iron pool (Fig. S11c). However, Gd³⁺ substituted for Fe³⁺ in the nanoemulsion displayed evidence of some metal leakage upon cell labeling; we observed ca. 25% reduction of ¹⁹F R₁ values after labeling (Fig. S11d).

Figure 5. Cell labeling with FETRIS nanoemulsion.

Cells (GL261) were labeled in culture using FETRIS nanoemulsion. a, Cell viability. b, Cell uptake of FETRIS as measured by ¹⁹F NMR. c, Correlation of uptake determined by ¹⁹F NMR with optical absorbance of cell lysate at 390 nm due to FETRIS. Error bars are standard deviations from three independent replicates.

Magnetic resonance imaging with FETRIS

Phantom ¹⁹F MRI studies demonstrated the feasibility of imaging FETRIS using conventional MRI methods. A phantom sample was prepared consisting of two NMR tubes containing FETRIS prepared with parameters R₁/R₂=32.5/170 s⁻¹ and the same emulsion without metal (R₁/R₂=2.2/3.7 s⁻¹); tubes were embedded in agarose. Images were acquired at 11.7 T using a spin-density weighted gradient echo (GRE) sequence, with scanning parameters set at the Ernst angle condition⁹ for optimal imaging of the FETRIS specimen, and a ~4 min image acquisition time. Figure 6a displays phantom MRI results, where the FETRIS sample appears hyperintense; the measured ¹⁹F image SNR for FETRIS and Fe-negative specimens were 8.6 and 1.7, respectively, yielding a SNR improvement of ~5 for the FETRIS sample, without Rician correction for low SNR regime³. If each capillary was imaged using its appropriate Ernst angle, the SNR improvement would be ~3.3 (see modeling results Fig. S12). To further minimize potential T₂ signal loss when imaging FETRIS agents, one could potentially use so called Ultrashort TE (UTE) or Zero TE (ZTE) pulse sequences⁷. Pulse sequences like GRE are commonplace on clinical scanners, whereas ZTE is not yet readily provided by MRI vendors.

Figure 6. MRI of FETRIS nanoemulsion.

a. Phantom comprised of two agarose-embedded NMR tubes containing FETRIS nanoemulsion (4.5 g/L ¹⁹F) with 0.5 mM Fe³⁺ (R₁/R₂ = 32.5/170 s⁻¹) and nanoemulsion without metal (R₁/R₂ 2.2/3.7 s⁻¹), denoted +Fe and −Fe, respectively. The top panel shows unthresholded ¹⁹F images, and below, the ¹⁹F image is thresholded, rendered in hot-iron pseudo-color (scale bar), and overlaid onto the grayscale ¹H image. The ¹⁹F/¹H MRI data were acquired using a GRE sequence. b, Displays mouse GL261 glioma cells (5×10⁶) labeled with FETRIS nanoemulsion ex vivo and injected subcutaneously into mouse flank. The ¹⁹F data is rendered in pseudo-color and placed on a grayscale slice from the ¹H data. After 24 hours, mice were imaged, and a cell ‘hot-spot’ is seen on the right flank in the axial view. Cells labeled with metal-free nanoemulsion and injected on the contralateral side could not be detected. Asterisk is adjacent chemical shift displacement artifact from hyperintense subcutaneous fat at 11.7T. The ¹⁹F and ¹H images were acquired using ZTE and GRE pulse sequences, respectively. For display, a co-registered 2D GRE slice was embedded into a 3D rendering of the ¹⁹F data.

Preliminary in vivo imaging of FETRIS-labeled cells was performed. Glioma cells were labeled with FETRIS nanoemulsion (50 wt.% pAn-FDK, 50 wt.% PFPE) ex vivo to a level of ~10^{12 19}F/cell. A second batch of glioma cells was labeled at comparable levels with PFPE emulsion without metal. Cells (5×10⁶ per side) were injected subcutaneously into left (no metal) and right (FETRIS) flanks in syngeneic C57BL/6 mice (N=3). After 24 hours, mice were imaged with ¹H/¹⁹F MRI at 11.7T (Fig. 6b). The ¹⁹F images were acquired using a three-dimensional ZTE sequence (Fig. 6b). Cells were readily visible (SNR~7) in the right injected flank (Fig. 6b), but not on the left side (no metal). Future in vivo studies will utilize FETRIS to image stem cells and immune cell populations in preclinical models.

Outlook

Here we present a unique approach for formulating nanoemulsions using PFPE-based β-diketones (FDK) as metal chelators. These ligands have previously been studied in the context of material science¹⁶, NMR spectroscopy²², and catalysis²³. We show that FDK is well-suited for incorporating large amounts of paramagnetic metal ions into the fluorous liquids. Formulated as stable PFPE-in-water nanoemulsions, FDK efficiently and irreversibly extract Fe³⁺ ions from aqueous solution into the fluorous phase, giving rise to cytocompatible FETRIS agent. These paramagnetic materials are useful for ¹⁹F MRI with enhanced sensitivity due to a dramatic reduction in T₁, a fundamental parameter limiting the speed of MRI data acquisitions. The ¹⁹F T₁ value reduction is magnetic field-strength dependent, but can potentially be accelerated to values approaching 80× at clinical field strengths, yielding a >8-fold sensitivity increase in ¹⁹F detection (Fig. S12). These sensitivity increases diminish at higher magnetic field strengths (Fig. S12). We show that FETRIS is effective for ¹⁹F MRI using conventional MRI pulse sequences.

Gd³⁺ and Fe³⁺ are at the heart of T₁- and T₂-based ¹H contrast agents, respectively, but for ¹⁹F MRI, the roles of these metal ions are reversed. Fe³⁺ was the optimal T₁ enhancer for perfluorocarbons, while analogous gadolinium (and manganese) chelates caused severe line broadening, essentially becoming ¹⁹F T₂ agents. Paramagnetic relaxation enhancement has been previously applied to ¹⁹F nuclei^{7, 20, 24, 25}. ¹⁹F MR probes based on macrocyclic lanthanide complexes with fluorinated substituents has been described^{7, 20}. However, these paramagnetic ¹⁹F tracers are not ideal for cell detection purposes. The relatively low ¹⁹F content of osmotically active macrocyclic chelates makes it difficult to reach MR-detectable cell loadings compared to highly-fluorinated PFC oils. In other approaches, Gd macrocyclic chelates bound to nanoemulsion surface can be used to provide a modest enhancement of ¹⁹F T₁, but these are unstable in the intracellular milieu, especially if they traffic to low pH compartments²⁸, which tends to separate the chelate from nanoemulsion droplet, thereby limiting long-term enhancement. In contrast, FETRIS complexes are characterized by very small rates of metal leakage even in the presence of EDTA in vitro and after cell labelling. The toxicity testing of FETRIS as reported here is viewed as preliminary; more rigorous in vitro cell studies, as well as animal testing, are needed to determine potential suitability for clinical trials. We note that emerging ¹H MRI techniques like PARACEST²⁶ and highly shifted proton MRI²⁷ have shown promise to detect multiple cell populations on standard MRI instrumentation with high specificity.

Overall, ¹⁹F MRI cell detection using PFC tracer agents is a rapidly emerging alternative to ¹H-based approaches using metal-ion-based contrast agents. The technical barriers associated with implementation of ¹⁹F MRI on a clinical scanner are surmountable, and clinical ¹⁹F cell detection has recently been demonstrated⁵. Future improvements in sensitivity of the probes will only accelerate adoption of this technology and open up new uses for this technology; towards this goal, the excellent stability and unique magnetic properties of FETRIS should advance this field.

Methods

Emulsion preparation

The fluorocarbon oil blends were prepared from PFPE, PFPE-DEA (Exfluor, Round Rock, TX), PFOB (Acros, Pittsburgh, PA), and pAn-FDK (see SI, Supplementary Methods for synthetic procedures) agents. Proportions (Fig. 2) were prepared gravimetrically in a 15 or 50 mL conical Falcon tube (Corning). Per 1 gram of PFC blend, 0.5 mL aqueous solution of Pluronic F68 (100 g/L) was added, and the mixture was vortexed at the highest speed. Water (8.5 mL) was added, followed by brief vortexing and ultrasonication (Omni Ruptor 250W, 30% power, 2 minutes, Omni International, Kennesaw, GA). The crude emulsion thus obtained was passed 4–6 times through LV1 microfluidizer (Microfluidics, Westwood, MA) operating at 20,000 psi and filtered through a 0.2 µm Supor membrane (Pall Corp. #4187, Port Washington, NY) into sterile glass vials.

NMR measurements

NMR spectra were obtained on Magritek Spinsolve (1.0 T), Bruker Avance 300 (7.0 T), Bruker Ascend 400 (9.4 T), Jeol ACA 500 (11.7 T), and Bruker DRX-600 (14.1 T) instruments. ¹⁹F NMR spectra of aqueous nanoemulsions were referenced to an internal standard (0.1 wt.% CF₃CO₂Na/D₂O, −76.00 ppm), which served as integration reference for quantitative NMR (see SI). Relaxation measurements were performed using a standard inversion recovery (with TI from 3⁻² to 3⁹ ms) pulse sequence and a Carr-Purcell-Meiboom-Gill sequence with TE values in 12 linear increments. R₁ and R₂ were obtained by non-linear fitting in MNova 6.0.2 software (Mestrelab, Escondido, CA). Fit errors were less than 5% for R₁ and 10% for R₂.

Cell labeling

Rat 9L or mouse GL261 glioma cells (3–5×10⁶, ATCC, Manassas, VA) were plated in 10 cm dishes and allowed to attach overnight. Immediately before cell labeling, FDK (B or D) or control (C or E) emulsion (0.5 mL) was mixed with freshly prepared FeCl₃ (50 mM in H₂O, 0.12 mL), protamine sulfate (1% in H₂O, 0.02 mL), and Tris base (1 M in H₂O, 0.25 mL). The dark-orange liquid was diluted to the desired PFPE content with DMEM (9L) or RPMI-1640 (GL261) media supplemented with 10% (v/v) fetal bovine serum (FBS). Labeling medium was added to cells at 5 mL/dish. After 16 h incubation at 37 °C, the cell labeling medium was removed, and cells were washed three times with phosphate buffered saline (PBS), detached by trypsinization, washed again in PBS, and resuspended in 1 mL of PBS. A portion of the cell suspension (~1/10) was used for cell number estimates by Cell Titer Glo (Promega, Madison, WI) or using a Countess II FL Cell Counter (Life Technologies, Carlsbad, CA). To assay nanoemulsion uptake, cells were pelleted and resuspended in 0.1 mL of lysis solution (0.5% Triton X, 100 mM NaCl, 20 mM Tris). A portion of this solution (6 µL) was used for absorbance measurements on NanoDrop 2000 spectrophotometer (Thermo Scientific, Pittsburgh, PA). The remainder was transferred to a 5 mm NMR tube, mixed with 0.15 mL of 0.1 wt.% CF₃CO₂Na/D₂O reference compound and ¹⁹F NMR spectra were obtained to measure ¹⁹F uptake, as previously described²⁹.

MRI

A phantom sample was prepared using 5 mm NMR tubes containing FETRIS (4.5 g/L ¹⁹F, 0.5 mM Fe³⁺, R₁/R₂ = 32.5/170 s⁻¹) and nanoemulsion without metal (R₁/R₂ = 2.2/3.7 s⁻¹); tubes were embedded in agarose. All images were acquired using a Bruker 11.7T BioSpec using a ¹⁹F/¹H double tuned volume coil. For ¹⁹F, a gradient echo (GRE) pulse sequence was used with parameters: TR/TE=15/0.83 ms (TR=recovery time), NA=256 (NA=number of averages), FOV=4×4 cm (FOV=field of view), 64×64 matrix, 8 mm thick slices, and a ~4 min data acquisition time. In this image, the echo time (TE) parameter was minimized to 0.83 ms, but at this value there is a residual amount of signal attenuation from T₂-effects in the FETRIS material (estimated ~12%). We used the Ernst angle condition⁹ for optimal ¹⁹F imaging of the FETRIS phantom. For ¹H, the GRE parameters were TR/TE=150/2 ms, NA=8, FOV=4×4 cm, 256×256 matrix, and 2 mm slices. The ¹⁹F image data was rendered in hot-iron pseudo-color using ImageJ software (NIH) and overlaid onto the grayscale ¹H image. For in vivo mouse studies, mouse GL261 glioma cells were labeled with FETRIS nanoemulsion (50 wt.% pAn-FDK, 50 wt.% PFPE) ex vivo to a level of ~10^{12 19}F/cell. A second batch of cells was similarly labeled but with unmetalated nanoemulsion. Cells (5×10⁶ per side) were injected subcutaneously into flanks in female syngeneic C57BL/6 mice (8–10 weeks old, N=3) using a vehicle of 0.2 ml Matrigel (BD Biosciences, Franklin Lakes, NJ) in PBS. The FETRIS labeled cells, and cells labeled with unmetalated nanoemulsion, were injected into the right and left sides, respectively. After 24 hours, mice were imaged using a three-dimensional ZTE sequence with parameters TR= 4 ms, receiver bandwidth 40 kHz, acquisition window 0.8 ms, number of projections 13030, NA=26, acquisition time 23 min, FOV=6×6×6 cm, and matrix size 64×64×64. Proton data were acquired using a two-dimensional spin-echo sequence with TR/TE=1500/14 ms, FOV=6×6 cm, and 256×256 matrix. ¹⁹F data were imported into Amira software (FEI, Hillsboro, OR) and rendered in color and a grayscale slice from the ¹H data was embedded for anatomical display purposes.